2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
85 /*---------------------------- Module parameters ----------------------------*/
87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 #define BOND_LINK_MON_INTERV 0
89 #define BOND_LINK_ARP_INTERV 0
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
94 static int miimon = BOND_LINK_MON_INTERV;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval = BOND_LINK_ARP_INTERV;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
114 module_param(max_bonds, int, 0);
115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
116 module_param(tx_queues, int, 0);
117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
118 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
119 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
120 "failover event (alias of num_unsol_na)");
121 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
122 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
123 "failover event (alias of num_grat_arp)");
124 module_param(miimon, int, 0);
125 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
126 module_param(updelay, int, 0);
127 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
128 module_param(downdelay, int, 0);
129 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
131 module_param(use_carrier, int, 0);
132 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
133 "0 for off, 1 for on (default)");
134 module_param(mode, charp, 0);
135 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
136 "1 for active-backup, 2 for balance-xor, "
137 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
138 "6 for balance-alb");
139 module_param(primary, charp, 0);
140 MODULE_PARM_DESC(primary, "Primary network device to use");
141 module_param(primary_reselect, charp, 0);
142 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
144 "0 for always (default), "
145 "1 for only if speed of primary is "
147 "2 for only on active slave "
149 module_param(lacp_rate, charp, 0);
150 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
151 "0 for slow, 1 for fast");
152 module_param(ad_select, charp, 0);
153 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
154 "0 for stable (default), 1 for bandwidth, "
156 module_param(min_links, int, 0);
157 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
159 module_param(xmit_hash_policy, charp, 0);
160 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
161 "0 for layer 2 (default), 1 for layer 3+4, "
163 module_param(arp_interval, int, 0);
164 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
165 module_param_array(arp_ip_target, charp, NULL, 0);
166 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
167 module_param(arp_validate, charp, 0);
168 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
169 "0 for none (default), 1 for active, "
170 "2 for backup, 3 for all");
171 module_param(arp_all_targets, charp, 0);
172 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
173 module_param(fail_over_mac, charp, 0);
174 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
175 "the same MAC; 0 for none (default), "
176 "1 for active, 2 for follow");
177 module_param(all_slaves_active, int, 0);
178 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
179 "by setting active flag for all slaves; "
180 "0 for never (default), 1 for always.");
181 module_param(resend_igmp, int, 0);
182 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
185 /*----------------------------- Global variables ----------------------------*/
187 #ifdef CONFIG_NET_POLL_CONTROLLER
188 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
191 int bond_net_id __read_mostly;
193 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
194 static int arp_ip_count;
195 static int bond_mode = BOND_MODE_ROUNDROBIN;
196 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
197 static int lacp_fast;
199 const struct bond_parm_tbl bond_lacp_tbl[] = {
200 { "slow", AD_LACP_SLOW},
201 { "fast", AD_LACP_FAST},
205 const struct bond_parm_tbl bond_mode_tbl[] = {
206 { "balance-rr", BOND_MODE_ROUNDROBIN},
207 { "active-backup", BOND_MODE_ACTIVEBACKUP},
208 { "balance-xor", BOND_MODE_XOR},
209 { "broadcast", BOND_MODE_BROADCAST},
210 { "802.3ad", BOND_MODE_8023AD},
211 { "balance-tlb", BOND_MODE_TLB},
212 { "balance-alb", BOND_MODE_ALB},
216 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
217 { "layer2", BOND_XMIT_POLICY_LAYER2},
218 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
219 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
223 const struct bond_parm_tbl arp_all_targets_tbl[] = {
224 { "any", BOND_ARP_TARGETS_ANY},
225 { "all", BOND_ARP_TARGETS_ALL},
229 const struct bond_parm_tbl arp_validate_tbl[] = {
230 { "none", BOND_ARP_VALIDATE_NONE},
231 { "active", BOND_ARP_VALIDATE_ACTIVE},
232 { "backup", BOND_ARP_VALIDATE_BACKUP},
233 { "all", BOND_ARP_VALIDATE_ALL},
237 const struct bond_parm_tbl fail_over_mac_tbl[] = {
238 { "none", BOND_FOM_NONE},
239 { "active", BOND_FOM_ACTIVE},
240 { "follow", BOND_FOM_FOLLOW},
244 const struct bond_parm_tbl pri_reselect_tbl[] = {
245 { "always", BOND_PRI_RESELECT_ALWAYS},
246 { "better", BOND_PRI_RESELECT_BETTER},
247 { "failure", BOND_PRI_RESELECT_FAILURE},
251 struct bond_parm_tbl ad_select_tbl[] = {
252 { "stable", BOND_AD_STABLE},
253 { "bandwidth", BOND_AD_BANDWIDTH},
254 { "count", BOND_AD_COUNT},
258 /*-------------------------- Forward declarations ---------------------------*/
260 static int bond_init(struct net_device *bond_dev);
261 static void bond_uninit(struct net_device *bond_dev);
263 /*---------------------------- General routines -----------------------------*/
265 const char *bond_mode_name(int mode)
267 static const char *names[] = {
268 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
269 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
270 [BOND_MODE_XOR] = "load balancing (xor)",
271 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
272 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
273 [BOND_MODE_TLB] = "transmit load balancing",
274 [BOND_MODE_ALB] = "adaptive load balancing",
277 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
283 /*---------------------------------- VLAN -----------------------------------*/
286 * bond_dev_queue_xmit - Prepare skb for xmit.
288 * @bond: bond device that got this skb for tx.
289 * @skb: hw accel VLAN tagged skb to transmit
290 * @slave_dev: slave that is supposed to xmit this skbuff
292 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
293 struct net_device *slave_dev)
295 skb->dev = slave_dev;
297 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
298 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
299 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
301 if (unlikely(netpoll_tx_running(bond->dev)))
302 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
310 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
311 * We don't protect the slave list iteration with a lock because:
312 * a. This operation is performed in IOCTL context,
313 * b. The operation is protected by the RTNL semaphore in the 8021q code,
314 * c. Holding a lock with BH disabled while directly calling a base driver
315 * entry point is generally a BAD idea.
317 * The design of synchronization/protection for this operation in the 8021q
318 * module is good for one or more VLAN devices over a single physical device
319 * and cannot be extended for a teaming solution like bonding, so there is a
320 * potential race condition here where a net device from the vlan group might
321 * be referenced (either by a base driver or the 8021q code) while it is being
322 * removed from the system. However, it turns out we're not making matters
323 * worse, and if it works for regular VLAN usage it will work here too.
327 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
328 * @bond_dev: bonding net device that got called
329 * @vid: vlan id being added
331 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
332 __be16 proto, u16 vid)
334 struct bonding *bond = netdev_priv(bond_dev);
338 bond_for_each_slave(bond, slave) {
339 res = vlan_vid_add(slave->dev, proto, vid);
347 /* unwind from the slave that failed */
348 bond_for_each_slave_continue_reverse(bond, slave)
349 vlan_vid_del(slave->dev, proto, vid);
355 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
356 * @bond_dev: bonding net device that got called
357 * @vid: vlan id being removed
359 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
360 __be16 proto, u16 vid)
362 struct bonding *bond = netdev_priv(bond_dev);
365 bond_for_each_slave(bond, slave)
366 vlan_vid_del(slave->dev, proto, vid);
368 if (bond_is_lb(bond))
369 bond_alb_clear_vlan(bond, vid);
374 /*------------------------------- Link status -------------------------------*/
377 * Set the carrier state for the master according to the state of its
378 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
379 * do special 802.3ad magic.
381 * Returns zero if carrier state does not change, nonzero if it does.
383 static int bond_set_carrier(struct bonding *bond)
387 if (list_empty(&bond->slave_list))
390 if (bond->params.mode == BOND_MODE_8023AD)
391 return bond_3ad_set_carrier(bond);
393 bond_for_each_slave(bond, slave) {
394 if (slave->link == BOND_LINK_UP) {
395 if (!netif_carrier_ok(bond->dev)) {
396 netif_carrier_on(bond->dev);
404 if (netif_carrier_ok(bond->dev)) {
405 netif_carrier_off(bond->dev);
412 * Get link speed and duplex from the slave's base driver
413 * using ethtool. If for some reason the call fails or the
414 * values are invalid, set speed and duplex to -1,
417 static void bond_update_speed_duplex(struct slave *slave)
419 struct net_device *slave_dev = slave->dev;
420 struct ethtool_cmd ecmd;
424 slave->speed = SPEED_UNKNOWN;
425 slave->duplex = DUPLEX_UNKNOWN;
427 res = __ethtool_get_settings(slave_dev, &ecmd);
431 slave_speed = ethtool_cmd_speed(&ecmd);
432 if (slave_speed == 0 || slave_speed == ((__u32) -1))
435 switch (ecmd.duplex) {
443 slave->speed = slave_speed;
444 slave->duplex = ecmd.duplex;
450 * if <dev> supports MII link status reporting, check its link status.
452 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
453 * depending upon the setting of the use_carrier parameter.
455 * Return either BMSR_LSTATUS, meaning that the link is up (or we
456 * can't tell and just pretend it is), or 0, meaning that the link is
459 * If reporting is non-zero, instead of faking link up, return -1 if
460 * both ETHTOOL and MII ioctls fail (meaning the device does not
461 * support them). If use_carrier is set, return whatever it says.
462 * It'd be nice if there was a good way to tell if a driver supports
463 * netif_carrier, but there really isn't.
465 static int bond_check_dev_link(struct bonding *bond,
466 struct net_device *slave_dev, int reporting)
468 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
469 int (*ioctl)(struct net_device *, struct ifreq *, int);
471 struct mii_ioctl_data *mii;
473 if (!reporting && !netif_running(slave_dev))
476 if (bond->params.use_carrier)
477 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
479 /* Try to get link status using Ethtool first. */
480 if (slave_dev->ethtool_ops->get_link)
481 return slave_dev->ethtool_ops->get_link(slave_dev) ?
484 /* Ethtool can't be used, fallback to MII ioctls. */
485 ioctl = slave_ops->ndo_do_ioctl;
487 /* TODO: set pointer to correct ioctl on a per team member */
488 /* bases to make this more efficient. that is, once */
489 /* we determine the correct ioctl, we will always */
490 /* call it and not the others for that team */
494 * We cannot assume that SIOCGMIIPHY will also read a
495 * register; not all network drivers (e.g., e100)
499 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
500 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
502 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
503 mii->reg_num = MII_BMSR;
504 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
505 return mii->val_out & BMSR_LSTATUS;
510 * If reporting, report that either there's no dev->do_ioctl,
511 * or both SIOCGMIIREG and get_link failed (meaning that we
512 * cannot report link status). If not reporting, pretend
515 return reporting ? -1 : BMSR_LSTATUS;
518 /*----------------------------- Multicast list ------------------------------*/
521 * Push the promiscuity flag down to appropriate slaves
523 static int bond_set_promiscuity(struct bonding *bond, int inc)
526 if (USES_PRIMARY(bond->params.mode)) {
527 /* write lock already acquired */
528 if (bond->curr_active_slave) {
529 err = dev_set_promiscuity(bond->curr_active_slave->dev,
535 bond_for_each_slave(bond, slave) {
536 err = dev_set_promiscuity(slave->dev, inc);
545 * Push the allmulti flag down to all slaves
547 static int bond_set_allmulti(struct bonding *bond, int inc)
550 if (USES_PRIMARY(bond->params.mode)) {
551 /* write lock already acquired */
552 if (bond->curr_active_slave) {
553 err = dev_set_allmulti(bond->curr_active_slave->dev,
559 bond_for_each_slave(bond, slave) {
560 err = dev_set_allmulti(slave->dev, inc);
569 * Retrieve the list of registered multicast addresses for the bonding
570 * device and retransmit an IGMP JOIN request to the current active
573 static void bond_resend_igmp_join_requests(struct bonding *bond)
575 if (!rtnl_trylock()) {
576 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
579 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
582 /* We use curr_slave_lock to protect against concurrent access to
583 * igmp_retrans from multiple running instances of this function and
584 * bond_change_active_slave
586 write_lock_bh(&bond->curr_slave_lock);
587 if (bond->igmp_retrans > 1) {
588 bond->igmp_retrans--;
589 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
591 write_unlock_bh(&bond->curr_slave_lock);
594 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
596 struct bonding *bond = container_of(work, struct bonding,
599 bond_resend_igmp_join_requests(bond);
602 /* Flush bond's hardware addresses from slave
604 static void bond_hw_addr_flush(struct net_device *bond_dev,
605 struct net_device *slave_dev)
607 struct bonding *bond = netdev_priv(bond_dev);
609 dev_uc_unsync(slave_dev, bond_dev);
610 dev_mc_unsync(slave_dev, bond_dev);
612 if (bond->params.mode == BOND_MODE_8023AD) {
613 /* del lacpdu mc addr from mc list */
614 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
616 dev_mc_del(slave_dev, lacpdu_multicast);
620 /*--------------------------- Active slave change ---------------------------*/
622 /* Update the hardware address list and promisc/allmulti for the new and
623 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
624 * slaves up date at all times; only the USES_PRIMARY modes need to call
625 * this function to swap these settings during a failover.
627 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
628 struct slave *old_active)
633 if (bond->dev->flags & IFF_PROMISC)
634 dev_set_promiscuity(old_active->dev, -1);
636 if (bond->dev->flags & IFF_ALLMULTI)
637 dev_set_allmulti(old_active->dev, -1);
639 bond_hw_addr_flush(bond->dev, old_active->dev);
643 /* FIXME: Signal errors upstream. */
644 if (bond->dev->flags & IFF_PROMISC)
645 dev_set_promiscuity(new_active->dev, 1);
647 if (bond->dev->flags & IFF_ALLMULTI)
648 dev_set_allmulti(new_active->dev, 1);
650 netif_addr_lock_bh(bond->dev);
651 dev_uc_sync(new_active->dev, bond->dev);
652 dev_mc_sync(new_active->dev, bond->dev);
653 netif_addr_unlock_bh(bond->dev);
658 * bond_set_dev_addr - clone slave's address to bond
659 * @bond_dev: bond net device
660 * @slave_dev: slave net device
662 * Should be called with RTNL held.
