2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
85 /*---------------------------- Module parameters ----------------------------*/
87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 #define BOND_LINK_MON_INTERV 0
89 #define BOND_LINK_ARP_INTERV 0
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
94 static int miimon = BOND_LINK_MON_INTERV;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval = BOND_LINK_ARP_INTERV;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
114 module_param(max_bonds, int, 0);
115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
116 module_param(tx_queues, int, 0);
117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
118 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
119 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
120 "failover event (alias of num_unsol_na)");
121 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
122 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
123 "failover event (alias of num_grat_arp)");
124 module_param(miimon, int, 0);
125 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
126 module_param(updelay, int, 0);
127 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
128 module_param(downdelay, int, 0);
129 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
131 module_param(use_carrier, int, 0);
132 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
133 "0 for off, 1 for on (default)");
134 module_param(mode, charp, 0);
135 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
136 "1 for active-backup, 2 for balance-xor, "
137 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
138 "6 for balance-alb");
139 module_param(primary, charp, 0);
140 MODULE_PARM_DESC(primary, "Primary network device to use");
141 module_param(primary_reselect, charp, 0);
142 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
144 "0 for always (default), "
145 "1 for only if speed of primary is "
147 "2 for only on active slave "
149 module_param(lacp_rate, charp, 0);
150 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
151 "0 for slow, 1 for fast");
152 module_param(ad_select, charp, 0);
153 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
154 "0 for stable (default), 1 for bandwidth, "
156 module_param(min_links, int, 0);
157 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
159 module_param(xmit_hash_policy, charp, 0);
160 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
161 "0 for layer 2 (default), 1 for layer 3+4, "
163 module_param(arp_interval, int, 0);
164 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
165 module_param_array(arp_ip_target, charp, NULL, 0);
166 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
167 module_param(arp_validate, charp, 0);
168 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
169 "0 for none (default), 1 for active, "
170 "2 for backup, 3 for all");
171 module_param(arp_all_targets, charp, 0);
172 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
173 module_param(fail_over_mac, charp, 0);
174 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
175 "the same MAC; 0 for none (default), "
176 "1 for active, 2 for follow");
177 module_param(all_slaves_active, int, 0);
178 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
179 "by setting active flag for all slaves; "
180 "0 for never (default), 1 for always.");
181 module_param(resend_igmp, int, 0);
182 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
185 /*----------------------------- Global variables ----------------------------*/
187 #ifdef CONFIG_NET_POLL_CONTROLLER
188 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
191 int bond_net_id __read_mostly;
193 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
194 static int arp_ip_count;
195 static int bond_mode = BOND_MODE_ROUNDROBIN;
196 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
197 static int lacp_fast;
199 const struct bond_parm_tbl bond_lacp_tbl[] = {
200 { "slow", AD_LACP_SLOW},
201 { "fast", AD_LACP_FAST},
205 const struct bond_parm_tbl bond_mode_tbl[] = {
206 { "balance-rr", BOND_MODE_ROUNDROBIN},
207 { "active-backup", BOND_MODE_ACTIVEBACKUP},
208 { "balance-xor", BOND_MODE_XOR},
209 { "broadcast", BOND_MODE_BROADCAST},
210 { "802.3ad", BOND_MODE_8023AD},
211 { "balance-tlb", BOND_MODE_TLB},
212 { "balance-alb", BOND_MODE_ALB},
216 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
217 { "layer2", BOND_XMIT_POLICY_LAYER2},
218 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
219 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
223 const struct bond_parm_tbl arp_all_targets_tbl[] = {
224 { "any", BOND_ARP_TARGETS_ANY},
225 { "all", BOND_ARP_TARGETS_ALL},
229 const struct bond_parm_tbl arp_validate_tbl[] = {
230 { "none", BOND_ARP_VALIDATE_NONE},
231 { "active", BOND_ARP_VALIDATE_ACTIVE},
232 { "backup", BOND_ARP_VALIDATE_BACKUP},
233 { "all", BOND_ARP_VALIDATE_ALL},
237 const struct bond_parm_tbl fail_over_mac_tbl[] = {
238 { "none", BOND_FOM_NONE},
239 { "active", BOND_FOM_ACTIVE},
240 { "follow", BOND_FOM_FOLLOW},
244 const struct bond_parm_tbl pri_reselect_tbl[] = {
245 { "always", BOND_PRI_RESELECT_ALWAYS},
246 { "better", BOND_PRI_RESELECT_BETTER},
247 { "failure", BOND_PRI_RESELECT_FAILURE},
251 struct bond_parm_tbl ad_select_tbl[] = {
252 { "stable", BOND_AD_STABLE},
253 { "bandwidth", BOND_AD_BANDWIDTH},
254 { "count", BOND_AD_COUNT},
258 /*-------------------------- Forward declarations ---------------------------*/
260 static int bond_init(struct net_device *bond_dev);
261 static void bond_uninit(struct net_device *bond_dev);
263 /*---------------------------- General routines -----------------------------*/
265 const char *bond_mode_name(int mode)
267 static const char *names[] = {
268 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
269 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
270 [BOND_MODE_XOR] = "load balancing (xor)",
271 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
272 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
273 [BOND_MODE_TLB] = "transmit load balancing",
274 [BOND_MODE_ALB] = "adaptive load balancing",
277 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
283 /*---------------------------------- VLAN -----------------------------------*/
286 * bond_dev_queue_xmit - Prepare skb for xmit.
288 * @bond: bond device that got this skb for tx.
289 * @skb: hw accel VLAN tagged skb to transmit
290 * @slave_dev: slave that is supposed to xmit this skbuff
292 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
293 struct net_device *slave_dev)
295 skb->dev = slave_dev;
297 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
298 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
299 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
301 if (unlikely(netpoll_tx_running(bond->dev)))
302 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
310 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
311 * We don't protect the slave list iteration with a lock because:
312 * a. This operation is performed in IOCTL context,
313 * b. The operation is protected by the RTNL semaphore in the 8021q code,
314 * c. Holding a lock with BH disabled while directly calling a base driver
315 * entry point is generally a BAD idea.
317 * The design of synchronization/protection for this operation in the 8021q
318 * module is good for one or more VLAN devices over a single physical device
319 * and cannot be extended for a teaming solution like bonding, so there is a
320 * potential race condition here where a net device from the vlan group might
321 * be referenced (either by a base driver or the 8021q code) while it is being
322 * removed from the system. However, it turns out we're not making matters
323 * worse, and if it works for regular VLAN usage it will work here too.
327 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
328 * @bond_dev: bonding net device that got called
329 * @vid: vlan id being added
331 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
332 __be16 proto, u16 vid)
334 struct bonding *bond = netdev_priv(bond_dev);
335 struct slave *slave, *rollback_slave;
336 struct list_head *iter;
339 bond_for_each_slave(bond, slave, iter) {
340 res = vlan_vid_add(slave->dev, proto, vid);
348 /* unwind to the slave that failed */
349 bond_for_each_slave(bond, rollback_slave, iter) {
350 if (rollback_slave == slave)
353 vlan_vid_del(rollback_slave->dev, proto, vid);
360 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
361 * @bond_dev: bonding net device that got called
362 * @vid: vlan id being removed
364 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
365 __be16 proto, u16 vid)
367 struct bonding *bond = netdev_priv(bond_dev);
368 struct list_head *iter;
371 bond_for_each_slave(bond, slave, iter)
372 vlan_vid_del(slave->dev, proto, vid);
374 if (bond_is_lb(bond))
375 bond_alb_clear_vlan(bond, vid);
380 /*------------------------------- Link status -------------------------------*/
383 * Set the carrier state for the master according to the state of its
384 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
385 * do special 802.3ad magic.
387 * Returns zero if carrier state does not change, nonzero if it does.
389 static int bond_set_carrier(struct bonding *bond)
391 struct list_head *iter;
394 if (!bond_has_slaves(bond))
397 if (bond->params.mode == BOND_MODE_8023AD)
398 return bond_3ad_set_carrier(bond);
400 bond_for_each_slave(bond, slave, iter) {
401 if (slave->link == BOND_LINK_UP) {
402 if (!netif_carrier_ok(bond->dev)) {
403 netif_carrier_on(bond->dev);
411 if (netif_carrier_ok(bond->dev)) {
412 netif_carrier_off(bond->dev);
419 * Get link speed and duplex from the slave's base driver
420 * using ethtool. If for some reason the call fails or the
421 * values are invalid, set speed and duplex to -1,
424 static void bond_update_speed_duplex(struct slave *slave)
426 struct net_device *slave_dev = slave->dev;
427 struct ethtool_cmd ecmd;
431 slave->speed = SPEED_UNKNOWN;
432 slave->duplex = DUPLEX_UNKNOWN;
434 res = __ethtool_get_settings(slave_dev, &ecmd);
438 slave_speed = ethtool_cmd_speed(&ecmd);
439 if (slave_speed == 0 || slave_speed == ((__u32) -1))
442 switch (ecmd.duplex) {
450 slave->speed = slave_speed;
451 slave->duplex = ecmd.duplex;
457 * if <dev> supports MII link status reporting, check its link status.
459 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
460 * depending upon the setting of the use_carrier parameter.
462 * Return either BMSR_LSTATUS, meaning that the link is up (or we
463 * can't tell and just pretend it is), or 0, meaning that the link is
466 * If reporting is non-zero, instead of faking link up, return -1 if
467 * both ETHTOOL and MII ioctls fail (meaning the device does not
468 * support them). If use_carrier is set, return whatever it says.
469 * It'd be nice if there was a good way to tell if a driver supports
470 * netif_carrier, but there really isn't.
472 static int bond_check_dev_link(struct bonding *bond,
473 struct net_device *slave_dev, int reporting)
475 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
476 int (*ioctl)(struct net_device *, struct ifreq *, int);
478 struct mii_ioctl_data *mii;
480 if (!reporting && !netif_running(slave_dev))
483 if (bond->params.use_carrier)
484 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
486 /* Try to get link status using Ethtool first. */
487 if (slave_dev->ethtool_ops->get_link)
488 return slave_dev->ethtool_ops->get_link(slave_dev) ?
491 /* Ethtool can't be used, fallback to MII ioctls. */
492 ioctl = slave_ops->ndo_do_ioctl;
494 /* TODO: set pointer to correct ioctl on a per team member */
495 /* bases to make this more efficient. that is, once */
496 /* we determine the correct ioctl, we will always */
497 /* call it and not the others for that team */
501 * We cannot assume that SIOCGMIIPHY will also read a
502 * register; not all network drivers (e.g., e100)
506 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
507 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
509 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
510 mii->reg_num = MII_BMSR;
511 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
512 return mii->val_out & BMSR_LSTATUS;
517 * If reporting, report that either there's no dev->do_ioctl,
518 * or both SIOCGMIIREG and get_link failed (meaning that we
519 * cannot report link status). If not reporting, pretend
522 return reporting ? -1 : BMSR_LSTATUS;
525 /*----------------------------- Multicast list ------------------------------*/
528 * Push the promiscuity flag down to appropriate slaves
530 static int bond_set_promiscuity(struct bonding *bond, int inc)
532 struct list_head *iter;
535 if (USES_PRIMARY(bond->params.mode)) {
536 /* write lock already acquired */
537 if (bond->curr_active_slave) {
538 err = dev_set_promiscuity(bond->curr_active_slave->dev,
544 bond_for_each_slave(bond, slave, iter) {
545 err = dev_set_promiscuity(slave->dev, inc);
554 * Push the allmulti flag down to all slaves
556 static int bond_set_allmulti(struct bonding *bond, int inc)
558 struct list_head *iter;
561 if (USES_PRIMARY(bond->params.mode)) {
562 /* write lock already acquired */
563 if (bond->curr_active_slave) {
564 err = dev_set_allmulti(bond->curr_active_slave->dev,
570 bond_for_each_slave(bond, slave, iter) {
571 err = dev_set_allmulti(slave->dev, inc);
580 * Retrieve the list of registered multicast addresses for the bonding
581 * device and retransmit an IGMP JOIN request to the current active
584 static void bond_resend_igmp_join_requests(struct bonding *bond)
586 if (!rtnl_trylock()) {
587 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
590 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
593 /* We use curr_slave_lock to protect against concurrent access to
594 * igmp_retrans from multiple running instances of this function and
595 * bond_change_active_slave
597 write_lock_bh(&bond->curr_slave_lock);
598 if (bond->igmp_retrans > 1) {
599 bond->igmp_retrans--;
600 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
602 write_unlock_bh(&bond->curr_slave_lock);
605 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
607 struct bonding *bond = container_of(work, struct bonding,
610 bond_resend_igmp_join_requests(bond);
613 /* Flush bond's hardware addresses from slave
615 static void bond_hw_addr_flush(struct net_device *bond_dev,
616 struct net_device *slave_dev)
618 struct bonding *bond = netdev_priv(bond_dev);
620 dev_uc_unsync(slave_dev, bond_dev);
621 dev_mc_unsync(slave_dev, bond_dev);
623 if (bond->params.mode == BOND_MODE_8023AD) {
624 /* del lacpdu mc addr from mc list */
625 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
627 dev_mc_del(slave_dev, lacpdu_multicast);
631 /*--------------------------- Active slave change ---------------------------*/
633 /* Update the hardware address list and promisc/allmulti for the new and
634 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
635 * slaves up date at all times; only the USES_PRIMARY modes need to call
636 * this function to swap these settings during a failover.
