2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.h>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV 0
90 #define BOND_LINK_ARP_INTERV 0
92 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
93 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
94 static int num_peer_notif = 1;
95 static int miimon = BOND_LINK_MON_INTERV;
98 static int use_carrier = 1;
100 static char *primary;
101 static char *primary_reselect;
102 static char *lacp_rate;
103 static int min_links;
104 static char *ad_select;
105 static char *xmit_hash_policy;
106 static int arp_interval = BOND_LINK_ARP_INTERV;
107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
108 static char *arp_validate;
109 static char *arp_all_targets;
110 static char *fail_over_mac;
111 static int all_slaves_active;
112 static struct bond_params bonding_defaults;
113 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125 module_param(miimon, int, 0);
126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127 module_param(updelay, int, 0);
128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129 module_param(downdelay, int, 0);
130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 module_param(use_carrier, int, 0);
133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135 module_param(mode, charp, 0);
136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140 module_param(primary, charp, 0);
141 MODULE_PARM_DESC(primary, "Primary network device to use");
142 module_param(primary_reselect, charp, 0);
143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
148 "2 for only on active slave "
150 module_param(lacp_rate, charp, 0);
151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153 module_param(ad_select, charp, 0);
154 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
157 module_param(min_links, int, 0);
158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160 module_param(xmit_hash_policy, charp, 0);
161 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165 module_param(arp_interval, int, 0);
166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167 module_param_array(arp_ip_target, charp, NULL, 0);
168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169 module_param(arp_validate, charp, 0);
170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173 module_param(arp_all_targets, charp, 0);
174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac, charp, 0);
176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179 module_param(all_slaves_active, int, 0);
180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183 module_param(resend_igmp, int, 0);
184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
187 /*----------------------------- Global variables ----------------------------*/
189 #ifdef CONFIG_NET_POLL_CONTROLLER
190 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
193 int bond_net_id __read_mostly;
195 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
196 static int arp_ip_count;
197 static int bond_mode = BOND_MODE_ROUNDROBIN;
198 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
199 static int lacp_fast;
201 const struct bond_parm_tbl bond_lacp_tbl[] = {
202 { "slow", AD_LACP_SLOW},
203 { "fast", AD_LACP_FAST},
207 const struct bond_parm_tbl bond_mode_tbl[] = {
208 { "balance-rr", BOND_MODE_ROUNDROBIN},
209 { "active-backup", BOND_MODE_ACTIVEBACKUP},
210 { "balance-xor", BOND_MODE_XOR},
211 { "broadcast", BOND_MODE_BROADCAST},
212 { "802.3ad", BOND_MODE_8023AD},
213 { "balance-tlb", BOND_MODE_TLB},
214 { "balance-alb", BOND_MODE_ALB},
218 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
219 { "layer2", BOND_XMIT_POLICY_LAYER2},
220 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
221 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
222 { "encap2+3", BOND_XMIT_POLICY_ENCAP23},
223 { "encap3+4", BOND_XMIT_POLICY_ENCAP34},
227 const struct bond_parm_tbl arp_all_targets_tbl[] = {
228 { "any", BOND_ARP_TARGETS_ANY},
229 { "all", BOND_ARP_TARGETS_ALL},
233 const struct bond_parm_tbl arp_validate_tbl[] = {
234 { "none", BOND_ARP_VALIDATE_NONE},
235 { "active", BOND_ARP_VALIDATE_ACTIVE},
236 { "backup", BOND_ARP_VALIDATE_BACKUP},
237 { "all", BOND_ARP_VALIDATE_ALL},
241 const struct bond_parm_tbl fail_over_mac_tbl[] = {
242 { "none", BOND_FOM_NONE},
243 { "active", BOND_FOM_ACTIVE},
244 { "follow", BOND_FOM_FOLLOW},
248 const struct bond_parm_tbl pri_reselect_tbl[] = {
249 { "always", BOND_PRI_RESELECT_ALWAYS},
250 { "better", BOND_PRI_RESELECT_BETTER},
251 { "failure", BOND_PRI_RESELECT_FAILURE},
255 struct bond_parm_tbl ad_select_tbl[] = {
256 { "stable", BOND_AD_STABLE},
257 { "bandwidth", BOND_AD_BANDWIDTH},
258 { "count", BOND_AD_COUNT},
262 /*-------------------------- Forward declarations ---------------------------*/
264 static int bond_init(struct net_device *bond_dev);
265 static void bond_uninit(struct net_device *bond_dev);
267 /*---------------------------- General routines -----------------------------*/
269 const char *bond_mode_name(int mode)
271 static const char *names[] = {
272 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
273 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
274 [BOND_MODE_XOR] = "load balancing (xor)",
275 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
276 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
277 [BOND_MODE_TLB] = "transmit load balancing",
278 [BOND_MODE_ALB] = "adaptive load balancing",
281 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
287 /*---------------------------------- VLAN -----------------------------------*/
290 * bond_dev_queue_xmit - Prepare skb for xmit.
292 * @bond: bond device that got this skb for tx.
293 * @skb: hw accel VLAN tagged skb to transmit
294 * @slave_dev: slave that is supposed to xmit this skbuff
296 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
297 struct net_device *slave_dev)
299 skb->dev = slave_dev;
301 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
302 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
303 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
305 if (unlikely(netpoll_tx_running(bond->dev)))
306 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
314 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
315 * We don't protect the slave list iteration with a lock because:
316 * a. This operation is performed in IOCTL context,
317 * b. The operation is protected by the RTNL semaphore in the 8021q code,
318 * c. Holding a lock with BH disabled while directly calling a base driver
319 * entry point is generally a BAD idea.
321 * The design of synchronization/protection for this operation in the 8021q
322 * module is good for one or more VLAN devices over a single physical device
323 * and cannot be extended for a teaming solution like bonding, so there is a
324 * potential race condition here where a net device from the vlan group might
325 * be referenced (either by a base driver or the 8021q code) while it is being
326 * removed from the system. However, it turns out we're not making matters
327 * worse, and if it works for regular VLAN usage it will work here too.
331 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
332 * @bond_dev: bonding net device that got called
333 * @vid: vlan id being added
335 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
336 __be16 proto, u16 vid)
338 struct bonding *bond = netdev_priv(bond_dev);
339 struct slave *slave, *rollback_slave;
340 struct list_head *iter;
343 bond_for_each_slave(bond, slave, iter) {
344 res = vlan_vid_add(slave->dev, proto, vid);
352 /* unwind to the slave that failed */
353 bond_for_each_slave(bond, rollback_slave, iter) {
354 if (rollback_slave == slave)
357 vlan_vid_del(rollback_slave->dev, proto, vid);
364 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
365 * @bond_dev: bonding net device that got called
366 * @vid: vlan id being removed
368 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
369 __be16 proto, u16 vid)
371 struct bonding *bond = netdev_priv(bond_dev);
372 struct list_head *iter;
375 bond_for_each_slave(bond, slave, iter)
376 vlan_vid_del(slave->dev, proto, vid);
378 if (bond_is_lb(bond))
379 bond_alb_clear_vlan(bond, vid);
384 /*------------------------------- Link status -------------------------------*/
387 * Set the carrier state for the master according to the state of its
388 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
389 * do special 802.3ad magic.
391 * Returns zero if carrier state does not change, nonzero if it does.
393 static int bond_set_carrier(struct bonding *bond)
395 struct list_head *iter;
398 if (!bond_has_slaves(bond))
401 if (bond->params.mode == BOND_MODE_8023AD)
402 return bond_3ad_set_carrier(bond);
404 bond_for_each_slave(bond, slave, iter) {
405 if (slave->link == BOND_LINK_UP) {
406 if (!netif_carrier_ok(bond->dev)) {
407 netif_carrier_on(bond->dev);
415 if (netif_carrier_ok(bond->dev)) {
416 netif_carrier_off(bond->dev);
423 * Get link speed and duplex from the slave's base driver
424 * using ethtool. If for some reason the call fails or the
425 * values are invalid, set speed and duplex to -1,
428 static void bond_update_speed_duplex(struct slave *slave)
430 struct net_device *slave_dev = slave->dev;
431 struct ethtool_cmd ecmd;
435 slave->speed = SPEED_UNKNOWN;
436 slave->duplex = DUPLEX_UNKNOWN;
438 res = __ethtool_get_settings(slave_dev, &ecmd);
442 slave_speed = ethtool_cmd_speed(&ecmd);
443 if (slave_speed == 0 || slave_speed == ((__u32) -1))
446 switch (ecmd.duplex) {
454 slave->speed = slave_speed;
455 slave->duplex = ecmd.duplex;
461 * if <dev> supports MII link status reporting, check its link status.
463 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
464 * depending upon the setting of the use_carrier parameter.
466 * Return either BMSR_LSTATUS, meaning that the link is up (or we
467 * can't tell and just pretend it is), or 0, meaning that the link is
470 * If reporting is non-zero, instead of faking link up, return -1 if
471 * both ETHTOOL and MII ioctls fail (meaning the device does not
472 * support them). If use_carrier is set, return whatever it says.
473 * It'd be nice if there was a good way to tell if a driver supports
474 * netif_carrier, but there really isn't.
476 static int bond_check_dev_link(struct bonding *bond,
477 struct net_device *slave_dev, int reporting)
479 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
480 int (*ioctl)(struct net_device *, struct ifreq *, int);
482 struct mii_ioctl_data *mii;
484 if (!reporting && !netif_running(slave_dev))
487 if (bond->params.use_carrier)
488 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
490 /* Try to get link status using Ethtool first. */
491 if (slave_dev->ethtool_ops->get_link)
492 return slave_dev->ethtool_ops->get_link(slave_dev) ?
495 /* Ethtool can't be used, fallback to MII ioctls. */
496 ioctl = slave_ops->ndo_do_ioctl;
498 /* TODO: set pointer to correct ioctl on a per team member */
499 /* bases to make this more efficient. that is, once */
500 /* we determine the correct ioctl, we will always */
501 /* call it and not the others for that team */
505 * We cannot assume that SIOCGMIIPHY will also read a
506 * register; not all network drivers (e.g., e100)
510 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
511 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
513 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
514 mii->reg_num = MII_BMSR;
515 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
516 return mii->val_out & BMSR_LSTATUS;
521 * If reporting, report that either there's no dev->do_ioctl,
522 * or both SIOCGMIIREG and get_link failed (meaning that we
523 * cannot report link status). If not reporting, pretend
526 return reporting ? -1 : BMSR_LSTATUS;
529 /*----------------------------- Multicast list ------------------------------*/
532 * Push the promiscuity flag down to appropriate slaves
534 static int bond_set_promiscuity(struct bonding *bond, int inc)
536 struct list_head *iter;
539 if (USES_PRIMARY(bond->params.mode)) {
540 /* write lock already acquired */
541 if (bond->curr_active_slave) {
542 err = dev_set_promiscuity(bond->curr_active_slave->dev,
548 bond_for_each_slave(bond, slave, iter) {
549 err = dev_set_promiscuity(slave->dev, inc);
558 * Push the allmulti flag down to all slaves
560 static int bond_set_allmulti(struct bonding *bond, int inc)
562 struct list_head *iter;
565 if (USES_PRIMARY(bond->params.mode)) {
566 /* write lock already acquired */
567 if (bond->curr_active_slave) {
568 err = dev_set_allmulti(bond->curr_active_slave->dev,
574 bond_for_each_slave(bond, slave, iter) {
575 err = dev_set_allmulti(slave->dev, inc);
584 * Retrieve the list of registered multicast addresses for the bonding
585 * device and retransmit an IGMP JOIN request to the current active
588 static void bond_resend_igmp_join_requests(struct bonding *bond)
590 if (!rtnl_trylock()) {
591 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
594 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
597 /* We use curr_slave_lock to protect against concurrent access to
598 * igmp_retrans from multiple running instances of this function and
599 * bond_change_active_slave
601 write_lock_bh(&bond->curr_slave_lock);
602 if (bond->igmp_retrans > 1) {
603 bond->igmp_retrans--;
604 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
606 write_unlock_bh(&bond->curr_slave_lock);
609 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
611 struct bonding *bond = container_of(work, struct bonding,
614 bond_resend_igmp_join_requests(bond);
617 /* Flush bond's hardware addresses from slave
619 static void bond_hw_addr_flush(struct net_device *bond_dev,
620 struct net_device *slave_dev)
622 struct bonding *bond = netdev_priv(bond_dev);
624 dev_uc_unsync(slave_dev, bond_dev);
625 dev_mc_unsync(slave_dev, bond_dev);
627 if (bond->params.mode == BOND_MODE_8023AD) {
628 /* del lacpdu mc addr from mc list */
629 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
631 dev_mc_del(slave_dev, lacpdu_multicast);
635 /*--------------------------- Active slave change ---------------------------*/
637 /* Update the hardware address list and promisc/allmulti for the new and
638 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
639 * slaves up date at all times; only the USES_PRIMARY modes need to call
640 * this function to swap these settings during a failover.
642 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
643 struct slave *old_active)
648 if (bond->dev->flags & IFF_PROMISC)
649 dev_set_promiscuity(old_active->dev, -1);
651 if (bond->dev->flags & IFF_ALLMULTI)
652 dev_set_allmulti(old_active->dev, -1);
654 bond_hw_addr_flush(bond->dev, old_active->dev);
658 /* FIXME: Signal errors upstream. */
659 if (bond->dev->flags & IFF_PROMISC)
660 dev_set_promiscuity(new_active->dev, 1);
662 if (bond->dev->flags & IFF_ALLMULTI)
663 dev_set_allmulti(new_active->dev, 1);
665 netif_addr_lock_bh(bond->dev);
666 dev_uc_sync(new_active->dev, bond->dev);
667 dev_mc_sync(new_active->dev, bond->dev);
668 netif_addr_unlock_bh(bond->dev);
673 * bond_set_dev_addr - clone slave's address to bond
674 * @bond_dev: bond net device
675 * @slave_dev: slave net device
677 * Should be called with RTNL held.