664 static void bond_set_dev_addr(struct net_device *bond_dev,
665 struct net_device *slave_dev)
667 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
668 bond_dev, slave_dev, slave_dev->addr_len);
669 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
670 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
671 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
675 * bond_do_fail_over_mac
677 * Perform special MAC address swapping for fail_over_mac settings
679 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
681 static void bond_do_fail_over_mac(struct bonding *bond,
682 struct slave *new_active,
683 struct slave *old_active)
684 __releases(&bond->curr_slave_lock)
685 __releases(&bond->lock)
686 __acquires(&bond->lock)
687 __acquires(&bond->curr_slave_lock)
689 u8 tmp_mac[ETH_ALEN];
690 struct sockaddr saddr;
693 switch (bond->params.fail_over_mac) {
694 case BOND_FOM_ACTIVE:
696 write_unlock_bh(&bond->curr_slave_lock);
697 read_unlock(&bond->lock);
698 bond_set_dev_addr(bond->dev, new_active->dev);
699 read_lock(&bond->lock);
700 write_lock_bh(&bond->curr_slave_lock);
703 case BOND_FOM_FOLLOW:
705 * if new_active && old_active, swap them
706 * if just old_active, do nothing (going to no active slave)
707 * if just new_active, set new_active to bond's MAC
712 write_unlock_bh(&bond->curr_slave_lock);
713 read_unlock(&bond->lock);
716 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
717 memcpy(saddr.sa_data, old_active->dev->dev_addr,
719 saddr.sa_family = new_active->dev->type;
721 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
722 saddr.sa_family = bond->dev->type;
725 rv = dev_set_mac_address(new_active->dev, &saddr);
727 pr_err("%s: Error %d setting MAC of slave %s\n",
728 bond->dev->name, -rv, new_active->dev->name);
735 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
736 saddr.sa_family = old_active->dev->type;
738 rv = dev_set_mac_address(old_active->dev, &saddr);
740 pr_err("%s: Error %d setting MAC of slave %s\n",
741 bond->dev->name, -rv, new_active->dev->name);
743 read_lock(&bond->lock);
744 write_lock_bh(&bond->curr_slave_lock);
747 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
748 bond->dev->name, bond->params.fail_over_mac);
754 static bool bond_should_change_active(struct bonding *bond)
756 struct slave *prim = bond->primary_slave;
757 struct slave *curr = bond->curr_active_slave;
759 if (!prim || !curr || curr->link != BOND_LINK_UP)
761 if (bond->force_primary) {
762 bond->force_primary = false;
765 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
766 (prim->speed < curr->speed ||
767 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
769 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
775 * find_best_interface - select the best available slave to be the active one
776 * @bond: our bonding struct
778 * Warning: Caller must hold curr_slave_lock for writing.
780 static struct slave *bond_find_best_slave(struct bonding *bond)
782 struct slave *new_active, *old_active;
783 struct slave *bestslave = NULL;
784 int mintime = bond->params.updelay;
787 new_active = bond->curr_active_slave;
789 if (!new_active) { /* there were no active slaves left */
790 new_active = bond_first_slave(bond);
792 return NULL; /* still no slave, return NULL */
795 if ((bond->primary_slave) &&
796 bond->primary_slave->link == BOND_LINK_UP &&
797 bond_should_change_active(bond)) {
798 new_active = bond->primary_slave;
801 /* remember where to stop iterating over the slaves */
802 old_active = new_active;
804 bond_for_each_slave_from(bond, new_active, i, old_active) {
805 if (new_active->link == BOND_LINK_UP) {
807 } else if (new_active->link == BOND_LINK_BACK &&
808 IS_UP(new_active->dev)) {
809 /* link up, but waiting for stabilization */
810 if (new_active->delay < mintime) {
811 mintime = new_active->delay;
812 bestslave = new_active;
820 static bool bond_should_notify_peers(struct bonding *bond)
822 struct slave *slave = bond->curr_active_slave;
824 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
825 bond->dev->name, slave ? slave->dev->name : "NULL");
827 if (!slave || !bond->send_peer_notif ||
828 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
831 bond->send_peer_notif--;
836 * change_active_interface - change the active slave into the specified one
837 * @bond: our bonding struct
838 * @new: the new slave to make the active one
840 * Set the new slave to the bond's settings and unset them on the old
842 * Setting include flags, mc-list, promiscuity, allmulti, etc.
844 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
845 * because it is apparently the best available slave we have, even though its
846 * updelay hasn't timed out yet.
848 * If new_active is not NULL, caller must hold bond->lock for read and
849 * curr_slave_lock for write_bh.
851 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
853 struct slave *old_active = bond->curr_active_slave;
855 if (old_active == new_active)
859 new_active->jiffies = jiffies;
861 if (new_active->link == BOND_LINK_BACK) {
862 if (USES_PRIMARY(bond->params.mode)) {
863 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
864 bond->dev->name, new_active->dev->name,
865 (bond->params.updelay - new_active->delay) * bond->params.miimon);
868 new_active->delay = 0;
869 new_active->link = BOND_LINK_UP;
871 if (bond->params.mode == BOND_MODE_8023AD)
872 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
874 if (bond_is_lb(bond))
875 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
877 if (USES_PRIMARY(bond->params.mode)) {
878 pr_info("%s: making interface %s the new active one.\n",
879 bond->dev->name, new_active->dev->name);
884 if (USES_PRIMARY(bond->params.mode))
885 bond_hw_addr_swap(bond, new_active, old_active);
887 if (bond_is_lb(bond)) {
888 bond_alb_handle_active_change(bond, new_active);
890 bond_set_slave_inactive_flags(old_active);
892 bond_set_slave_active_flags(new_active);
894 rcu_assign_pointer(bond->curr_active_slave, new_active);
897 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
899 bond_set_slave_inactive_flags(old_active);
902 bool should_notify_peers = false;
904 bond_set_slave_active_flags(new_active);
906 if (bond->params.fail_over_mac)
907 bond_do_fail_over_mac(bond, new_active,
910 if (netif_running(bond->dev)) {
911 bond->send_peer_notif =
912 bond->params.num_peer_notif;
913 should_notify_peers =
914 bond_should_notify_peers(bond);
917 write_unlock_bh(&bond->curr_slave_lock);
918 read_unlock(&bond->lock);
920 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
921 if (should_notify_peers)
922 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
925 read_lock(&bond->lock);
926 write_lock_bh(&bond->curr_slave_lock);
930 /* resend IGMP joins since active slave has changed or
931 * all were sent on curr_active_slave.
932 * resend only if bond is brought up with the affected
933 * bonding modes and the retransmission is enabled */
934 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
935 ((USES_PRIMARY(bond->params.mode) && new_active) ||
936 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
937 bond->igmp_retrans = bond->params.resend_igmp;
938 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
943 * bond_select_active_slave - select a new active slave, if needed
944 * @bond: our bonding struct
946 * This functions should be called when one of the following occurs:
947 * - The old curr_active_slave has been released or lost its link.
948 * - The primary_slave has got its link back.
949 * - A slave has got its link back and there's no old curr_active_slave.
951 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
953 void bond_select_active_slave(struct bonding *bond)
955 struct slave *best_slave;
958 best_slave = bond_find_best_slave(bond);
959 if (best_slave != bond->curr_active_slave) {
960 bond_change_active_slave(bond, best_slave);
961 rv = bond_set_carrier(bond);
965 if (netif_carrier_ok(bond->dev)) {
966 pr_info("%s: first active interface up!\n",
969 pr_info("%s: now running without any active interface !\n",
975 /*--------------------------- slave list handling ---------------------------*/
978 * This function attaches the slave to the end of list.
980 * bond->lock held for writing by caller.
982 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
984 list_add_tail_rcu(&new_slave->list, &bond->slave_list);
989 * This function detaches the slave from the list.
990 * WARNING: no check is made to verify if the slave effectively
992 * Nothing is freed on return, structures are just unchained.
993 * If any slave pointer in bond was pointing to <slave>,
994 * it should be changed by the calling function.
996 * bond->lock held for writing by caller.
998 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1000 list_del_rcu(&slave->list);
1004 #ifdef CONFIG_NET_POLL_CONTROLLER
1005 static inline int slave_enable_netpoll(struct slave *slave)
1010 np = kzalloc(sizeof(*np), GFP_ATOMIC);
1015 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1024 static inline void slave_disable_netpoll(struct slave *slave)
1026 struct netpoll *np = slave->np;
1032 __netpoll_free_async(np);
1034 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1036 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1038 if (!slave_dev->netdev_ops->ndo_poll_controller)
1043 static void bond_poll_controller(struct net_device *bond_dev)
1047 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1049 struct bonding *bond = netdev_priv(bond_dev);
1050 struct slave *slave;
1052 bond_for_each_slave(bond, slave)
1053 if (IS_UP(slave->dev))
1054 slave_disable_netpoll(slave);
1057 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1059 struct bonding *bond = netdev_priv(dev);
1060 struct slave *slave;
1063 bond_for_each_slave(bond, slave) {
1064 err = slave_enable_netpoll(slave);
1066 bond_netpoll_cleanup(dev);
1073 static inline int slave_enable_netpoll(struct slave *slave)
1077 static inline void slave_disable_netpoll(struct slave *slave)
1080 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1085 /*---------------------------------- IOCTL ----------------------------------*/
1087 static netdev_features_t bond_fix_features(struct net_device *dev,
1088 netdev_features_t features)
1090 struct slave *slave;
1091 struct bonding *bond = netdev_priv(dev);
1092 netdev_features_t mask;
1094 read_lock(&bond->lock);
1096 if (list_empty(&bond->slave_list)) {
1097 /* Disable adding VLANs to empty bond. But why? --mq */
1098 features |= NETIF_F_VLAN_CHALLENGED;
1103 features &= ~NETIF_F_ONE_FOR_ALL;
1104 features |= NETIF_F_ALL_FOR_ALL;
1106 bond_for_each_slave(bond, slave) {
1107 features = netdev_increment_features(features,
1108 slave->dev->features,
1111 features = netdev_add_tso_features(features, mask);
1114 read_unlock(&bond->lock);
1118 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1119 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1120 NETIF_F_HIGHDMA | NETIF_F_LRO)
1122 static void bond_compute_features(struct bonding *bond)
1124 struct slave *slave;
1125 struct net_device *bond_dev = bond->dev;
1126 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1127 unsigned short max_hard_header_len = ETH_HLEN;
1128 unsigned int gso_max_size = GSO_MAX_SIZE;
1129 u16 gso_max_segs = GSO_MAX_SEGS;
1130 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1132 read_lock(&bond->lock);
1134 if (list_empty(&bond->slave_list))
1137 bond_for_each_slave(bond, slave) {
1138 vlan_features = netdev_increment_features(vlan_features,
1139 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1141 dst_release_flag &= slave->dev->priv_flags;
1142 if (slave->dev->hard_header_len > max_hard_header_len)
1143 max_hard_header_len = slave->dev->hard_header_len;
1145 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1146 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1150 bond_dev->vlan_features = vlan_features;
1151 bond_dev->hard_header_len = max_hard_header_len;
1152 bond_dev->gso_max_segs = gso_max_segs;
1153 netif_set_gso_max_size(bond_dev, gso_max_size);
1155 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1156 bond_dev->priv_flags = flags | dst_release_flag;
1158 read_unlock(&bond->lock);
1160 netdev_change_features(bond_dev);
1163 static void bond_setup_by_slave(struct net_device *bond_dev,
1164 struct net_device *slave_dev)
1166 bond_dev->header_ops = slave_dev->header_ops;
1168 bond_dev->type = slave_dev->type;
1169 bond_dev->hard_header_len = slave_dev->hard_header_len;
1170 bond_dev->addr_len = slave_dev->addr_len;
1172 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1173 slave_dev->addr_len);
1176 /* On bonding slaves other than the currently active slave, suppress
1177 * duplicates except for alb non-mcast/bcast.
1179 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1180 struct slave *slave,
1181 struct bonding *bond)
1183 if (bond_is_slave_inactive(slave)) {
1184 if (bond->params.mode == BOND_MODE_ALB &&
1185 skb->pkt_type != PACKET_BROADCAST &&
1186 skb->pkt_type != PACKET_MULTICAST)
1193 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1195 struct sk_buff *skb = *pskb;
1196 struct slave *slave;
1197 struct bonding *bond;
1198 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1200 int ret = RX_HANDLER_ANOTHER;
1202 skb = skb_share_check(skb, GFP_ATOMIC);
1204 return RX_HANDLER_CONSUMED;
1208 slave = bond_slave_get_rcu(skb->dev);
1211 if (bond->params.arp_interval)
1212 slave->dev->last_rx = jiffies;
1214 recv_probe = ACCESS_ONCE(bond->recv_probe);
1216 ret = recv_probe(skb, bond, slave);
1217 if (ret == RX_HANDLER_CONSUMED) {
1223 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1224 return RX_HANDLER_EXACT;
1227 skb->dev = bond->dev;
1229 if (bond->params.mode == BOND_MODE_ALB &&
1230 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1231 skb->pkt_type == PACKET_HOST) {
1233 if (unlikely(skb_cow_head(skb,
1234 skb->data - skb_mac_header(skb)))) {
1236 return RX_HANDLER_CONSUMED;
1238 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1244 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1245 struct net_device *slave_dev)
1249 err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1252 slave_dev->flags |= IFF_SLAVE;
1253 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1257 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1258 struct net_device *slave_dev)
1260 netdev_upper_dev_unlink(slave_dev, bond_dev);
1261 slave_dev->flags &= ~IFF_SLAVE;
1262 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1265 /* enslave device <slave> to bond device <master> */
1266 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1268 struct bonding *bond = netdev_priv(bond_dev);
1269 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1270 struct slave *new_slave = NULL;
1271 struct sockaddr addr;
1275 if (!bond->params.use_carrier &&
1276 slave_dev->ethtool_ops->get_link == NULL &&
1277 slave_ops->ndo_do_ioctl == NULL) {
1278 pr_warning("%s: Warning: no link monitoring support for %s\n",
1279 bond_dev->name, slave_dev->name);
1282 /* already enslaved */
1283 if (slave_dev->flags & IFF_SLAVE) {
1284 pr_debug("Error, Device was already enslaved\n");
1288 /* vlan challenged mutual exclusion */
1289 /* no need to lock since we're protected by rtnl_lock */
1290 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1291 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1292 if (vlan_uses_dev(bond_dev)) {
1293 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1294 bond_dev->name, slave_dev->name, bond_dev->name);
1297 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1298 bond_dev->name, slave_dev->name,
1299 slave_dev->name, bond_dev->name);
1302 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1306 * Old ifenslave binaries are no longer supported. These can
1307 * be identified with moderate accuracy by the state of the slave:
1308 * the current ifenslave will set the interface down prior to
1309 * enslaving it; the old ifenslave will not.