638 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
639 struct slave *old_active)
644 if (bond->dev->flags & IFF_PROMISC)
645 dev_set_promiscuity(old_active->dev, -1);
647 if (bond->dev->flags & IFF_ALLMULTI)
648 dev_set_allmulti(old_active->dev, -1);
650 bond_hw_addr_flush(bond->dev, old_active->dev);
654 /* FIXME: Signal errors upstream. */
655 if (bond->dev->flags & IFF_PROMISC)
656 dev_set_promiscuity(new_active->dev, 1);
658 if (bond->dev->flags & IFF_ALLMULTI)
659 dev_set_allmulti(new_active->dev, 1);
661 netif_addr_lock_bh(bond->dev);
662 dev_uc_sync(new_active->dev, bond->dev);
663 dev_mc_sync(new_active->dev, bond->dev);
664 netif_addr_unlock_bh(bond->dev);
669 * bond_set_dev_addr - clone slave's address to bond
670 * @bond_dev: bond net device
671 * @slave_dev: slave net device
673 * Should be called with RTNL held.
675 static void bond_set_dev_addr(struct net_device *bond_dev,
676 struct net_device *slave_dev)
678 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
679 bond_dev, slave_dev, slave_dev->addr_len);
680 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
681 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
682 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
686 * bond_do_fail_over_mac
688 * Perform special MAC address swapping for fail_over_mac settings
690 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
692 static void bond_do_fail_over_mac(struct bonding *bond,
693 struct slave *new_active,
694 struct slave *old_active)
695 __releases(&bond->curr_slave_lock)
696 __releases(&bond->lock)
697 __acquires(&bond->lock)
698 __acquires(&bond->curr_slave_lock)
700 u8 tmp_mac[ETH_ALEN];
701 struct sockaddr saddr;
704 switch (bond->params.fail_over_mac) {
705 case BOND_FOM_ACTIVE:
707 write_unlock_bh(&bond->curr_slave_lock);
708 read_unlock(&bond->lock);
709 bond_set_dev_addr(bond->dev, new_active->dev);
710 read_lock(&bond->lock);
711 write_lock_bh(&bond->curr_slave_lock);
714 case BOND_FOM_FOLLOW:
716 * if new_active && old_active, swap them
717 * if just old_active, do nothing (going to no active slave)
718 * if just new_active, set new_active to bond's MAC
723 write_unlock_bh(&bond->curr_slave_lock);
724 read_unlock(&bond->lock);
727 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
728 memcpy(saddr.sa_data, old_active->dev->dev_addr,
730 saddr.sa_family = new_active->dev->type;
732 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
733 saddr.sa_family = bond->dev->type;
736 rv = dev_set_mac_address(new_active->dev, &saddr);
738 pr_err("%s: Error %d setting MAC of slave %s\n",
739 bond->dev->name, -rv, new_active->dev->name);
746 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
747 saddr.sa_family = old_active->dev->type;
749 rv = dev_set_mac_address(old_active->dev, &saddr);
751 pr_err("%s: Error %d setting MAC of slave %s\n",
752 bond->dev->name, -rv, new_active->dev->name);
754 read_lock(&bond->lock);
755 write_lock_bh(&bond->curr_slave_lock);
758 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
759 bond->dev->name, bond->params.fail_over_mac);
765 static bool bond_should_change_active(struct bonding *bond)
767 struct slave *prim = bond->primary_slave;
768 struct slave *curr = bond->curr_active_slave;
770 if (!prim || !curr || curr->link != BOND_LINK_UP)
772 if (bond->force_primary) {
773 bond->force_primary = false;
776 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
777 (prim->speed < curr->speed ||
778 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
780 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
786 * find_best_interface - select the best available slave to be the active one
787 * @bond: our bonding struct
789 static struct slave *bond_find_best_slave(struct bonding *bond)
791 struct slave *slave, *bestslave = NULL;
792 struct list_head *iter;
793 int mintime = bond->params.updelay;
795 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
796 bond_should_change_active(bond))
797 return bond->primary_slave;
799 bond_for_each_slave(bond, slave, iter) {
800 if (slave->link == BOND_LINK_UP)
802 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
803 slave->delay < mintime) {
804 mintime = slave->delay;
812 static bool bond_should_notify_peers(struct bonding *bond)
814 struct slave *slave = bond->curr_active_slave;
816 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
817 bond->dev->name, slave ? slave->dev->name : "NULL");
819 if (!slave || !bond->send_peer_notif ||
820 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
827 * change_active_interface - change the active slave into the specified one
828 * @bond: our bonding struct
829 * @new: the new slave to make the active one
831 * Set the new slave to the bond's settings and unset them on the old
833 * Setting include flags, mc-list, promiscuity, allmulti, etc.
835 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
836 * because it is apparently the best available slave we have, even though its
837 * updelay hasn't timed out yet.
839 * If new_active is not NULL, caller must hold bond->lock for read and
840 * curr_slave_lock for write_bh.
842 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
844 struct slave *old_active = bond->curr_active_slave;
846 if (old_active == new_active)
850 new_active->jiffies = jiffies;
852 if (new_active->link == BOND_LINK_BACK) {
853 if (USES_PRIMARY(bond->params.mode)) {
854 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
855 bond->dev->name, new_active->dev->name,
856 (bond->params.updelay - new_active->delay) * bond->params.miimon);
859 new_active->delay = 0;
860 new_active->link = BOND_LINK_UP;
862 if (bond->params.mode == BOND_MODE_8023AD)
863 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
865 if (bond_is_lb(bond))
866 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
868 if (USES_PRIMARY(bond->params.mode)) {
869 pr_info("%s: making interface %s the new active one.\n",
870 bond->dev->name, new_active->dev->name);
875 if (USES_PRIMARY(bond->params.mode))
876 bond_hw_addr_swap(bond, new_active, old_active);
878 if (bond_is_lb(bond)) {
879 bond_alb_handle_active_change(bond, new_active);
881 bond_set_slave_inactive_flags(old_active);
883 bond_set_slave_active_flags(new_active);
885 rcu_assign_pointer(bond->curr_active_slave, new_active);
888 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
890 bond_set_slave_inactive_flags(old_active);
893 bool should_notify_peers = false;
895 bond_set_slave_active_flags(new_active);
897 if (bond->params.fail_over_mac)
898 bond_do_fail_over_mac(bond, new_active,
901 if (netif_running(bond->dev)) {
902 bond->send_peer_notif =
903 bond->params.num_peer_notif;
904 should_notify_peers =
905 bond_should_notify_peers(bond);
908 write_unlock_bh(&bond->curr_slave_lock);
909 read_unlock(&bond->lock);
911 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
912 if (should_notify_peers)
913 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
916 read_lock(&bond->lock);
917 write_lock_bh(&bond->curr_slave_lock);
921 /* resend IGMP joins since active slave has changed or
922 * all were sent on curr_active_slave.
923 * resend only if bond is brought up with the affected
924 * bonding modes and the retransmission is enabled */
925 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
926 ((USES_PRIMARY(bond->params.mode) && new_active) ||
927 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
928 bond->igmp_retrans = bond->params.resend_igmp;
929 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
934 * bond_select_active_slave - select a new active slave, if needed
935 * @bond: our bonding struct
937 * This functions should be called when one of the following occurs:
938 * - The old curr_active_slave has been released or lost its link.
939 * - The primary_slave has got its link back.
940 * - A slave has got its link back and there's no old curr_active_slave.
942 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
944 void bond_select_active_slave(struct bonding *bond)
946 struct slave *best_slave;
949 best_slave = bond_find_best_slave(bond);
950 if (best_slave != bond->curr_active_slave) {
951 bond_change_active_slave(bond, best_slave);
952 rv = bond_set_carrier(bond);
956 if (netif_carrier_ok(bond->dev)) {
957 pr_info("%s: first active interface up!\n",
960 pr_info("%s: now running without any active interface !\n",
966 /*--------------------------- slave list handling ---------------------------*/
969 * This function attaches the slave to the end of list.
971 * bond->lock held for writing by caller.
973 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
979 * This function detaches the slave from the list.
980 * WARNING: no check is made to verify if the slave effectively
982 * Nothing is freed on return, structures are just unchained.
983 * If any slave pointer in bond was pointing to <slave>,
984 * it should be changed by the calling function.
986 * bond->lock held for writing by caller.
988 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
993 #ifdef CONFIG_NET_POLL_CONTROLLER
994 static inline int slave_enable_netpoll(struct slave *slave)
999 np = kzalloc(sizeof(*np), GFP_ATOMIC);
1004 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1013 static inline void slave_disable_netpoll(struct slave *slave)
1015 struct netpoll *np = slave->np;
1021 __netpoll_free_async(np);
1023 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1025 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1027 if (!slave_dev->netdev_ops->ndo_poll_controller)
1032 static void bond_poll_controller(struct net_device *bond_dev)
1036 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1038 struct bonding *bond = netdev_priv(bond_dev);
1039 struct list_head *iter;
1040 struct slave *slave;
1042 bond_for_each_slave(bond, slave, iter)
1043 if (IS_UP(slave->dev))
1044 slave_disable_netpoll(slave);
1047 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1049 struct bonding *bond = netdev_priv(dev);
1050 struct list_head *iter;
1051 struct slave *slave;
1054 bond_for_each_slave(bond, slave, iter) {
1055 err = slave_enable_netpoll(slave);
1057 bond_netpoll_cleanup(dev);
1064 static inline int slave_enable_netpoll(struct slave *slave)
1068 static inline void slave_disable_netpoll(struct slave *slave)
1071 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1076 /*---------------------------------- IOCTL ----------------------------------*/
1078 static netdev_features_t bond_fix_features(struct net_device *dev,
1079 netdev_features_t features)
1081 struct bonding *bond = netdev_priv(dev);
1082 struct list_head *iter;
1083 netdev_features_t mask;
1084 struct slave *slave;
1086 if (!bond_has_slaves(bond)) {
1087 /* Disable adding VLANs to empty bond. But why? --mq */
1088 features |= NETIF_F_VLAN_CHALLENGED;
1093 features &= ~NETIF_F_ONE_FOR_ALL;
1094 features |= NETIF_F_ALL_FOR_ALL;
1096 bond_for_each_slave(bond, slave, iter) {
1097 features = netdev_increment_features(features,
1098 slave->dev->features,
1101 features = netdev_add_tso_features(features, mask);
1106 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1107 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1108 NETIF_F_HIGHDMA | NETIF_F_LRO)
1110 static void bond_compute_features(struct bonding *bond)
1112 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1113 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1114 struct net_device *bond_dev = bond->dev;
1115 struct list_head *iter;
1116 struct slave *slave;
1117 unsigned short max_hard_header_len = ETH_HLEN;
1118 unsigned int gso_max_size = GSO_MAX_SIZE;
1119 u16 gso_max_segs = GSO_MAX_SEGS;
1121 if (!bond_has_slaves(bond))
1124 bond_for_each_slave(bond, slave, iter) {
1125 vlan_features = netdev_increment_features(vlan_features,
1126 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1128 dst_release_flag &= slave->dev->priv_flags;
1129 if (slave->dev->hard_header_len > max_hard_header_len)
1130 max_hard_header_len = slave->dev->hard_header_len;
1132 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1133 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1137 bond_dev->vlan_features = vlan_features;
1138 bond_dev->hard_header_len = max_hard_header_len;
1139 bond_dev->gso_max_segs = gso_max_segs;
1140 netif_set_gso_max_size(bond_dev, gso_max_size);
1142 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1143 bond_dev->priv_flags = flags | dst_release_flag;
1145 netdev_change_features(bond_dev);
1148 static void bond_setup_by_slave(struct net_device *bond_dev,
1149 struct net_device *slave_dev)
1151 bond_dev->header_ops = slave_dev->header_ops;
1153 bond_dev->type = slave_dev->type;
1154 bond_dev->hard_header_len = slave_dev->hard_header_len;
1155 bond_dev->addr_len = slave_dev->addr_len;
1157 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1158 slave_dev->addr_len);
1161 /* On bonding slaves other than the currently active slave, suppress
1162 * duplicates except for alb non-mcast/bcast.
1164 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1165 struct slave *slave,
1166 struct bonding *bond)
1168 if (bond_is_slave_inactive(slave)) {
1169 if (bond->params.mode == BOND_MODE_ALB &&
1170 skb->pkt_type != PACKET_BROADCAST &&
1171 skb->pkt_type != PACKET_MULTICAST)
1178 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1180 struct sk_buff *skb = *pskb;
1181 struct slave *slave;
1182 struct bonding *bond;
1183 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1185 int ret = RX_HANDLER_ANOTHER;
1187 skb = skb_share_check(skb, GFP_ATOMIC);
1189 return RX_HANDLER_CONSUMED;
1193 slave = bond_slave_get_rcu(skb->dev);
1196 if (bond->params.arp_interval)
1197 slave->dev->last_rx = jiffies;
1199 recv_probe = ACCESS_ONCE(bond->recv_probe);
1201 ret = recv_probe(skb, bond, slave);
1202 if (ret == RX_HANDLER_CONSUMED) {
1208 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1209 return RX_HANDLER_EXACT;
1212 skb->dev = bond->dev;
1214 if (bond->params.mode == BOND_MODE_ALB &&
1215 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1216 skb->pkt_type == PACKET_HOST) {
1218 if (unlikely(skb_cow_head(skb,
1219 skb->data - skb_mac_header(skb)))) {
1221 return RX_HANDLER_CONSUMED;
1223 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1229 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1230 struct net_device *slave_dev,
1231 struct slave *slave)
1235 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1238 slave_dev->flags |= IFF_SLAVE;
1239 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1243 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1244 struct net_device *slave_dev)
1246 netdev_upper_dev_unlink(slave_dev, bond_dev);
1247 slave_dev->flags &= ~IFF_SLAVE;
1248 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1251 /* enslave device <slave> to bond device <master> */
1252 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1254 struct bonding *bond = netdev_priv(bond_dev);
1255 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1256 struct slave *new_slave = NULL, *prev_slave;
1257 struct sockaddr addr;
1261 if (!bond->params.use_carrier &&
1262 slave_dev->ethtool_ops->get_link == NULL &&
1263 slave_ops->ndo_do_ioctl == NULL) {
1264 pr_warning("%s: Warning: no link monitoring support for %s\n",
1265 bond_dev->name, slave_dev->name);
1268 /* already enslaved */
1269 if (slave_dev->flags & IFF_SLAVE) {
1270 pr_debug("Error, Device was already enslaved\n");
1274 /* vlan challenged mutual exclusion */
1275 /* no need to lock since we're protected by rtnl_lock */
1276 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1277 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1278 if (vlan_uses_dev(bond_dev)) {
1279 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1280 bond_dev->name, slave_dev->name, bond_dev->name);
1283 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1284 bond_dev->name, slave_dev->name,
1285 slave_dev->name, bond_dev->name);
1288 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1292 * Old ifenslave binaries are no longer supported. These can
1293 * be identified with moderate accuracy by the state of the slave:
1294 * the current ifenslave will set the interface down prior to
1295 * enslaving it; the old ifenslave will not.