679 static void bond_set_dev_addr(struct net_device *bond_dev,
680 struct net_device *slave_dev)
682 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
683 bond_dev, slave_dev, slave_dev->addr_len);
684 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
685 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
686 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
690 * bond_do_fail_over_mac
692 * Perform special MAC address swapping for fail_over_mac settings
694 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
696 static void bond_do_fail_over_mac(struct bonding *bond,
697 struct slave *new_active,
698 struct slave *old_active)
699 __releases(&bond->curr_slave_lock)
700 __releases(&bond->lock)
701 __acquires(&bond->lock)
702 __acquires(&bond->curr_slave_lock)
704 u8 tmp_mac[ETH_ALEN];
705 struct sockaddr saddr;
708 switch (bond->params.fail_over_mac) {
709 case BOND_FOM_ACTIVE:
711 write_unlock_bh(&bond->curr_slave_lock);
712 read_unlock(&bond->lock);
713 bond_set_dev_addr(bond->dev, new_active->dev);
714 read_lock(&bond->lock);
715 write_lock_bh(&bond->curr_slave_lock);
718 case BOND_FOM_FOLLOW:
720 * if new_active && old_active, swap them
721 * if just old_active, do nothing (going to no active slave)
722 * if just new_active, set new_active to bond's MAC
727 write_unlock_bh(&bond->curr_slave_lock);
728 read_unlock(&bond->lock);
731 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
732 memcpy(saddr.sa_data, old_active->dev->dev_addr,
734 saddr.sa_family = new_active->dev->type;
736 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
737 saddr.sa_family = bond->dev->type;
740 rv = dev_set_mac_address(new_active->dev, &saddr);
742 pr_err("%s: Error %d setting MAC of slave %s\n",
743 bond->dev->name, -rv, new_active->dev->name);
750 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
751 saddr.sa_family = old_active->dev->type;
753 rv = dev_set_mac_address(old_active->dev, &saddr);
755 pr_err("%s: Error %d setting MAC of slave %s\n",
756 bond->dev->name, -rv, new_active->dev->name);
758 read_lock(&bond->lock);
759 write_lock_bh(&bond->curr_slave_lock);
762 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
763 bond->dev->name, bond->params.fail_over_mac);
769 static bool bond_should_change_active(struct bonding *bond)
771 struct slave *prim = bond->primary_slave;
772 struct slave *curr = bond->curr_active_slave;
774 if (!prim || !curr || curr->link != BOND_LINK_UP)
776 if (bond->force_primary) {
777 bond->force_primary = false;
780 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
781 (prim->speed < curr->speed ||
782 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
784 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
790 * find_best_interface - select the best available slave to be the active one
791 * @bond: our bonding struct
793 static struct slave *bond_find_best_slave(struct bonding *bond)
795 struct slave *slave, *bestslave = NULL;
796 struct list_head *iter;
797 int mintime = bond->params.updelay;
799 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
800 bond_should_change_active(bond))
801 return bond->primary_slave;
803 bond_for_each_slave(bond, slave, iter) {
804 if (slave->link == BOND_LINK_UP)
806 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
807 slave->delay < mintime) {
808 mintime = slave->delay;
816 static bool bond_should_notify_peers(struct bonding *bond)
818 struct slave *slave = bond->curr_active_slave;
820 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
821 bond->dev->name, slave ? slave->dev->name : "NULL");
823 if (!slave || !bond->send_peer_notif ||
824 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
831 * change_active_interface - change the active slave into the specified one
832 * @bond: our bonding struct
833 * @new: the new slave to make the active one
835 * Set the new slave to the bond's settings and unset them on the old
837 * Setting include flags, mc-list, promiscuity, allmulti, etc.
839 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
840 * because it is apparently the best available slave we have, even though its
841 * updelay hasn't timed out yet.
843 * If new_active is not NULL, caller must hold bond->lock for read and
844 * curr_slave_lock for write_bh.
846 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
848 struct slave *old_active = bond->curr_active_slave;
850 if (old_active == new_active)
854 new_active->jiffies = jiffies;
856 if (new_active->link == BOND_LINK_BACK) {
857 if (USES_PRIMARY(bond->params.mode)) {
858 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
859 bond->dev->name, new_active->dev->name,
860 (bond->params.updelay - new_active->delay) * bond->params.miimon);
863 new_active->delay = 0;
864 new_active->link = BOND_LINK_UP;
866 if (bond->params.mode == BOND_MODE_8023AD)
867 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
869 if (bond_is_lb(bond))
870 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
872 if (USES_PRIMARY(bond->params.mode)) {
873 pr_info("%s: making interface %s the new active one.\n",
874 bond->dev->name, new_active->dev->name);
879 if (USES_PRIMARY(bond->params.mode))
880 bond_hw_addr_swap(bond, new_active, old_active);
882 if (bond_is_lb(bond)) {
883 bond_alb_handle_active_change(bond, new_active);
885 bond_set_slave_inactive_flags(old_active);
887 bond_set_slave_active_flags(new_active);
889 rcu_assign_pointer(bond->curr_active_slave, new_active);
892 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
894 bond_set_slave_inactive_flags(old_active);
897 bool should_notify_peers = false;
899 bond_set_slave_active_flags(new_active);
901 if (bond->params.fail_over_mac)
902 bond_do_fail_over_mac(bond, new_active,
905 if (netif_running(bond->dev)) {
906 bond->send_peer_notif =
907 bond->params.num_peer_notif;
908 should_notify_peers =
909 bond_should_notify_peers(bond);
912 write_unlock_bh(&bond->curr_slave_lock);
913 read_unlock(&bond->lock);
915 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
916 if (should_notify_peers)
917 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
920 read_lock(&bond->lock);
921 write_lock_bh(&bond->curr_slave_lock);
925 /* resend IGMP joins since active slave has changed or
926 * all were sent on curr_active_slave.
927 * resend only if bond is brought up with the affected
928 * bonding modes and the retransmission is enabled */
929 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
930 ((USES_PRIMARY(bond->params.mode) && new_active) ||
931 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
932 bond->igmp_retrans = bond->params.resend_igmp;
933 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
938 * bond_select_active_slave - select a new active slave, if needed
939 * @bond: our bonding struct
941 * This functions should be called when one of the following occurs:
942 * - The old curr_active_slave has been released or lost its link.
943 * - The primary_slave has got its link back.
944 * - A slave has got its link back and there's no old curr_active_slave.
946 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
948 void bond_select_active_slave(struct bonding *bond)
950 struct slave *best_slave;
953 best_slave = bond_find_best_slave(bond);
954 if (best_slave != bond->curr_active_slave) {
955 bond_change_active_slave(bond, best_slave);
956 rv = bond_set_carrier(bond);
960 if (netif_carrier_ok(bond->dev)) {
961 pr_info("%s: first active interface up!\n",
964 pr_info("%s: now running without any active interface !\n",
970 /*--------------------------- slave list handling ---------------------------*/
973 * This function attaches the slave to the end of list.
975 * bond->lock held for writing by caller.
977 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
983 * This function detaches the slave from the list.
984 * WARNING: no check is made to verify if the slave effectively
986 * Nothing is freed on return, structures are just unchained.
987 * If any slave pointer in bond was pointing to <slave>,
988 * it should be changed by the calling function.
990 * bond->lock held for writing by caller.
992 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
997 #ifdef CONFIG_NET_POLL_CONTROLLER
998 static inline int slave_enable_netpoll(struct slave *slave)
1003 np = kzalloc(sizeof(*np), GFP_ATOMIC);
1008 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1017 static inline void slave_disable_netpoll(struct slave *slave)
1019 struct netpoll *np = slave->np;
1025 __netpoll_free_async(np);
1027 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1029 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1031 if (!slave_dev->netdev_ops->ndo_poll_controller)
1036 static void bond_poll_controller(struct net_device *bond_dev)
1040 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1042 struct bonding *bond = netdev_priv(bond_dev);
1043 struct list_head *iter;
1044 struct slave *slave;
1046 bond_for_each_slave(bond, slave, iter)
1047 if (IS_UP(slave->dev))
1048 slave_disable_netpoll(slave);
1051 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1053 struct bonding *bond = netdev_priv(dev);
1054 struct list_head *iter;
1055 struct slave *slave;
1058 bond_for_each_slave(bond, slave, iter) {
1059 err = slave_enable_netpoll(slave);
1061 bond_netpoll_cleanup(dev);
1068 static inline int slave_enable_netpoll(struct slave *slave)
1072 static inline void slave_disable_netpoll(struct slave *slave)
1075 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1080 /*---------------------------------- IOCTL ----------------------------------*/
1082 static netdev_features_t bond_fix_features(struct net_device *dev,
1083 netdev_features_t features)
1085 struct bonding *bond = netdev_priv(dev);
1086 struct list_head *iter;
1087 netdev_features_t mask;
1088 struct slave *slave;
1090 if (!bond_has_slaves(bond)) {
1091 /* Disable adding VLANs to empty bond. But why? --mq */
1092 features |= NETIF_F_VLAN_CHALLENGED;
1097 features &= ~NETIF_F_ONE_FOR_ALL;
1098 features |= NETIF_F_ALL_FOR_ALL;
1100 bond_for_each_slave(bond, slave, iter) {
1101 features = netdev_increment_features(features,
1102 slave->dev->features,
1105 features = netdev_add_tso_features(features, mask);
1110 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1111 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1112 NETIF_F_HIGHDMA | NETIF_F_LRO)
1114 static void bond_compute_features(struct bonding *bond)
1116 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1117 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1118 struct net_device *bond_dev = bond->dev;
1119 struct list_head *iter;
1120 struct slave *slave;
1121 unsigned short max_hard_header_len = ETH_HLEN;
1122 unsigned int gso_max_size = GSO_MAX_SIZE;
1123 u16 gso_max_segs = GSO_MAX_SEGS;
1125 if (!bond_has_slaves(bond))
1128 bond_for_each_slave(bond, slave, iter) {
1129 vlan_features = netdev_increment_features(vlan_features,
1130 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1132 dst_release_flag &= slave->dev->priv_flags;
1133 if (slave->dev->hard_header_len > max_hard_header_len)
1134 max_hard_header_len = slave->dev->hard_header_len;
1136 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1137 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1141 bond_dev->vlan_features = vlan_features;
1142 bond_dev->hard_header_len = max_hard_header_len;
1143 bond_dev->gso_max_segs = gso_max_segs;
1144 netif_set_gso_max_size(bond_dev, gso_max_size);
1146 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1147 bond_dev->priv_flags = flags | dst_release_flag;
1149 netdev_change_features(bond_dev);
1152 static void bond_setup_by_slave(struct net_device *bond_dev,
1153 struct net_device *slave_dev)
1155 bond_dev->header_ops = slave_dev->header_ops;
1157 bond_dev->type = slave_dev->type;
1158 bond_dev->hard_header_len = slave_dev->hard_header_len;
1159 bond_dev->addr_len = slave_dev->addr_len;
1161 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1162 slave_dev->addr_len);
1165 /* On bonding slaves other than the currently active slave, suppress
1166 * duplicates except for alb non-mcast/bcast.
1168 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1169 struct slave *slave,
1170 struct bonding *bond)
1172 if (bond_is_slave_inactive(slave)) {
1173 if (bond->params.mode == BOND_MODE_ALB &&
1174 skb->pkt_type != PACKET_BROADCAST &&
1175 skb->pkt_type != PACKET_MULTICAST)
1182 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1184 struct sk_buff *skb = *pskb;
1185 struct slave *slave;
1186 struct bonding *bond;
1187 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1189 int ret = RX_HANDLER_ANOTHER;
1191 skb = skb_share_check(skb, GFP_ATOMIC);
1193 return RX_HANDLER_CONSUMED;
1197 slave = bond_slave_get_rcu(skb->dev);
1200 if (bond->params.arp_interval)
1201 slave->dev->last_rx = jiffies;
1203 recv_probe = ACCESS_ONCE(bond->recv_probe);
1205 ret = recv_probe(skb, bond, slave);
1206 if (ret == RX_HANDLER_CONSUMED) {
1212 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1213 return RX_HANDLER_EXACT;
1216 skb->dev = bond->dev;
1218 if (bond->params.mode == BOND_MODE_ALB &&
1219 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1220 skb->pkt_type == PACKET_HOST) {
1222 if (unlikely(skb_cow_head(skb,
1223 skb->data - skb_mac_header(skb)))) {
1225 return RX_HANDLER_CONSUMED;
1227 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1233 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1234 struct net_device *slave_dev,
1235 struct slave *slave)
1239 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1242 slave_dev->flags |= IFF_SLAVE;
1243 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1247 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1248 struct net_device *slave_dev)
1250 netdev_upper_dev_unlink(slave_dev, bond_dev);
1251 slave_dev->flags &= ~IFF_SLAVE;
1252 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1255 /* enslave device <slave> to bond device <master> */
1256 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1258 struct bonding *bond = netdev_priv(bond_dev);
1259 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1260 struct slave *new_slave = NULL, *prev_slave;
1261 struct sockaddr addr;
1265 if (!bond->params.use_carrier &&
1266 slave_dev->ethtool_ops->get_link == NULL &&
1267 slave_ops->ndo_do_ioctl == NULL) {
1268 pr_warning("%s: Warning: no link monitoring support for %s\n",
1269 bond_dev->name, slave_dev->name);
1272 /* already enslaved */
1273 if (slave_dev->flags & IFF_SLAVE) {
1274 pr_debug("Error, Device was already enslaved\n");
1278 /* vlan challenged mutual exclusion */
1279 /* no need to lock since we're protected by rtnl_lock */
1280 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1281 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1282 if (vlan_uses_dev(bond_dev)) {
1283 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1284 bond_dev->name, slave_dev->name, bond_dev->name);
1287 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1288 bond_dev->name, slave_dev->name,
1289 slave_dev->name, bond_dev->name);
1292 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1296 * Old ifenslave binaries are no longer supported. These can
1297 * be identified with moderate accuracy by the state of the slave:
1298 * the current ifenslave will set the interface down prior to
1299 * enslaving it; the old ifenslave will not.