1311 if ((slave_dev->flags & IFF_UP)) {
1312 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1315 goto err_undo_flags;
1318 /* set bonding device ether type by slave - bonding netdevices are
1319 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1320 * there is a need to override some of the type dependent attribs/funcs.
1322 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1323 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1325 if (list_empty(&bond->slave_list)) {
1326 if (bond_dev->type != slave_dev->type) {
1327 pr_debug("%s: change device type from %d to %d\n",
1329 bond_dev->type, slave_dev->type);
1331 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1333 res = notifier_to_errno(res);
1335 pr_err("%s: refused to change device type\n",
1338 goto err_undo_flags;
1341 /* Flush unicast and multicast addresses */
1342 dev_uc_flush(bond_dev);
1343 dev_mc_flush(bond_dev);
1345 if (slave_dev->type != ARPHRD_ETHER)
1346 bond_setup_by_slave(bond_dev, slave_dev);
1348 ether_setup(bond_dev);
1349 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1352 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1355 } else if (bond_dev->type != slave_dev->type) {
1356 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1358 slave_dev->type, bond_dev->type);
1360 goto err_undo_flags;
1363 if (slave_ops->ndo_set_mac_address == NULL) {
1364 if (list_empty(&bond->slave_list)) {
1365 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1367 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1368 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1369 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",
1372 goto err_undo_flags;
1376 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1378 /* If this is the first slave, then we need to set the master's hardware
1379 * address to be the same as the slave's. */
1380 if (list_empty(&bond->slave_list) &&
1381 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1382 bond_set_dev_addr(bond->dev, slave_dev);
1384 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1387 goto err_undo_flags;
1389 INIT_LIST_HEAD(&new_slave->list);
1391 * Set the new_slave's queue_id to be zero. Queue ID mapping
1392 * is set via sysfs or module option if desired.
1394 new_slave->queue_id = 0;
1396 /* Save slave's original mtu and then set it to match the bond */
1397 new_slave->original_mtu = slave_dev->mtu;
1398 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1400 pr_debug("Error %d calling dev_set_mtu\n", res);
1405 * Save slave's original ("permanent") mac address for modes
1406 * that need it, and for restoring it upon release, and then
1407 * set it to the master's address
1409 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1411 if (!bond->params.fail_over_mac) {
1413 * Set slave to master's mac address. The application already
1414 * set the master's mac address to that of the first slave
1416 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1417 addr.sa_family = slave_dev->type;
1418 res = dev_set_mac_address(slave_dev, &addr);
1420 pr_debug("Error %d calling set_mac_address\n", res);
1421 goto err_restore_mtu;
1425 res = bond_master_upper_dev_link(bond_dev, slave_dev);
1427 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1428 goto err_restore_mac;
1431 /* open the slave since the application closed it */
1432 res = dev_open(slave_dev);
1434 pr_debug("Opening slave %s failed\n", slave_dev->name);
1435 goto err_unset_master;
1438 new_slave->bond = bond;
1439 new_slave->dev = slave_dev;
1440 slave_dev->priv_flags |= IFF_BONDING;
1442 if (bond_is_lb(bond)) {
1443 /* bond_alb_init_slave() must be called before all other stages since
1444 * it might fail and we do not want to have to undo everything
1446 res = bond_alb_init_slave(bond, new_slave);
1451 /* If the mode USES_PRIMARY, then the following is handled by
1452 * bond_change_active_slave().
1454 if (!USES_PRIMARY(bond->params.mode)) {
1455 /* set promiscuity level to new slave */
1456 if (bond_dev->flags & IFF_PROMISC) {
1457 res = dev_set_promiscuity(slave_dev, 1);
1462 /* set allmulti level to new slave */
1463 if (bond_dev->flags & IFF_ALLMULTI) {
1464 res = dev_set_allmulti(slave_dev, 1);
1469 netif_addr_lock_bh(bond_dev);
1471 dev_mc_sync_multiple(slave_dev, bond_dev);
1472 dev_uc_sync_multiple(slave_dev, bond_dev);
1474 netif_addr_unlock_bh(bond_dev);
1477 if (bond->params.mode == BOND_MODE_8023AD) {
1478 /* add lacpdu mc addr to mc list */
1479 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1481 dev_mc_add(slave_dev, lacpdu_multicast);
1484 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1486 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1487 bond_dev->name, slave_dev->name);
1491 write_lock_bh(&bond->lock);
1493 bond_attach_slave(bond, new_slave);
1495 new_slave->delay = 0;
1496 new_slave->link_failure_count = 0;
1498 write_unlock_bh(&bond->lock);
1500 bond_compute_features(bond);
1502 bond_update_speed_duplex(new_slave);
1504 read_lock(&bond->lock);
1506 new_slave->last_arp_rx = jiffies -
1507 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1508 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1509 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1511 if (bond->params.miimon && !bond->params.use_carrier) {
1512 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1514 if ((link_reporting == -1) && !bond->params.arp_interval) {
1516 * miimon is set but a bonded network driver
1517 * does not support ETHTOOL/MII and
1518 * arp_interval is not set. Note: if
1519 * use_carrier is enabled, we will never go
1520 * here (because netif_carrier is always
1521 * supported); thus, we don't need to change
1522 * the messages for netif_carrier.
1524 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",
1525 bond_dev->name, slave_dev->name);
1526 } else if (link_reporting == -1) {
1527 /* unable get link status using mii/ethtool */
1528 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",
1529 bond_dev->name, slave_dev->name);
1533 /* check for initial state */
1534 if (bond->params.miimon) {
1535 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1536 if (bond->params.updelay) {
1537 new_slave->link = BOND_LINK_BACK;
1538 new_slave->delay = bond->params.updelay;
1540 new_slave->link = BOND_LINK_UP;
1543 new_slave->link = BOND_LINK_DOWN;
1545 } else if (bond->params.arp_interval) {
1546 new_slave->link = (netif_carrier_ok(slave_dev) ?
1547 BOND_LINK_UP : BOND_LINK_DOWN);
1549 new_slave->link = BOND_LINK_UP;
1552 if (new_slave->link != BOND_LINK_DOWN)
1553 new_slave->jiffies = jiffies;
1554 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1555 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1556 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1558 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1559 /* if there is a primary slave, remember it */
1560 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1561 bond->primary_slave = new_slave;
1562 bond->force_primary = true;
1566 write_lock_bh(&bond->curr_slave_lock);
1568 switch (bond->params.mode) {
1569 case BOND_MODE_ACTIVEBACKUP:
1570 bond_set_slave_inactive_flags(new_slave);
1571 bond_select_active_slave(bond);
1573 case BOND_MODE_8023AD:
1574 /* in 802.3ad mode, the internal mechanism
1575 * will activate the slaves in the selected
1578 bond_set_slave_inactive_flags(new_slave);
1579 /* if this is the first slave */
1580 if (bond_first_slave(bond) == new_slave) {
1581 SLAVE_AD_INFO(new_slave).id = 1;
1582 /* Initialize AD with the number of times that the AD timer is called in 1 second
1583 * can be called only after the mac address of the bond is set
1585 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1587 struct slave *prev_slave;
1589 prev_slave = bond_prev_slave(bond, new_slave);
1590 SLAVE_AD_INFO(new_slave).id =
1591 SLAVE_AD_INFO(prev_slave).id + 1;
1594 bond_3ad_bind_slave(new_slave);
1598 bond_set_active_slave(new_slave);
1599 bond_set_slave_inactive_flags(new_slave);
1600 bond_select_active_slave(bond);
1603 pr_debug("This slave is always active in trunk mode\n");
1605 /* always active in trunk mode */
1606 bond_set_active_slave(new_slave);
1608 /* In trunking mode there is little meaning to curr_active_slave
1609 * anyway (it holds no special properties of the bond device),
1610 * so we can change it without calling change_active_interface()
1612 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1613 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1616 } /* switch(bond_mode) */
1618 write_unlock_bh(&bond->curr_slave_lock);
1620 bond_set_carrier(bond);
1622 #ifdef CONFIG_NET_POLL_CONTROLLER
1623 slave_dev->npinfo = bond->dev->npinfo;
1624 if (slave_dev->npinfo) {
1625 if (slave_enable_netpoll(new_slave)) {
1626 read_unlock(&bond->lock);
1627 pr_info("Error, %s: master_dev is using netpoll, "
1628 "but new slave device does not support netpoll.\n",
1636 read_unlock(&bond->lock);
1638 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1642 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1645 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1646 goto err_dest_symlinks;
1649 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1650 bond_dev->name, slave_dev->name,
1651 bond_is_active_slave(new_slave) ? "n active" : " backup",
1652 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1654 /* enslave is successful */
1657 /* Undo stages on error */
1659 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1662 if (!USES_PRIMARY(bond->params.mode))
1663 bond_hw_addr_flush(bond_dev, slave_dev);
1665 vlan_vids_del_by_dev(slave_dev, bond_dev);
1666 write_lock_bh(&bond->lock);
1667 bond_detach_slave(bond, new_slave);
1668 if (bond->primary_slave == new_slave)
1669 bond->primary_slave = NULL;
1670 if (bond->curr_active_slave == new_slave) {
1671 bond_change_active_slave(bond, NULL);
1672 write_unlock_bh(&bond->lock);
1673 read_lock(&bond->lock);
1674 write_lock_bh(&bond->curr_slave_lock);
1675 bond_select_active_slave(bond);
1676 write_unlock_bh(&bond->curr_slave_lock);
1677 read_unlock(&bond->lock);
1679 write_unlock_bh(&bond->lock);
1681 slave_disable_netpoll(new_slave);
1684 slave_dev->priv_flags &= ~IFF_BONDING;
1685 dev_close(slave_dev);
1688 bond_upper_dev_unlink(bond_dev, slave_dev);
1691 if (!bond->params.fail_over_mac) {
1692 /* XXX TODO - fom follow mode needs to change master's
1693 * MAC if this slave's MAC is in use by the bond, or at
1694 * least print a warning.
1696 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1697 addr.sa_family = slave_dev->type;
1698 dev_set_mac_address(slave_dev, &addr);
1702 dev_set_mtu(slave_dev, new_slave->original_mtu);
1708 bond_compute_features(bond);
1709 /* Enslave of first slave has failed and we need to fix master's mac */
1710 if (list_empty(&bond->slave_list) &&
1711 ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1712 eth_hw_addr_random(bond_dev);
1718 * Try to release the slave device <slave> from the bond device <master>
1719 * It is legal to access curr_active_slave without a lock because all the function
1720 * is write-locked. If "all" is true it means that the function is being called
1721 * while destroying a bond interface and all slaves are being released.
1723 * The rules for slave state should be:
1724 * for Active/Backup:
1725 * Active stays on all backups go down
1726 * for Bonded connections:
1727 * The first up interface should be left on and all others downed.
1729 static int __bond_release_one(struct net_device *bond_dev,
1730 struct net_device *slave_dev,
1733 struct bonding *bond = netdev_priv(bond_dev);
1734 struct slave *slave, *oldcurrent;
1735 struct sockaddr addr;
1736 netdev_features_t old_features = bond_dev->features;
1738 /* slave is not a slave or master is not master of this slave */
1739 if (!(slave_dev->flags & IFF_SLAVE) ||
1740 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1741 pr_err("%s: Error: cannot release %s.\n",
1742 bond_dev->name, slave_dev->name);
1747 write_lock_bh(&bond->lock);
1749 slave = bond_get_slave_by_dev(bond, slave_dev);
1751 /* not a slave of this bond */
1752 pr_info("%s: %s not enslaved\n",
1753 bond_dev->name, slave_dev->name);
1754 write_unlock_bh(&bond->lock);
1755 unblock_netpoll_tx();
1759 write_unlock_bh(&bond->lock);
1760 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1761 * for this slave anymore.
1763 netdev_rx_handler_unregister(slave_dev);
1764 write_lock_bh(&bond->lock);
1766 /* Inform AD package of unbinding of slave. */
1767 if (bond->params.mode == BOND_MODE_8023AD) {
1768 /* must be called before the slave is
1769 * detached from the list
1771 bond_3ad_unbind_slave(slave);
1774 pr_info("%s: releasing %s interface %s\n",
1776 bond_is_active_slave(slave) ? "active" : "backup",
1779 oldcurrent = bond->curr_active_slave;
1781 bond->current_arp_slave = NULL;
1783 /* release the slave from its bond */
1784 bond_detach_slave(bond, slave);
1786 if (!all && !bond->params.fail_over_mac) {
1787 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1788 !list_empty(&bond->slave_list))
1789 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",
1790 bond_dev->name, slave_dev->name,
1792 bond_dev->name, slave_dev->name);
1795 if (bond->primary_slave == slave)
1796 bond->primary_slave = NULL;
1798 if (oldcurrent == slave)
1799 bond_change_active_slave(bond, NULL);
1801 if (bond_is_lb(bond)) {
1802 /* Must be called only after the slave has been
1803 * detached from the list and the curr_active_slave
1804 * has been cleared (if our_slave == old_current),
1805 * but before a new active slave is selected.