1297 if ((slave_dev->flags & IFF_UP)) {
1298 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1301 goto err_undo_flags;
1304 /* set bonding device ether type by slave - bonding netdevices are
1305 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1306 * there is a need to override some of the type dependent attribs/funcs.
1308 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1309 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1311 if (!bond_has_slaves(bond)) {
1312 if (bond_dev->type != slave_dev->type) {
1313 pr_debug("%s: change device type from %d to %d\n",
1315 bond_dev->type, slave_dev->type);
1317 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1319 res = notifier_to_errno(res);
1321 pr_err("%s: refused to change device type\n",
1324 goto err_undo_flags;
1327 /* Flush unicast and multicast addresses */
1328 dev_uc_flush(bond_dev);
1329 dev_mc_flush(bond_dev);
1331 if (slave_dev->type != ARPHRD_ETHER)
1332 bond_setup_by_slave(bond_dev, slave_dev);
1334 ether_setup(bond_dev);
1335 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1338 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1341 } else if (bond_dev->type != slave_dev->type) {
1342 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1344 slave_dev->type, bond_dev->type);
1346 goto err_undo_flags;
1349 if (slave_ops->ndo_set_mac_address == NULL) {
1350 if (!bond_has_slaves(bond)) {
1351 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1353 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1354 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1355 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",
1358 goto err_undo_flags;
1362 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1364 /* If this is the first slave, then we need to set the master's hardware
1365 * address to be the same as the slave's. */
1366 if (!bond_has_slaves(bond) &&
1367 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1368 bond_set_dev_addr(bond->dev, slave_dev);
1370 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1373 goto err_undo_flags;
1376 * Set the new_slave's queue_id to be zero. Queue ID mapping
1377 * is set via sysfs or module option if desired.
1379 new_slave->queue_id = 0;
1381 /* Save slave's original mtu and then set it to match the bond */
1382 new_slave->original_mtu = slave_dev->mtu;
1383 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1385 pr_debug("Error %d calling dev_set_mtu\n", res);
1390 * Save slave's original ("permanent") mac address for modes
1391 * that need it, and for restoring it upon release, and then
1392 * set it to the master's address
1394 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1396 if (!bond->params.fail_over_mac) {
1398 * Set slave to master's mac address. The application already
1399 * set the master's mac address to that of the first slave
1401 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1402 addr.sa_family = slave_dev->type;
1403 res = dev_set_mac_address(slave_dev, &addr);
1405 pr_debug("Error %d calling set_mac_address\n", res);
1406 goto err_restore_mtu;
1410 /* open the slave since the application closed it */
1411 res = dev_open(slave_dev);
1413 pr_debug("Opening slave %s failed\n", slave_dev->name);
1414 goto err_restore_mac;
1417 new_slave->bond = bond;
1418 new_slave->dev = slave_dev;
1419 slave_dev->priv_flags |= IFF_BONDING;
1421 if (bond_is_lb(bond)) {
1422 /* bond_alb_init_slave() must be called before all other stages since
1423 * it might fail and we do not want to have to undo everything
1425 res = bond_alb_init_slave(bond, new_slave);
1430 /* If the mode USES_PRIMARY, then the following is handled by
1431 * bond_change_active_slave().
1433 if (!USES_PRIMARY(bond->params.mode)) {
1434 /* set promiscuity level to new slave */
1435 if (bond_dev->flags & IFF_PROMISC) {
1436 res = dev_set_promiscuity(slave_dev, 1);
1441 /* set allmulti level to new slave */
1442 if (bond_dev->flags & IFF_ALLMULTI) {
1443 res = dev_set_allmulti(slave_dev, 1);
1448 netif_addr_lock_bh(bond_dev);
1450 dev_mc_sync_multiple(slave_dev, bond_dev);
1451 dev_uc_sync_multiple(slave_dev, bond_dev);
1453 netif_addr_unlock_bh(bond_dev);
1456 if (bond->params.mode == BOND_MODE_8023AD) {
1457 /* add lacpdu mc addr to mc list */
1458 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1460 dev_mc_add(slave_dev, lacpdu_multicast);
1463 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1465 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1466 bond_dev->name, slave_dev->name);
1470 write_lock_bh(&bond->lock);
1472 prev_slave = bond_last_slave(bond);
1473 bond_attach_slave(bond, new_slave);
1475 new_slave->delay = 0;
1476 new_slave->link_failure_count = 0;
1478 write_unlock_bh(&bond->lock);
1480 bond_compute_features(bond);
1482 bond_update_speed_duplex(new_slave);
1484 read_lock(&bond->lock);
1486 new_slave->last_arp_rx = jiffies -
1487 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1488 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1489 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1491 if (bond->params.miimon && !bond->params.use_carrier) {
1492 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1494 if ((link_reporting == -1) && !bond->params.arp_interval) {
1496 * miimon is set but a bonded network driver
1497 * does not support ETHTOOL/MII and
1498 * arp_interval is not set. Note: if
1499 * use_carrier is enabled, we will never go
1500 * here (because netif_carrier is always
1501 * supported); thus, we don't need to change
1502 * the messages for netif_carrier.
1504 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",
1505 bond_dev->name, slave_dev->name);
1506 } else if (link_reporting == -1) {
1507 /* unable get link status using mii/ethtool */
1508 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",
1509 bond_dev->name, slave_dev->name);
1513 /* check for initial state */
1514 if (bond->params.miimon) {
1515 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1516 if (bond->params.updelay) {
1517 new_slave->link = BOND_LINK_BACK;
1518 new_slave->delay = bond->params.updelay;
1520 new_slave->link = BOND_LINK_UP;
1523 new_slave->link = BOND_LINK_DOWN;
1525 } else if (bond->params.arp_interval) {
1526 new_slave->link = (netif_carrier_ok(slave_dev) ?
1527 BOND_LINK_UP : BOND_LINK_DOWN);
1529 new_slave->link = BOND_LINK_UP;
1532 if (new_slave->link != BOND_LINK_DOWN)
1533 new_slave->jiffies = jiffies;
1534 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1535 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1536 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1538 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1539 /* if there is a primary slave, remember it */
1540 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1541 bond->primary_slave = new_slave;
1542 bond->force_primary = true;
1546 write_lock_bh(&bond->curr_slave_lock);
1548 switch (bond->params.mode) {
1549 case BOND_MODE_ACTIVEBACKUP:
1550 bond_set_slave_inactive_flags(new_slave);
1551 bond_select_active_slave(bond);
1553 case BOND_MODE_8023AD:
1554 /* in 802.3ad mode, the internal mechanism
1555 * will activate the slaves in the selected
1558 bond_set_slave_inactive_flags(new_slave);
1559 /* if this is the first slave */
1561 SLAVE_AD_INFO(new_slave).id = 1;
1562 /* Initialize AD with the number of times that the AD timer is called in 1 second
1563 * can be called only after the mac address of the bond is set
1565 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1567 SLAVE_AD_INFO(new_slave).id =
1568 SLAVE_AD_INFO(prev_slave).id + 1;
1571 bond_3ad_bind_slave(new_slave);
1575 bond_set_active_slave(new_slave);
1576 bond_set_slave_inactive_flags(new_slave);
1577 bond_select_active_slave(bond);
1580 pr_debug("This slave is always active in trunk mode\n");
1582 /* always active in trunk mode */
1583 bond_set_active_slave(new_slave);
1585 /* In trunking mode there is little meaning to curr_active_slave
1586 * anyway (it holds no special properties of the bond device),
1587 * so we can change it without calling change_active_interface()
1589 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1590 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1593 } /* switch(bond_mode) */
1595 write_unlock_bh(&bond->curr_slave_lock);
1597 bond_set_carrier(bond);
1599 #ifdef CONFIG_NET_POLL_CONTROLLER
1600 slave_dev->npinfo = bond->dev->npinfo;
1601 if (slave_dev->npinfo) {
1602 if (slave_enable_netpoll(new_slave)) {
1603 read_unlock(&bond->lock);
1604 pr_info("Error, %s: master_dev is using netpoll, "
1605 "but new slave device does not support netpoll.\n",
1613 read_unlock(&bond->lock);
1615 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1618 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1622 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1624 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1625 goto err_unregister;
1629 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1630 bond_dev->name, slave_dev->name,
1631 bond_is_active_slave(new_slave) ? "n active" : " backup",
1632 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1634 /* enslave is successful */
1637 /* Undo stages on error */
1639 netdev_rx_handler_unregister(slave_dev);
1642 if (!USES_PRIMARY(bond->params.mode))
1643 bond_hw_addr_flush(bond_dev, slave_dev);
1645 vlan_vids_del_by_dev(slave_dev, bond_dev);
1646 write_lock_bh(&bond->lock);
1647 bond_detach_slave(bond, new_slave);
1648 if (bond->primary_slave == new_slave)
1649 bond->primary_slave = NULL;
1650 if (bond->curr_active_slave == new_slave) {
1651 bond_change_active_slave(bond, NULL);
1652 write_unlock_bh(&bond->lock);
1653 read_lock(&bond->lock);
1654 write_lock_bh(&bond->curr_slave_lock);
1655 bond_select_active_slave(bond);
1656 write_unlock_bh(&bond->curr_slave_lock);
1657 read_unlock(&bond->lock);
1659 write_unlock_bh(&bond->lock);
1661 slave_disable_netpoll(new_slave);
1664 slave_dev->priv_flags &= ~IFF_BONDING;
1665 dev_close(slave_dev);
1668 if (!bond->params.fail_over_mac) {
1669 /* XXX TODO - fom follow mode needs to change master's
1670 * MAC if this slave's MAC is in use by the bond, or at
1671 * least print a warning.
1673 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1674 addr.sa_family = slave_dev->type;
1675 dev_set_mac_address(slave_dev, &addr);
1679 dev_set_mtu(slave_dev, new_slave->original_mtu);
1685 bond_compute_features(bond);
1686 /* Enslave of first slave has failed and we need to fix master's mac */
1687 if (!bond_has_slaves(bond) &&
1688 ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1689 eth_hw_addr_random(bond_dev);
1695 * Try to release the slave device <slave> from the bond device <master>
1696 * It is legal to access curr_active_slave without a lock because all the function
1697 * is write-locked. If "all" is true it means that the function is being called
1698 * while destroying a bond interface and all slaves are being released.
1700 * The rules for slave state should be:
1701 * for Active/Backup:
1702 * Active stays on all backups go down
1703 * for Bonded connections:
1704 * The first up interface should be left on and all others downed.
1706 static int __bond_release_one(struct net_device *bond_dev,
1707 struct net_device *slave_dev,
1710 struct bonding *bond = netdev_priv(bond_dev);
1711 struct slave *slave, *oldcurrent;
1712 struct sockaddr addr;
1713 netdev_features_t old_features = bond_dev->features;
1715 /* slave is not a slave or master is not master of this slave */
1716 if (!(slave_dev->flags & IFF_SLAVE) ||
1717 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1718 pr_err("%s: Error: cannot release %s.\n",
1719 bond_dev->name, slave_dev->name);
1724 write_lock_bh(&bond->lock);
1726 slave = bond_get_slave_by_dev(bond, slave_dev);
1728 /* not a slave of this bond */
1729 pr_info("%s: %s not enslaved\n",
1730 bond_dev->name, slave_dev->name);
1731 write_unlock_bh(&bond->lock);
1732 unblock_netpoll_tx();
1736 write_unlock_bh(&bond->lock);
1738 bond_upper_dev_unlink(bond_dev, slave_dev);
1739 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1740 * for this slave anymore.
1742 netdev_rx_handler_unregister(slave_dev);
1743 write_lock_bh(&bond->lock);
1745 /* Inform AD package of unbinding of slave. */
1746 if (bond->params.mode == BOND_MODE_8023AD) {
1747 /* must be called before the slave is
1748 * detached from the list
1750 bond_3ad_unbind_slave(slave);
1753 pr_info("%s: releasing %s interface %s\n",
1755 bond_is_active_slave(slave) ? "active" : "backup",
1758 oldcurrent = bond->curr_active_slave;
1760 bond->current_arp_slave = NULL;
1762 /* release the slave from its bond */
1763 bond_detach_slave(bond, slave);
1765 if (!all && !bond->params.fail_over_mac) {
1766 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1767 bond_has_slaves(bond))
1768 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",
1769 bond_dev->name, slave_dev->name,
1771 bond_dev->name, slave_dev->name);
1774 if (bond->primary_slave == slave)
1775 bond->primary_slave = NULL;
1777 if (oldcurrent == slave)
1778 bond_change_active_slave(bond, NULL);
1780 if (bond_is_lb(bond)) {
1781 /* Must be called only after the slave has been
1782 * detached from the list and the curr_active_slave
1783 * has been cleared (if our_slave == old_current),
1784 * but before a new active slave is selected.