1301 if ((slave_dev->flags & IFF_UP)) {
1302 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1305 goto err_undo_flags;
1308 /* set bonding device ether type by slave - bonding netdevices are
1309 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1310 * there is a need to override some of the type dependent attribs/funcs.
1312 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1313 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1315 if (!bond_has_slaves(bond)) {
1316 if (bond_dev->type != slave_dev->type) {
1317 pr_debug("%s: change device type from %d to %d\n",
1319 bond_dev->type, slave_dev->type);
1321 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1323 res = notifier_to_errno(res);
1325 pr_err("%s: refused to change device type\n",
1328 goto err_undo_flags;
1331 /* Flush unicast and multicast addresses */
1332 dev_uc_flush(bond_dev);
1333 dev_mc_flush(bond_dev);
1335 if (slave_dev->type != ARPHRD_ETHER)
1336 bond_setup_by_slave(bond_dev, slave_dev);
1338 ether_setup(bond_dev);
1339 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1342 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1345 } else if (bond_dev->type != slave_dev->type) {
1346 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1348 slave_dev->type, bond_dev->type);
1350 goto err_undo_flags;
1353 if (slave_ops->ndo_set_mac_address == NULL) {
1354 if (!bond_has_slaves(bond)) {
1355 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1357 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1358 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1359 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",
1362 goto err_undo_flags;
1366 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1368 /* If this is the first slave, then we need to set the master's hardware
1369 * address to be the same as the slave's. */
1370 if (!bond_has_slaves(bond) &&
1371 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1372 bond_set_dev_addr(bond->dev, slave_dev);
1374 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1377 goto err_undo_flags;
1380 * Set the new_slave's queue_id to be zero. Queue ID mapping
1381 * is set via sysfs or module option if desired.
1383 new_slave->queue_id = 0;
1385 /* Save slave's original mtu and then set it to match the bond */
1386 new_slave->original_mtu = slave_dev->mtu;
1387 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1389 pr_debug("Error %d calling dev_set_mtu\n", res);
1394 * Save slave's original ("permanent") mac address for modes
1395 * that need it, and for restoring it upon release, and then
1396 * set it to the master's address
1398 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1400 if (!bond->params.fail_over_mac) {
1402 * Set slave to master's mac address. The application already
1403 * set the master's mac address to that of the first slave
1405 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1406 addr.sa_family = slave_dev->type;
1407 res = dev_set_mac_address(slave_dev, &addr);
1409 pr_debug("Error %d calling set_mac_address\n", res);
1410 goto err_restore_mtu;
1414 /* open the slave since the application closed it */
1415 res = dev_open(slave_dev);
1417 pr_debug("Opening slave %s failed\n", slave_dev->name);
1418 goto err_restore_mac;
1421 new_slave->bond = bond;
1422 new_slave->dev = slave_dev;
1423 slave_dev->priv_flags |= IFF_BONDING;
1425 if (bond_is_lb(bond)) {
1426 /* bond_alb_init_slave() must be called before all other stages since
1427 * it might fail and we do not want to have to undo everything
1429 res = bond_alb_init_slave(bond, new_slave);
1434 /* If the mode USES_PRIMARY, then the following is handled by
1435 * bond_change_active_slave().
1437 if (!USES_PRIMARY(bond->params.mode)) {
1438 /* set promiscuity level to new slave */
1439 if (bond_dev->flags & IFF_PROMISC) {
1440 res = dev_set_promiscuity(slave_dev, 1);
1445 /* set allmulti level to new slave */
1446 if (bond_dev->flags & IFF_ALLMULTI) {
1447 res = dev_set_allmulti(slave_dev, 1);
1452 netif_addr_lock_bh(bond_dev);
1454 dev_mc_sync_multiple(slave_dev, bond_dev);
1455 dev_uc_sync_multiple(slave_dev, bond_dev);
1457 netif_addr_unlock_bh(bond_dev);
1460 if (bond->params.mode == BOND_MODE_8023AD) {
1461 /* add lacpdu mc addr to mc list */
1462 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1464 dev_mc_add(slave_dev, lacpdu_multicast);
1467 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1469 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1470 bond_dev->name, slave_dev->name);
1474 write_lock_bh(&bond->lock);
1476 prev_slave = bond_last_slave(bond);
1477 bond_attach_slave(bond, new_slave);
1479 new_slave->delay = 0;
1480 new_slave->link_failure_count = 0;
1482 write_unlock_bh(&bond->lock);
1484 bond_compute_features(bond);
1486 bond_update_speed_duplex(new_slave);
1488 read_lock(&bond->lock);
1490 new_slave->last_arp_rx = jiffies -
1491 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1492 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1493 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1495 if (bond->params.miimon && !bond->params.use_carrier) {
1496 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1498 if ((link_reporting == -1) && !bond->params.arp_interval) {
1500 * miimon is set but a bonded network driver
1501 * does not support ETHTOOL/MII and
1502 * arp_interval is not set. Note: if
1503 * use_carrier is enabled, we will never go
1504 * here (because netif_carrier is always
1505 * supported); thus, we don't need to change
1506 * the messages for netif_carrier.
1508 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",
1509 bond_dev->name, slave_dev->name);
1510 } else if (link_reporting == -1) {
1511 /* unable get link status using mii/ethtool */
1512 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",
1513 bond_dev->name, slave_dev->name);
1517 /* check for initial state */
1518 if (bond->params.miimon) {
1519 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1520 if (bond->params.updelay) {
1521 new_slave->link = BOND_LINK_BACK;
1522 new_slave->delay = bond->params.updelay;
1524 new_slave->link = BOND_LINK_UP;
1527 new_slave->link = BOND_LINK_DOWN;
1529 } else if (bond->params.arp_interval) {
1530 new_slave->link = (netif_carrier_ok(slave_dev) ?
1531 BOND_LINK_UP : BOND_LINK_DOWN);
1533 new_slave->link = BOND_LINK_UP;
1536 if (new_slave->link != BOND_LINK_DOWN)
1537 new_slave->jiffies = jiffies;
1538 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1539 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1540 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1542 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1543 /* if there is a primary slave, remember it */
1544 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1545 bond->primary_slave = new_slave;
1546 bond->force_primary = true;
1550 write_lock_bh(&bond->curr_slave_lock);
1552 switch (bond->params.mode) {
1553 case BOND_MODE_ACTIVEBACKUP:
1554 bond_set_slave_inactive_flags(new_slave);
1555 bond_select_active_slave(bond);
1557 case BOND_MODE_8023AD:
1558 /* in 802.3ad mode, the internal mechanism
1559 * will activate the slaves in the selected
1562 bond_set_slave_inactive_flags(new_slave);
1563 /* if this is the first slave */
1565 SLAVE_AD_INFO(new_slave).id = 1;
1566 /* Initialize AD with the number of times that the AD timer is called in 1 second
1567 * can be called only after the mac address of the bond is set
1569 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1571 SLAVE_AD_INFO(new_slave).id =
1572 SLAVE_AD_INFO(prev_slave).id + 1;
1575 bond_3ad_bind_slave(new_slave);
1579 bond_set_active_slave(new_slave);
1580 bond_set_slave_inactive_flags(new_slave);
1581 bond_select_active_slave(bond);
1584 pr_debug("This slave is always active in trunk mode\n");
1586 /* always active in trunk mode */
1587 bond_set_active_slave(new_slave);
1589 /* In trunking mode there is little meaning to curr_active_slave
1590 * anyway (it holds no special properties of the bond device),
1591 * so we can change it without calling change_active_interface()
1593 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1594 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1597 } /* switch(bond_mode) */
1599 write_unlock_bh(&bond->curr_slave_lock);
1601 bond_set_carrier(bond);
1603 #ifdef CONFIG_NET_POLL_CONTROLLER
1604 slave_dev->npinfo = bond->dev->npinfo;
1605 if (slave_dev->npinfo) {
1606 if (slave_enable_netpoll(new_slave)) {
1607 read_unlock(&bond->lock);
1608 pr_info("Error, %s: master_dev is using netpoll, "
1609 "but new slave device does not support netpoll.\n",
1617 read_unlock(&bond->lock);
1619 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1622 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1626 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1628 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1629 goto err_unregister;
1633 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1634 bond_dev->name, slave_dev->name,
1635 bond_is_active_slave(new_slave) ? "n active" : " backup",
1636 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1638 /* enslave is successful */
1641 /* Undo stages on error */
1643 netdev_rx_handler_unregister(slave_dev);
1646 if (!USES_PRIMARY(bond->params.mode))
1647 bond_hw_addr_flush(bond_dev, slave_dev);
1649 vlan_vids_del_by_dev(slave_dev, bond_dev);
1650 write_lock_bh(&bond->lock);
1651 bond_detach_slave(bond, new_slave);
1652 if (bond->primary_slave == new_slave)
1653 bond->primary_slave = NULL;
1654 if (bond->curr_active_slave == new_slave) {
1655 bond_change_active_slave(bond, NULL);
1656 write_unlock_bh(&bond->lock);
1657 read_lock(&bond->lock);
1658 write_lock_bh(&bond->curr_slave_lock);
1659 bond_select_active_slave(bond);
1660 write_unlock_bh(&bond->curr_slave_lock);
1661 read_unlock(&bond->lock);
1663 write_unlock_bh(&bond->lock);
1665 slave_disable_netpoll(new_slave);
1668 slave_dev->priv_flags &= ~IFF_BONDING;
1669 dev_close(slave_dev);
1672 if (!bond->params.fail_over_mac) {
1673 /* XXX TODO - fom follow mode needs to change master's
1674 * MAC if this slave's MAC is in use by the bond, or at
1675 * least print a warning.
1677 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1678 addr.sa_family = slave_dev->type;
1679 dev_set_mac_address(slave_dev, &addr);
1683 dev_set_mtu(slave_dev, new_slave->original_mtu);
1689 bond_compute_features(bond);
1690 /* Enslave of first slave has failed and we need to fix master's mac */
1691 if (!bond_has_slaves(bond) &&
1692 ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1693 eth_hw_addr_random(bond_dev);
1699 * Try to release the slave device <slave> from the bond device <master>
1700 * It is legal to access curr_active_slave without a lock because all the function
1701 * is write-locked. If "all" is true it means that the function is being called
1702 * while destroying a bond interface and all slaves are being released.
1704 * The rules for slave state should be:
1705 * for Active/Backup:
1706 * Active stays on all backups go down
1707 * for Bonded connections:
1708 * The first up interface should be left on and all others downed.
1710 static int __bond_release_one(struct net_device *bond_dev,
1711 struct net_device *slave_dev,
1714 struct bonding *bond = netdev_priv(bond_dev);
1715 struct slave *slave, *oldcurrent;
1716 struct sockaddr addr;
1717 int old_flags = bond_dev->flags;
1718 netdev_features_t old_features = bond_dev->features;
1720 /* slave is not a slave or master is not master of this slave */
1721 if (!(slave_dev->flags & IFF_SLAVE) ||
1722 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1723 pr_err("%s: Error: cannot release %s.\n",
1724 bond_dev->name, slave_dev->name);
1729 write_lock_bh(&bond->lock);
1731 slave = bond_get_slave_by_dev(bond, slave_dev);
1733 /* not a slave of this bond */
1734 pr_info("%s: %s not enslaved\n",
1735 bond_dev->name, slave_dev->name);
1736 write_unlock_bh(&bond->lock);
1737 unblock_netpoll_tx();
1741 write_unlock_bh(&bond->lock);
1743 bond_upper_dev_unlink(bond_dev, slave_dev);
1744 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1745 * for this slave anymore.
1747 netdev_rx_handler_unregister(slave_dev);
1748 write_lock_bh(&bond->lock);
1750 /* Inform AD package of unbinding of slave. */
1751 if (bond->params.mode == BOND_MODE_8023AD) {
1752 /* must be called before the slave is
1753 * detached from the list
1755 bond_3ad_unbind_slave(slave);
1758 pr_info("%s: releasing %s interface %s\n",
1760 bond_is_active_slave(slave) ? "active" : "backup",
1763 oldcurrent = bond->curr_active_slave;
1765 bond->current_arp_slave = NULL;
1767 /* release the slave from its bond */
1768 bond_detach_slave(bond, slave);
1770 if (!all && !bond->params.fail_over_mac) {
1771 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1772 bond_has_slaves(bond))
1773 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",
1774 bond_dev->name, slave_dev->name,
1776 bond_dev->name, slave_dev->name);
1779 if (bond->primary_slave == slave)
1780 bond->primary_slave = NULL;
1782 if (oldcurrent == slave)
1783 bond_change_active_slave(bond, NULL);
1785 if (bond_is_lb(bond)) {
1786 /* Must be called only after the slave has been
1787 * detached from the list and the curr_active_slave
1788 * has been cleared (if our_slave == old_current),
1789 * but before a new active slave is selected.