1807 write_unlock_bh(&bond->lock);
1808 bond_alb_deinit_slave(bond, slave);
1809 write_lock_bh(&bond->lock);
1813 rcu_assign_pointer(bond->curr_active_slave, NULL);
1814 } else if (oldcurrent == slave) {
1816 * Note that we hold RTNL over this sequence, so there
1817 * is no concern that another slave add/remove event
1820 write_unlock_bh(&bond->lock);
1821 read_lock(&bond->lock);
1822 write_lock_bh(&bond->curr_slave_lock);
1824 bond_select_active_slave(bond);
1826 write_unlock_bh(&bond->curr_slave_lock);
1827 read_unlock(&bond->lock);
1828 write_lock_bh(&bond->lock);
1831 if (list_empty(&bond->slave_list)) {
1832 bond_set_carrier(bond);
1833 eth_hw_addr_random(bond_dev);
1835 if (vlan_uses_dev(bond_dev)) {
1836 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1837 bond_dev->name, bond_dev->name);
1838 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1843 write_unlock_bh(&bond->lock);
1844 unblock_netpoll_tx();
1847 if (list_empty(&bond->slave_list)) {
1848 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1849 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1852 bond_compute_features(bond);
1853 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1854 (old_features & NETIF_F_VLAN_CHALLENGED))
1855 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1856 bond_dev->name, slave_dev->name, bond_dev->name);
1858 /* must do this from outside any spinlocks */
1859 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1861 vlan_vids_del_by_dev(slave_dev, bond_dev);
1863 /* If the mode USES_PRIMARY, then this cases was handled above by
1864 * bond_change_active_slave(..., NULL)
1866 if (!USES_PRIMARY(bond->params.mode)) {
1867 /* unset promiscuity level from slave */
1868 if (bond_dev->flags & IFF_PROMISC)
1869 dev_set_promiscuity(slave_dev, -1);
1871 /* unset allmulti level from slave */
1872 if (bond_dev->flags & IFF_ALLMULTI)
1873 dev_set_allmulti(slave_dev, -1);
1875 bond_hw_addr_flush(bond_dev, slave_dev);
1878 bond_upper_dev_unlink(bond_dev, slave_dev);
1880 slave_disable_netpoll(slave);
1882 /* close slave before restoring its mac address */
1883 dev_close(slave_dev);
1885 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1886 /* restore original ("permanent") mac address */
1887 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1888 addr.sa_family = slave_dev->type;
1889 dev_set_mac_address(slave_dev, &addr);
1892 dev_set_mtu(slave_dev, slave->original_mtu);
1894 slave_dev->priv_flags &= ~IFF_BONDING;
1898 return 0; /* deletion OK */
1901 /* A wrapper used because of ndo_del_link */
1902 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1904 return __bond_release_one(bond_dev, slave_dev, false);
1908 * First release a slave and then destroy the bond if no more slaves are left.
1909 * Must be under rtnl_lock when this function is called.
1911 static int bond_release_and_destroy(struct net_device *bond_dev,
1912 struct net_device *slave_dev)
1914 struct bonding *bond = netdev_priv(bond_dev);
1917 ret = bond_release(bond_dev, slave_dev);
1918 if (ret == 0 && list_empty(&bond->slave_list)) {
1919 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1920 pr_info("%s: destroying bond %s.\n",
1921 bond_dev->name, bond_dev->name);
1922 unregister_netdevice(bond_dev);
1928 * This function changes the active slave to slave <slave_dev>.
1929 * It returns -EINVAL in the following cases.
1930 * - <slave_dev> is not found in the list.
1931 * - There is not active slave now.
1932 * - <slave_dev> is already active.
1933 * - The link state of <slave_dev> is not BOND_LINK_UP.
1934 * - <slave_dev> is not running.
1935 * In these cases, this function does nothing.
1936 * In the other cases, current_slave pointer is changed and 0 is returned.
1938 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1940 struct bonding *bond = netdev_priv(bond_dev);
1941 struct slave *old_active = NULL;
1942 struct slave *new_active = NULL;
1945 if (!USES_PRIMARY(bond->params.mode))
1948 /* Verify that bond_dev is indeed the master of slave_dev */
1949 if (!(slave_dev->flags & IFF_SLAVE) ||
1950 !netdev_has_upper_dev(slave_dev, bond_dev))
1953 read_lock(&bond->lock);
1955 old_active = bond->curr_active_slave;
1956 new_active = bond_get_slave_by_dev(bond, slave_dev);
1958 * Changing to the current active: do nothing; return success.
1960 if (new_active && new_active == old_active) {
1961 read_unlock(&bond->lock);
1967 new_active->link == BOND_LINK_UP &&
1968 IS_UP(new_active->dev)) {
1970 write_lock_bh(&bond->curr_slave_lock);
1971 bond_change_active_slave(bond, new_active);
1972 write_unlock_bh(&bond->curr_slave_lock);
1973 unblock_netpoll_tx();
1977 read_unlock(&bond->lock);
1982 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1984 struct bonding *bond = netdev_priv(bond_dev);
1986 info->bond_mode = bond->params.mode;
1987 info->miimon = bond->params.miimon;
1989 read_lock(&bond->lock);
1990 info->num_slaves = bond->slave_cnt;
1991 read_unlock(&bond->lock);
1996 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1998 struct bonding *bond = netdev_priv(bond_dev);
1999 int i = 0, res = -ENODEV;
2000 struct slave *slave;
2002 read_lock(&bond->lock);
2003 bond_for_each_slave(bond, slave) {
2004 if (i++ == (int)info->slave_id) {
2006 strcpy(info->slave_name, slave->dev->name);
2007 info->link = slave->link;
2008 info->state = bond_slave_state(slave);
2009 info->link_failure_count = slave->link_failure_count;
2013 read_unlock(&bond->lock);
2018 /*-------------------------------- Monitoring -------------------------------*/
2021 static int bond_miimon_inspect(struct bonding *bond)
2023 int link_state, commit = 0;
2024 struct slave *slave;
2025 bool ignore_updelay;
2027 ignore_updelay = !bond->curr_active_slave ? true : false;
2029 bond_for_each_slave(bond, slave) {
2030 slave->new_link = BOND_LINK_NOCHANGE;
2032 link_state = bond_check_dev_link(bond, slave->dev, 0);
2034 switch (slave->link) {
2039 slave->link = BOND_LINK_FAIL;
2040 slave->delay = bond->params.downdelay;
2042 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2044 (bond->params.mode ==
2045 BOND_MODE_ACTIVEBACKUP) ?
2046 (bond_is_active_slave(slave) ?
2047 "active " : "backup ") : "",
2049 bond->params.downdelay * bond->params.miimon);
2052 case BOND_LINK_FAIL:
2055 * recovered before downdelay expired
2057 slave->link = BOND_LINK_UP;
2058 slave->jiffies = jiffies;
2059 pr_info("%s: link status up again after %d ms for interface %s.\n",
2061 (bond->params.downdelay - slave->delay) *
2062 bond->params.miimon,
2067 if (slave->delay <= 0) {
2068 slave->new_link = BOND_LINK_DOWN;
2076 case BOND_LINK_DOWN:
2080 slave->link = BOND_LINK_BACK;
2081 slave->delay = bond->params.updelay;
2084 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2085 bond->dev->name, slave->dev->name,
2086 ignore_updelay ? 0 :
2087 bond->params.updelay *
2088 bond->params.miimon);
2091 case BOND_LINK_BACK:
2093 slave->link = BOND_LINK_DOWN;
2094 pr_info("%s: link status down again after %d ms for interface %s.\n",
2096 (bond->params.updelay - slave->delay) *
2097 bond->params.miimon,
2106 if (slave->delay <= 0) {
2107 slave->new_link = BOND_LINK_UP;
2109 ignore_updelay = false;
2121 static void bond_miimon_commit(struct bonding *bond)
2123 struct slave *slave;
2125 bond_for_each_slave(bond, slave) {
2126 switch (slave->new_link) {
2127 case BOND_LINK_NOCHANGE:
2131 slave->link = BOND_LINK_UP;
2132 slave->jiffies = jiffies;
2134 if (bond->params.mode == BOND_MODE_8023AD) {
2135 /* prevent it from being the active one */
2136 bond_set_backup_slave(slave);
2137 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2138 /* make it immediately active */
2139 bond_set_active_slave(slave);
2140 } else if (slave != bond->primary_slave) {
2141 /* prevent it from being the active one */
2142 bond_set_backup_slave(slave);
2145 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2146 bond->dev->name, slave->dev->name,
2147 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2148 slave->duplex ? "full" : "half");
2150 /* notify ad that the link status has changed */
2151 if (bond->params.mode == BOND_MODE_8023AD)
2152 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2154 if (bond_is_lb(bond))
2155 bond_alb_handle_link_change(bond, slave,
2158 if (!bond->curr_active_slave ||
2159 (slave == bond->primary_slave))
2164 case BOND_LINK_DOWN:
2165 if (slave->link_failure_count < UINT_MAX)
2166 slave->link_failure_count++;
2168 slave->link = BOND_LINK_DOWN;
2170 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2171 bond->params.mode == BOND_MODE_8023AD)
2172 bond_set_slave_inactive_flags(slave);
2174 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2175 bond->dev->name, slave->dev->name);
2177 if (bond->params.mode == BOND_MODE_8023AD)
2178 bond_3ad_handle_link_change(slave,
2181 if (bond_is_lb(bond))
2182 bond_alb_handle_link_change(bond, slave,
2185 if (slave == bond->curr_active_slave)
2191 pr_err("%s: invalid new link %d on slave %s\n",
2192 bond->dev->name, slave->new_link,
2194 slave->new_link = BOND_LINK_NOCHANGE;
2202 write_lock_bh(&bond->curr_slave_lock);
2203 bond_select_active_slave(bond);
2204 write_unlock_bh(&bond->curr_slave_lock);
2205 unblock_netpoll_tx();
2208 bond_set_carrier(bond);
2214 * Really a wrapper that splits the mii monitor into two phases: an
2215 * inspection, then (if inspection indicates something needs to be done)
2216 * an acquisition of appropriate locks followed by a commit phase to
2217 * implement whatever link state changes are indicated.
2219 void bond_mii_monitor(struct work_struct *work)
2221 struct bonding *bond = container_of(work, struct bonding,
2223 bool should_notify_peers = false;
2224 unsigned long delay;
2226 read_lock(&bond->lock);
2228 delay = msecs_to_jiffies(bond->params.miimon);
2230 if (list_empty(&bond->slave_list))
2233 should_notify_peers = bond_should_notify_peers(bond);
2235 if (bond_miimon_inspect(bond)) {
2236 read_unlock(&bond->lock);
2238 /* Race avoidance with bond_close cancel of workqueue */
2239 if (!rtnl_trylock()) {
2240 read_lock(&bond->lock);
2242 should_notify_peers = false;
2246 read_lock(&bond->lock);
2248 bond_miimon_commit(bond);
2250 read_unlock(&bond->lock);
2251 rtnl_unlock(); /* might sleep, hold no other locks */
2252 read_lock(&bond->lock);
2256 if (bond->params.miimon)
2257 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2259 read_unlock(&bond->lock);
2261 if (should_notify_peers) {
2262 if (!rtnl_trylock()) {
2263 read_lock(&bond->lock);
2264 bond->send_peer_notif++;
2265 read_unlock(&bond->lock);
2268 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2273 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2275 struct net_device *upper;
2276 struct list_head *iter;
2279 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2283 netdev_for_each_upper_dev_rcu(bond->dev, upper, iter) {
2284 if (ip == bond_confirm_addr(upper, 0, ip)) {
2295 * We go to the (large) trouble of VLAN tagging ARP frames because
2296 * switches in VLAN mode (especially if ports are configured as
2297 * "native" to a VLAN) might not pass non-tagged frames.
2299 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2301 struct sk_buff *skb;
2303 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2304 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2306 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2307 NULL, slave_dev->dev_addr, NULL);
2310 pr_err("ARP packet allocation failed\n");
2314 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2316 pr_err("failed to insert VLAN tag\n");
2324 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2326 struct net_device *upper, *vlan_upper;
2327 struct list_head *iter, *vlan_iter;
2329 __be32 *targets = bond->params.arp_targets, addr;
2332 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2333 pr_debug("basa: target %pI4\n", &targets[i]);
2335 /* Find out through which dev should the packet go */
2336 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2339 if (net_ratelimit()) {
2340 pr_warning("%s: no route to arp_ip_target %pI4\n",
2341 bond->dev->name, &targets[i]);
2348 /* bond device itself */
2349 if (rt->dst.dev == bond->dev)
2353 /* first we search only for vlan devices. for every vlan
2354 * found we verify its upper dev list, searching for the
2355 * rt->dst.dev. If found we save the tag of the vlan and
2356 * proceed to send the packet.