1786 write_unlock_bh(&bond->lock);
1787 bond_alb_deinit_slave(bond, slave);
1788 write_lock_bh(&bond->lock);
1792 rcu_assign_pointer(bond->curr_active_slave, NULL);
1793 } else if (oldcurrent == slave) {
1795 * Note that we hold RTNL over this sequence, so there
1796 * is no concern that another slave add/remove event
1799 write_unlock_bh(&bond->lock);
1800 read_lock(&bond->lock);
1801 write_lock_bh(&bond->curr_slave_lock);
1803 bond_select_active_slave(bond);
1805 write_unlock_bh(&bond->curr_slave_lock);
1806 read_unlock(&bond->lock);
1807 write_lock_bh(&bond->lock);
1810 if (!bond_has_slaves(bond)) {
1811 bond_set_carrier(bond);
1812 eth_hw_addr_random(bond_dev);
1814 if (vlan_uses_dev(bond_dev)) {
1815 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1816 bond_dev->name, bond_dev->name);
1817 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1822 write_unlock_bh(&bond->lock);
1823 unblock_netpoll_tx();
1826 if (!bond_has_slaves(bond)) {
1827 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1828 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1831 bond_compute_features(bond);
1832 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1833 (old_features & NETIF_F_VLAN_CHALLENGED))
1834 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1835 bond_dev->name, slave_dev->name, bond_dev->name);
1837 /* must do this from outside any spinlocks */
1838 vlan_vids_del_by_dev(slave_dev, bond_dev);
1840 /* If the mode USES_PRIMARY, then this cases was handled above by
1841 * bond_change_active_slave(..., NULL)
1843 if (!USES_PRIMARY(bond->params.mode)) {
1844 /* unset promiscuity level from slave */
1845 if (bond_dev->flags & IFF_PROMISC)
1846 dev_set_promiscuity(slave_dev, -1);
1848 /* unset allmulti level from slave */
1849 if (bond_dev->flags & IFF_ALLMULTI)
1850 dev_set_allmulti(slave_dev, -1);
1852 bond_hw_addr_flush(bond_dev, slave_dev);
1855 slave_disable_netpoll(slave);
1857 /* close slave before restoring its mac address */
1858 dev_close(slave_dev);
1860 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1861 /* restore original ("permanent") mac address */
1862 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1863 addr.sa_family = slave_dev->type;
1864 dev_set_mac_address(slave_dev, &addr);
1867 dev_set_mtu(slave_dev, slave->original_mtu);
1869 slave_dev->priv_flags &= ~IFF_BONDING;
1873 return 0; /* deletion OK */
1876 /* A wrapper used because of ndo_del_link */
1877 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1879 return __bond_release_one(bond_dev, slave_dev, false);
1883 * First release a slave and then destroy the bond if no more slaves are left.
1884 * Must be under rtnl_lock when this function is called.
1886 static int bond_release_and_destroy(struct net_device *bond_dev,
1887 struct net_device *slave_dev)
1889 struct bonding *bond = netdev_priv(bond_dev);
1892 ret = bond_release(bond_dev, slave_dev);
1893 if (ret == 0 && !bond_has_slaves(bond)) {
1894 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1895 pr_info("%s: destroying bond %s.\n",
1896 bond_dev->name, bond_dev->name);
1897 unregister_netdevice(bond_dev);
1903 * This function changes the active slave to slave <slave_dev>.
1904 * It returns -EINVAL in the following cases.
1905 * - <slave_dev> is not found in the list.
1906 * - There is not active slave now.
1907 * - <slave_dev> is already active.
1908 * - The link state of <slave_dev> is not BOND_LINK_UP.
1909 * - <slave_dev> is not running.
1910 * In these cases, this function does nothing.
1911 * In the other cases, current_slave pointer is changed and 0 is returned.
1913 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1915 struct bonding *bond = netdev_priv(bond_dev);
1916 struct slave *old_active = NULL;
1917 struct slave *new_active = NULL;
1920 if (!USES_PRIMARY(bond->params.mode))
1923 /* Verify that bond_dev is indeed the master of slave_dev */
1924 if (!(slave_dev->flags & IFF_SLAVE) ||
1925 !netdev_has_upper_dev(slave_dev, bond_dev))
1928 read_lock(&bond->lock);
1930 old_active = bond->curr_active_slave;
1931 new_active = bond_get_slave_by_dev(bond, slave_dev);
1933 * Changing to the current active: do nothing; return success.
1935 if (new_active && new_active == old_active) {
1936 read_unlock(&bond->lock);
1942 new_active->link == BOND_LINK_UP &&
1943 IS_UP(new_active->dev)) {
1945 write_lock_bh(&bond->curr_slave_lock);
1946 bond_change_active_slave(bond, new_active);
1947 write_unlock_bh(&bond->curr_slave_lock);
1948 unblock_netpoll_tx();
1952 read_unlock(&bond->lock);
1957 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1959 struct bonding *bond = netdev_priv(bond_dev);
1961 info->bond_mode = bond->params.mode;
1962 info->miimon = bond->params.miimon;
1964 read_lock(&bond->lock);
1965 info->num_slaves = bond->slave_cnt;
1966 read_unlock(&bond->lock);
1971 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1973 struct bonding *bond = netdev_priv(bond_dev);
1974 struct list_head *iter;
1975 int i = 0, res = -ENODEV;
1976 struct slave *slave;
1978 read_lock(&bond->lock);
1979 bond_for_each_slave(bond, slave, iter) {
1980 if (i++ == (int)info->slave_id) {
1982 strcpy(info->slave_name, slave->dev->name);
1983 info->link = slave->link;
1984 info->state = bond_slave_state(slave);
1985 info->link_failure_count = slave->link_failure_count;
1989 read_unlock(&bond->lock);
1994 /*-------------------------------- Monitoring -------------------------------*/
1997 static int bond_miimon_inspect(struct bonding *bond)
1999 int link_state, commit = 0;
2000 struct list_head *iter;
2001 struct slave *slave;
2002 bool ignore_updelay;
2004 ignore_updelay = !bond->curr_active_slave ? true : false;
2006 bond_for_each_slave(bond, slave, iter) {
2007 slave->new_link = BOND_LINK_NOCHANGE;
2009 link_state = bond_check_dev_link(bond, slave->dev, 0);
2011 switch (slave->link) {
2016 slave->link = BOND_LINK_FAIL;
2017 slave->delay = bond->params.downdelay;
2019 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2021 (bond->params.mode ==
2022 BOND_MODE_ACTIVEBACKUP) ?
2023 (bond_is_active_slave(slave) ?
2024 "active " : "backup ") : "",
2026 bond->params.downdelay * bond->params.miimon);
2029 case BOND_LINK_FAIL:
2032 * recovered before downdelay expired
2034 slave->link = BOND_LINK_UP;
2035 slave->jiffies = jiffies;
2036 pr_info("%s: link status up again after %d ms for interface %s.\n",
2038 (bond->params.downdelay - slave->delay) *
2039 bond->params.miimon,
2044 if (slave->delay <= 0) {
2045 slave->new_link = BOND_LINK_DOWN;
2053 case BOND_LINK_DOWN:
2057 slave->link = BOND_LINK_BACK;
2058 slave->delay = bond->params.updelay;
2061 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2062 bond->dev->name, slave->dev->name,
2063 ignore_updelay ? 0 :
2064 bond->params.updelay *
2065 bond->params.miimon);
2068 case BOND_LINK_BACK:
2070 slave->link = BOND_LINK_DOWN;
2071 pr_info("%s: link status down again after %d ms for interface %s.\n",
2073 (bond->params.updelay - slave->delay) *
2074 bond->params.miimon,
2083 if (slave->delay <= 0) {
2084 slave->new_link = BOND_LINK_UP;
2086 ignore_updelay = false;
2098 static void bond_miimon_commit(struct bonding *bond)
2100 struct list_head *iter;
2101 struct slave *slave;
2103 bond_for_each_slave(bond, slave, iter) {
2104 switch (slave->new_link) {
2105 case BOND_LINK_NOCHANGE:
2109 slave->link = BOND_LINK_UP;
2110 slave->jiffies = jiffies;
2112 if (bond->params.mode == BOND_MODE_8023AD) {
2113 /* prevent it from being the active one */
2114 bond_set_backup_slave(slave);
2115 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2116 /* make it immediately active */
2117 bond_set_active_slave(slave);
2118 } else if (slave != bond->primary_slave) {
2119 /* prevent it from being the active one */
2120 bond_set_backup_slave(slave);
2123 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2124 bond->dev->name, slave->dev->name,
2125 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2126 slave->duplex ? "full" : "half");
2128 /* notify ad that the link status has changed */
2129 if (bond->params.mode == BOND_MODE_8023AD)
2130 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2132 if (bond_is_lb(bond))
2133 bond_alb_handle_link_change(bond, slave,
2136 if (!bond->curr_active_slave ||
2137 (slave == bond->primary_slave))
2142 case BOND_LINK_DOWN:
2143 if (slave->link_failure_count < UINT_MAX)
2144 slave->link_failure_count++;
2146 slave->link = BOND_LINK_DOWN;
2148 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2149 bond->params.mode == BOND_MODE_8023AD)
2150 bond_set_slave_inactive_flags(slave);
2152 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2153 bond->dev->name, slave->dev->name);
2155 if (bond->params.mode == BOND_MODE_8023AD)
2156 bond_3ad_handle_link_change(slave,
2159 if (bond_is_lb(bond))
2160 bond_alb_handle_link_change(bond, slave,
2163 if (slave == bond->curr_active_slave)
2169 pr_err("%s: invalid new link %d on slave %s\n",
2170 bond->dev->name, slave->new_link,
2172 slave->new_link = BOND_LINK_NOCHANGE;
2180 write_lock_bh(&bond->curr_slave_lock);
2181 bond_select_active_slave(bond);
2182 write_unlock_bh(&bond->curr_slave_lock);
2183 unblock_netpoll_tx();
2186 bond_set_carrier(bond);
2192 * Really a wrapper that splits the mii monitor into two phases: an
2193 * inspection, then (if inspection indicates something needs to be done)
2194 * an acquisition of appropriate locks followed by a commit phase to
2195 * implement whatever link state changes are indicated.
2197 void bond_mii_monitor(struct work_struct *work)
2199 struct bonding *bond = container_of(work, struct bonding,
2201 bool should_notify_peers = false;
2202 unsigned long delay;
2204 read_lock(&bond->lock);
2206 delay = msecs_to_jiffies(bond->params.miimon);
2208 if (!bond_has_slaves(bond))
2211 should_notify_peers = bond_should_notify_peers(bond);
2213 if (bond_miimon_inspect(bond)) {
2214 read_unlock(&bond->lock);
2216 /* Race avoidance with bond_close cancel of workqueue */
2217 if (!rtnl_trylock()) {
2218 read_lock(&bond->lock);
2220 should_notify_peers = false;
2224 read_lock(&bond->lock);
2226 bond_miimon_commit(bond);
2228 read_unlock(&bond->lock);
2229 rtnl_unlock(); /* might sleep, hold no other locks */
2230 read_lock(&bond->lock);
2234 if (bond->params.miimon)
2235 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2237 read_unlock(&bond->lock);
2239 if (should_notify_peers) {
2240 if (!rtnl_trylock())
2242 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2247 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2249 struct net_device *upper;
2250 struct list_head *iter;
2253 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2257 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2258 if (ip == bond_confirm_addr(upper, 0, ip)) {
2269 * We go to the (large) trouble of VLAN tagging ARP frames because
2270 * switches in VLAN mode (especially if ports are configured as
2271 * "native" to a VLAN) might not pass non-tagged frames.
2273 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2275 struct sk_buff *skb;
2277 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2278 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2280 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2281 NULL, slave_dev->dev_addr, NULL);
2284 pr_err("ARP packet allocation failed\n");
2288 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2290 pr_err("failed to insert VLAN tag\n");
2298 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2300 struct net_device *upper, *vlan_upper;
2301 struct list_head *iter, *vlan_iter;
2303 __be32 *targets = bond->params.arp_targets, addr;
2306 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2307 pr_debug("basa: target %pI4\n", &targets[i]);
2309 /* Find out through which dev should the packet go */
2310 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2313 pr_debug("%s: no route to arp_ip_target %pI4\n",
2314 bond->dev->name, &targets[i]);
2320 /* bond device itself */
2321 if (rt->dst.dev == bond->dev)
2325 /* first we search only for vlan devices. for every vlan
2326 * found we verify its upper dev list, searching for the
2327 * rt->dst.dev. If found we save the tag of the vlan and
2328 * proceed to send the packet.
2332 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2334 if (!is_vlan_dev(vlan_upper))
2336 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2338 if (upper == rt->dst.dev) {
2339 vlan_id = vlan_dev_vlan_id(vlan_upper);
2346 /* if the device we're looking for is not on top of any of
2347 * our upper vlans, then just search for any dev that
2348 * matches, and in case it's a vlan - save the id
2350 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2351 if (upper == rt->dst.dev) {
2352 /* if it's a vlan - get its VID */
2353 if (is_vlan_dev(upper))
2354 vlan_id = vlan_dev_vlan_id(upper);
2362 /* Not our device - skip */
2363 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2364 bond->dev->name, &targets[i],
2365 rt->dst.dev ? rt->dst.dev->name : "NULL");
2371 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2373 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2378 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2382 if (!sip || !bond_has_this_ip(bond, tip)) {
2383 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2387 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2389 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2392 slave->last_arp_rx = jiffies;
2393 slave->target_last_arp_rx[i] = jiffies;
2396 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2397 struct slave *slave)
2399 struct arphdr *arp = (struct arphdr *)skb->data;
2400 unsigned char *arp_ptr;
2404 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2405 return RX_HANDLER_ANOTHER;
2407 read_lock(&bond->lock);
2409 if (!slave_do_arp_validate(bond, slave))
2412 alen = arp_hdr_len(bond->dev);
2414 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2415 bond->dev->name, skb->dev->name);
2417 if (alen > skb_headlen(skb)) {
2418 arp = kmalloc(alen, GFP_ATOMIC);
2421 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2425 if (arp->ar_hln != bond->dev->addr_len ||
2426 skb->pkt_type == PACKET_OTHERHOST ||
2427 skb->pkt_type == PACKET_LOOPBACK ||
2428 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2429 arp->ar_pro != htons(ETH_P_IP) ||
2433 arp_ptr = (unsigned char *)(arp + 1);
2434 arp_ptr += bond->dev->addr_len;
2435 memcpy(&sip, arp_ptr, 4);
2436 arp_ptr += 4 + bond->dev->addr_len;
2437 memcpy(&tip, arp_ptr, 4);
2439 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2440 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2441 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2445 * Backup slaves won't see the ARP reply, but do come through
2446 * here for each ARP probe (so we swap the sip/tip to validate
2447 * the probe). In a "redundant switch, common router" type of
2448 * configuration, the ARP probe will (hopefully) travel from
2449 * the active, through one switch, the router, then the other
2450 * switch before reaching the backup.
