2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.h>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
96 static int use_carrier = 1;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *arp_all_targets;
108 static char *fail_over_mac;
109 static int all_slaves_active;
110 static struct bond_params bonding_defaults;
111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 static int packets_per_slave = 1;
113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125 module_param(miimon, int, 0);
126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127 module_param(updelay, int, 0);
128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129 module_param(downdelay, int, 0);
130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 module_param(use_carrier, int, 0);
133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135 module_param(mode, charp, 0);
136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140 module_param(primary, charp, 0);
141 MODULE_PARM_DESC(primary, "Primary network device to use");
142 module_param(primary_reselect, charp, 0);
143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
148 "2 for only on active slave "
150 module_param(lacp_rate, charp, 0);
151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153 module_param(ad_select, charp, 0);
154 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
157 module_param(min_links, int, 0);
158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160 module_param(xmit_hash_policy, charp, 0);
161 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165 module_param(arp_interval, int, 0);
166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167 module_param_array(arp_ip_target, charp, NULL, 0);
168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169 module_param(arp_validate, charp, 0);
170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173 module_param(arp_all_targets, charp, 0);
174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac, charp, 0);
176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179 module_param(all_slaves_active, int, 0);
180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183 module_param(resend_igmp, int, 0);
184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 module_param(packets_per_slave, int, 0);
187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190 module_param(lp_interval, uint, 0);
191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
195 /*----------------------------- Global variables ----------------------------*/
197 #ifdef CONFIG_NET_POLL_CONTROLLER
198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
201 int bond_net_id __read_mostly;
203 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
204 static int arp_ip_count;
205 static int bond_mode = BOND_MODE_ROUNDROBIN;
206 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
207 static int lacp_fast;
209 const struct bond_parm_tbl pri_reselect_tbl[] = {
210 { "always", BOND_PRI_RESELECT_ALWAYS},
211 { "better", BOND_PRI_RESELECT_BETTER},
212 { "failure", BOND_PRI_RESELECT_FAILURE},
216 /*-------------------------- Forward declarations ---------------------------*/
218 static int bond_init(struct net_device *bond_dev);
219 static void bond_uninit(struct net_device *bond_dev);
221 /*---------------------------- General routines -----------------------------*/
223 const char *bond_mode_name(int mode)
225 static const char *names[] = {
226 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
227 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
228 [BOND_MODE_XOR] = "load balancing (xor)",
229 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
230 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
231 [BOND_MODE_TLB] = "transmit load balancing",
232 [BOND_MODE_ALB] = "adaptive load balancing",
235 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
241 /*---------------------------------- VLAN -----------------------------------*/
244 * bond_dev_queue_xmit - Prepare skb for xmit.
246 * @bond: bond device that got this skb for tx.
247 * @skb: hw accel VLAN tagged skb to transmit
248 * @slave_dev: slave that is supposed to xmit this skbuff
250 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
251 struct net_device *slave_dev)
253 skb->dev = slave_dev;
255 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
256 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
257 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
259 if (unlikely(netpoll_tx_running(bond->dev)))
260 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
266 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
267 * We don't protect the slave list iteration with a lock because:
268 * a. This operation is performed in IOCTL context,
269 * b. The operation is protected by the RTNL semaphore in the 8021q code,
270 * c. Holding a lock with BH disabled while directly calling a base driver
271 * entry point is generally a BAD idea.
273 * The design of synchronization/protection for this operation in the 8021q
274 * module is good for one or more VLAN devices over a single physical device
275 * and cannot be extended for a teaming solution like bonding, so there is a
276 * potential race condition here where a net device from the vlan group might
277 * be referenced (either by a base driver or the 8021q code) while it is being
278 * removed from the system. However, it turns out we're not making matters
279 * worse, and if it works for regular VLAN usage it will work here too.