2360 netdev_for_each_upper_dev_rcu(bond->dev, vlan_upper, vlan_iter) {
2361 if (!is_vlan_dev(vlan_upper))
2363 netdev_for_each_upper_dev_rcu(vlan_upper, upper, iter) {
2364 if (upper == rt->dst.dev) {
2365 vlan_id = vlan_dev_vlan_id(vlan_upper);
2372 /* if the device we're looking for is not on top of any of
2373 * our upper vlans, then just search for any dev that
2374 * matches, and in case it's a vlan - save the id
2376 netdev_for_each_upper_dev_rcu(bond->dev, upper, iter) {
2377 if (upper == rt->dst.dev) {
2378 /* if it's a vlan - get its VID */
2379 if (is_vlan_dev(upper))
2380 vlan_id = vlan_dev_vlan_id(upper);
2388 /* Not our device - skip */
2389 if (net_ratelimit())
2390 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2391 bond->dev->name, &targets[i],
2392 rt->dst.dev ? rt->dst.dev->name : "NULL");
2397 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2399 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2404 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2408 if (!sip || !bond_has_this_ip(bond, tip)) {
2409 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2413 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2415 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2418 slave->last_arp_rx = jiffies;
2419 slave->target_last_arp_rx[i] = jiffies;
2422 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2423 struct slave *slave)
2425 struct arphdr *arp = (struct arphdr *)skb->data;
2426 unsigned char *arp_ptr;
2430 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2431 return RX_HANDLER_ANOTHER;
2433 read_lock(&bond->lock);
2435 if (!slave_do_arp_validate(bond, slave))
2438 alen = arp_hdr_len(bond->dev);
2440 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2441 bond->dev->name, skb->dev->name);
2443 if (alen > skb_headlen(skb)) {
2444 arp = kmalloc(alen, GFP_ATOMIC);
2447 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2451 if (arp->ar_hln != bond->dev->addr_len ||
2452 skb->pkt_type == PACKET_OTHERHOST ||
2453 skb->pkt_type == PACKET_LOOPBACK ||
2454 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2455 arp->ar_pro != htons(ETH_P_IP) ||
2459 arp_ptr = (unsigned char *)(arp + 1);
2460 arp_ptr += bond->dev->addr_len;
2461 memcpy(&sip, arp_ptr, 4);
2462 arp_ptr += 4 + bond->dev->addr_len;
2463 memcpy(&tip, arp_ptr, 4);
2465 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2466 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2467 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2471 * Backup slaves won't see the ARP reply, but do come through
2472 * here for each ARP probe (so we swap the sip/tip to validate
2473 * the probe). In a "redundant switch, common router" type of
2474 * configuration, the ARP probe will (hopefully) travel from
2475 * the active, through one switch, the router, then the other
2476 * switch before reaching the backup.
2478 * We 'trust' the arp requests if there is an active slave and
2479 * it received valid arp reply(s) after it became active. This
2480 * is done to avoid endless looping when we can't reach the
2481 * arp_ip_target and fool ourselves with our own arp requests.
2483 if (bond_is_active_slave(slave))
2484 bond_validate_arp(bond, slave, sip, tip);
2485 else if (bond->curr_active_slave &&
2486 time_after(slave_last_rx(bond, bond->curr_active_slave),
2487 bond->curr_active_slave->jiffies))
2488 bond_validate_arp(bond, slave, tip, sip);
2491 read_unlock(&bond->lock);
2492 if (arp != (struct arphdr *)skb->data)
2494 return RX_HANDLER_ANOTHER;
2497 /* function to verify if we're in the arp_interval timeslice, returns true if
2498 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2499 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2501 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2504 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2506 return time_in_range(jiffies,
2507 last_act - delta_in_ticks,
2508 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2512 * this function is called regularly to monitor each slave's link
2513 * ensuring that traffic is being sent and received when arp monitoring
2514 * is used in load-balancing mode. if the adapter has been dormant, then an
2515 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2516 * arp monitoring in active backup mode.
2518 void bond_loadbalance_arp_mon(struct work_struct *work)
2520 struct bonding *bond = container_of(work, struct bonding,
2522 struct slave *slave, *oldcurrent;
2523 int do_failover = 0;
2525 read_lock(&bond->lock);
2527 if (list_empty(&bond->slave_list))
2530 oldcurrent = bond->curr_active_slave;
2531 /* see if any of the previous devices are up now (i.e. they have
2532 * xmt and rcv traffic). the curr_active_slave does not come into
2533 * the picture unless it is null. also, slave->jiffies is not needed
2534 * here because we send an arp on each slave and give a slave as
2535 * long as it needs to get the tx/rx within the delta.
2536 * TODO: what about up/down delay in arp mode? it wasn't here before
2539 bond_for_each_slave(bond, slave) {
2540 unsigned long trans_start = dev_trans_start(slave->dev);
2542 if (slave->link != BOND_LINK_UP) {
2543 if (bond_time_in_interval(bond, trans_start, 1) &&
2544 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2546 slave->link = BOND_LINK_UP;
2547 bond_set_active_slave(slave);
2549 /* primary_slave has no meaning in round-robin
2550 * mode. the window of a slave being up and
2551 * curr_active_slave being null after enslaving
2555 pr_info("%s: link status definitely up for interface %s, ",
2560 pr_info("%s: interface %s is now up\n",
2566 /* slave->link == BOND_LINK_UP */
2568 /* not all switches will respond to an arp request
2569 * when the source ip is 0, so don't take the link down
2570 * if we don't know our ip yet
2572 if (!bond_time_in_interval(bond, trans_start, 2) ||
2573 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2575 slave->link = BOND_LINK_DOWN;
2576 bond_set_backup_slave(slave);
2578 if (slave->link_failure_count < UINT_MAX)
2579 slave->link_failure_count++;
2581 pr_info("%s: interface %s is now down.\n",
2585 if (slave == oldcurrent)
2590 /* note: if switch is in round-robin mode, all links
2591 * must tx arp to ensure all links rx an arp - otherwise
2592 * links may oscillate or not come up at all; if switch is
2593 * in something like xor mode, there is nothing we can
2594 * do - all replies will be rx'ed on same link causing slaves
2595 * to be unstable during low/no traffic periods
2597 if (IS_UP(slave->dev))
2598 bond_arp_send_all(bond, slave);
2603 write_lock_bh(&bond->curr_slave_lock);
2605 bond_select_active_slave(bond);
2607 write_unlock_bh(&bond->curr_slave_lock);
2608 unblock_netpoll_tx();
2612 if (bond->params.arp_interval)
2613 queue_delayed_work(bond->wq, &bond->arp_work,
2614 msecs_to_jiffies(bond->params.arp_interval));
2616 read_unlock(&bond->lock);
2620 * Called to inspect slaves for active-backup mode ARP monitor link state
2621 * changes. Sets new_link in slaves to specify what action should take
2622 * place for the slave. Returns 0 if no changes are found, >0 if changes
2623 * to link states must be committed.
2625 * Called with bond->lock held for read.
2627 static int bond_ab_arp_inspect(struct bonding *bond)
2629 unsigned long trans_start, last_rx;
2630 struct slave *slave;
2633 bond_for_each_slave(bond, slave) {
2634 slave->new_link = BOND_LINK_NOCHANGE;
2635 last_rx = slave_last_rx(bond, slave);
2637 if (slave->link != BOND_LINK_UP) {
2638 if (bond_time_in_interval(bond, last_rx, 1)) {
2639 slave->new_link = BOND_LINK_UP;
2646 * Give slaves 2*delta after being enslaved or made
2647 * active. This avoids bouncing, as the last receive
2648 * times need a full ARP monitor cycle to be updated.
2650 if (bond_time_in_interval(bond, slave->jiffies, 2))
2654 * Backup slave is down if:
2655 * - No current_arp_slave AND
2656 * - more than 3*delta since last receive AND
2657 * - the bond has an IP address
2659 * Note: a non-null current_arp_slave indicates
2660 * the curr_active_slave went down and we are
2661 * searching for a new one; under this condition
2662 * we only take the curr_active_slave down - this
2663 * gives each slave a chance to tx/rx traffic
2664 * before being taken out
2666 if (!bond_is_active_slave(slave) &&
2667 !bond->current_arp_slave &&
2668 !bond_time_in_interval(bond, last_rx, 3)) {
2669 slave->new_link = BOND_LINK_DOWN;
2674 * Active slave is down if:
2675 * - more than 2*delta since transmitting OR
2676 * - (more than 2*delta since receive AND
2677 * the bond has an IP address)
2679 trans_start = dev_trans_start(slave->dev);
2680 if (bond_is_active_slave(slave) &&
2681 (!bond_time_in_interval(bond, trans_start, 2) ||
2682 !bond_time_in_interval(bond, last_rx, 2))) {
2683 slave->new_link = BOND_LINK_DOWN;
2692 * Called to commit link state changes noted by inspection step of
2693 * active-backup mode ARP monitor.
2695 * Called with RTNL and bond->lock for read.
2697 static void bond_ab_arp_commit(struct bonding *bond)
2699 unsigned long trans_start;
2700 struct slave *slave;
2702 bond_for_each_slave(bond, slave) {
2703 switch (slave->new_link) {
2704 case BOND_LINK_NOCHANGE:
2708 trans_start = dev_trans_start(slave->dev);
2709 if (bond->curr_active_slave != slave ||
2710 (!bond->curr_active_slave &&
2711 bond_time_in_interval(bond, trans_start, 1))) {
2712 slave->link = BOND_LINK_UP;
2713 if (bond->current_arp_slave) {
2714 bond_set_slave_inactive_flags(
2715 bond->current_arp_slave);
2716 bond->current_arp_slave = NULL;
2719 pr_info("%s: link status definitely up for interface %s.\n",
2720 bond->dev->name, slave->dev->name);
2722 if (!bond->curr_active_slave ||
2723 (slave == bond->primary_slave))
2730 case BOND_LINK_DOWN:
2731 if (slave->link_failure_count < UINT_MAX)
2732 slave->link_failure_count++;
2734 slave->link = BOND_LINK_DOWN;
2735 bond_set_slave_inactive_flags(slave);
2737 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2738 bond->dev->name, slave->dev->name);
2740 if (slave == bond->curr_active_slave) {
2741 bond->current_arp_slave = NULL;
2748 pr_err("%s: impossible: new_link %d on slave %s\n",
2749 bond->dev->name, slave->new_link,
2757 write_lock_bh(&bond->curr_slave_lock);
2758 bond_select_active_slave(bond);
2759 write_unlock_bh(&bond->curr_slave_lock);
2760 unblock_netpoll_tx();
2763 bond_set_carrier(bond);
2767 * Send ARP probes for active-backup mode ARP monitor.
2769 * Called with bond->lock held for read.
2771 static void bond_ab_arp_probe(struct bonding *bond)
2773 struct slave *slave, *next_slave;
2776 read_lock(&bond->curr_slave_lock);
2778 if (bond->current_arp_slave && bond->curr_active_slave)
2779 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2780 bond->current_arp_slave->dev->name,
2781 bond->curr_active_slave->dev->name);
2783 if (bond->curr_active_slave) {
2784 bond_arp_send_all(bond, bond->curr_active_slave);
2785 read_unlock(&bond->curr_slave_lock);
2789 read_unlock(&bond->curr_slave_lock);
2791 /* if we don't have a curr_active_slave, search for the next available
2792 * backup slave from the current_arp_slave and make it the candidate
2793 * for becoming the curr_active_slave
2796 if (!bond->current_arp_slave) {
2797 bond->current_arp_slave = bond_first_slave(bond);
2798 if (!bond->current_arp_slave)
2802 bond_set_slave_inactive_flags(bond->current_arp_slave);
2804 /* search for next candidate */
2805 next_slave = bond_next_slave(bond, bond->current_arp_slave);
2806 bond_for_each_slave_from(bond, slave, i, next_slave) {
2807 if (IS_UP(slave->dev)) {
2808 slave->link = BOND_LINK_BACK;
2809 bond_set_slave_active_flags(slave);
2810 bond_arp_send_all(bond, slave);
2811 slave->jiffies = jiffies;
2812 bond->current_arp_slave = slave;
2816 /* if the link state is up at this point, we
2817 * mark it down - this can happen if we have
2818 * simultaneous link failures and
2819 * reselect_active_interface doesn't make this
2820 * one the current slave so it is still marked
2821 * up when it is actually down
2823 if (slave->link == BOND_LINK_UP) {
2824 slave->link = BOND_LINK_DOWN;
2825 if (slave->link_failure_count < UINT_MAX)
2826 slave->link_failure_count++;
2828 bond_set_slave_inactive_flags(slave);
2830 pr_info("%s: backup interface %s is now down.\n",
2831 bond->dev->name, slave->dev->name);
2836 void bond_activebackup_arp_mon(struct work_struct *work)
2838 struct bonding *bond = container_of(work, struct bonding,
2840 bool should_notify_peers = false;
2843 read_lock(&bond->lock);
2845 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2847 if (list_empty(&bond->slave_list))
2850 should_notify_peers = bond_should_notify_peers(bond);
2852 if (bond_ab_arp_inspect(bond)) {
2853 read_unlock(&bond->lock);
2855 /* Race avoidance with bond_close flush of workqueue */
2856 if (!rtnl_trylock()) {
2857 read_lock(&bond->lock);
2859 should_notify_peers = false;
2863 read_lock(&bond->lock);
2865 bond_ab_arp_commit(bond);
2867 read_unlock(&bond->lock);
2869 read_lock(&bond->lock);
2872 bond_ab_arp_probe(bond);
2875 if (bond->params.arp_interval)
2876 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2878 read_unlock(&bond->lock);
2880 if (should_notify_peers) {
2881 if (!rtnl_trylock()) {
2882 read_lock(&bond->lock);
2883 bond->send_peer_notif++;
2884 read_unlock(&bond->lock);
2887 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2892 /*-------------------------- netdev event handling --------------------------*/
2895 * Change device name
2897 static int bond_event_changename(struct bonding *bond)
2899 bond_remove_proc_entry(bond);
2900 bond_create_proc_entry(bond);
2902 bond_debug_reregister(bond);
2907 static int bond_master_netdev_event(unsigned long event,
2908 struct net_device *bond_dev)
2910 struct bonding *event_bond = netdev_priv(bond_dev);
2913 case NETDEV_CHANGENAME:
2914 return bond_event_changename(event_bond);
2915 case NETDEV_UNREGISTER:
2916 bond_remove_proc_entry(event_bond);
2918 case NETDEV_REGISTER:
2919 bond_create_proc_entry(event_bond);
2928 static int bond_slave_netdev_event(unsigned long event,
2929 struct net_device *slave_dev)
2931 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2932 struct bonding *bond;
2933 struct net_device *bond_dev;
2937 /* A netdev event can be generated while enslaving a device
2938 * before netdev_rx_handler_register is called in which case
2939 * slave will be NULL
2943 bond_dev = slave->bond->dev;
2947 case NETDEV_UNREGISTER:
2948 if (bond_dev->type != ARPHRD_ETHER)
2949 bond_release_and_destroy(bond_dev, slave_dev);
2951 bond_release(bond_dev, slave_dev);
2955 old_speed = slave->speed;
2956 old_duplex = slave->duplex;
2958 bond_update_speed_duplex(slave);
2960 if (bond->params.mode == BOND_MODE_8023AD) {
2961 if (old_speed != slave->speed)
2962 bond_3ad_adapter_speed_changed(slave);
2963 if (old_duplex != slave->duplex)
2964 bond_3ad_adapter_duplex_changed(slave);
2969 * ... Or is it this?