2452 * We 'trust' the arp requests if there is an active slave and
2453 * it received valid arp reply(s) after it became active. This
2454 * is done to avoid endless looping when we can't reach the
2455 * arp_ip_target and fool ourselves with our own arp requests.
2457 if (bond_is_active_slave(slave))
2458 bond_validate_arp(bond, slave, sip, tip);
2459 else if (bond->curr_active_slave &&
2460 time_after(slave_last_rx(bond, bond->curr_active_slave),
2461 bond->curr_active_slave->jiffies))
2462 bond_validate_arp(bond, slave, tip, sip);
2465 read_unlock(&bond->lock);
2466 if (arp != (struct arphdr *)skb->data)
2468 return RX_HANDLER_ANOTHER;
2471 /* function to verify if we're in the arp_interval timeslice, returns true if
2472 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2473 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2475 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2478 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2480 return time_in_range(jiffies,
2481 last_act - delta_in_ticks,
2482 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2486 * this function is called regularly to monitor each slave's link
2487 * ensuring that traffic is being sent and received when arp monitoring
2488 * is used in load-balancing mode. if the adapter has been dormant, then an
2489 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2490 * arp monitoring in active backup mode.
2492 void bond_loadbalance_arp_mon(struct work_struct *work)
2494 struct bonding *bond = container_of(work, struct bonding,
2496 struct slave *slave, *oldcurrent;
2497 struct list_head *iter;
2498 int do_failover = 0;
2500 read_lock(&bond->lock);
2502 if (!bond_has_slaves(bond))
2505 oldcurrent = bond->curr_active_slave;
2506 /* see if any of the previous devices are up now (i.e. they have
2507 * xmt and rcv traffic). the curr_active_slave does not come into
2508 * the picture unless it is null. also, slave->jiffies is not needed
2509 * here because we send an arp on each slave and give a slave as
2510 * long as it needs to get the tx/rx within the delta.
2511 * TODO: what about up/down delay in arp mode? it wasn't here before
2514 bond_for_each_slave(bond, slave, iter) {
2515 unsigned long trans_start = dev_trans_start(slave->dev);
2517 if (slave->link != BOND_LINK_UP) {
2518 if (bond_time_in_interval(bond, trans_start, 1) &&
2519 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2521 slave->link = BOND_LINK_UP;
2522 bond_set_active_slave(slave);
2524 /* primary_slave has no meaning in round-robin
2525 * mode. the window of a slave being up and
2526 * curr_active_slave being null after enslaving
2530 pr_info("%s: link status definitely up for interface %s, ",
2535 pr_info("%s: interface %s is now up\n",
2541 /* slave->link == BOND_LINK_UP */
2543 /* not all switches will respond to an arp request
2544 * when the source ip is 0, so don't take the link down
2545 * if we don't know our ip yet
2547 if (!bond_time_in_interval(bond, trans_start, 2) ||
2548 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2550 slave->link = BOND_LINK_DOWN;
2551 bond_set_backup_slave(slave);
2553 if (slave->link_failure_count < UINT_MAX)
2554 slave->link_failure_count++;
2556 pr_info("%s: interface %s is now down.\n",
2560 if (slave == oldcurrent)
2565 /* note: if switch is in round-robin mode, all links
2566 * must tx arp to ensure all links rx an arp - otherwise
2567 * links may oscillate or not come up at all; if switch is
2568 * in something like xor mode, there is nothing we can
2569 * do - all replies will be rx'ed on same link causing slaves
2570 * to be unstable during low/no traffic periods
2572 if (IS_UP(slave->dev))
2573 bond_arp_send_all(bond, slave);
2578 write_lock_bh(&bond->curr_slave_lock);
2580 bond_select_active_slave(bond);
2582 write_unlock_bh(&bond->curr_slave_lock);
2583 unblock_netpoll_tx();
2587 if (bond->params.arp_interval)
2588 queue_delayed_work(bond->wq, &bond->arp_work,
2589 msecs_to_jiffies(bond->params.arp_interval));
2591 read_unlock(&bond->lock);
2595 * Called to inspect slaves for active-backup mode ARP monitor link state
2596 * changes. Sets new_link in slaves to specify what action should take
2597 * place for the slave. Returns 0 if no changes are found, >0 if changes
2598 * to link states must be committed.
2600 * Called with bond->lock held for read.
2602 static int bond_ab_arp_inspect(struct bonding *bond)
2604 unsigned long trans_start, last_rx;
2605 struct list_head *iter;
2606 struct slave *slave;
2609 bond_for_each_slave(bond, slave, iter) {
2610 slave->new_link = BOND_LINK_NOCHANGE;
2611 last_rx = slave_last_rx(bond, slave);
2613 if (slave->link != BOND_LINK_UP) {
2614 if (bond_time_in_interval(bond, last_rx, 1)) {
2615 slave->new_link = BOND_LINK_UP;
2622 * Give slaves 2*delta after being enslaved or made
2623 * active. This avoids bouncing, as the last receive
2624 * times need a full ARP monitor cycle to be updated.
2626 if (bond_time_in_interval(bond, slave->jiffies, 2))
2630 * Backup slave is down if:
2631 * - No current_arp_slave AND
2632 * - more than 3*delta since last receive AND
2633 * - the bond has an IP address
2635 * Note: a non-null current_arp_slave indicates
2636 * the curr_active_slave went down and we are
2637 * searching for a new one; under this condition
2638 * we only take the curr_active_slave down - this
2639 * gives each slave a chance to tx/rx traffic
2640 * before being taken out
2642 if (!bond_is_active_slave(slave) &&
2643 !bond->current_arp_slave &&
2644 !bond_time_in_interval(bond, last_rx, 3)) {
2645 slave->new_link = BOND_LINK_DOWN;
2650 * Active slave is down if:
2651 * - more than 2*delta since transmitting OR
2652 * - (more than 2*delta since receive AND
2653 * the bond has an IP address)
2655 trans_start = dev_trans_start(slave->dev);
2656 if (bond_is_active_slave(slave) &&
2657 (!bond_time_in_interval(bond, trans_start, 2) ||
2658 !bond_time_in_interval(bond, last_rx, 2))) {
2659 slave->new_link = BOND_LINK_DOWN;
2668 * Called to commit link state changes noted by inspection step of
2669 * active-backup mode ARP monitor.
2671 * Called with RTNL and bond->lock for read.
2673 static void bond_ab_arp_commit(struct bonding *bond)
2675 unsigned long trans_start;
2676 struct list_head *iter;
2677 struct slave *slave;
2679 bond_for_each_slave(bond, slave, iter) {
2680 switch (slave->new_link) {
2681 case BOND_LINK_NOCHANGE:
2685 trans_start = dev_trans_start(slave->dev);
2686 if (bond->curr_active_slave != slave ||
2687 (!bond->curr_active_slave &&
2688 bond_time_in_interval(bond, trans_start, 1))) {
2689 slave->link = BOND_LINK_UP;
2690 if (bond->current_arp_slave) {
2691 bond_set_slave_inactive_flags(
2692 bond->current_arp_slave);
2693 bond->current_arp_slave = NULL;
2696 pr_info("%s: link status definitely up for interface %s.\n",
2697 bond->dev->name, slave->dev->name);
2699 if (!bond->curr_active_slave ||
2700 (slave == bond->primary_slave))
2707 case BOND_LINK_DOWN:
2708 if (slave->link_failure_count < UINT_MAX)
2709 slave->link_failure_count++;
2711 slave->link = BOND_LINK_DOWN;
2712 bond_set_slave_inactive_flags(slave);
2714 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2715 bond->dev->name, slave->dev->name);
2717 if (slave == bond->curr_active_slave) {
2718 bond->current_arp_slave = NULL;
2725 pr_err("%s: impossible: new_link %d on slave %s\n",
2726 bond->dev->name, slave->new_link,
2734 write_lock_bh(&bond->curr_slave_lock);
2735 bond_select_active_slave(bond);
2736 write_unlock_bh(&bond->curr_slave_lock);
2737 unblock_netpoll_tx();
2740 bond_set_carrier(bond);
2744 * Send ARP probes for active-backup mode ARP monitor.
2746 * Called with bond->lock held for read.
2748 static void bond_ab_arp_probe(struct bonding *bond)
2750 struct slave *slave, *before = NULL, *new_slave = NULL;
2751 struct list_head *iter;
2754 read_lock(&bond->curr_slave_lock);
2756 if (bond->current_arp_slave && bond->curr_active_slave)
2757 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2758 bond->current_arp_slave->dev->name,
2759 bond->curr_active_slave->dev->name);
2761 if (bond->curr_active_slave) {
2762 bond_arp_send_all(bond, bond->curr_active_slave);
2763 read_unlock(&bond->curr_slave_lock);
2767 read_unlock(&bond->curr_slave_lock);
2769 /* if we don't have a curr_active_slave, search for the next available
2770 * backup slave from the current_arp_slave and make it the candidate
2771 * for becoming the curr_active_slave
2774 if (!bond->current_arp_slave) {
2775 bond->current_arp_slave = bond_first_slave(bond);
2776 if (!bond->current_arp_slave)
2780 bond_set_slave_inactive_flags(bond->current_arp_slave);
2782 bond_for_each_slave(bond, slave, iter) {
2783 if (!found && !before && IS_UP(slave->dev))
2786 if (found && !new_slave && IS_UP(slave->dev))
2788 /* if the link state is up at this point, we
2789 * mark it down - this can happen if we have
2790 * simultaneous link failures and
2791 * reselect_active_interface doesn't make this
2792 * one the current slave so it is still marked
2793 * up when it is actually down
2795 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2796 slave->link = BOND_LINK_DOWN;
2797 if (slave->link_failure_count < UINT_MAX)
2798 slave->link_failure_count++;
2800 bond_set_slave_inactive_flags(slave);
2802 pr_info("%s: backup interface %s is now down.\n",
2803 bond->dev->name, slave->dev->name);
2805 if (slave == bond->current_arp_slave)
2809 if (!new_slave && before)
2815 new_slave->link = BOND_LINK_BACK;
2816 bond_set_slave_active_flags(new_slave);
2817 bond_arp_send_all(bond, new_slave);
2818 new_slave->jiffies = jiffies;
2819 bond->current_arp_slave = new_slave;
2823 void bond_activebackup_arp_mon(struct work_struct *work)
2825 struct bonding *bond = container_of(work, struct bonding,
2827 bool should_notify_peers = false;
2830 read_lock(&bond->lock);
2832 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2834 if (!bond_has_slaves(bond))
2837 should_notify_peers = bond_should_notify_peers(bond);
2839 if (bond_ab_arp_inspect(bond)) {
2840 read_unlock(&bond->lock);
2842 /* Race avoidance with bond_close flush of workqueue */
2843 if (!rtnl_trylock()) {
2844 read_lock(&bond->lock);
2846 should_notify_peers = false;
2850 read_lock(&bond->lock);
2852 bond_ab_arp_commit(bond);
2854 read_unlock(&bond->lock);
2856 read_lock(&bond->lock);
2859 bond_ab_arp_probe(bond);
2862 if (bond->params.arp_interval)
2863 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2865 read_unlock(&bond->lock);
2867 if (should_notify_peers) {
2868 if (!rtnl_trylock())
2870 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2875 /*-------------------------- netdev event handling --------------------------*/
2878 * Change device name
2880 static int bond_event_changename(struct bonding *bond)
2882 bond_remove_proc_entry(bond);
2883 bond_create_proc_entry(bond);
2885 bond_debug_reregister(bond);
2890 static int bond_master_netdev_event(unsigned long event,
2891 struct net_device *bond_dev)
2893 struct bonding *event_bond = netdev_priv(bond_dev);
2896 case NETDEV_CHANGENAME:
2897 return bond_event_changename(event_bond);
2898 case NETDEV_UNREGISTER:
2899 bond_remove_proc_entry(event_bond);
2901 case NETDEV_REGISTER:
2902 bond_create_proc_entry(event_bond);
2904 case NETDEV_NOTIFY_PEERS:
2905 if (event_bond->send_peer_notif)
2906 event_bond->send_peer_notif--;
2915 static int bond_slave_netdev_event(unsigned long event,
2916 struct net_device *slave_dev)
2918 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2919 struct bonding *bond;
2920 struct net_device *bond_dev;
2924 /* A netdev event can be generated while enslaving a device
2925 * before netdev_rx_handler_register is called in which case
2926 * slave will be NULL
2930 bond_dev = slave->bond->dev;
2934 case NETDEV_UNREGISTER:
2935 if (bond_dev->type != ARPHRD_ETHER)
2936 bond_release_and_destroy(bond_dev, slave_dev);
2938 bond_release(bond_dev, slave_dev);
2942 old_speed = slave->speed;
2943 old_duplex = slave->duplex;
2945 bond_update_speed_duplex(slave);
2947 if (bond->params.mode == BOND_MODE_8023AD) {
2948 if (old_speed != slave->speed)
2949 bond_3ad_adapter_speed_changed(slave);
2950 if (old_duplex != slave->duplex)
2951 bond_3ad_adapter_duplex_changed(slave);
2956 * ... Or is it this?