1791 write_unlock_bh(&bond->lock);
1792 bond_alb_deinit_slave(bond, slave);
1793 write_lock_bh(&bond->lock);
1797 rcu_assign_pointer(bond->curr_active_slave, NULL);
1798 } else if (oldcurrent == slave) {
1800 * Note that we hold RTNL over this sequence, so there
1801 * is no concern that another slave add/remove event
1804 write_unlock_bh(&bond->lock);
1805 read_lock(&bond->lock);
1806 write_lock_bh(&bond->curr_slave_lock);
1808 bond_select_active_slave(bond);
1810 write_unlock_bh(&bond->curr_slave_lock);
1811 read_unlock(&bond->lock);
1812 write_lock_bh(&bond->lock);
1815 if (!bond_has_slaves(bond)) {
1816 bond_set_carrier(bond);
1817 eth_hw_addr_random(bond_dev);
1819 if (vlan_uses_dev(bond_dev)) {
1820 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1821 bond_dev->name, bond_dev->name);
1822 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1827 write_unlock_bh(&bond->lock);
1828 unblock_netpoll_tx();
1831 if (!bond_has_slaves(bond)) {
1832 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1833 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1836 bond_compute_features(bond);
1837 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1838 (old_features & NETIF_F_VLAN_CHALLENGED))
1839 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1840 bond_dev->name, slave_dev->name, bond_dev->name);
1842 /* must do this from outside any spinlocks */
1843 vlan_vids_del_by_dev(slave_dev, bond_dev);
1845 /* If the mode USES_PRIMARY, then this cases was handled above by
1846 * bond_change_active_slave(..., NULL)
1848 if (!USES_PRIMARY(bond->params.mode)) {
1849 /* unset promiscuity level from slave
1850 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1851 * of the IFF_PROMISC flag in the bond_dev, but we need the
1852 * value of that flag before that change, as that was the value
1853 * when this slave was attached, so we cache at the start of the
1854 * function and use it here. Same goes for ALLMULTI below
1856 if (old_flags & IFF_PROMISC)
1857 dev_set_promiscuity(slave_dev, -1);
1859 /* unset allmulti level from slave */
1860 if (old_flags & IFF_ALLMULTI)
1861 dev_set_allmulti(slave_dev, -1);
1863 bond_hw_addr_flush(bond_dev, slave_dev);
1866 slave_disable_netpoll(slave);
1868 /* close slave before restoring its mac address */
1869 dev_close(slave_dev);
1871 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1872 /* restore original ("permanent") mac address */
1873 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1874 addr.sa_family = slave_dev->type;
1875 dev_set_mac_address(slave_dev, &addr);
1878 dev_set_mtu(slave_dev, slave->original_mtu);
1880 slave_dev->priv_flags &= ~IFF_BONDING;
1884 return 0; /* deletion OK */
1887 /* A wrapper used because of ndo_del_link */
1888 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1890 return __bond_release_one(bond_dev, slave_dev, false);
1894 * First release a slave and then destroy the bond if no more slaves are left.
1895 * Must be under rtnl_lock when this function is called.
1897 static int bond_release_and_destroy(struct net_device *bond_dev,
1898 struct net_device *slave_dev)
1900 struct bonding *bond = netdev_priv(bond_dev);
1903 ret = bond_release(bond_dev, slave_dev);
1904 if (ret == 0 && !bond_has_slaves(bond)) {
1905 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1906 pr_info("%s: destroying bond %s.\n",
1907 bond_dev->name, bond_dev->name);
1908 unregister_netdevice(bond_dev);
1914 * This function changes the active slave to slave <slave_dev>.
1915 * It returns -EINVAL in the following cases.
1916 * - <slave_dev> is not found in the list.
1917 * - There is not active slave now.
1918 * - <slave_dev> is already active.
1919 * - The link state of <slave_dev> is not BOND_LINK_UP.
1920 * - <slave_dev> is not running.
1921 * In these cases, this function does nothing.
1922 * In the other cases, current_slave pointer is changed and 0 is returned.
1924 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1926 struct bonding *bond = netdev_priv(bond_dev);
1927 struct slave *old_active = NULL;
1928 struct slave *new_active = NULL;
1931 if (!USES_PRIMARY(bond->params.mode))
1934 /* Verify that bond_dev is indeed the master of slave_dev */
1935 if (!(slave_dev->flags & IFF_SLAVE) ||
1936 !netdev_has_upper_dev(slave_dev, bond_dev))
1939 read_lock(&bond->lock);
1941 old_active = bond->curr_active_slave;
1942 new_active = bond_get_slave_by_dev(bond, slave_dev);
1944 * Changing to the current active: do nothing; return success.
1946 if (new_active && new_active == old_active) {
1947 read_unlock(&bond->lock);
1953 new_active->link == BOND_LINK_UP &&
1954 IS_UP(new_active->dev)) {
1956 write_lock_bh(&bond->curr_slave_lock);
1957 bond_change_active_slave(bond, new_active);
1958 write_unlock_bh(&bond->curr_slave_lock);
1959 unblock_netpoll_tx();
1963 read_unlock(&bond->lock);
1968 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1970 struct bonding *bond = netdev_priv(bond_dev);
1972 info->bond_mode = bond->params.mode;
1973 info->miimon = bond->params.miimon;
1975 read_lock(&bond->lock);
1976 info->num_slaves = bond->slave_cnt;
1977 read_unlock(&bond->lock);
1982 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1984 struct bonding *bond = netdev_priv(bond_dev);
1985 struct list_head *iter;
1986 int i = 0, res = -ENODEV;
1987 struct slave *slave;
1989 read_lock(&bond->lock);
1990 bond_for_each_slave(bond, slave, iter) {
1991 if (i++ == (int)info->slave_id) {
1993 strcpy(info->slave_name, slave->dev->name);
1994 info->link = slave->link;
1995 info->state = bond_slave_state(slave);
1996 info->link_failure_count = slave->link_failure_count;
2000 read_unlock(&bond->lock);
2005 /*-------------------------------- Monitoring -------------------------------*/
2008 static int bond_miimon_inspect(struct bonding *bond)
2010 int link_state, commit = 0;
2011 struct list_head *iter;
2012 struct slave *slave;
2013 bool ignore_updelay;
2015 ignore_updelay = !bond->curr_active_slave ? true : false;
2017 bond_for_each_slave(bond, slave, iter) {
2018 slave->new_link = BOND_LINK_NOCHANGE;
2020 link_state = bond_check_dev_link(bond, slave->dev, 0);
2022 switch (slave->link) {
2027 slave->link = BOND_LINK_FAIL;
2028 slave->delay = bond->params.downdelay;
2030 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2032 (bond->params.mode ==
2033 BOND_MODE_ACTIVEBACKUP) ?
2034 (bond_is_active_slave(slave) ?
2035 "active " : "backup ") : "",
2037 bond->params.downdelay * bond->params.miimon);
2040 case BOND_LINK_FAIL:
2043 * recovered before downdelay expired
2045 slave->link = BOND_LINK_UP;
2046 slave->jiffies = jiffies;
2047 pr_info("%s: link status up again after %d ms for interface %s.\n",
2049 (bond->params.downdelay - slave->delay) *
2050 bond->params.miimon,
2055 if (slave->delay <= 0) {
2056 slave->new_link = BOND_LINK_DOWN;
2064 case BOND_LINK_DOWN:
2068 slave->link = BOND_LINK_BACK;
2069 slave->delay = bond->params.updelay;
2072 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2073 bond->dev->name, slave->dev->name,
2074 ignore_updelay ? 0 :
2075 bond->params.updelay *
2076 bond->params.miimon);
2079 case BOND_LINK_BACK:
2081 slave->link = BOND_LINK_DOWN;
2082 pr_info("%s: link status down again after %d ms for interface %s.\n",
2084 (bond->params.updelay - slave->delay) *
2085 bond->params.miimon,
2094 if (slave->delay <= 0) {
2095 slave->new_link = BOND_LINK_UP;
2097 ignore_updelay = false;
2109 static void bond_miimon_commit(struct bonding *bond)
2111 struct list_head *iter;
2112 struct slave *slave;
2114 bond_for_each_slave(bond, slave, iter) {
2115 switch (slave->new_link) {
2116 case BOND_LINK_NOCHANGE:
2120 slave->link = BOND_LINK_UP;
2121 slave->jiffies = jiffies;
2123 if (bond->params.mode == BOND_MODE_8023AD) {
2124 /* prevent it from being the active one */
2125 bond_set_backup_slave(slave);
2126 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2127 /* make it immediately active */
2128 bond_set_active_slave(slave);
2129 } else if (slave != bond->primary_slave) {
2130 /* prevent it from being the active one */
2131 bond_set_backup_slave(slave);
2134 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2135 bond->dev->name, slave->dev->name,
2136 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2137 slave->duplex ? "full" : "half");
2139 /* notify ad that the link status has changed */
2140 if (bond->params.mode == BOND_MODE_8023AD)
2141 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2143 if (bond_is_lb(bond))
2144 bond_alb_handle_link_change(bond, slave,
2147 if (!bond->curr_active_slave ||
2148 (slave == bond->primary_slave))
2153 case BOND_LINK_DOWN:
2154 if (slave->link_failure_count < UINT_MAX)
2155 slave->link_failure_count++;
2157 slave->link = BOND_LINK_DOWN;
2159 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2160 bond->params.mode == BOND_MODE_8023AD)
2161 bond_set_slave_inactive_flags(slave);
2163 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2164 bond->dev->name, slave->dev->name);
2166 if (bond->params.mode == BOND_MODE_8023AD)
2167 bond_3ad_handle_link_change(slave,
2170 if (bond_is_lb(bond))
2171 bond_alb_handle_link_change(bond, slave,
2174 if (slave == bond->curr_active_slave)
2180 pr_err("%s: invalid new link %d on slave %s\n",
2181 bond->dev->name, slave->new_link,
2183 slave->new_link = BOND_LINK_NOCHANGE;
2191 write_lock_bh(&bond->curr_slave_lock);
2192 bond_select_active_slave(bond);
2193 write_unlock_bh(&bond->curr_slave_lock);
2194 unblock_netpoll_tx();
2197 bond_set_carrier(bond);
2203 * Really a wrapper that splits the mii monitor into two phases: an
2204 * inspection, then (if inspection indicates something needs to be done)
2205 * an acquisition of appropriate locks followed by a commit phase to
2206 * implement whatever link state changes are indicated.
2208 void bond_mii_monitor(struct work_struct *work)
2210 struct bonding *bond = container_of(work, struct bonding,
2212 bool should_notify_peers = false;
2213 unsigned long delay;
2215 read_lock(&bond->lock);
2217 delay = msecs_to_jiffies(bond->params.miimon);
2219 if (!bond_has_slaves(bond))
2222 should_notify_peers = bond_should_notify_peers(bond);
2224 if (bond_miimon_inspect(bond)) {
2225 read_unlock(&bond->lock);
2227 /* Race avoidance with bond_close cancel of workqueue */
2228 if (!rtnl_trylock()) {
2229 read_lock(&bond->lock);
2231 should_notify_peers = false;
2235 read_lock(&bond->lock);
2237 bond_miimon_commit(bond);
2239 read_unlock(&bond->lock);
2240 rtnl_unlock(); /* might sleep, hold no other locks */
2241 read_lock(&bond->lock);
2245 if (bond->params.miimon)
2246 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2248 read_unlock(&bond->lock);
2250 if (should_notify_peers) {
2251 if (!rtnl_trylock())
2253 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2258 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2260 struct net_device *upper;
2261 struct list_head *iter;
2264 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2268 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2269 if (ip == bond_confirm_addr(upper, 0, ip)) {
2280 * We go to the (large) trouble of VLAN tagging ARP frames because
2281 * switches in VLAN mode (especially if ports are configured as
2282 * "native" to a VLAN) might not pass non-tagged frames.
2284 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2286 struct sk_buff *skb;
2288 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2289 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2291 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2292 NULL, slave_dev->dev_addr, NULL);
2295 pr_err("ARP packet allocation failed\n");
2299 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2301 pr_err("failed to insert VLAN tag\n");
2309 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2311 struct net_device *upper, *vlan_upper;
2312 struct list_head *iter, *vlan_iter;
2314 __be32 *targets = bond->params.arp_targets, addr;
2317 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2318 pr_debug("basa: target %pI4\n", &targets[i]);
2320 /* Find out through which dev should the packet go */
2321 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2324 pr_debug("%s: no route to arp_ip_target %pI4\n",
2325 bond->dev->name, &targets[i]);
2331 /* bond device itself */
2332 if (rt->dst.dev == bond->dev)
2336 /* first we search only for vlan devices. for every vlan
2337 * found we verify its upper dev list, searching for the
2338 * rt->dst.dev. If found we save the tag of the vlan and
2339 * proceed to send the packet.