2972 case NETDEV_CHANGEMTU:
2974 * TODO: Should slaves be allowed to
2975 * independently alter their MTU? For
2976 * an active-backup bond, slaves need
2977 * not be the same type of device, so
2978 * MTUs may vary. For other modes,
2979 * slaves arguably should have the
2980 * same MTUs. To do this, we'd need to
2981 * take over the slave's change_mtu
2982 * function for the duration of their
2986 case NETDEV_CHANGENAME:
2988 * TODO: handle changing the primary's name
2991 case NETDEV_FEAT_CHANGE:
2992 bond_compute_features(bond);
2994 case NETDEV_RESEND_IGMP:
2995 /* Propagate to master device */
2996 call_netdevice_notifiers(event, slave->bond->dev);
3006 * bond_netdev_event: handle netdev notifier chain events.
3008 * This function receives events for the netdev chain. The caller (an
3009 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3010 * locks for us to safely manipulate the slave devices (RTNL lock,
3013 static int bond_netdev_event(struct notifier_block *this,
3014 unsigned long event, void *ptr)
3016 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3018 pr_debug("event_dev: %s, event: %lx\n",
3019 event_dev ? event_dev->name : "None",
3022 if (!(event_dev->priv_flags & IFF_BONDING))
3025 if (event_dev->flags & IFF_MASTER) {
3026 pr_debug("IFF_MASTER\n");
3027 return bond_master_netdev_event(event, event_dev);
3030 if (event_dev->flags & IFF_SLAVE) {
3031 pr_debug("IFF_SLAVE\n");
3032 return bond_slave_netdev_event(event, event_dev);
3038 static struct notifier_block bond_netdev_notifier = {
3039 .notifier_call = bond_netdev_event,
3042 /*---------------------------- Hashing Policies -----------------------------*/
3045 * Hash for the output device based upon layer 2 data
3047 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3049 struct ethhdr *data = (struct ethhdr *)skb->data;
3051 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3052 return (data->h_dest[5] ^ data->h_source[5]) % count;
3058 * Hash for the output device based upon layer 2 and layer 3 data. If
3059 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3061 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3063 const struct ethhdr *data;
3064 const struct iphdr *iph;
3065 const struct ipv6hdr *ipv6h;
3067 const __be32 *s, *d;
3069 if (skb->protocol == htons(ETH_P_IP) &&
3070 pskb_network_may_pull(skb, sizeof(*iph))) {
3072 data = (struct ethhdr *)skb->data;
3073 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3074 (data->h_dest[5] ^ data->h_source[5])) % count;
3075 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3076 pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3077 ipv6h = ipv6_hdr(skb);
3078 data = (struct ethhdr *)skb->data;
3079 s = &ipv6h->saddr.s6_addr32[0];
3080 d = &ipv6h->daddr.s6_addr32[0];
3081 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3082 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3083 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3086 return bond_xmit_hash_policy_l2(skb, count);
3090 * Hash for the output device based upon layer 3 and layer 4 data. If
3091 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3092 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3094 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3097 const struct iphdr *iph;
3098 const struct ipv6hdr *ipv6h;
3099 const __be32 *s, *d;
3100 const __be16 *l4 = NULL;
3102 int noff = skb_network_offset(skb);
3105 if (skb->protocol == htons(ETH_P_IP) &&
3106 pskb_may_pull(skb, noff + sizeof(*iph))) {
3108 poff = proto_ports_offset(iph->protocol);
3110 if (!ip_is_fragment(iph) && poff >= 0) {
3111 l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3114 layer4_xor = ntohs(l4[0] ^ l4[1]);
3116 return (layer4_xor ^
3117 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3118 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3119 pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3120 ipv6h = ipv6_hdr(skb);
3121 poff = proto_ports_offset(ipv6h->nexthdr);
3123 l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3126 layer4_xor = ntohs(l4[0] ^ l4[1]);
3128 s = &ipv6h->saddr.s6_addr32[0];
3129 d = &ipv6h->daddr.s6_addr32[0];
3130 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3131 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3133 return layer4_xor % count;
3136 return bond_xmit_hash_policy_l2(skb, count);
3139 /*-------------------------- Device entry points ----------------------------*/
3141 static void bond_work_init_all(struct bonding *bond)
3143 INIT_DELAYED_WORK(&bond->mcast_work,
3144 bond_resend_igmp_join_requests_delayed);
3145 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3146 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3147 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3148 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3150 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3151 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3154 static void bond_work_cancel_all(struct bonding *bond)
3156 cancel_delayed_work_sync(&bond->mii_work);
3157 cancel_delayed_work_sync(&bond->arp_work);
3158 cancel_delayed_work_sync(&bond->alb_work);
3159 cancel_delayed_work_sync(&bond->ad_work);
3160 cancel_delayed_work_sync(&bond->mcast_work);
3163 static int bond_open(struct net_device *bond_dev)
3165 struct bonding *bond = netdev_priv(bond_dev);
3166 struct slave *slave;
3168 /* reset slave->backup and slave->inactive */
3169 read_lock(&bond->lock);
3170 if (!list_empty(&bond->slave_list)) {
3171 read_lock(&bond->curr_slave_lock);
3172 bond_for_each_slave(bond, slave) {
3173 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3174 && (slave != bond->curr_active_slave)) {
3175 bond_set_slave_inactive_flags(slave);
3177 bond_set_slave_active_flags(slave);
3180 read_unlock(&bond->curr_slave_lock);
3182 read_unlock(&bond->lock);
3184 bond_work_init_all(bond);
3186 if (bond_is_lb(bond)) {
3187 /* bond_alb_initialize must be called before the timer
3190 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3192 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3195 if (bond->params.miimon) /* link check interval, in milliseconds. */
3196 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3198 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3199 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3200 if (bond->params.arp_validate)
3201 bond->recv_probe = bond_arp_rcv;
3204 if (bond->params.mode == BOND_MODE_8023AD) {
3205 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3206 /* register to receive LACPDUs */
3207 bond->recv_probe = bond_3ad_lacpdu_recv;
3208 bond_3ad_initiate_agg_selection(bond, 1);
3214 static int bond_close(struct net_device *bond_dev)
3216 struct bonding *bond = netdev_priv(bond_dev);
3218 write_lock_bh(&bond->lock);
3219 bond->send_peer_notif = 0;
3220 write_unlock_bh(&bond->lock);
3222 bond_work_cancel_all(bond);
3223 if (bond_is_lb(bond)) {
3224 /* Must be called only after all
3225 * slaves have been released
3227 bond_alb_deinitialize(bond);
3229 bond->recv_probe = NULL;
3234 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3235 struct rtnl_link_stats64 *stats)
3237 struct bonding *bond = netdev_priv(bond_dev);
3238 struct rtnl_link_stats64 temp;
3239 struct slave *slave;
3241 memset(stats, 0, sizeof(*stats));
3243 read_lock_bh(&bond->lock);
3244 bond_for_each_slave(bond, slave) {
3245 const struct rtnl_link_stats64 *sstats =
3246 dev_get_stats(slave->dev, &temp);
3248 stats->rx_packets += sstats->rx_packets;
3249 stats->rx_bytes += sstats->rx_bytes;
3250 stats->rx_errors += sstats->rx_errors;
3251 stats->rx_dropped += sstats->rx_dropped;
3253 stats->tx_packets += sstats->tx_packets;
3254 stats->tx_bytes += sstats->tx_bytes;
3255 stats->tx_errors += sstats->tx_errors;
3256 stats->tx_dropped += sstats->tx_dropped;
3258 stats->multicast += sstats->multicast;
3259 stats->collisions += sstats->collisions;
3261 stats->rx_length_errors += sstats->rx_length_errors;
3262 stats->rx_over_errors += sstats->rx_over_errors;
3263 stats->rx_crc_errors += sstats->rx_crc_errors;
3264 stats->rx_frame_errors += sstats->rx_frame_errors;
3265 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3266 stats->rx_missed_errors += sstats->rx_missed_errors;
3268 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3269 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3270 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3271 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3272 stats->tx_window_errors += sstats->tx_window_errors;
3274 read_unlock_bh(&bond->lock);
3279 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3281 struct net_device *slave_dev = NULL;
3282 struct ifbond k_binfo;
3283 struct ifbond __user *u_binfo = NULL;
3284 struct ifslave k_sinfo;
3285 struct ifslave __user *u_sinfo = NULL;
3286 struct mii_ioctl_data *mii = NULL;
3290 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3302 * We do this again just in case we were called by SIOCGMIIREG
3303 * instead of SIOCGMIIPHY.
3310 if (mii->reg_num == 1) {
3311 struct bonding *bond = netdev_priv(bond_dev);
3313 read_lock(&bond->lock);
3314 read_lock(&bond->curr_slave_lock);
3315 if (netif_carrier_ok(bond->dev))
3316 mii->val_out = BMSR_LSTATUS;
3318 read_unlock(&bond->curr_slave_lock);
3319 read_unlock(&bond->lock);
3323 case BOND_INFO_QUERY_OLD:
3324 case SIOCBONDINFOQUERY:
3325 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3327 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3330 res = bond_info_query(bond_dev, &k_binfo);
3332 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3336 case BOND_SLAVE_INFO_QUERY_OLD:
3337 case SIOCBONDSLAVEINFOQUERY:
3338 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3340 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3343 res = bond_slave_info_query(bond_dev, &k_sinfo);
3345 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3354 net = dev_net(bond_dev);
3356 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3359 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3361 pr_debug("slave_dev=%p:\n", slave_dev);
3366 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3368 case BOND_ENSLAVE_OLD:
3369 case SIOCBONDENSLAVE:
3370 res = bond_enslave(bond_dev, slave_dev);
3372 case BOND_RELEASE_OLD:
3373 case SIOCBONDRELEASE:
3374 res = bond_release(bond_dev, slave_dev);
3376 case BOND_SETHWADDR_OLD:
3377 case SIOCBONDSETHWADDR:
3378 bond_set_dev_addr(bond_dev, slave_dev);
3381 case BOND_CHANGE_ACTIVE_OLD:
3382 case SIOCBONDCHANGEACTIVE:
3383 res = bond_ioctl_change_active(bond_dev, slave_dev);
3395 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3397 struct bonding *bond = netdev_priv(bond_dev);
3399 if (change & IFF_PROMISC)
3400 bond_set_promiscuity(bond,
3401 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3403 if (change & IFF_ALLMULTI)
3404 bond_set_allmulti(bond,
3405 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3408 static void bond_set_rx_mode(struct net_device *bond_dev)
3410 struct bonding *bond = netdev_priv(bond_dev);
3411 struct slave *slave;
3415 if (USES_PRIMARY(bond->params.mode)) {
3416 slave = rtnl_dereference(bond->curr_active_slave);
3418 dev_uc_sync(slave->dev, bond_dev);
3419 dev_mc_sync(slave->dev, bond_dev);
3422 bond_for_each_slave(bond, slave) {
3423 dev_uc_sync_multiple(slave->dev, bond_dev);
3424 dev_mc_sync_multiple(slave->dev, bond_dev);
3429 static int bond_neigh_init(struct neighbour *n)
3431 struct bonding *bond = netdev_priv(n->dev);
3432 const struct net_device_ops *slave_ops;
3433 struct neigh_parms parms;
3434 struct slave *slave;
3437 slave = bond_first_slave(bond);
3440 slave_ops = slave->dev->netdev_ops;
3441 if (!slave_ops->ndo_neigh_setup)
3444 parms.neigh_setup = NULL;
3445 parms.neigh_cleanup = NULL;
3446 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3451 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3452 * after the last slave has been detached. Assumes that all slaves
3453 * utilize the same neigh_cleanup (true at this writing as only user
3456 n->parms->neigh_cleanup = parms.neigh_cleanup;
3458 if (!parms.neigh_setup)
3461 return parms.neigh_setup(n);
3465 * The bonding ndo_neigh_setup is called at init time beofre any
3466 * slave exists. So we must declare proxy setup function which will
3467 * be used at run time to resolve the actual slave neigh param setup.
3469 * It's also called by master devices (such as vlans) to setup their
3470 * underlying devices. In that case - do nothing, we're already set up from
3473 static int bond_neigh_setup(struct net_device *dev,
3474 struct neigh_parms *parms)
3476 /* modify only our neigh_parms */
3477 if (parms->dev == dev)
3478 parms->neigh_setup = bond_neigh_init;
3484 * Change the MTU of all of a master's slaves to match the master
3486 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3488 struct bonding *bond = netdev_priv(bond_dev);
3489 struct slave *slave;
3492 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3493 (bond_dev ? bond_dev->name : "None"), new_mtu);
3495 /* Can't hold bond->lock with bh disabled here since
3496 * some base drivers panic. On the other hand we can't
3497 * hold bond->lock without bh disabled because we'll
3498 * deadlock. The only solution is to rely on the fact
3499 * that we're under rtnl_lock here, and the slaves
3500 * list won't change. This doesn't solve the problem
3501 * of setting the slave's MTU while it is
3502 * transmitting, but the assumption is that the base
3503 * driver can handle that.
3505 * TODO: figure out a way to safely iterate the slaves
3506 * list, but without holding a lock around the actual
3507 * call to the base driver.