2959 case NETDEV_CHANGEMTU:
2961 * TODO: Should slaves be allowed to
2962 * independently alter their MTU? For
2963 * an active-backup bond, slaves need
2964 * not be the same type of device, so
2965 * MTUs may vary. For other modes,
2966 * slaves arguably should have the
2967 * same MTUs. To do this, we'd need to
2968 * take over the slave's change_mtu
2969 * function for the duration of their
2973 case NETDEV_CHANGENAME:
2975 * TODO: handle changing the primary's name
2978 case NETDEV_FEAT_CHANGE:
2979 bond_compute_features(bond);
2981 case NETDEV_RESEND_IGMP:
2982 /* Propagate to master device */
2983 call_netdevice_notifiers(event, slave->bond->dev);
2993 * bond_netdev_event: handle netdev notifier chain events.
2995 * This function receives events for the netdev chain. The caller (an
2996 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2997 * locks for us to safely manipulate the slave devices (RTNL lock,
3000 static int bond_netdev_event(struct notifier_block *this,
3001 unsigned long event, void *ptr)
3003 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3005 pr_debug("event_dev: %s, event: %lx\n",
3006 event_dev ? event_dev->name : "None",
3009 if (!(event_dev->priv_flags & IFF_BONDING))
3012 if (event_dev->flags & IFF_MASTER) {
3013 pr_debug("IFF_MASTER\n");
3014 return bond_master_netdev_event(event, event_dev);
3017 if (event_dev->flags & IFF_SLAVE) {
3018 pr_debug("IFF_SLAVE\n");
3019 return bond_slave_netdev_event(event, event_dev);
3025 static struct notifier_block bond_netdev_notifier = {
3026 .notifier_call = bond_netdev_event,
3029 /*---------------------------- Hashing Policies -----------------------------*/
3032 * Hash for the output device based upon layer 2 data
3034 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3036 struct ethhdr *data = (struct ethhdr *)skb->data;
3038 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3039 return (data->h_dest[5] ^ data->h_source[5]) % count;
3045 * Hash for the output device based upon layer 2 and layer 3 data. If
3046 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3048 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3050 const struct ethhdr *data;
3051 const struct iphdr *iph;
3052 const struct ipv6hdr *ipv6h;
3054 const __be32 *s, *d;
3056 if (skb->protocol == htons(ETH_P_IP) &&
3057 pskb_network_may_pull(skb, sizeof(*iph))) {
3059 data = (struct ethhdr *)skb->data;
3060 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3061 (data->h_dest[5] ^ data->h_source[5])) % count;
3062 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3063 pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3064 ipv6h = ipv6_hdr(skb);
3065 data = (struct ethhdr *)skb->data;
3066 s = &ipv6h->saddr.s6_addr32[0];
3067 d = &ipv6h->daddr.s6_addr32[0];
3068 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3069 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3070 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3073 return bond_xmit_hash_policy_l2(skb, count);
3077 * Hash for the output device based upon layer 3 and layer 4 data. If
3078 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3079 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3081 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3084 const struct iphdr *iph;
3085 const struct ipv6hdr *ipv6h;
3086 const __be32 *s, *d;
3087 const __be16 *l4 = NULL;
3089 int noff = skb_network_offset(skb);
3092 if (skb->protocol == htons(ETH_P_IP) &&
3093 pskb_may_pull(skb, noff + sizeof(*iph))) {
3095 poff = proto_ports_offset(iph->protocol);
3097 if (!ip_is_fragment(iph) && poff >= 0) {
3098 l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3101 layer4_xor = ntohs(l4[0] ^ l4[1]);
3103 return (layer4_xor ^
3104 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3105 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3106 pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3107 ipv6h = ipv6_hdr(skb);
3108 poff = proto_ports_offset(ipv6h->nexthdr);
3110 l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3113 layer4_xor = ntohs(l4[0] ^ l4[1]);
3115 s = &ipv6h->saddr.s6_addr32[0];
3116 d = &ipv6h->daddr.s6_addr32[0];
3117 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3118 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3120 return layer4_xor % count;
3123 return bond_xmit_hash_policy_l2(skb, count);
3126 /*-------------------------- Device entry points ----------------------------*/
3128 static void bond_work_init_all(struct bonding *bond)
3130 INIT_DELAYED_WORK(&bond->mcast_work,
3131 bond_resend_igmp_join_requests_delayed);
3132 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3133 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3134 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3135 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3137 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3138 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3141 static void bond_work_cancel_all(struct bonding *bond)
3143 cancel_delayed_work_sync(&bond->mii_work);
3144 cancel_delayed_work_sync(&bond->arp_work);
3145 cancel_delayed_work_sync(&bond->alb_work);
3146 cancel_delayed_work_sync(&bond->ad_work);
3147 cancel_delayed_work_sync(&bond->mcast_work);
3150 static int bond_open(struct net_device *bond_dev)
3152 struct bonding *bond = netdev_priv(bond_dev);
3153 struct list_head *iter;
3154 struct slave *slave;
3156 /* reset slave->backup and slave->inactive */
3157 read_lock(&bond->lock);
3158 if (bond_has_slaves(bond)) {
3159 read_lock(&bond->curr_slave_lock);
3160 bond_for_each_slave(bond, slave, iter) {
3161 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3162 && (slave != bond->curr_active_slave)) {
3163 bond_set_slave_inactive_flags(slave);
3165 bond_set_slave_active_flags(slave);
3168 read_unlock(&bond->curr_slave_lock);
3170 read_unlock(&bond->lock);
3172 bond_work_init_all(bond);
3174 if (bond_is_lb(bond)) {
3175 /* bond_alb_initialize must be called before the timer
3178 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3180 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3183 if (bond->params.miimon) /* link check interval, in milliseconds. */
3184 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3186 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3187 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3188 if (bond->params.arp_validate)
3189 bond->recv_probe = bond_arp_rcv;
3192 if (bond->params.mode == BOND_MODE_8023AD) {
3193 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3194 /* register to receive LACPDUs */
3195 bond->recv_probe = bond_3ad_lacpdu_recv;
3196 bond_3ad_initiate_agg_selection(bond, 1);
3202 static int bond_close(struct net_device *bond_dev)
3204 struct bonding *bond = netdev_priv(bond_dev);
3206 bond_work_cancel_all(bond);
3207 bond->send_peer_notif = 0;
3208 if (bond_is_lb(bond))
3209 bond_alb_deinitialize(bond);
3210 bond->recv_probe = NULL;
3215 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3216 struct rtnl_link_stats64 *stats)
3218 struct bonding *bond = netdev_priv(bond_dev);
3219 struct rtnl_link_stats64 temp;
3220 struct list_head *iter;
3221 struct slave *slave;
3223 memset(stats, 0, sizeof(*stats));
3225 read_lock_bh(&bond->lock);
3226 bond_for_each_slave(bond, slave, iter) {
3227 const struct rtnl_link_stats64 *sstats =
3228 dev_get_stats(slave->dev, &temp);
3230 stats->rx_packets += sstats->rx_packets;
3231 stats->rx_bytes += sstats->rx_bytes;
3232 stats->rx_errors += sstats->rx_errors;
3233 stats->rx_dropped += sstats->rx_dropped;
3235 stats->tx_packets += sstats->tx_packets;
3236 stats->tx_bytes += sstats->tx_bytes;
3237 stats->tx_errors += sstats->tx_errors;
3238 stats->tx_dropped += sstats->tx_dropped;
3240 stats->multicast += sstats->multicast;
3241 stats->collisions += sstats->collisions;
3243 stats->rx_length_errors += sstats->rx_length_errors;
3244 stats->rx_over_errors += sstats->rx_over_errors;
3245 stats->rx_crc_errors += sstats->rx_crc_errors;
3246 stats->rx_frame_errors += sstats->rx_frame_errors;
3247 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3248 stats->rx_missed_errors += sstats->rx_missed_errors;
3250 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3251 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3252 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3253 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3254 stats->tx_window_errors += sstats->tx_window_errors;
3256 read_unlock_bh(&bond->lock);
3261 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3263 struct net_device *slave_dev = NULL;
3264 struct ifbond k_binfo;
3265 struct ifbond __user *u_binfo = NULL;
3266 struct ifslave k_sinfo;
3267 struct ifslave __user *u_sinfo = NULL;
3268 struct mii_ioctl_data *mii = NULL;
3272 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3284 * We do this again just in case we were called by SIOCGMIIREG
3285 * instead of SIOCGMIIPHY.
3292 if (mii->reg_num == 1) {
3293 struct bonding *bond = netdev_priv(bond_dev);
3295 read_lock(&bond->lock);
3296 read_lock(&bond->curr_slave_lock);
3297 if (netif_carrier_ok(bond->dev))
3298 mii->val_out = BMSR_LSTATUS;
3300 read_unlock(&bond->curr_slave_lock);
3301 read_unlock(&bond->lock);
3305 case BOND_INFO_QUERY_OLD:
3306 case SIOCBONDINFOQUERY:
3307 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3309 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3312 res = bond_info_query(bond_dev, &k_binfo);
3314 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3318 case BOND_SLAVE_INFO_QUERY_OLD:
3319 case SIOCBONDSLAVEINFOQUERY:
3320 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3322 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3325 res = bond_slave_info_query(bond_dev, &k_sinfo);
3327 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3336 net = dev_net(bond_dev);
3338 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3341 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3343 pr_debug("slave_dev=%p:\n", slave_dev);
3348 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3350 case BOND_ENSLAVE_OLD:
3351 case SIOCBONDENSLAVE:
3352 res = bond_enslave(bond_dev, slave_dev);
3354 case BOND_RELEASE_OLD:
3355 case SIOCBONDRELEASE:
3356 res = bond_release(bond_dev, slave_dev);
3358 case BOND_SETHWADDR_OLD:
3359 case SIOCBONDSETHWADDR:
3360 bond_set_dev_addr(bond_dev, slave_dev);
3363 case BOND_CHANGE_ACTIVE_OLD:
3364 case SIOCBONDCHANGEACTIVE:
3365 res = bond_ioctl_change_active(bond_dev, slave_dev);
3377 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3379 struct bonding *bond = netdev_priv(bond_dev);
3381 if (change & IFF_PROMISC)
3382 bond_set_promiscuity(bond,
3383 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3385 if (change & IFF_ALLMULTI)
3386 bond_set_allmulti(bond,
3387 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3390 static void bond_set_rx_mode(struct net_device *bond_dev)
3392 struct bonding *bond = netdev_priv(bond_dev);
3393 struct list_head *iter;
3394 struct slave *slave;
3398 if (USES_PRIMARY(bond->params.mode)) {
3399 slave = rtnl_dereference(bond->curr_active_slave);
3401 dev_uc_sync(slave->dev, bond_dev);
3402 dev_mc_sync(slave->dev, bond_dev);
3405 bond_for_each_slave(bond, slave, iter) {
3406 dev_uc_sync_multiple(slave->dev, bond_dev);
3407 dev_mc_sync_multiple(slave->dev, bond_dev);
3412 static int bond_neigh_init(struct neighbour *n)
3414 struct bonding *bond = netdev_priv(n->dev);
3415 const struct net_device_ops *slave_ops;
3416 struct neigh_parms parms;
3417 struct slave *slave;
3420 slave = bond_first_slave(bond);
3423 slave_ops = slave->dev->netdev_ops;
3424 if (!slave_ops->ndo_neigh_setup)
3427 parms.neigh_setup = NULL;
3428 parms.neigh_cleanup = NULL;
3429 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3434 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3435 * after the last slave has been detached. Assumes that all slaves
3436 * utilize the same neigh_cleanup (true at this writing as only user
3439 n->parms->neigh_cleanup = parms.neigh_cleanup;
3441 if (!parms.neigh_setup)
3444 return parms.neigh_setup(n);
3448 * The bonding ndo_neigh_setup is called at init time beofre any
3449 * slave exists. So we must declare proxy setup function which will
3450 * be used at run time to resolve the actual slave neigh param setup.
3452 * It's also called by master devices (such as vlans) to setup their
3453 * underlying devices. In that case - do nothing, we're already set up from
3456 static int bond_neigh_setup(struct net_device *dev,
3457 struct neigh_parms *parms)
3459 /* modify only our neigh_parms */
3460 if (parms->dev == dev)
3461 parms->neigh_setup = bond_neigh_init;
3467 * Change the MTU of all of a master's slaves to match the master
3469 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3471 struct bonding *bond = netdev_priv(bond_dev);
3472 struct slave *slave, *rollback_slave;
3473 struct list_head *iter;
3476 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3477 (bond_dev ? bond_dev->name : "None"), new_mtu);
3479 /* Can't hold bond->lock with bh disabled here since
3480 * some base drivers panic. On the other hand we can't
3481 * hold bond->lock without bh disabled because we'll
3482 * deadlock. The only solution is to rely on the fact
3483 * that we're under rtnl_lock here, and the slaves
3484 * list won't change. This doesn't solve the problem
3485 * of setting the slave's MTU while it is
3486 * transmitting, but the assumption is that the base
3487 * driver can handle that.
3489 * TODO: figure out a way to safely iterate the slaves
3490 * list, but without holding a lock around the actual
3491 * call to the base driver.
3494 bond_for_each_slave(bond, slave, iter) {
3495 pr_debug("s %p c_m %p\n",
3497 slave->dev->netdev_ops->ndo_change_mtu);
3499 res = dev_set_mtu(slave->dev, new_mtu);
3502 /* If we failed to set the slave's mtu to the new value
3503 * we must abort the operation even in ACTIVE_BACKUP
3504 * mode, because if we allow the backup slaves to have
3505 * different mtu values than the active slave we'll
3506 * need to change their mtu when doing a failover. That
3507 * means changing their mtu from timer context, which
3508 * is probably not a good idea.