2343 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2345 if (!is_vlan_dev(vlan_upper))
2347 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2349 if (upper == rt->dst.dev) {
2350 vlan_id = vlan_dev_vlan_id(vlan_upper);
2357 /* if the device we're looking for is not on top of any of
2358 * our upper vlans, then just search for any dev that
2359 * matches, and in case it's a vlan - save the id
2361 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2362 if (upper == rt->dst.dev) {
2363 /* if it's a vlan - get its VID */
2364 if (is_vlan_dev(upper))
2365 vlan_id = vlan_dev_vlan_id(upper);
2373 /* Not our device - skip */
2374 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2375 bond->dev->name, &targets[i],
2376 rt->dst.dev ? rt->dst.dev->name : "NULL");
2382 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2384 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2389 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2393 if (!sip || !bond_has_this_ip(bond, tip)) {
2394 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2398 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2400 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2403 slave->last_arp_rx = jiffies;
2404 slave->target_last_arp_rx[i] = jiffies;
2407 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2408 struct slave *slave)
2410 struct arphdr *arp = (struct arphdr *)skb->data;
2411 unsigned char *arp_ptr;
2415 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2416 return RX_HANDLER_ANOTHER;
2418 read_lock(&bond->lock);
2420 if (!slave_do_arp_validate(bond, slave))
2423 alen = arp_hdr_len(bond->dev);
2425 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2426 bond->dev->name, skb->dev->name);
2428 if (alen > skb_headlen(skb)) {
2429 arp = kmalloc(alen, GFP_ATOMIC);
2432 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2436 if (arp->ar_hln != bond->dev->addr_len ||
2437 skb->pkt_type == PACKET_OTHERHOST ||
2438 skb->pkt_type == PACKET_LOOPBACK ||
2439 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2440 arp->ar_pro != htons(ETH_P_IP) ||
2444 arp_ptr = (unsigned char *)(arp + 1);
2445 arp_ptr += bond->dev->addr_len;
2446 memcpy(&sip, arp_ptr, 4);
2447 arp_ptr += 4 + bond->dev->addr_len;
2448 memcpy(&tip, arp_ptr, 4);
2450 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2451 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2452 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2456 * Backup slaves won't see the ARP reply, but do come through
2457 * here for each ARP probe (so we swap the sip/tip to validate
2458 * the probe). In a "redundant switch, common router" type of
2459 * configuration, the ARP probe will (hopefully) travel from
2460 * the active, through one switch, the router, then the other
2461 * switch before reaching the backup.
2463 * We 'trust' the arp requests if there is an active slave and
2464 * it received valid arp reply(s) after it became active. This
2465 * is done to avoid endless looping when we can't reach the
2466 * arp_ip_target and fool ourselves with our own arp requests.
2468 if (bond_is_active_slave(slave))
2469 bond_validate_arp(bond, slave, sip, tip);
2470 else if (bond->curr_active_slave &&
2471 time_after(slave_last_rx(bond, bond->curr_active_slave),
2472 bond->curr_active_slave->jiffies))
2473 bond_validate_arp(bond, slave, tip, sip);
2476 read_unlock(&bond->lock);
2477 if (arp != (struct arphdr *)skb->data)
2479 return RX_HANDLER_ANOTHER;
2482 /* function to verify if we're in the arp_interval timeslice, returns true if
2483 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2484 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2486 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2489 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2491 return time_in_range(jiffies,
2492 last_act - delta_in_ticks,
2493 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2497 * this function is called regularly to monitor each slave's link
2498 * ensuring that traffic is being sent and received when arp monitoring
2499 * is used in load-balancing mode. if the adapter has been dormant, then an
2500 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2501 * arp monitoring in active backup mode.
2503 void bond_loadbalance_arp_mon(struct work_struct *work)
2505 struct bonding *bond = container_of(work, struct bonding,
2507 struct slave *slave, *oldcurrent;
2508 struct list_head *iter;
2509 int do_failover = 0;
2511 read_lock(&bond->lock);
2513 if (!bond_has_slaves(bond))
2516 oldcurrent = bond->curr_active_slave;
2517 /* see if any of the previous devices are up now (i.e. they have
2518 * xmt and rcv traffic). the curr_active_slave does not come into
2519 * the picture unless it is null. also, slave->jiffies is not needed
2520 * here because we send an arp on each slave and give a slave as
2521 * long as it needs to get the tx/rx within the delta.
2522 * TODO: what about up/down delay in arp mode? it wasn't here before
2525 bond_for_each_slave(bond, slave, iter) {
2526 unsigned long trans_start = dev_trans_start(slave->dev);
2528 if (slave->link != BOND_LINK_UP) {
2529 if (bond_time_in_interval(bond, trans_start, 1) &&
2530 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2532 slave->link = BOND_LINK_UP;
2533 bond_set_active_slave(slave);
2535 /* primary_slave has no meaning in round-robin
2536 * mode. the window of a slave being up and
2537 * curr_active_slave being null after enslaving
2541 pr_info("%s: link status definitely up for interface %s, ",
2546 pr_info("%s: interface %s is now up\n",
2552 /* slave->link == BOND_LINK_UP */
2554 /* not all switches will respond to an arp request
2555 * when the source ip is 0, so don't take the link down
2556 * if we don't know our ip yet
2558 if (!bond_time_in_interval(bond, trans_start, 2) ||
2559 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2561 slave->link = BOND_LINK_DOWN;
2562 bond_set_backup_slave(slave);
2564 if (slave->link_failure_count < UINT_MAX)
2565 slave->link_failure_count++;
2567 pr_info("%s: interface %s is now down.\n",
2571 if (slave == oldcurrent)
2576 /* note: if switch is in round-robin mode, all links
2577 * must tx arp to ensure all links rx an arp - otherwise
2578 * links may oscillate or not come up at all; if switch is
2579 * in something like xor mode, there is nothing we can
2580 * do - all replies will be rx'ed on same link causing slaves
2581 * to be unstable during low/no traffic periods
2583 if (IS_UP(slave->dev))
2584 bond_arp_send_all(bond, slave);
2589 write_lock_bh(&bond->curr_slave_lock);
2591 bond_select_active_slave(bond);
2593 write_unlock_bh(&bond->curr_slave_lock);
2594 unblock_netpoll_tx();
2598 if (bond->params.arp_interval)
2599 queue_delayed_work(bond->wq, &bond->arp_work,
2600 msecs_to_jiffies(bond->params.arp_interval));
2602 read_unlock(&bond->lock);
2606 * Called to inspect slaves for active-backup mode ARP monitor link state
2607 * changes. Sets new_link in slaves to specify what action should take
2608 * place for the slave. Returns 0 if no changes are found, >0 if changes
2609 * to link states must be committed.
2611 * Called with bond->lock held for read.
2613 static int bond_ab_arp_inspect(struct bonding *bond)
2615 unsigned long trans_start, last_rx;
2616 struct list_head *iter;
2617 struct slave *slave;
2620 bond_for_each_slave(bond, slave, iter) {
2621 slave->new_link = BOND_LINK_NOCHANGE;
2622 last_rx = slave_last_rx(bond, slave);
2624 if (slave->link != BOND_LINK_UP) {
2625 if (bond_time_in_interval(bond, last_rx, 1)) {
2626 slave->new_link = BOND_LINK_UP;
2633 * Give slaves 2*delta after being enslaved or made
2634 * active. This avoids bouncing, as the last receive
2635 * times need a full ARP monitor cycle to be updated.
2637 if (bond_time_in_interval(bond, slave->jiffies, 2))
2641 * Backup slave is down if:
2642 * - No current_arp_slave AND
2643 * - more than 3*delta since last receive AND
2644 * - the bond has an IP address
2646 * Note: a non-null current_arp_slave indicates
2647 * the curr_active_slave went down and we are
2648 * searching for a new one; under this condition
2649 * we only take the curr_active_slave down - this
2650 * gives each slave a chance to tx/rx traffic
2651 * before being taken out
2653 if (!bond_is_active_slave(slave) &&
2654 !bond->current_arp_slave &&
2655 !bond_time_in_interval(bond, last_rx, 3)) {
2656 slave->new_link = BOND_LINK_DOWN;
2661 * Active slave is down if:
2662 * - more than 2*delta since transmitting OR
2663 * - (more than 2*delta since receive AND
2664 * the bond has an IP address)
2666 trans_start = dev_trans_start(slave->dev);
2667 if (bond_is_active_slave(slave) &&
2668 (!bond_time_in_interval(bond, trans_start, 2) ||
2669 !bond_time_in_interval(bond, last_rx, 2))) {
2670 slave->new_link = BOND_LINK_DOWN;
2679 * Called to commit link state changes noted by inspection step of
2680 * active-backup mode ARP monitor.
2682 * Called with RTNL and bond->lock for read.
2684 static void bond_ab_arp_commit(struct bonding *bond)
2686 unsigned long trans_start;
2687 struct list_head *iter;
2688 struct slave *slave;
2690 bond_for_each_slave(bond, slave, iter) {
2691 switch (slave->new_link) {
2692 case BOND_LINK_NOCHANGE:
2696 trans_start = dev_trans_start(slave->dev);
2697 if (bond->curr_active_slave != slave ||
2698 (!bond->curr_active_slave &&
2699 bond_time_in_interval(bond, trans_start, 1))) {
2700 slave->link = BOND_LINK_UP;
2701 if (bond->current_arp_slave) {
2702 bond_set_slave_inactive_flags(
2703 bond->current_arp_slave);
2704 bond->current_arp_slave = NULL;
2707 pr_info("%s: link status definitely up for interface %s.\n",
2708 bond->dev->name, slave->dev->name);
2710 if (!bond->curr_active_slave ||
2711 (slave == bond->primary_slave))
2718 case BOND_LINK_DOWN:
2719 if (slave->link_failure_count < UINT_MAX)
2720 slave->link_failure_count++;
2722 slave->link = BOND_LINK_DOWN;
2723 bond_set_slave_inactive_flags(slave);
2725 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2726 bond->dev->name, slave->dev->name);
2728 if (slave == bond->curr_active_slave) {
2729 bond->current_arp_slave = NULL;
2736 pr_err("%s: impossible: new_link %d on slave %s\n",
2737 bond->dev->name, slave->new_link,
2745 write_lock_bh(&bond->curr_slave_lock);
2746 bond_select_active_slave(bond);
2747 write_unlock_bh(&bond->curr_slave_lock);
2748 unblock_netpoll_tx();
2751 bond_set_carrier(bond);
2755 * Send ARP probes for active-backup mode ARP monitor.
2757 * Called with bond->lock held for read.
2759 static void bond_ab_arp_probe(struct bonding *bond)
2761 struct slave *slave, *before = NULL, *new_slave = NULL;
2762 struct list_head *iter;
2765 read_lock(&bond->curr_slave_lock);
2767 if (bond->current_arp_slave && bond->curr_active_slave)
2768 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2769 bond->current_arp_slave->dev->name,
2770 bond->curr_active_slave->dev->name);
2772 if (bond->curr_active_slave) {
2773 bond_arp_send_all(bond, bond->curr_active_slave);
2774 read_unlock(&bond->curr_slave_lock);
2778 read_unlock(&bond->curr_slave_lock);
2780 /* if we don't have a curr_active_slave, search for the next available
2781 * backup slave from the current_arp_slave and make it the candidate
2782 * for becoming the curr_active_slave
2785 if (!bond->current_arp_slave) {
2786 bond->current_arp_slave = bond_first_slave(bond);
2787 if (!bond->current_arp_slave)
2791 bond_set_slave_inactive_flags(bond->current_arp_slave);
2793 bond_for_each_slave(bond, slave, iter) {
2794 if (!found && !before && IS_UP(slave->dev))
2797 if (found && !new_slave && IS_UP(slave->dev))
2799 /* if the link state is up at this point, we
2800 * mark it down - this can happen if we have
2801 * simultaneous link failures and
2802 * reselect_active_interface doesn't make this
2803 * one the current slave so it is still marked
2804 * up when it is actually down
2806 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2807 slave->link = BOND_LINK_DOWN;
2808 if (slave->link_failure_count < UINT_MAX)
2809 slave->link_failure_count++;
2811 bond_set_slave_inactive_flags(slave);
2813 pr_info("%s: backup interface %s is now down.\n",
2814 bond->dev->name, slave->dev->name);
2816 if (slave == bond->current_arp_slave)
2820 if (!new_slave && before)
2826 new_slave->link = BOND_LINK_BACK;
2827 bond_set_slave_active_flags(new_slave);
2828 bond_arp_send_all(bond, new_slave);
2829 new_slave->jiffies = jiffies;
2830 bond->current_arp_slave = new_slave;
2834 void bond_activebackup_arp_mon(struct work_struct *work)
2836 struct bonding *bond = container_of(work, struct bonding,
2838 bool should_notify_peers = false;
2841 read_lock(&bond->lock);
2843 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2845 if (!bond_has_slaves(bond))
2848 should_notify_peers = bond_should_notify_peers(bond);
2850 if (bond_ab_arp_inspect(bond)) {
2851 read_unlock(&bond->lock);
2853 /* Race avoidance with bond_close flush of workqueue */
2854 if (!rtnl_trylock()) {
2855 read_lock(&bond->lock);
2857 should_notify_peers = false;
2861 read_lock(&bond->lock);
2863 bond_ab_arp_commit(bond);
2865 read_unlock(&bond->lock);
2867 read_lock(&bond->lock);
2870 bond_ab_arp_probe(bond);
2873 if (bond->params.arp_interval)
2874 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2876 read_unlock(&bond->lock);
2878 if (should_notify_peers) {
2879 if (!rtnl_trylock())
2881 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2886 /*-------------------------- netdev event handling --------------------------*/
2889 * Change device name
2891 static int bond_event_changename(struct bonding *bond)
2893 bond_remove_proc_entry(bond);
2894 bond_create_proc_entry(bond);
2896 bond_debug_reregister(bond);
2901 static int bond_master_netdev_event(unsigned long event,
2902 struct net_device *bond_dev)
2904 struct bonding *event_bond = netdev_priv(bond_dev);
2907 case NETDEV_CHANGENAME:
2908 return bond_event_changename(event_bond);
2909 case NETDEV_UNREGISTER:
2910 bond_remove_proc_entry(event_bond);
2912 case NETDEV_REGISTER:
2913 bond_create_proc_entry(event_bond);
2915 case NETDEV_NOTIFY_PEERS:
2916 if (event_bond->send_peer_notif)
2917 event_bond->send_peer_notif--;
2926 static int bond_slave_netdev_event(unsigned long event,
2927 struct net_device *slave_dev)
2929 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2930 struct bonding *bond;
2931 struct net_device *bond_dev;
2935 /* A netdev event can be generated while enslaving a device
2936 * before netdev_rx_handler_register is called in which case
2937 * slave will be NULL
2941 bond_dev = slave->bond->dev;
2945 case NETDEV_UNREGISTER:
2946 if (bond_dev->type != ARPHRD_ETHER)
2947 bond_release_and_destroy(bond_dev, slave_dev);
2949 bond_release(bond_dev, slave_dev);
2953 old_speed = slave->speed;
2954 old_duplex = slave->duplex;
2956 bond_update_speed_duplex(slave);
2958 if (bond->params.mode == BOND_MODE_8023AD) {
2959 if (old_speed != slave->speed)
2960 bond_3ad_adapter_speed_changed(slave);
2961 if (old_duplex != slave->duplex)
2962 bond_3ad_adapter_duplex_changed(slave);
2967 * ... Or is it this?