3510 bond_for_each_slave(bond, slave) {
3511 pr_debug("s %p s->p %p c_m %p\n",
3513 bond_prev_slave(bond, slave),
3514 slave->dev->netdev_ops->ndo_change_mtu);
3516 res = dev_set_mtu(slave->dev, new_mtu);
3519 /* If we failed to set the slave's mtu to the new value
3520 * we must abort the operation even in ACTIVE_BACKUP
3521 * mode, because if we allow the backup slaves to have
3522 * different mtu values than the active slave we'll
3523 * need to change their mtu when doing a failover. That
3524 * means changing their mtu from timer context, which
3525 * is probably not a good idea.
3527 pr_debug("err %d %s\n", res, slave->dev->name);
3532 bond_dev->mtu = new_mtu;
3537 /* unwind from head to the slave that failed */
3538 bond_for_each_slave_continue_reverse(bond, slave) {
3541 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3543 pr_debug("unwind err %d dev %s\n",
3544 tmp_res, slave->dev->name);
3554 * Note that many devices must be down to change the HW address, and
3555 * downing the master releases all slaves. We can make bonds full of
3556 * bonding devices to test this, however.
3558 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3560 struct bonding *bond = netdev_priv(bond_dev);
3561 struct sockaddr *sa = addr, tmp_sa;
3562 struct slave *slave;
3565 if (bond->params.mode == BOND_MODE_ALB)
3566 return bond_alb_set_mac_address(bond_dev, addr);
3569 pr_debug("bond=%p, name=%s\n",
3570 bond, bond_dev ? bond_dev->name : "None");
3572 /* If fail_over_mac is enabled, do nothing and return success.
3573 * Returning an error causes ifenslave to fail.
3575 if (bond->params.fail_over_mac)
3578 if (!is_valid_ether_addr(sa->sa_data))
3579 return -EADDRNOTAVAIL;
3581 /* Can't hold bond->lock with bh disabled here since
3582 * some base drivers panic. On the other hand we can't
3583 * hold bond->lock without bh disabled because we'll
3584 * deadlock. The only solution is to rely on the fact
3585 * that we're under rtnl_lock here, and the slaves
3586 * list won't change. This doesn't solve the problem
3587 * of setting the slave's hw address while it is
3588 * transmitting, but the assumption is that the base
3589 * driver can handle that.
3591 * TODO: figure out a way to safely iterate the slaves
3592 * list, but without holding a lock around the actual
3593 * call to the base driver.
3596 bond_for_each_slave(bond, slave) {
3597 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3598 pr_debug("slave %p %s\n", slave, slave->dev->name);
3600 if (slave_ops->ndo_set_mac_address == NULL) {
3602 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3606 res = dev_set_mac_address(slave->dev, addr);
3608 /* TODO: consider downing the slave
3610 * User should expect communications
3611 * breakage anyway until ARP finish
3614 pr_debug("err %d %s\n", res, slave->dev->name);
3620 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3624 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3625 tmp_sa.sa_family = bond_dev->type;
3627 /* unwind from head to the slave that failed */
3628 bond_for_each_slave_continue_reverse(bond, slave) {
3631 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3633 pr_debug("unwind err %d dev %s\n",
3634 tmp_res, slave->dev->name);
3642 * bond_xmit_slave_id - transmit skb through slave with slave_id
3643 * @bond: bonding device that is transmitting
3644 * @skb: buffer to transmit
3645 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3647 * This function tries to transmit through slave with slave_id but in case
3648 * it fails, it tries to find the first available slave for transmission.
3649 * The skb is consumed in all cases, thus the function is void.
3651 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3653 struct slave *slave;
3656 /* Here we start from the slave with slave_id */
3657 bond_for_each_slave_rcu(bond, slave) {
3659 if (slave_can_tx(slave)) {
3660 bond_dev_queue_xmit(bond, skb, slave->dev);
3666 /* Here we start from the first slave up to slave_id */
3668 bond_for_each_slave_rcu(bond, slave) {
3671 if (slave_can_tx(slave)) {
3672 bond_dev_queue_xmit(bond, skb, slave->dev);
3676 /* no slave that can tx has been found */
3680 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3682 struct bonding *bond = netdev_priv(bond_dev);
3683 struct iphdr *iph = ip_hdr(skb);
3684 struct slave *slave;
3687 * Start with the curr_active_slave that joined the bond as the
3688 * default for sending IGMP traffic. For failover purposes one
3689 * needs to maintain some consistency for the interface that will
3690 * send the join/membership reports. The curr_active_slave found
3691 * will send all of this type of traffic.
3693 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3694 slave = rcu_dereference(bond->curr_active_slave);
3695 if (slave && slave_can_tx(slave))
3696 bond_dev_queue_xmit(bond, skb, slave->dev);
3698 bond_xmit_slave_id(bond, skb, 0);
3700 bond_xmit_slave_id(bond, skb,
3701 bond->rr_tx_counter++ % bond->slave_cnt);
3704 return NETDEV_TX_OK;
3708 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3709 * the bond has a usable interface.
3711 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3713 struct bonding *bond = netdev_priv(bond_dev);
3714 struct slave *slave;
3716 slave = rcu_dereference(bond->curr_active_slave);
3718 bond_dev_queue_xmit(bond, skb, slave->dev);
3722 return NETDEV_TX_OK;
3726 * In bond_xmit_xor() , we determine the output device by using a pre-
3727 * determined xmit_hash_policy(), If the selected device is not enabled,
3728 * find the next active slave.
3730 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3732 struct bonding *bond = netdev_priv(bond_dev);
3734 bond_xmit_slave_id(bond, skb,
3735 bond->xmit_hash_policy(skb, bond->slave_cnt));
3737 return NETDEV_TX_OK;
3740 /* in broadcast mode, we send everything to all usable interfaces. */
3741 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3743 struct bonding *bond = netdev_priv(bond_dev);
3744 struct slave *slave = NULL;
3746 bond_for_each_slave_rcu(bond, slave) {
3747 if (bond_is_last_slave(bond, slave))
3749 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3750 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3753 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3757 /* bond_dev_queue_xmit always returns 0 */
3758 bond_dev_queue_xmit(bond, skb2, slave->dev);
3761 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3762 bond_dev_queue_xmit(bond, skb, slave->dev);
3766 return NETDEV_TX_OK;
3769 /*------------------------- Device initialization ---------------------------*/
3771 static void bond_set_xmit_hash_policy(struct bonding *bond)
3773 switch (bond->params.xmit_policy) {
3774 case BOND_XMIT_POLICY_LAYER23:
3775 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
3777 case BOND_XMIT_POLICY_LAYER34:
3778 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
3780 case BOND_XMIT_POLICY_LAYER2:
3782 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
3788 * Lookup the slave that corresponds to a qid
3790 static inline int bond_slave_override(struct bonding *bond,
3791 struct sk_buff *skb)
3793 struct slave *slave = NULL;
3794 struct slave *check_slave;
3797 if (!skb->queue_mapping)
3800 /* Find out if any slaves have the same mapping as this skb. */
3801 bond_for_each_slave_rcu(bond, check_slave) {
3802 if (check_slave->queue_id == skb->queue_mapping) {
3803 slave = check_slave;
3808 /* If the slave isn't UP, use default transmit policy. */
3809 if (slave && slave->queue_id && IS_UP(slave->dev) &&
3810 (slave->link == BOND_LINK_UP)) {
3811 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3818 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3821 * This helper function exists to help dev_pick_tx get the correct
3822 * destination queue. Using a helper function skips a call to
3823 * skb_tx_hash and will put the skbs in the queue we expect on their
3824 * way down to the bonding driver.
3826 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3829 * Save the original txq to restore before passing to the driver
3831 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3833 if (unlikely(txq >= dev->real_num_tx_queues)) {
3835 txq -= dev->real_num_tx_queues;
3836 } while (txq >= dev->real_num_tx_queues);
3841 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3843 struct bonding *bond = netdev_priv(dev);
3845 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3846 if (!bond_slave_override(bond, skb))
3847 return NETDEV_TX_OK;
3850 switch (bond->params.mode) {
3851 case BOND_MODE_ROUNDROBIN:
3852 return bond_xmit_roundrobin(skb, dev);
3853 case BOND_MODE_ACTIVEBACKUP:
3854 return bond_xmit_activebackup(skb, dev);
3856 return bond_xmit_xor(skb, dev);
3857 case BOND_MODE_BROADCAST:
3858 return bond_xmit_broadcast(skb, dev);
3859 case BOND_MODE_8023AD:
3860 return bond_3ad_xmit_xor(skb, dev);
3863 return bond_alb_xmit(skb, dev);
3865 /* Should never happen, mode already checked */
3866 pr_err("%s: Error: Unknown bonding mode %d\n",
3867 dev->name, bond->params.mode);
3870 return NETDEV_TX_OK;
3874 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3876 struct bonding *bond = netdev_priv(dev);
3877 netdev_tx_t ret = NETDEV_TX_OK;
3880 * If we risk deadlock from transmitting this in the
3881 * netpoll path, tell netpoll to queue the frame for later tx
3883 if (is_netpoll_tx_blocked(dev))
3884 return NETDEV_TX_BUSY;
3887 if (!list_empty(&bond->slave_list))
3888 ret = __bond_start_xmit(skb, dev);
3897 * set bond mode specific net device operations
3899 void bond_set_mode_ops(struct bonding *bond, int mode)
3901 struct net_device *bond_dev = bond->dev;
3904 case BOND_MODE_ROUNDROBIN:
3906 case BOND_MODE_ACTIVEBACKUP:
3909 bond_set_xmit_hash_policy(bond);
3911 case BOND_MODE_BROADCAST:
3913 case BOND_MODE_8023AD:
3914 bond_set_xmit_hash_policy(bond);
3921 /* Should never happen, mode already checked */
3922 pr_err("%s: Error: Unknown bonding mode %d\n",
3923 bond_dev->name, mode);
3928 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3929 struct ethtool_cmd *ecmd)
3931 struct bonding *bond = netdev_priv(bond_dev);
3932 unsigned long speed = 0;
3933 struct slave *slave;
3935 ecmd->duplex = DUPLEX_UNKNOWN;
3936 ecmd->port = PORT_OTHER;
3938 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3939 * do not need to check mode. Though link speed might not represent
3940 * the true receive or transmit bandwidth (not all modes are symmetric)
3941 * this is an accurate maximum.
3943 read_lock(&bond->lock);
3944 bond_for_each_slave(bond, slave) {
3945 if (SLAVE_IS_OK(slave)) {
3946 if (slave->speed != SPEED_UNKNOWN)
3947 speed += slave->speed;
3948 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3949 slave->duplex != DUPLEX_UNKNOWN)
3950 ecmd->duplex = slave->duplex;
3953 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3954 read_unlock(&bond->lock);
3959 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3960 struct ethtool_drvinfo *drvinfo)
3962 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3963 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3964 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3968 static const struct ethtool_ops bond_ethtool_ops = {
3969 .get_drvinfo = bond_ethtool_get_drvinfo,
3970 .get_settings = bond_ethtool_get_settings,
3971 .get_link = ethtool_op_get_link,
3974 static const struct net_device_ops bond_netdev_ops = {
3975 .ndo_init = bond_init,
3976 .ndo_uninit = bond_uninit,
3977 .ndo_open = bond_open,
3978 .ndo_stop = bond_close,
3979 .ndo_start_xmit = bond_start_xmit,
3980 .ndo_select_queue = bond_select_queue,
3981 .ndo_get_stats64 = bond_get_stats,
3982 .ndo_do_ioctl = bond_do_ioctl,
3983 .ndo_change_rx_flags = bond_change_rx_flags,
3984 .ndo_set_rx_mode = bond_set_rx_mode,
3985 .ndo_change_mtu = bond_change_mtu,
3986 .ndo_set_mac_address = bond_set_mac_address,
3987 .ndo_neigh_setup = bond_neigh_setup,
3988 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3989 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3990 #ifdef CONFIG_NET_POLL_CONTROLLER
3991 .ndo_netpoll_setup = bond_netpoll_setup,
3992 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3993 .ndo_poll_controller = bond_poll_controller,
3995 .ndo_add_slave = bond_enslave,
3996 .ndo_del_slave = bond_release,
3997 .ndo_fix_features = bond_fix_features,
4000 static const struct device_type bond_type = {
4004 static void bond_destructor(struct net_device *bond_dev)
4006 struct bonding *bond = netdev_priv(bond_dev);
4008 destroy_workqueue(bond->wq);
4009 free_netdev(bond_dev);
4012 static void bond_setup(struct net_device *bond_dev)
4014 struct bonding *bond = netdev_priv(bond_dev);
4016 /* initialize rwlocks */
4017 rwlock_init(&bond->lock);
4018 rwlock_init(&bond->curr_slave_lock);
4019 INIT_LIST_HEAD(&bond->slave_list);
4020 bond->params = bonding_defaults;
4022 /* Initialize pointers */
4023 bond->dev = bond_dev;
4025 /* Initialize the device entry points */
4026 ether_setup(bond_dev);
4027 bond_dev->netdev_ops = &bond_netdev_ops;
4028 bond_dev->ethtool_ops = &bond_ethtool_ops;
4029 bond_set_mode_ops(bond, bond->params.mode);
4031 bond_dev->destructor = bond_destructor;
4033 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4035 /* Initialize the device options */
4036 bond_dev->tx_queue_len = 0;
4037 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4038 bond_dev->priv_flags |= IFF_BONDING;
4039 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4041 /* At first, we block adding VLANs. That's the only way to
4042 * prevent problems that occur when adding VLANs over an
4043 * empty bond. The block will be removed once non-challenged
4044 * slaves are enslaved.