3510 pr_debug("err %d %s\n", res, slave->dev->name);
3515 bond_dev->mtu = new_mtu;
3520 /* unwind from head to the slave that failed */
3521 bond_for_each_slave(bond, rollback_slave, iter) {
3524 if (rollback_slave == slave)
3527 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3529 pr_debug("unwind err %d dev %s\n",
3530 tmp_res, rollback_slave->dev->name);
3540 * Note that many devices must be down to change the HW address, and
3541 * downing the master releases all slaves. We can make bonds full of
3542 * bonding devices to test this, however.
3544 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3546 struct bonding *bond = netdev_priv(bond_dev);
3547 struct slave *slave, *rollback_slave;
3548 struct sockaddr *sa = addr, tmp_sa;
3549 struct list_head *iter;
3552 if (bond->params.mode == BOND_MODE_ALB)
3553 return bond_alb_set_mac_address(bond_dev, addr);
3556 pr_debug("bond=%p, name=%s\n",
3557 bond, bond_dev ? bond_dev->name : "None");
3559 /* If fail_over_mac is enabled, do nothing and return success.
3560 * Returning an error causes ifenslave to fail.
3562 if (bond->params.fail_over_mac)
3565 if (!is_valid_ether_addr(sa->sa_data))
3566 return -EADDRNOTAVAIL;
3568 /* Can't hold bond->lock with bh disabled here since
3569 * some base drivers panic. On the other hand we can't
3570 * hold bond->lock without bh disabled because we'll
3571 * deadlock. The only solution is to rely on the fact
3572 * that we're under rtnl_lock here, and the slaves
3573 * list won't change. This doesn't solve the problem
3574 * of setting the slave's hw address while it is
3575 * transmitting, but the assumption is that the base
3576 * driver can handle that.
3578 * TODO: figure out a way to safely iterate the slaves
3579 * list, but without holding a lock around the actual
3580 * call to the base driver.
3583 bond_for_each_slave(bond, slave, iter) {
3584 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3585 pr_debug("slave %p %s\n", slave, slave->dev->name);
3587 if (slave_ops->ndo_set_mac_address == NULL) {
3589 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3593 res = dev_set_mac_address(slave->dev, addr);
3595 /* TODO: consider downing the slave
3597 * User should expect communications
3598 * breakage anyway until ARP finish
3601 pr_debug("err %d %s\n", res, slave->dev->name);
3607 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3611 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3612 tmp_sa.sa_family = bond_dev->type;
3614 /* unwind from head to the slave that failed */
3615 bond_for_each_slave(bond, rollback_slave, iter) {
3618 if (rollback_slave == slave)
3621 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3623 pr_debug("unwind err %d dev %s\n",
3624 tmp_res, rollback_slave->dev->name);
3632 * bond_xmit_slave_id - transmit skb through slave with slave_id
3633 * @bond: bonding device that is transmitting
3634 * @skb: buffer to transmit
3635 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3637 * This function tries to transmit through slave with slave_id but in case
3638 * it fails, it tries to find the first available slave for transmission.
3639 * The skb is consumed in all cases, thus the function is void.
3641 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3643 struct list_head *iter;
3644 struct slave *slave;
3647 /* Here we start from the slave with slave_id */
3648 bond_for_each_slave_rcu(bond, slave, iter) {
3650 if (slave_can_tx(slave)) {
3651 bond_dev_queue_xmit(bond, skb, slave->dev);
3657 /* Here we start from the first slave up to slave_id */
3659 bond_for_each_slave_rcu(bond, slave, iter) {
3662 if (slave_can_tx(slave)) {
3663 bond_dev_queue_xmit(bond, skb, slave->dev);
3667 /* no slave that can tx has been found */
3671 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3673 struct bonding *bond = netdev_priv(bond_dev);
3674 struct iphdr *iph = ip_hdr(skb);
3675 struct slave *slave;
3678 * Start with the curr_active_slave that joined the bond as the
3679 * default for sending IGMP traffic. For failover purposes one
3680 * needs to maintain some consistency for the interface that will
3681 * send the join/membership reports. The curr_active_slave found
3682 * will send all of this type of traffic.
3684 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3685 slave = rcu_dereference(bond->curr_active_slave);
3686 if (slave && slave_can_tx(slave))
3687 bond_dev_queue_xmit(bond, skb, slave->dev);
3689 bond_xmit_slave_id(bond, skb, 0);
3691 bond_xmit_slave_id(bond, skb,
3692 bond->rr_tx_counter++ % bond->slave_cnt);
3695 return NETDEV_TX_OK;
3699 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3700 * the bond has a usable interface.
3702 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3704 struct bonding *bond = netdev_priv(bond_dev);
3705 struct slave *slave;
3707 slave = rcu_dereference(bond->curr_active_slave);
3709 bond_dev_queue_xmit(bond, skb, slave->dev);
3713 return NETDEV_TX_OK;
3717 * In bond_xmit_xor() , we determine the output device by using a pre-
3718 * determined xmit_hash_policy(), If the selected device is not enabled,
3719 * find the next active slave.
3721 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3723 struct bonding *bond = netdev_priv(bond_dev);
3725 bond_xmit_slave_id(bond, skb,
3726 bond->xmit_hash_policy(skb, bond->slave_cnt));
3728 return NETDEV_TX_OK;
3731 /* in broadcast mode, we send everything to all usable interfaces. */
3732 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3734 struct bonding *bond = netdev_priv(bond_dev);
3735 struct slave *slave = NULL;
3736 struct list_head *iter;
3738 bond_for_each_slave_rcu(bond, slave, iter) {
3739 if (bond_is_last_slave(bond, slave))
3741 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3742 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3745 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3749 /* bond_dev_queue_xmit always returns 0 */
3750 bond_dev_queue_xmit(bond, skb2, slave->dev);
3753 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3754 bond_dev_queue_xmit(bond, skb, slave->dev);
3758 return NETDEV_TX_OK;
3761 /*------------------------- Device initialization ---------------------------*/
3763 static void bond_set_xmit_hash_policy(struct bonding *bond)
3765 switch (bond->params.xmit_policy) {
3766 case BOND_XMIT_POLICY_LAYER23:
3767 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
3769 case BOND_XMIT_POLICY_LAYER34:
3770 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
3772 case BOND_XMIT_POLICY_LAYER2:
3774 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
3780 * Lookup the slave that corresponds to a qid
3782 static inline int bond_slave_override(struct bonding *bond,
3783 struct sk_buff *skb)
3785 struct slave *slave = NULL;
3786 struct slave *check_slave;
3787 struct list_head *iter;
3790 if (!skb->queue_mapping)
3793 /* Find out if any slaves have the same mapping as this skb. */
3794 bond_for_each_slave_rcu(bond, check_slave, iter) {
3795 if (check_slave->queue_id == skb->queue_mapping) {
3796 slave = check_slave;
3801 /* If the slave isn't UP, use default transmit policy. */
3802 if (slave && slave->queue_id && IS_UP(slave->dev) &&
3803 (slave->link == BOND_LINK_UP)) {
3804 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3811 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3814 * This helper function exists to help dev_pick_tx get the correct
3815 * destination queue. Using a helper function skips a call to
3816 * skb_tx_hash and will put the skbs in the queue we expect on their
3817 * way down to the bonding driver.
3819 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3822 * Save the original txq to restore before passing to the driver
3824 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3826 if (unlikely(txq >= dev->real_num_tx_queues)) {
3828 txq -= dev->real_num_tx_queues;
3829 } while (txq >= dev->real_num_tx_queues);
3834 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3836 struct bonding *bond = netdev_priv(dev);
3838 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3839 if (!bond_slave_override(bond, skb))
3840 return NETDEV_TX_OK;
3843 switch (bond->params.mode) {
3844 case BOND_MODE_ROUNDROBIN:
3845 return bond_xmit_roundrobin(skb, dev);
3846 case BOND_MODE_ACTIVEBACKUP:
3847 return bond_xmit_activebackup(skb, dev);
3849 return bond_xmit_xor(skb, dev);
3850 case BOND_MODE_BROADCAST:
3851 return bond_xmit_broadcast(skb, dev);
3852 case BOND_MODE_8023AD:
3853 return bond_3ad_xmit_xor(skb, dev);
3856 return bond_alb_xmit(skb, dev);
3858 /* Should never happen, mode already checked */
3859 pr_err("%s: Error: Unknown bonding mode %d\n",
3860 dev->name, bond->params.mode);
3863 return NETDEV_TX_OK;
3867 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3869 struct bonding *bond = netdev_priv(dev);
3870 netdev_tx_t ret = NETDEV_TX_OK;
3873 * If we risk deadlock from transmitting this in the
3874 * netpoll path, tell netpoll to queue the frame for later tx
3876 if (is_netpoll_tx_blocked(dev))
3877 return NETDEV_TX_BUSY;
3880 if (bond_has_slaves(bond))
3881 ret = __bond_start_xmit(skb, dev);
3890 * set bond mode specific net device operations
3892 void bond_set_mode_ops(struct bonding *bond, int mode)
3894 struct net_device *bond_dev = bond->dev;
3897 case BOND_MODE_ROUNDROBIN:
3899 case BOND_MODE_ACTIVEBACKUP:
3902 bond_set_xmit_hash_policy(bond);
3904 case BOND_MODE_BROADCAST:
3906 case BOND_MODE_8023AD:
3907 bond_set_xmit_hash_policy(bond);
3914 /* Should never happen, mode already checked */
3915 pr_err("%s: Error: Unknown bonding mode %d\n",
3916 bond_dev->name, mode);
3921 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3922 struct ethtool_cmd *ecmd)
3924 struct bonding *bond = netdev_priv(bond_dev);
3925 unsigned long speed = 0;
3926 struct list_head *iter;
3927 struct slave *slave;
3929 ecmd->duplex = DUPLEX_UNKNOWN;
3930 ecmd->port = PORT_OTHER;
3932 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3933 * do not need to check mode. Though link speed might not represent
3934 * the true receive or transmit bandwidth (not all modes are symmetric)
3935 * this is an accurate maximum.
3937 read_lock(&bond->lock);
3938 bond_for_each_slave(bond, slave, iter) {
3939 if (SLAVE_IS_OK(slave)) {
3940 if (slave->speed != SPEED_UNKNOWN)
3941 speed += slave->speed;
3942 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3943 slave->duplex != DUPLEX_UNKNOWN)
3944 ecmd->duplex = slave->duplex;
3947 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3948 read_unlock(&bond->lock);
3953 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3954 struct ethtool_drvinfo *drvinfo)
3956 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3957 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3958 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3962 static const struct ethtool_ops bond_ethtool_ops = {
3963 .get_drvinfo = bond_ethtool_get_drvinfo,
3964 .get_settings = bond_ethtool_get_settings,
3965 .get_link = ethtool_op_get_link,
3968 static const struct net_device_ops bond_netdev_ops = {
3969 .ndo_init = bond_init,
3970 .ndo_uninit = bond_uninit,
3971 .ndo_open = bond_open,
3972 .ndo_stop = bond_close,
3973 .ndo_start_xmit = bond_start_xmit,
3974 .ndo_select_queue = bond_select_queue,
3975 .ndo_get_stats64 = bond_get_stats,
3976 .ndo_do_ioctl = bond_do_ioctl,
3977 .ndo_change_rx_flags = bond_change_rx_flags,
3978 .ndo_set_rx_mode = bond_set_rx_mode,
3979 .ndo_change_mtu = bond_change_mtu,
3980 .ndo_set_mac_address = bond_set_mac_address,
3981 .ndo_neigh_setup = bond_neigh_setup,
3982 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3983 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3984 #ifdef CONFIG_NET_POLL_CONTROLLER
3985 .ndo_netpoll_setup = bond_netpoll_setup,
3986 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3987 .ndo_poll_controller = bond_poll_controller,
3989 .ndo_add_slave = bond_enslave,
3990 .ndo_del_slave = bond_release,
3991 .ndo_fix_features = bond_fix_features,
3994 static const struct device_type bond_type = {
3998 static void bond_destructor(struct net_device *bond_dev)
4000 struct bonding *bond = netdev_priv(bond_dev);
4002 destroy_workqueue(bond->wq);
4003 free_netdev(bond_dev);
4006 static void bond_setup(struct net_device *bond_dev)
4008 struct bonding *bond = netdev_priv(bond_dev);
4010 /* initialize rwlocks */
4011 rwlock_init(&bond->lock);
4012 rwlock_init(&bond->curr_slave_lock);
4013 bond->params = bonding_defaults;
4015 /* Initialize pointers */
4016 bond->dev = bond_dev;
4018 /* Initialize the device entry points */
4019 ether_setup(bond_dev);
4020 bond_dev->netdev_ops = &bond_netdev_ops;
4021 bond_dev->ethtool_ops = &bond_ethtool_ops;
4022 bond_set_mode_ops(bond, bond->params.mode);
4024 bond_dev->destructor = bond_destructor;
4026 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4028 /* Initialize the device options */
4029 bond_dev->tx_queue_len = 0;
4030 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4031 bond_dev->priv_flags |= IFF_BONDING;
4032 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4034 /* At first, we block adding VLANs. That's the only way to
4035 * prevent problems that occur when adding VLANs over an
4036 * empty bond. The block will be removed once non-challenged
4037 * slaves are enslaved.