2970 case NETDEV_CHANGEMTU:
2972 * TODO: Should slaves be allowed to
2973 * independently alter their MTU? For
2974 * an active-backup bond, slaves need
2975 * not be the same type of device, so
2976 * MTUs may vary. For other modes,
2977 * slaves arguably should have the
2978 * same MTUs. To do this, we'd need to
2979 * take over the slave's change_mtu
2980 * function for the duration of their
2984 case NETDEV_CHANGENAME:
2986 * TODO: handle changing the primary's name
2989 case NETDEV_FEAT_CHANGE:
2990 bond_compute_features(bond);
2992 case NETDEV_RESEND_IGMP:
2993 /* Propagate to master device */
2994 call_netdevice_notifiers(event, slave->bond->dev);
3004 * bond_netdev_event: handle netdev notifier chain events.
3006 * This function receives events for the netdev chain. The caller (an
3007 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3008 * locks for us to safely manipulate the slave devices (RTNL lock,
3011 static int bond_netdev_event(struct notifier_block *this,
3012 unsigned long event, void *ptr)
3014 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3016 pr_debug("event_dev: %s, event: %lx\n",
3017 event_dev ? event_dev->name : "None",
3020 if (!(event_dev->priv_flags & IFF_BONDING))
3023 if (event_dev->flags & IFF_MASTER) {
3024 pr_debug("IFF_MASTER\n");
3025 return bond_master_netdev_event(event, event_dev);
3028 if (event_dev->flags & IFF_SLAVE) {
3029 pr_debug("IFF_SLAVE\n");
3030 return bond_slave_netdev_event(event, event_dev);
3036 static struct notifier_block bond_netdev_notifier = {
3037 .notifier_call = bond_netdev_event,
3040 /*---------------------------- Hashing Policies -----------------------------*/
3042 /* L2 hash helper */
3043 static inline u32 bond_eth_hash(struct sk_buff *skb)
3045 struct ethhdr *data = (struct ethhdr *)skb->data;
3047 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3048 return data->h_dest[5] ^ data->h_source[5];
3053 /* Extract the appropriate headers based on bond's xmit policy */
3054 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3055 struct flow_keys *fk)
3057 const struct ipv6hdr *iph6;
3058 const struct iphdr *iph;
3059 int noff, proto = -1;
3061 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3062 return skb_flow_dissect(skb, fk);
3065 noff = skb_network_offset(skb);
3066 if (skb->protocol == htons(ETH_P_IP)) {
3067 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
3070 fk->src = iph->saddr;
3071 fk->dst = iph->daddr;
3072 noff += iph->ihl << 2;
3073 if (!ip_is_fragment(iph))
3074 proto = iph->protocol;
3075 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3076 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
3078 iph6 = ipv6_hdr(skb);
3079 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3080 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3081 noff += sizeof(*iph6);
3082 proto = iph6->nexthdr;
3086 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3087 fk->ports = skb_flow_get_ports(skb, noff, proto);
3093 * bond_xmit_hash - generate a hash value based on the xmit policy
3094 * @bond: bonding device
3095 * @skb: buffer to use for headers
3096 * @count: modulo value
3098 * This function will extract the necessary headers from the skb buffer and use
3099 * them to generate a hash based on the xmit_policy set in the bonding device
3100 * which will be reduced modulo count before returning.
3102 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
3104 struct flow_keys flow;
3107 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3108 !bond_flow_dissect(bond, skb, &flow))
3109 return bond_eth_hash(skb) % count;
3111 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3112 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3113 hash = bond_eth_hash(skb);
3115 hash = (__force u32)flow.ports;
3116 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3117 hash ^= (hash >> 16);
3118 hash ^= (hash >> 8);
3120 return hash % count;
3123 /*-------------------------- Device entry points ----------------------------*/
3125 static void bond_work_init_all(struct bonding *bond)
3127 INIT_DELAYED_WORK(&bond->mcast_work,
3128 bond_resend_igmp_join_requests_delayed);
3129 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3130 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3131 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3132 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3134 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3135 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3138 static void bond_work_cancel_all(struct bonding *bond)
3140 cancel_delayed_work_sync(&bond->mii_work);
3141 cancel_delayed_work_sync(&bond->arp_work);
3142 cancel_delayed_work_sync(&bond->alb_work);
3143 cancel_delayed_work_sync(&bond->ad_work);
3144 cancel_delayed_work_sync(&bond->mcast_work);
3147 static int bond_open(struct net_device *bond_dev)
3149 struct bonding *bond = netdev_priv(bond_dev);
3150 struct list_head *iter;
3151 struct slave *slave;
3153 /* reset slave->backup and slave->inactive */
3154 read_lock(&bond->lock);
3155 if (bond_has_slaves(bond)) {
3156 read_lock(&bond->curr_slave_lock);
3157 bond_for_each_slave(bond, slave, iter) {
3158 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3159 && (slave != bond->curr_active_slave)) {
3160 bond_set_slave_inactive_flags(slave);
3162 bond_set_slave_active_flags(slave);
3165 read_unlock(&bond->curr_slave_lock);
3167 read_unlock(&bond->lock);
3169 bond_work_init_all(bond);
3171 if (bond_is_lb(bond)) {
3172 /* bond_alb_initialize must be called before the timer
3175 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3177 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3180 if (bond->params.miimon) /* link check interval, in milliseconds. */
3181 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3183 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3184 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3185 if (bond->params.arp_validate)
3186 bond->recv_probe = bond_arp_rcv;
3189 if (bond->params.mode == BOND_MODE_8023AD) {
3190 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3191 /* register to receive LACPDUs */
3192 bond->recv_probe = bond_3ad_lacpdu_recv;
3193 bond_3ad_initiate_agg_selection(bond, 1);
3199 static int bond_close(struct net_device *bond_dev)
3201 struct bonding *bond = netdev_priv(bond_dev);
3203 bond_work_cancel_all(bond);
3204 bond->send_peer_notif = 0;
3205 if (bond_is_lb(bond))
3206 bond_alb_deinitialize(bond);
3207 bond->recv_probe = NULL;
3212 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3213 struct rtnl_link_stats64 *stats)
3215 struct bonding *bond = netdev_priv(bond_dev);
3216 struct rtnl_link_stats64 temp;
3217 struct list_head *iter;
3218 struct slave *slave;
3220 memset(stats, 0, sizeof(*stats));
3222 read_lock_bh(&bond->lock);
3223 bond_for_each_slave(bond, slave, iter) {
3224 const struct rtnl_link_stats64 *sstats =
3225 dev_get_stats(slave->dev, &temp);
3227 stats->rx_packets += sstats->rx_packets;
3228 stats->rx_bytes += sstats->rx_bytes;
3229 stats->rx_errors += sstats->rx_errors;
3230 stats->rx_dropped += sstats->rx_dropped;
3232 stats->tx_packets += sstats->tx_packets;
3233 stats->tx_bytes += sstats->tx_bytes;
3234 stats->tx_errors += sstats->tx_errors;
3235 stats->tx_dropped += sstats->tx_dropped;
3237 stats->multicast += sstats->multicast;
3238 stats->collisions += sstats->collisions;
3240 stats->rx_length_errors += sstats->rx_length_errors;
3241 stats->rx_over_errors += sstats->rx_over_errors;
3242 stats->rx_crc_errors += sstats->rx_crc_errors;
3243 stats->rx_frame_errors += sstats->rx_frame_errors;
3244 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3245 stats->rx_missed_errors += sstats->rx_missed_errors;
3247 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3248 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3249 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3250 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3251 stats->tx_window_errors += sstats->tx_window_errors;
3253 read_unlock_bh(&bond->lock);
3258 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3260 struct net_device *slave_dev = NULL;
3261 struct ifbond k_binfo;
3262 struct ifbond __user *u_binfo = NULL;
3263 struct ifslave k_sinfo;
3264 struct ifslave __user *u_sinfo = NULL;
3265 struct mii_ioctl_data *mii = NULL;
3269 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3281 * We do this again just in case we were called by SIOCGMIIREG
3282 * instead of SIOCGMIIPHY.
3289 if (mii->reg_num == 1) {
3290 struct bonding *bond = netdev_priv(bond_dev);
3292 read_lock(&bond->lock);
3293 read_lock(&bond->curr_slave_lock);
3294 if (netif_carrier_ok(bond->dev))
3295 mii->val_out = BMSR_LSTATUS;
3297 read_unlock(&bond->curr_slave_lock);
3298 read_unlock(&bond->lock);
3302 case BOND_INFO_QUERY_OLD:
3303 case SIOCBONDINFOQUERY:
3304 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3306 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3309 res = bond_info_query(bond_dev, &k_binfo);
3311 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3315 case BOND_SLAVE_INFO_QUERY_OLD:
3316 case SIOCBONDSLAVEINFOQUERY:
3317 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3319 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3322 res = bond_slave_info_query(bond_dev, &k_sinfo);
3324 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3333 net = dev_net(bond_dev);
3335 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3338 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3340 pr_debug("slave_dev=%p:\n", slave_dev);
3345 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3347 case BOND_ENSLAVE_OLD:
3348 case SIOCBONDENSLAVE:
3349 res = bond_enslave(bond_dev, slave_dev);
3351 case BOND_RELEASE_OLD:
3352 case SIOCBONDRELEASE:
3353 res = bond_release(bond_dev, slave_dev);
3355 case BOND_SETHWADDR_OLD:
3356 case SIOCBONDSETHWADDR:
3357 bond_set_dev_addr(bond_dev, slave_dev);
3360 case BOND_CHANGE_ACTIVE_OLD:
3361 case SIOCBONDCHANGEACTIVE:
3362 res = bond_ioctl_change_active(bond_dev, slave_dev);
3374 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3376 struct bonding *bond = netdev_priv(bond_dev);
3378 if (change & IFF_PROMISC)
3379 bond_set_promiscuity(bond,
3380 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3382 if (change & IFF_ALLMULTI)
3383 bond_set_allmulti(bond,
3384 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3387 static void bond_set_rx_mode(struct net_device *bond_dev)
3389 struct bonding *bond = netdev_priv(bond_dev);
3390 struct list_head *iter;
3391 struct slave *slave;
3395 if (USES_PRIMARY(bond->params.mode)) {
3396 slave = rcu_dereference(bond->curr_active_slave);
3398 dev_uc_sync(slave->dev, bond_dev);
3399 dev_mc_sync(slave->dev, bond_dev);
3402 bond_for_each_slave_rcu(bond, slave, iter) {
3403 dev_uc_sync_multiple(slave->dev, bond_dev);
3404 dev_mc_sync_multiple(slave->dev, bond_dev);
3410 static int bond_neigh_init(struct neighbour *n)
3412 struct bonding *bond = netdev_priv(n->dev);
3413 const struct net_device_ops *slave_ops;
3414 struct neigh_parms parms;
3415 struct slave *slave;
3418 slave = bond_first_slave(bond);
3421 slave_ops = slave->dev->netdev_ops;
3422 if (!slave_ops->ndo_neigh_setup)
3425 parms.neigh_setup = NULL;
3426 parms.neigh_cleanup = NULL;
3427 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3432 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3433 * after the last slave has been detached. Assumes that all slaves
3434 * utilize the same neigh_cleanup (true at this writing as only user
3437 n->parms->neigh_cleanup = parms.neigh_cleanup;
3439 if (!parms.neigh_setup)
3442 return parms.neigh_setup(n);
3446 * The bonding ndo_neigh_setup is called at init time beofre any
3447 * slave exists. So we must declare proxy setup function which will
3448 * be used at run time to resolve the actual slave neigh param setup.