4046 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4048 /* don't acquire bond device's netif_tx_lock when
4050 bond_dev->features |= NETIF_F_LLTX;
4052 /* By default, we declare the bond to be fully
4053 * VLAN hardware accelerated capable. Special
4054 * care is taken in the various xmit functions
4055 * when there are slaves that are not hw accel
4059 bond_dev->hw_features = BOND_VLAN_FEATURES |
4060 NETIF_F_HW_VLAN_CTAG_TX |
4061 NETIF_F_HW_VLAN_CTAG_RX |
4062 NETIF_F_HW_VLAN_CTAG_FILTER;
4064 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4065 bond_dev->features |= bond_dev->hw_features;
4069 * Destroy a bonding device.
4070 * Must be under rtnl_lock when this function is called.
4072 static void bond_uninit(struct net_device *bond_dev)
4074 struct bonding *bond = netdev_priv(bond_dev);
4075 struct slave *slave, *tmp_slave;
4077 bond_netpoll_cleanup(bond_dev);
4079 /* Release the bonded slaves */
4080 list_for_each_entry_safe(slave, tmp_slave, &bond->slave_list, list)
4081 __bond_release_one(bond_dev, slave->dev, true);
4082 pr_info("%s: released all slaves\n", bond_dev->name);
4084 list_del(&bond->bond_list);
4086 bond_debug_unregister(bond);
4089 /*------------------------- Module initialization ---------------------------*/
4092 * Convert string input module parms. Accept either the
4093 * number of the mode or its string name. A bit complicated because
4094 * some mode names are substrings of other names, and calls from sysfs
4095 * may have whitespace in the name (trailing newlines, for example).
4097 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4099 int modeint = -1, i, rv;
4100 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4102 for (p = (char *)buf; *p; p++)
4103 if (!(isdigit(*p) || isspace(*p)))
4107 rv = sscanf(buf, "%20s", modestr);
4109 rv = sscanf(buf, "%d", &modeint);
4114 for (i = 0; tbl[i].modename; i++) {
4115 if (modeint == tbl[i].mode)
4117 if (strcmp(modestr, tbl[i].modename) == 0)
4124 static int bond_check_params(struct bond_params *params)
4126 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4127 int arp_all_targets_value;
4130 * Convert string parameters.
4133 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4134 if (bond_mode == -1) {
4135 pr_err("Error: Invalid bonding mode \"%s\"\n",
4136 mode == NULL ? "NULL" : mode);
4141 if (xmit_hash_policy) {
4142 if ((bond_mode != BOND_MODE_XOR) &&
4143 (bond_mode != BOND_MODE_8023AD)) {
4144 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4145 bond_mode_name(bond_mode));
4147 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4149 if (xmit_hashtype == -1) {
4150 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4151 xmit_hash_policy == NULL ? "NULL" :
4159 if (bond_mode != BOND_MODE_8023AD) {
4160 pr_info("lacp_rate param is irrelevant in mode %s\n",
4161 bond_mode_name(bond_mode));
4163 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4164 if (lacp_fast == -1) {
4165 pr_err("Error: Invalid lacp rate \"%s\"\n",
4166 lacp_rate == NULL ? "NULL" : lacp_rate);
4173 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4174 if (params->ad_select == -1) {
4175 pr_err("Error: Invalid ad_select \"%s\"\n",
4176 ad_select == NULL ? "NULL" : ad_select);
4180 if (bond_mode != BOND_MODE_8023AD) {
4181 pr_warning("ad_select param only affects 802.3ad mode\n");
4184 params->ad_select = BOND_AD_STABLE;
4187 if (max_bonds < 0) {
4188 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4189 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4190 max_bonds = BOND_DEFAULT_MAX_BONDS;
4194 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4195 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4196 miimon = BOND_LINK_MON_INTERV;
4200 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4205 if (downdelay < 0) {
4206 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4207 downdelay, INT_MAX);
4211 if ((use_carrier != 0) && (use_carrier != 1)) {
4212 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4217 if (num_peer_notif < 0 || num_peer_notif > 255) {
4218 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4223 /* reset values for 802.3ad */
4224 if (bond_mode == BOND_MODE_8023AD) {
4226 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");
4227 pr_warning("Forcing miimon to 100msec\n");
4232 if (tx_queues < 1 || tx_queues > 255) {
4233 pr_warning("Warning: tx_queues (%d) should be between "
4234 "1 and 255, resetting to %d\n",
4235 tx_queues, BOND_DEFAULT_TX_QUEUES);
4236 tx_queues = BOND_DEFAULT_TX_QUEUES;
4239 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4240 pr_warning("Warning: all_slaves_active module parameter (%d), "
4241 "not of valid value (0/1), so it was set to "
4242 "0\n", all_slaves_active);
4243 all_slaves_active = 0;
4246 if (resend_igmp < 0 || resend_igmp > 255) {
4247 pr_warning("Warning: resend_igmp (%d) should be between "
4248 "0 and 255, resetting to %d\n",
4249 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4250 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4253 /* reset values for TLB/ALB */
4254 if ((bond_mode == BOND_MODE_TLB) ||
4255 (bond_mode == BOND_MODE_ALB)) {
4257 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");
4258 pr_warning("Forcing miimon to 100msec\n");
4263 if (bond_mode == BOND_MODE_ALB) {
4264 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",
4269 if (updelay || downdelay) {
4270 /* just warn the user the up/down delay will have
4271 * no effect since miimon is zero...
4273 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",
4274 updelay, downdelay);
4277 /* don't allow arp monitoring */
4279 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4280 miimon, arp_interval);
4284 if ((updelay % miimon) != 0) {
4285 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4287 (updelay / miimon) * miimon);
4292 if ((downdelay % miimon) != 0) {
4293 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4295 (downdelay / miimon) * miimon);
4298 downdelay /= miimon;
4301 if (arp_interval < 0) {
4302 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4303 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4304 arp_interval = BOND_LINK_ARP_INTERV;
4307 for (arp_ip_count = 0, i = 0;
4308 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4309 /* not complete check, but should be good enough to
4311 __be32 ip = in_aton(arp_ip_target[i]);
4312 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4313 ip == htonl(INADDR_BROADCAST)) {
4314 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4318 if (bond_get_targets_ip(arp_target, ip) == -1)
4319 arp_target[arp_ip_count++] = ip;
4321 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4326 if (arp_interval && !arp_ip_count) {
4327 /* don't allow arping if no arp_ip_target given... */
4328 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4334 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4335 pr_err("arp_validate only supported in active-backup mode\n");
4338 if (!arp_interval) {
4339 pr_err("arp_validate requires arp_interval\n");
4343 arp_validate_value = bond_parse_parm(arp_validate,
4345 if (arp_validate_value == -1) {
4346 pr_err("Error: invalid arp_validate \"%s\"\n",
4347 arp_validate == NULL ? "NULL" : arp_validate);
4351 arp_validate_value = 0;
4353 arp_all_targets_value = 0;
4354 if (arp_all_targets) {
4355 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4356 arp_all_targets_tbl);
4358 if (arp_all_targets_value == -1) {
4359 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4361 arp_all_targets_value = 0;
4366 pr_info("MII link monitoring set to %d ms\n", miimon);
4367 } else if (arp_interval) {
4368 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4370 arp_validate_tbl[arp_validate_value].modename,
4373 for (i = 0; i < arp_ip_count; i++)
4374 pr_info(" %s", arp_ip_target[i]);
4378 } else if (max_bonds) {
4379 /* miimon and arp_interval not set, we need one so things
4380 * work as expected, see bonding.txt for details
4382 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");
4385 if (primary && !USES_PRIMARY(bond_mode)) {
4386 /* currently, using a primary only makes sense
4387 * in active backup, TLB or ALB modes
4389 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4390 primary, bond_mode_name(bond_mode));
4394 if (primary && primary_reselect) {
4395 primary_reselect_value = bond_parse_parm(primary_reselect,
4397 if (primary_reselect_value == -1) {
4398 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4400 NULL ? "NULL" : primary_reselect);
4404 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4407 if (fail_over_mac) {
4408 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4410 if (fail_over_mac_value == -1) {
4411 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4412 arp_validate == NULL ? "NULL" : arp_validate);
4416 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4417 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4419 fail_over_mac_value = BOND_FOM_NONE;
4422 /* fill params struct with the proper values */
4423 params->mode = bond_mode;
4424 params->xmit_policy = xmit_hashtype;
4425 params->miimon = miimon;
4426 params->num_peer_notif = num_peer_notif;
4427 params->arp_interval = arp_interval;
4428 params->arp_validate = arp_validate_value;
4429 params->arp_all_targets = arp_all_targets_value;
4430 params->updelay = updelay;
4431 params->downdelay = downdelay;
4432 params->use_carrier = use_carrier;
4433 params->lacp_fast = lacp_fast;
4434 params->primary[0] = 0;
4435 params->primary_reselect = primary_reselect_value;
4436 params->fail_over_mac = fail_over_mac_value;
4437 params->tx_queues = tx_queues;
4438 params->all_slaves_active = all_slaves_active;
4439 params->resend_igmp = resend_igmp;
4440 params->min_links = min_links;
4443 strncpy(params->primary, primary, IFNAMSIZ);
4444 params->primary[IFNAMSIZ - 1] = 0;
4447 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4452 static struct lock_class_key bonding_netdev_xmit_lock_key;
4453 static struct lock_class_key bonding_netdev_addr_lock_key;
4454 static struct lock_class_key bonding_tx_busylock_key;
4456 static void bond_set_lockdep_class_one(struct net_device *dev,
4457 struct netdev_queue *txq,
4460 lockdep_set_class(&txq->_xmit_lock,
4461 &bonding_netdev_xmit_lock_key);
4464 static void bond_set_lockdep_class(struct net_device *dev)
4466 lockdep_set_class(&dev->addr_list_lock,
4467 &bonding_netdev_addr_lock_key);
4468 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4469 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4473 * Called from registration process
4475 static int bond_init(struct net_device *bond_dev)
4477 struct bonding *bond = netdev_priv(bond_dev);
4478 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4479 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4481 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4484 * Initialize locks that may be required during
4485 * en/deslave operations. All of the bond_open work
4486 * (of which this is part) should really be moved to
4487 * a phase prior to dev_open
4489 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4490 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4492 bond->wq = create_singlethread_workqueue(bond_dev->name);
4496 bond_set_lockdep_class(bond_dev);
4498 list_add_tail(&bond->bond_list, &bn->dev_list);
4500 bond_prepare_sysfs_group(bond);
4502 bond_debug_register(bond);
4504 /* Ensure valid dev_addr */
4505 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4506 bond_dev->addr_assign_type == NET_ADDR_PERM)
4507 eth_hw_addr_random(bond_dev);
4512 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4514 if (tb[IFLA_ADDRESS]) {
4515 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4517 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4518 return -EADDRNOTAVAIL;
4523 static unsigned int bond_get_num_tx_queues(void)
4528 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4530 .priv_size = sizeof(struct bonding),
4531 .setup = bond_setup,
4532 .validate = bond_validate,
4533 .get_num_tx_queues = bond_get_num_tx_queues,
4534 .get_num_rx_queues = bond_get_num_tx_queues, /* Use the same number
4538 /* Create a new bond based on the specified name and bonding parameters.
4539 * If name is NULL, obtain a suitable "bond%d" name for us.
4540 * Caller must NOT hold rtnl_lock; we need to release it here before we
4541 * set up our sysfs entries.
4543 int bond_create(struct net *net, const char *name)
4545 struct net_device *bond_dev;
4550 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4551 name ? name : "bond%d",
4552 bond_setup, tx_queues);
4554 pr_err("%s: eek! can't alloc netdev!\n", name);
4559 dev_net_set(bond_dev, net);
4560 bond_dev->rtnl_link_ops = &bond_link_ops;
4562 res = register_netdevice(bond_dev);
4564 netif_carrier_off(bond_dev);
4568 bond_destructor(bond_dev);
4572 static int __net_init bond_net_init(struct net *net)
4574 struct bond_net *bn = net_generic(net, bond_net_id);
4577 INIT_LIST_HEAD(&bn->dev_list);
4579 bond_create_proc_dir(bn);
4580 bond_create_sysfs(bn);
4585 static void __net_exit bond_net_exit(struct net *net)
4587 struct bond_net *bn = net_generic(net, bond_net_id);
4588 struct bonding *bond, *tmp_bond;
4591 bond_destroy_sysfs(bn);
4592 bond_destroy_proc_dir(bn);
4594 /* Kill off any bonds created after unregistering bond rtnl ops */
4596 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4597 unregister_netdevice_queue(bond->dev, &list);
4598 unregister_netdevice_many(&list);
4602 static struct pernet_operations bond_net_ops = {
4603 .init = bond_net_init,
4604 .exit = bond_net_exit,
4606 .size = sizeof(struct bond_net),
4609 static int __init bonding_init(void)
4614 pr_info("%s", bond_version);
4616 res = bond_check_params(&bonding_defaults);
4620 res = register_pernet_subsys(&bond_net_ops);
4624 res = rtnl_link_register(&bond_link_ops);
4628 bond_create_debugfs();
4630 for (i = 0; i < max_bonds; i++) {
4631 res = bond_create(&init_net, NULL);
4636 register_netdevice_notifier(&bond_netdev_notifier);
4640 rtnl_link_unregister(&bond_link_ops);
4642 unregister_pernet_subsys(&bond_net_ops);
4647 static void __exit bonding_exit(void)
4649 unregister_netdevice_notifier(&bond_netdev_notifier);
4651 bond_destroy_debugfs();
4653 rtnl_link_unregister(&bond_link_ops);
4654 unregister_pernet_subsys(&bond_net_ops);
4656 #ifdef CONFIG_NET_POLL_CONTROLLER
4658 * Make sure we don't have an imbalance on our netpoll blocking
4660 WARN_ON(atomic_read(&netpoll_block_tx));
4664 module_init(bonding_init);
4665 module_exit(bonding_exit);
4666 MODULE_LICENSE("GPL");
4667 MODULE_VERSION(DRV_VERSION);
4668 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4669 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4670 MODULE_ALIAS_RTNL_LINK("bond");