4039 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4041 /* don't acquire bond device's netif_tx_lock when
4043 bond_dev->features |= NETIF_F_LLTX;
4045 /* By default, we declare the bond to be fully
4046 * VLAN hardware accelerated capable. Special
4047 * care is taken in the various xmit functions
4048 * when there are slaves that are not hw accel
4052 bond_dev->hw_features = BOND_VLAN_FEATURES |
4053 NETIF_F_HW_VLAN_CTAG_TX |
4054 NETIF_F_HW_VLAN_CTAG_RX |
4055 NETIF_F_HW_VLAN_CTAG_FILTER;
4057 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4058 bond_dev->features |= bond_dev->hw_features;
4062 * Destroy a bonding device.
4063 * Must be under rtnl_lock when this function is called.
4065 static void bond_uninit(struct net_device *bond_dev)
4067 struct bonding *bond = netdev_priv(bond_dev);
4068 struct list_head *iter;
4069 struct slave *slave;
4071 bond_netpoll_cleanup(bond_dev);
4073 /* Release the bonded slaves */
4074 bond_for_each_slave(bond, slave, iter)
4075 __bond_release_one(bond_dev, slave->dev, true);
4076 pr_info("%s: released all slaves\n", bond_dev->name);
4078 list_del(&bond->bond_list);
4080 bond_debug_unregister(bond);
4083 /*------------------------- Module initialization ---------------------------*/
4086 * Convert string input module parms. Accept either the
4087 * number of the mode or its string name. A bit complicated because
4088 * some mode names are substrings of other names, and calls from sysfs
4089 * may have whitespace in the name (trailing newlines, for example).
4091 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4093 int modeint = -1, i, rv;
4094 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4096 for (p = (char *)buf; *p; p++)
4097 if (!(isdigit(*p) || isspace(*p)))
4101 rv = sscanf(buf, "%20s", modestr);
4103 rv = sscanf(buf, "%d", &modeint);
4108 for (i = 0; tbl[i].modename; i++) {
4109 if (modeint == tbl[i].mode)
4111 if (strcmp(modestr, tbl[i].modename) == 0)
4118 static int bond_check_params(struct bond_params *params)
4120 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4121 int arp_all_targets_value;
4124 * Convert string parameters.
4127 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4128 if (bond_mode == -1) {
4129 pr_err("Error: Invalid bonding mode \"%s\"\n",
4130 mode == NULL ? "NULL" : mode);
4135 if (xmit_hash_policy) {
4136 if ((bond_mode != BOND_MODE_XOR) &&
4137 (bond_mode != BOND_MODE_8023AD)) {
4138 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4139 bond_mode_name(bond_mode));
4141 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4143 if (xmit_hashtype == -1) {
4144 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4145 xmit_hash_policy == NULL ? "NULL" :
4153 if (bond_mode != BOND_MODE_8023AD) {
4154 pr_info("lacp_rate param is irrelevant in mode %s\n",
4155 bond_mode_name(bond_mode));
4157 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4158 if (lacp_fast == -1) {
4159 pr_err("Error: Invalid lacp rate \"%s\"\n",
4160 lacp_rate == NULL ? "NULL" : lacp_rate);
4167 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4168 if (params->ad_select == -1) {
4169 pr_err("Error: Invalid ad_select \"%s\"\n",
4170 ad_select == NULL ? "NULL" : ad_select);
4174 if (bond_mode != BOND_MODE_8023AD) {
4175 pr_warning("ad_select param only affects 802.3ad mode\n");
4178 params->ad_select = BOND_AD_STABLE;
4181 if (max_bonds < 0) {
4182 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4183 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4184 max_bonds = BOND_DEFAULT_MAX_BONDS;
4188 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4189 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4190 miimon = BOND_LINK_MON_INTERV;
4194 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4199 if (downdelay < 0) {
4200 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4201 downdelay, INT_MAX);
4205 if ((use_carrier != 0) && (use_carrier != 1)) {
4206 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4211 if (num_peer_notif < 0 || num_peer_notif > 255) {
4212 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4217 /* reset values for 802.3ad */
4218 if (bond_mode == BOND_MODE_8023AD) {
4220 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");
4221 pr_warning("Forcing miimon to 100msec\n");
4226 if (tx_queues < 1 || tx_queues > 255) {
4227 pr_warning("Warning: tx_queues (%d) should be between "
4228 "1 and 255, resetting to %d\n",
4229 tx_queues, BOND_DEFAULT_TX_QUEUES);
4230 tx_queues = BOND_DEFAULT_TX_QUEUES;
4233 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4234 pr_warning("Warning: all_slaves_active module parameter (%d), "
4235 "not of valid value (0/1), so it was set to "
4236 "0\n", all_slaves_active);
4237 all_slaves_active = 0;
4240 if (resend_igmp < 0 || resend_igmp > 255) {
4241 pr_warning("Warning: resend_igmp (%d) should be between "
4242 "0 and 255, resetting to %d\n",
4243 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4244 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4247 /* reset values for TLB/ALB */
4248 if ((bond_mode == BOND_MODE_TLB) ||
4249 (bond_mode == BOND_MODE_ALB)) {
4251 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");
4252 pr_warning("Forcing miimon to 100msec\n");
4257 if (bond_mode == BOND_MODE_ALB) {
4258 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",
4263 if (updelay || downdelay) {
4264 /* just warn the user the up/down delay will have
4265 * no effect since miimon is zero...
4267 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",
4268 updelay, downdelay);
4271 /* don't allow arp monitoring */
4273 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4274 miimon, arp_interval);
4278 if ((updelay % miimon) != 0) {
4279 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4281 (updelay / miimon) * miimon);
4286 if ((downdelay % miimon) != 0) {
4287 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4289 (downdelay / miimon) * miimon);
4292 downdelay /= miimon;
4295 if (arp_interval < 0) {
4296 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4297 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4298 arp_interval = BOND_LINK_ARP_INTERV;
4301 for (arp_ip_count = 0, i = 0;
4302 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4303 /* not complete check, but should be good enough to
4305 __be32 ip = in_aton(arp_ip_target[i]);
4306 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4307 ip == htonl(INADDR_BROADCAST)) {
4308 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4312 if (bond_get_targets_ip(arp_target, ip) == -1)
4313 arp_target[arp_ip_count++] = ip;
4315 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4320 if (arp_interval && !arp_ip_count) {
4321 /* don't allow arping if no arp_ip_target given... */
4322 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4328 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4329 pr_err("arp_validate only supported in active-backup mode\n");
4332 if (!arp_interval) {
4333 pr_err("arp_validate requires arp_interval\n");
4337 arp_validate_value = bond_parse_parm(arp_validate,
4339 if (arp_validate_value == -1) {
4340 pr_err("Error: invalid arp_validate \"%s\"\n",
4341 arp_validate == NULL ? "NULL" : arp_validate);
4345 arp_validate_value = 0;
4347 arp_all_targets_value = 0;
4348 if (arp_all_targets) {
4349 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4350 arp_all_targets_tbl);
4352 if (arp_all_targets_value == -1) {
4353 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4355 arp_all_targets_value = 0;
4360 pr_info("MII link monitoring set to %d ms\n", miimon);
4361 } else if (arp_interval) {
4362 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4364 arp_validate_tbl[arp_validate_value].modename,
4367 for (i = 0; i < arp_ip_count; i++)
4368 pr_info(" %s", arp_ip_target[i]);
4372 } else if (max_bonds) {
4373 /* miimon and arp_interval not set, we need one so things
4374 * work as expected, see bonding.txt for details
4376 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");
4379 if (primary && !USES_PRIMARY(bond_mode)) {
4380 /* currently, using a primary only makes sense
4381 * in active backup, TLB or ALB modes
4383 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4384 primary, bond_mode_name(bond_mode));
4388 if (primary && primary_reselect) {
4389 primary_reselect_value = bond_parse_parm(primary_reselect,
4391 if (primary_reselect_value == -1) {
4392 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4394 NULL ? "NULL" : primary_reselect);
4398 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4401 if (fail_over_mac) {
4402 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4404 if (fail_over_mac_value == -1) {
4405 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4406 arp_validate == NULL ? "NULL" : arp_validate);
4410 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4411 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4413 fail_over_mac_value = BOND_FOM_NONE;
4416 /* fill params struct with the proper values */
4417 params->mode = bond_mode;
4418 params->xmit_policy = xmit_hashtype;
4419 params->miimon = miimon;
4420 params->num_peer_notif = num_peer_notif;
4421 params->arp_interval = arp_interval;
4422 params->arp_validate = arp_validate_value;
4423 params->arp_all_targets = arp_all_targets_value;
4424 params->updelay = updelay;
4425 params->downdelay = downdelay;
4426 params->use_carrier = use_carrier;
4427 params->lacp_fast = lacp_fast;
4428 params->primary[0] = 0;
4429 params->primary_reselect = primary_reselect_value;
4430 params->fail_over_mac = fail_over_mac_value;
4431 params->tx_queues = tx_queues;
4432 params->all_slaves_active = all_slaves_active;
4433 params->resend_igmp = resend_igmp;
4434 params->min_links = min_links;
4435 params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4438 strncpy(params->primary, primary, IFNAMSIZ);
4439 params->primary[IFNAMSIZ - 1] = 0;
4442 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4447 static struct lock_class_key bonding_netdev_xmit_lock_key;
4448 static struct lock_class_key bonding_netdev_addr_lock_key;
4449 static struct lock_class_key bonding_tx_busylock_key;
4451 static void bond_set_lockdep_class_one(struct net_device *dev,
4452 struct netdev_queue *txq,
4455 lockdep_set_class(&txq->_xmit_lock,
4456 &bonding_netdev_xmit_lock_key);
4459 static void bond_set_lockdep_class(struct net_device *dev)
4461 lockdep_set_class(&dev->addr_list_lock,
4462 &bonding_netdev_addr_lock_key);
4463 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4464 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4468 * Called from registration process
4470 static int bond_init(struct net_device *bond_dev)
4472 struct bonding *bond = netdev_priv(bond_dev);
4473 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4474 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4476 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4479 * Initialize locks that may be required during
4480 * en/deslave operations. All of the bond_open work
4481 * (of which this is part) should really be moved to
4482 * a phase prior to dev_open
4484 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4485 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4487 bond->wq = create_singlethread_workqueue(bond_dev->name);
4491 bond_set_lockdep_class(bond_dev);
4493 list_add_tail(&bond->bond_list, &bn->dev_list);
4495 bond_prepare_sysfs_group(bond);
4497 bond_debug_register(bond);
4499 /* Ensure valid dev_addr */
4500 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4501 bond_dev->addr_assign_type == NET_ADDR_PERM)
4502 eth_hw_addr_random(bond_dev);
4507 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4509 if (tb[IFLA_ADDRESS]) {
4510 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4512 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4513 return -EADDRNOTAVAIL;
4518 static unsigned int bond_get_num_tx_queues(void)
4523 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4525 .priv_size = sizeof(struct bonding),
4526 .setup = bond_setup,
4527 .validate = bond_validate,
4528 .get_num_tx_queues = bond_get_num_tx_queues,
4529 .get_num_rx_queues = bond_get_num_tx_queues, /* Use the same number
4533 /* Create a new bond based on the specified name and bonding parameters.
4534 * If name is NULL, obtain a suitable "bond%d" name for us.
4535 * Caller must NOT hold rtnl_lock; we need to release it here before we
4536 * set up our sysfs entries.
4538 int bond_create(struct net *net, const char *name)
4540 struct net_device *bond_dev;
4545 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4546 name ? name : "bond%d",
4547 bond_setup, tx_queues);
4549 pr_err("%s: eek! can't alloc netdev!\n", name);
4554 dev_net_set(bond_dev, net);
4555 bond_dev->rtnl_link_ops = &bond_link_ops;
4557 res = register_netdevice(bond_dev);
4559 netif_carrier_off(bond_dev);
4563 bond_destructor(bond_dev);
4567 static int __net_init bond_net_init(struct net *net)
4569 struct bond_net *bn = net_generic(net, bond_net_id);
4572 INIT_LIST_HEAD(&bn->dev_list);
4574 bond_create_proc_dir(bn);
4575 bond_create_sysfs(bn);
4580 static void __net_exit bond_net_exit(struct net *net)
4582 struct bond_net *bn = net_generic(net, bond_net_id);
4583 struct bonding *bond, *tmp_bond;
4586 bond_destroy_sysfs(bn);
4587 bond_destroy_proc_dir(bn);
4589 /* Kill off any bonds created after unregistering bond rtnl ops */
4591 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4592 unregister_netdevice_queue(bond->dev, &list);
4593 unregister_netdevice_many(&list);
4597 static struct pernet_operations bond_net_ops = {
4598 .init = bond_net_init,
4599 .exit = bond_net_exit,
4601 .size = sizeof(struct bond_net),
4604 static int __init bonding_init(void)
4609 pr_info("%s", bond_version);
4611 res = bond_check_params(&bonding_defaults);
4615 res = register_pernet_subsys(&bond_net_ops);
4619 res = rtnl_link_register(&bond_link_ops);
4623 bond_create_debugfs();
4625 for (i = 0; i < max_bonds; i++) {
4626 res = bond_create(&init_net, NULL);
4631 register_netdevice_notifier(&bond_netdev_notifier);
4635 rtnl_link_unregister(&bond_link_ops);
4637 unregister_pernet_subsys(&bond_net_ops);
4642 static void __exit bonding_exit(void)
4644 unregister_netdevice_notifier(&bond_netdev_notifier);
4646 bond_destroy_debugfs();
4648 rtnl_link_unregister(&bond_link_ops);
4649 unregister_pernet_subsys(&bond_net_ops);
4651 #ifdef CONFIG_NET_POLL_CONTROLLER
4653 * Make sure we don't have an imbalance on our netpoll blocking
4655 WARN_ON(atomic_read(&netpoll_block_tx));
4659 module_init(bonding_init);
4660 module_exit(bonding_exit);
4661 MODULE_LICENSE("GPL");
4662 MODULE_VERSION(DRV_VERSION);
4663 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4664 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4665 MODULE_ALIAS_RTNL_LINK("bond");