3450 * It's also called by master devices (such as vlans) to setup their
3451 * underlying devices. In that case - do nothing, we're already set up from
3454 static int bond_neigh_setup(struct net_device *dev,
3455 struct neigh_parms *parms)
3457 /* modify only our neigh_parms */
3458 if (parms->dev == dev)
3459 parms->neigh_setup = bond_neigh_init;
3465 * Change the MTU of all of a master's slaves to match the master
3467 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3469 struct bonding *bond = netdev_priv(bond_dev);
3470 struct slave *slave, *rollback_slave;
3471 struct list_head *iter;
3474 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3475 (bond_dev ? bond_dev->name : "None"), new_mtu);
3477 /* Can't hold bond->lock with bh disabled here since
3478 * some base drivers panic. On the other hand we can't
3479 * hold bond->lock without bh disabled because we'll
3480 * deadlock. The only solution is to rely on the fact
3481 * that we're under rtnl_lock here, and the slaves
3482 * list won't change. This doesn't solve the problem
3483 * of setting the slave's MTU while it is
3484 * transmitting, but the assumption is that the base
3485 * driver can handle that.
3487 * TODO: figure out a way to safely iterate the slaves
3488 * list, but without holding a lock around the actual
3489 * call to the base driver.
3492 bond_for_each_slave(bond, slave, iter) {
3493 pr_debug("s %p c_m %p\n",
3495 slave->dev->netdev_ops->ndo_change_mtu);
3497 res = dev_set_mtu(slave->dev, new_mtu);
3500 /* If we failed to set the slave's mtu to the new value
3501 * we must abort the operation even in ACTIVE_BACKUP
3502 * mode, because if we allow the backup slaves to have
3503 * different mtu values than the active slave we'll
3504 * need to change their mtu when doing a failover. That
3505 * means changing their mtu from timer context, which
3506 * is probably not a good idea.
3508 pr_debug("err %d %s\n", res, slave->dev->name);
3513 bond_dev->mtu = new_mtu;
3518 /* unwind from head to the slave that failed */
3519 bond_for_each_slave(bond, rollback_slave, iter) {
3522 if (rollback_slave == slave)
3525 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3527 pr_debug("unwind err %d dev %s\n",
3528 tmp_res, rollback_slave->dev->name);
3538 * Note that many devices must be down to change the HW address, and
3539 * downing the master releases all slaves. We can make bonds full of
3540 * bonding devices to test this, however.
3542 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3544 struct bonding *bond = netdev_priv(bond_dev);
3545 struct slave *slave, *rollback_slave;
3546 struct sockaddr *sa = addr, tmp_sa;
3547 struct list_head *iter;
3550 if (bond->params.mode == BOND_MODE_ALB)
3551 return bond_alb_set_mac_address(bond_dev, addr);
3554 pr_debug("bond=%p, name=%s\n",
3555 bond, bond_dev ? bond_dev->name : "None");
3557 /* If fail_over_mac is enabled, do nothing and return success.
3558 * Returning an error causes ifenslave to fail.
3560 if (bond->params.fail_over_mac)
3563 if (!is_valid_ether_addr(sa->sa_data))
3564 return -EADDRNOTAVAIL;
3566 /* Can't hold bond->lock with bh disabled here since
3567 * some base drivers panic. On the other hand we can't
3568 * hold bond->lock without bh disabled because we'll
3569 * deadlock. The only solution is to rely on the fact
3570 * that we're under rtnl_lock here, and the slaves
3571 * list won't change. This doesn't solve the problem
3572 * of setting the slave's hw address while it is
3573 * transmitting, but the assumption is that the base
3574 * driver can handle that.
3576 * TODO: figure out a way to safely iterate the slaves
3577 * list, but without holding a lock around the actual
3578 * call to the base driver.
3581 bond_for_each_slave(bond, slave, iter) {
3582 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3583 pr_debug("slave %p %s\n", slave, slave->dev->name);
3585 if (slave_ops->ndo_set_mac_address == NULL) {
3587 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3591 res = dev_set_mac_address(slave->dev, addr);
3593 /* TODO: consider downing the slave
3595 * User should expect communications
3596 * breakage anyway until ARP finish
3599 pr_debug("err %d %s\n", res, slave->dev->name);
3605 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3609 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3610 tmp_sa.sa_family = bond_dev->type;
3612 /* unwind from head to the slave that failed */
3613 bond_for_each_slave(bond, rollback_slave, iter) {
3616 if (rollback_slave == slave)
3619 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3621 pr_debug("unwind err %d dev %s\n",
3622 tmp_res, rollback_slave->dev->name);
3630 * bond_xmit_slave_id - transmit skb through slave with slave_id
3631 * @bond: bonding device that is transmitting
3632 * @skb: buffer to transmit
3633 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3635 * This function tries to transmit through slave with slave_id but in case
3636 * it fails, it tries to find the first available slave for transmission.
3637 * The skb is consumed in all cases, thus the function is void.
3639 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3641 struct list_head *iter;
3642 struct slave *slave;
3645 /* Here we start from the slave with slave_id */
3646 bond_for_each_slave_rcu(bond, slave, iter) {
3648 if (slave_can_tx(slave)) {
3649 bond_dev_queue_xmit(bond, skb, slave->dev);
3655 /* Here we start from the first slave up to slave_id */
3657 bond_for_each_slave_rcu(bond, slave, iter) {
3660 if (slave_can_tx(slave)) {
3661 bond_dev_queue_xmit(bond, skb, slave->dev);
3665 /* no slave that can tx has been found */
3669 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3671 struct bonding *bond = netdev_priv(bond_dev);
3672 struct iphdr *iph = ip_hdr(skb);
3673 struct slave *slave;
3676 * Start with the curr_active_slave that joined the bond as the
3677 * default for sending IGMP traffic. For failover purposes one
3678 * needs to maintain some consistency for the interface that will
3679 * send the join/membership reports. The curr_active_slave found
3680 * will send all of this type of traffic.
3682 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3683 slave = rcu_dereference(bond->curr_active_slave);
3684 if (slave && slave_can_tx(slave))
3685 bond_dev_queue_xmit(bond, skb, slave->dev);
3687 bond_xmit_slave_id(bond, skb, 0);
3689 bond_xmit_slave_id(bond, skb,
3690 bond->rr_tx_counter++ % bond->slave_cnt);
3693 return NETDEV_TX_OK;
3697 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3698 * the bond has a usable interface.
3700 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3702 struct bonding *bond = netdev_priv(bond_dev);
3703 struct slave *slave;
3705 slave = rcu_dereference(bond->curr_active_slave);
3707 bond_dev_queue_xmit(bond, skb, slave->dev);
3711 return NETDEV_TX_OK;
3714 /* In bond_xmit_xor() , we determine the output device by using a pre-
3715 * determined xmit_hash_policy(), If the selected device is not enabled,
3716 * find the next active slave.
3718 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3720 struct bonding *bond = netdev_priv(bond_dev);
3722 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3724 return NETDEV_TX_OK;
3727 /* in broadcast mode, we send everything to all usable interfaces. */
3728 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3730 struct bonding *bond = netdev_priv(bond_dev);
3731 struct slave *slave = NULL;
3732 struct list_head *iter;
3734 bond_for_each_slave_rcu(bond, slave, iter) {
3735 if (bond_is_last_slave(bond, slave))
3737 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3738 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3741 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3745 /* bond_dev_queue_xmit always returns 0 */
3746 bond_dev_queue_xmit(bond, skb2, slave->dev);
3749 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3750 bond_dev_queue_xmit(bond, skb, slave->dev);
3754 return NETDEV_TX_OK;
3757 /*------------------------- Device initialization ---------------------------*/
3760 * Lookup the slave that corresponds to a qid
3762 static inline int bond_slave_override(struct bonding *bond,
3763 struct sk_buff *skb)
3765 struct slave *slave = NULL;
3766 struct slave *check_slave;
3767 struct list_head *iter;
3770 if (!skb->queue_mapping)
3773 /* Find out if any slaves have the same mapping as this skb. */
3774 bond_for_each_slave_rcu(bond, check_slave, iter) {
3775 if (check_slave->queue_id == skb->queue_mapping) {
3776 slave = check_slave;
3781 /* If the slave isn't UP, use default transmit policy. */
3782 if (slave && slave->queue_id && IS_UP(slave->dev) &&
3783 (slave->link == BOND_LINK_UP)) {
3784 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3791 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3794 * This helper function exists to help dev_pick_tx get the correct
3795 * destination queue. Using a helper function skips a call to
3796 * skb_tx_hash and will put the skbs in the queue we expect on their
3797 * way down to the bonding driver.
3799 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3802 * Save the original txq to restore before passing to the driver
3804 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3806 if (unlikely(txq >= dev->real_num_tx_queues)) {
3808 txq -= dev->real_num_tx_queues;
3809 } while (txq >= dev->real_num_tx_queues);
3814 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3816 struct bonding *bond = netdev_priv(dev);
3818 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3819 if (!bond_slave_override(bond, skb))
3820 return NETDEV_TX_OK;
3823 switch (bond->params.mode) {
3824 case BOND_MODE_ROUNDROBIN:
3825 return bond_xmit_roundrobin(skb, dev);
3826 case BOND_MODE_ACTIVEBACKUP:
3827 return bond_xmit_activebackup(skb, dev);
3829 return bond_xmit_xor(skb, dev);
3830 case BOND_MODE_BROADCAST:
3831 return bond_xmit_broadcast(skb, dev);
3832 case BOND_MODE_8023AD:
3833 return bond_3ad_xmit_xor(skb, dev);
3836 return bond_alb_xmit(skb, dev);
3838 /* Should never happen, mode already checked */
3839 pr_err("%s: Error: Unknown bonding mode %d\n",
3840 dev->name, bond->params.mode);
3843 return NETDEV_TX_OK;
3847 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3849 struct bonding *bond = netdev_priv(dev);
3850 netdev_tx_t ret = NETDEV_TX_OK;
3853 * If we risk deadlock from transmitting this in the
3854 * netpoll path, tell netpoll to queue the frame for later tx
3856 if (is_netpoll_tx_blocked(dev))
3857 return NETDEV_TX_BUSY;
3860 if (bond_has_slaves(bond))
3861 ret = __bond_start_xmit(skb, dev);
3869 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3870 struct ethtool_cmd *ecmd)
3872 struct bonding *bond = netdev_priv(bond_dev);
3873 unsigned long speed = 0;
3874 struct list_head *iter;
3875 struct slave *slave;
3877 ecmd->duplex = DUPLEX_UNKNOWN;
3878 ecmd->port = PORT_OTHER;
3880 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3881 * do not need to check mode. Though link speed might not represent
3882 * the true receive or transmit bandwidth (not all modes are symmetric)
3883 * this is an accurate maximum.
3885 read_lock(&bond->lock);
3886 bond_for_each_slave(bond, slave, iter) {
3887 if (SLAVE_IS_OK(slave)) {
3888 if (slave->speed != SPEED_UNKNOWN)
3889 speed += slave->speed;
3890 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3891 slave->duplex != DUPLEX_UNKNOWN)
3892 ecmd->duplex = slave->duplex;
3895 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3896 read_unlock(&bond->lock);
3901 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3902 struct ethtool_drvinfo *drvinfo)
3904 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3905 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3906 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3910 static const struct ethtool_ops bond_ethtool_ops = {
3911 .get_drvinfo = bond_ethtool_get_drvinfo,
3912 .get_settings = bond_ethtool_get_settings,
3913 .get_link = ethtool_op_get_link,
3916 static const struct net_device_ops bond_netdev_ops = {
3917 .ndo_init = bond_init,
3918 .ndo_uninit = bond_uninit,
3919 .ndo_open = bond_open,
3920 .ndo_stop = bond_close,
3921 .ndo_start_xmit = bond_start_xmit,
3922 .ndo_select_queue = bond_select_queue,
3923 .ndo_get_stats64 = bond_get_stats,
3924 .ndo_do_ioctl = bond_do_ioctl,
3925 .ndo_change_rx_flags = bond_change_rx_flags,
3926 .ndo_set_rx_mode = bond_set_rx_mode,
3927 .ndo_change_mtu = bond_change_mtu,
3928 .ndo_set_mac_address = bond_set_mac_address,
3929 .ndo_neigh_setup = bond_neigh_setup,
3930 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3931 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3932 #ifdef CONFIG_NET_POLL_CONTROLLER
3933 .ndo_netpoll_setup = bond_netpoll_setup,
3934 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3935 .ndo_poll_controller = bond_poll_controller,
3937 .ndo_add_slave = bond_enslave,
3938 .ndo_del_slave = bond_release,
3939 .ndo_fix_features = bond_fix_features,
3942 static const struct device_type bond_type = {
3946 static void bond_destructor(struct net_device *bond_dev)
3948 struct bonding *bond = netdev_priv(bond_dev);
3950 destroy_workqueue(bond->wq);
3951 free_netdev(bond_dev);
3954 void bond_setup(struct net_device *bond_dev)
3956 struct bonding *bond = netdev_priv(bond_dev);
3958 /* initialize rwlocks */
3959 rwlock_init(&bond->lock);
3960 rwlock_init(&bond->curr_slave_lock);
3961 bond->params = bonding_defaults;
3963 /* Initialize pointers */
3964 bond->dev = bond_dev;
3966 /* Initialize the device entry points */
3967 ether_setup(bond_dev);
3968 bond_dev->netdev_ops = &bond_netdev_ops;
3969 bond_dev->ethtool_ops = &bond_ethtool_ops;
3971 bond_dev->destructor = bond_destructor;
3973 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3975 /* Initialize the device options */
3976 bond_dev->tx_queue_len = 0;
3977 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3978 bond_dev->priv_flags |= IFF_BONDING;
3979 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3981 /* At first, we block adding VLANs. That's the only way to
3982 * prevent problems that occur when adding VLANs over an
3983 * empty bond. The block will be removed once non-challenged
3984 * slaves are enslaved.
3986 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3988 /* don't acquire bond device's netif_tx_lock when
3990 bond_dev->features |= NETIF_F_LLTX;
3992 /* By default, we declare the bond to be fully
3993 * VLAN hardware accelerated capable. Special
3994 * care is taken in the various xmit functions
3995 * when there are slaves that are not hw accel
3999 bond_dev->hw_features = BOND_VLAN_FEATURES |
4000 NETIF_F_HW_VLAN_CTAG_TX |
4001 NETIF_F_HW_VLAN_CTAG_RX |
4002 NETIF_F_HW_VLAN_CTAG_FILTER;
4004 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4005 bond_dev->features |= bond_dev->hw_features;
4009 * Destroy a bonding device.
4010 * Must be under rtnl_lock when this function is called.
4012 static void bond_uninit(struct net_device *bond_dev)
4014 struct bonding *bond = netdev_priv(bond_dev);
4015 struct list_head *iter;
4016 struct slave *slave;
4018 bond_netpoll_cleanup(bond_dev);
4020 /* Release the bonded slaves */
4021 bond_for_each_slave(bond, slave, iter)
4022 __bond_release_one(bond_dev, slave->dev, true);
4023 pr_info("%s: released all slaves\n", bond_dev->name);
4025 list_del(&bond->bond_list);
4027 bond_debug_unregister(bond);
4030 /*------------------------- Module initialization ---------------------------*/
4033 * Convert string input module parms. Accept either the
4034 * number of the mode or its string name. A bit complicated because
4035 * some mode names are substrings of other names, and calls from sysfs
4036 * may have whitespace in the name (trailing newlines, for example).
4038 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4040 int modeint = -1, i, rv;
4041 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4043 for (p = (char *)buf; *p; p++)
4044 if (!(isdigit(*p) || isspace(*p)))
4048 rv = sscanf(buf, "%20s", modestr);
4050 rv = sscanf(buf, "%d", &modeint);
4055 for (i = 0; tbl[i].modename; i++) {
4056 if (modeint == tbl[i].mode)
4058 if (strcmp(modestr, tbl[i].modename) == 0)
4065 static int bond_check_params(struct bond_params *params)
4067 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4068 int arp_all_targets_value;
4071 * Convert string parameters.
4074 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4075 if (bond_mode == -1) {
4076 pr_err("Error: Invalid bonding mode \"%s\"\n",
4077 mode == NULL ? "NULL" : mode);
4082 if (xmit_hash_policy) {
4083 if ((bond_mode != BOND_MODE_XOR) &&
4084 (bond_mode != BOND_MODE_8023AD)) {
4085 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4086 bond_mode_name(bond_mode));
4088 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4090 if (xmit_hashtype == -1) {
4091 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4092 xmit_hash_policy == NULL ? "NULL" :
4100 if (bond_mode != BOND_MODE_8023AD) {
4101 pr_info("lacp_rate param is irrelevant in mode %s\n",
4102 bond_mode_name(bond_mode));
4104 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4105 if (lacp_fast == -1) {
4106 pr_err("Error: Invalid lacp rate \"%s\"\n",
4107 lacp_rate == NULL ? "NULL" : lacp_rate);
4114 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4115 if (params->ad_select == -1) {
4116 pr_err("Error: Invalid ad_select \"%s\"\n",
4117 ad_select == NULL ? "NULL" : ad_select);
4121 if (bond_mode != BOND_MODE_8023AD) {
4122 pr_warning("ad_select param only affects 802.3ad mode\n");
4125 params->ad_select = BOND_AD_STABLE;
4128 if (max_bonds < 0) {
4129 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4130 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4131 max_bonds = BOND_DEFAULT_MAX_BONDS;
4135 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4136 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4137 miimon = BOND_LINK_MON_INTERV;
4141 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4146 if (downdelay < 0) {
4147 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4148 downdelay, INT_MAX);
4152 if ((use_carrier != 0) && (use_carrier != 1)) {
4153 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4158 if (num_peer_notif < 0 || num_peer_notif > 255) {
4159 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4164 /* reset values for 802.3ad */
4165 if (bond_mode == BOND_MODE_8023AD) {
4167 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");
4168 pr_warning("Forcing miimon to 100msec\n");
4173 if (tx_queues < 1 || tx_queues > 255) {
4174 pr_warning("Warning: tx_queues (%d) should be between "
4175 "1 and 255, resetting to %d\n",
4176 tx_queues, BOND_DEFAULT_TX_QUEUES);
4177 tx_queues = BOND_DEFAULT_TX_QUEUES;
4180 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4181 pr_warning("Warning: all_slaves_active module parameter (%d), "
4182 "not of valid value (0/1), so it was set to "
4183 "0\n", all_slaves_active);
4184 all_slaves_active = 0;
4187 if (resend_igmp < 0 || resend_igmp > 255) {
4188 pr_warning("Warning: resend_igmp (%d) should be between "
4189 "0 and 255, resetting to %d\n",
4190 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4191 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4194 /* reset values for TLB/ALB */
4195 if ((bond_mode == BOND_MODE_TLB) ||
4196 (bond_mode == BOND_MODE_ALB)) {
4198 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");
4199 pr_warning("Forcing miimon to 100msec\n");
4204 if (bond_mode == BOND_MODE_ALB) {
4205 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",
4210 if (updelay || downdelay) {
4211 /* just warn the user the up/down delay will have
4212 * no effect since miimon is zero...
4214 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",
4215 updelay, downdelay);
4218 /* don't allow arp monitoring */
4220 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4221 miimon, arp_interval);
4225 if ((updelay % miimon) != 0) {
4226 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4228 (updelay / miimon) * miimon);
4233 if ((downdelay % miimon) != 0) {
4234 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4236 (downdelay / miimon) * miimon);
4239 downdelay /= miimon;
4242 if (arp_interval < 0) {
4243 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4244 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4245 arp_interval = BOND_LINK_ARP_INTERV;
4248 for (arp_ip_count = 0, i = 0;
4249 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4250 /* not complete check, but should be good enough to
4252 __be32 ip = in_aton(arp_ip_target[i]);
4253 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4254 ip == htonl(INADDR_BROADCAST)) {
4255 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4259 if (bond_get_targets_ip(arp_target, ip) == -1)
4260 arp_target[arp_ip_count++] = ip;
4262 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4267 if (arp_interval && !arp_ip_count) {
4268 /* don't allow arping if no arp_ip_target given... */
4269 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4275 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4276 pr_err("arp_validate only supported in active-backup mode\n");
4279 if (!arp_interval) {
4280 pr_err("arp_validate requires arp_interval\n");
4284 arp_validate_value = bond_parse_parm(arp_validate,
4286 if (arp_validate_value == -1) {
4287 pr_err("Error: invalid arp_validate \"%s\"\n",
4288 arp_validate == NULL ? "NULL" : arp_validate);
4292 arp_validate_value = 0;
4294 arp_all_targets_value = 0;
4295 if (arp_all_targets) {
4296 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4297 arp_all_targets_tbl);
4299 if (arp_all_targets_value == -1) {
4300 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4302 arp_all_targets_value = 0;
4307 pr_info("MII link monitoring set to %d ms\n", miimon);
4308 } else if (arp_interval) {
4309 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4311 arp_validate_tbl[arp_validate_value].modename,
4314 for (i = 0; i < arp_ip_count; i++)
4315 pr_info(" %s", arp_ip_target[i]);
4319 } else if (max_bonds) {
4320 /* miimon and arp_interval not set, we need one so things
4321 * work as expected, see bonding.txt for details
4323 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");
4326 if (primary && !USES_PRIMARY(bond_mode)) {
4327 /* currently, using a primary only makes sense
4328 * in active backup, TLB or ALB modes
4330 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4331 primary, bond_mode_name(bond_mode));
4335 if (primary && primary_reselect) {
4336 primary_reselect_value = bond_parse_parm(primary_reselect,
4338 if (primary_reselect_value == -1) {
4339 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4341 NULL ? "NULL" : primary_reselect);
4345 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4348 if (fail_over_mac) {
4349 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4351 if (fail_over_mac_value == -1) {
4352 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4353 arp_validate == NULL ? "NULL" : arp_validate);
4357 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4358 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4360 fail_over_mac_value = BOND_FOM_NONE;
4363 /* fill params struct with the proper values */
4364 params->mode = bond_mode;
4365 params->xmit_policy = xmit_hashtype;
4366 params->miimon = miimon;
4367 params->num_peer_notif = num_peer_notif;
4368 params->arp_interval = arp_interval;
4369 params->arp_validate = arp_validate_value;
4370 params->arp_all_targets = arp_all_targets_value;
4371 params->updelay = updelay;
4372 params->downdelay = downdelay;
4373 params->use_carrier = use_carrier;
4374 params->lacp_fast = lacp_fast;
4375 params->primary[0] = 0;
4376 params->primary_reselect = primary_reselect_value;
4377 params->fail_over_mac = fail_over_mac_value;
4378 params->tx_queues = tx_queues;
4379 params->all_slaves_active = all_slaves_active;
4380 params->resend_igmp = resend_igmp;
4381 params->min_links = min_links;
4382 params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4385 strncpy(params->primary, primary, IFNAMSIZ);
4386 params->primary[IFNAMSIZ - 1] = 0;
4389 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4394 static struct lock_class_key bonding_netdev_xmit_lock_key;
4395 static struct lock_class_key bonding_netdev_addr_lock_key;
4396 static struct lock_class_key bonding_tx_busylock_key;
4398 static void bond_set_lockdep_class_one(struct net_device *dev,
4399 struct netdev_queue *txq,
4402 lockdep_set_class(&txq->_xmit_lock,
4403 &bonding_netdev_xmit_lock_key);
4406 static void bond_set_lockdep_class(struct net_device *dev)
4408 lockdep_set_class(&dev->addr_list_lock,
4409 &bonding_netdev_addr_lock_key);
4410 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4411 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4415 * Called from registration process
4417 static int bond_init(struct net_device *bond_dev)
4419 struct bonding *bond = netdev_priv(bond_dev);
4420 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4421 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4423 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4426 * Initialize locks that may be required during
4427 * en/deslave operations. All of the bond_open work
4428 * (of which this is part) should really be moved to
4429 * a phase prior to dev_open
4431 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4432 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4434 bond->wq = create_singlethread_workqueue(bond_dev->name);
4438 bond_set_lockdep_class(bond_dev);
4440 list_add_tail(&bond->bond_list, &bn->dev_list);
4442 bond_prepare_sysfs_group(bond);
4444 bond_debug_register(bond);
4446 /* Ensure valid dev_addr */
4447 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4448 bond_dev->addr_assign_type == NET_ADDR_PERM)
4449 eth_hw_addr_random(bond_dev);
4454 unsigned int bond_get_num_tx_queues(void)
4459 /* Create a new bond based on the specified name and bonding parameters.
4460 * If name is NULL, obtain a suitable "bond%d" name for us.
4461 * Caller must NOT hold rtnl_lock; we need to release it here before we
4462 * set up our sysfs entries.
4464 int bond_create(struct net *net, const char *name)
4466 struct net_device *bond_dev;
4471 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4472 name ? name : "bond%d",
4473 bond_setup, tx_queues);
4475 pr_err("%s: eek! can't alloc netdev!\n", name);
4480 dev_net_set(bond_dev, net);
4481 bond_dev->rtnl_link_ops = &bond_link_ops;
4483 res = register_netdevice(bond_dev);
4485 netif_carrier_off(bond_dev);
4489 bond_destructor(bond_dev);
4493 static int __net_init bond_net_init(struct net *net)
4495 struct bond_net *bn = net_generic(net, bond_net_id);
4498 INIT_LIST_HEAD(&bn->dev_list);
4500 bond_create_proc_dir(bn);
4501 bond_create_sysfs(bn);
4506 static void __net_exit bond_net_exit(struct net *net)
4508 struct bond_net *bn = net_generic(net, bond_net_id);
4509 struct bonding *bond, *tmp_bond;
4512 bond_destroy_sysfs(bn);
4513 bond_destroy_proc_dir(bn);
4515 /* Kill off any bonds created after unregistering bond rtnl ops */
4517 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4518 unregister_netdevice_queue(bond->dev, &list);
4519 unregister_netdevice_many(&list);
4523 static struct pernet_operations bond_net_ops = {
4524 .init = bond_net_init,
4525 .exit = bond_net_exit,
4527 .size = sizeof(struct bond_net),
4530 static int __init bonding_init(void)
4535 pr_info("%s", bond_version);
4537 res = bond_check_params(&bonding_defaults);
4541 res = register_pernet_subsys(&bond_net_ops);
4545 res = bond_netlink_init();
4549 bond_create_debugfs();
4551 for (i = 0; i < max_bonds; i++) {
4552 res = bond_create(&init_net, NULL);
4557 register_netdevice_notifier(&bond_netdev_notifier);
4561 bond_netlink_fini();
4563 unregister_pernet_subsys(&bond_net_ops);
4568 static void __exit bonding_exit(void)
4570 unregister_netdevice_notifier(&bond_netdev_notifier);
4572 bond_destroy_debugfs();
4574 bond_netlink_fini();
4575 unregister_pernet_subsys(&bond_net_ops);
4577 #ifdef CONFIG_NET_POLL_CONTROLLER
4579 * Make sure we don't have an imbalance on our netpoll blocking
4581 WARN_ON(atomic_read(&netpoll_block_tx));
4585 module_init(bonding_init);
4586 module_exit(bonding_exit);
4587 MODULE_LICENSE("GPL");
4588 MODULE_VERSION(DRV_VERSION);
4589 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4590 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");