283 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
284 * @bond_dev: bonding net device that got called
285 * @vid: vlan id being added
287 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
288 __be16 proto, u16 vid)
290 struct bonding *bond = netdev_priv(bond_dev);
291 struct slave *slave, *rollback_slave;
292 struct list_head *iter;
295 bond_for_each_slave(bond, slave, iter) {
296 res = vlan_vid_add(slave->dev, proto, vid);
304 /* unwind to the slave that failed */
305 bond_for_each_slave(bond, rollback_slave, iter) {
306 if (rollback_slave == slave)
309 vlan_vid_del(rollback_slave->dev, proto, vid);
316 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
317 * @bond_dev: bonding net device that got called
318 * @vid: vlan id being removed
320 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
321 __be16 proto, u16 vid)
323 struct bonding *bond = netdev_priv(bond_dev);
324 struct list_head *iter;
327 bond_for_each_slave(bond, slave, iter)
328 vlan_vid_del(slave->dev, proto, vid);
330 if (bond_is_lb(bond))
331 bond_alb_clear_vlan(bond, vid);
336 /*------------------------------- Link status -------------------------------*/
339 * Set the carrier state for the master according to the state of its
340 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
341 * do special 802.3ad magic.
343 * Returns zero if carrier state does not change, nonzero if it does.
345 static int bond_set_carrier(struct bonding *bond)
347 struct list_head *iter;
350 if (!bond_has_slaves(bond))
353 if (bond->params.mode == BOND_MODE_8023AD)
354 return bond_3ad_set_carrier(bond);
356 bond_for_each_slave(bond, slave, iter) {
357 if (slave->link == BOND_LINK_UP) {
358 if (!netif_carrier_ok(bond->dev)) {
359 netif_carrier_on(bond->dev);
367 if (netif_carrier_ok(bond->dev)) {
368 netif_carrier_off(bond->dev);
375 * Get link speed and duplex from the slave's base driver
376 * using ethtool. If for some reason the call fails or the
377 * values are invalid, set speed and duplex to -1,
380 static void bond_update_speed_duplex(struct slave *slave)
382 struct net_device *slave_dev = slave->dev;
383 struct ethtool_cmd ecmd;
387 slave->speed = SPEED_UNKNOWN;
388 slave->duplex = DUPLEX_UNKNOWN;
390 res = __ethtool_get_settings(slave_dev, &ecmd);
394 slave_speed = ethtool_cmd_speed(&ecmd);
395 if (slave_speed == 0 || slave_speed == ((__u32) -1))
398 switch (ecmd.duplex) {
406 slave->speed = slave_speed;
407 slave->duplex = ecmd.duplex;
412 const char *bond_slave_link_status(s8 link)
429 * if <dev> supports MII link status reporting, check its link status.
431 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
432 * depending upon the setting of the use_carrier parameter.
434 * Return either BMSR_LSTATUS, meaning that the link is up (or we
435 * can't tell and just pretend it is), or 0, meaning that the link is
438 * If reporting is non-zero, instead of faking link up, return -1 if
439 * both ETHTOOL and MII ioctls fail (meaning the device does not
440 * support them). If use_carrier is set, return whatever it says.
441 * It'd be nice if there was a good way to tell if a driver supports
442 * netif_carrier, but there really isn't.
444 static int bond_check_dev_link(struct bonding *bond,
445 struct net_device *slave_dev, int reporting)
447 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
448 int (*ioctl)(struct net_device *, struct ifreq *, int);
450 struct mii_ioctl_data *mii;
452 if (!reporting && !netif_running(slave_dev))
455 if (bond->params.use_carrier)
456 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
458 /* Try to get link status using Ethtool first. */
459 if (slave_dev->ethtool_ops->get_link)
460 return slave_dev->ethtool_ops->get_link(slave_dev) ?
463 /* Ethtool can't be used, fallback to MII ioctls. */
464 ioctl = slave_ops->ndo_do_ioctl;
466 /* TODO: set pointer to correct ioctl on a per team member */
467 /* bases to make this more efficient. that is, once */
468 /* we determine the correct ioctl, we will always */
469 /* call it and not the others for that team */
473 * We cannot assume that SIOCGMIIPHY will also read a
474 * register; not all network drivers (e.g., e100)
478 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
479 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
481 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
482 mii->reg_num = MII_BMSR;
483 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
484 return mii->val_out & BMSR_LSTATUS;
489 * If reporting, report that either there's no dev->do_ioctl,
490 * or both SIOCGMIIREG and get_link failed (meaning that we
491 * cannot report link status). If not reporting, pretend
494 return reporting ? -1 : BMSR_LSTATUS;
497 /*----------------------------- Multicast list ------------------------------*/
500 * Push the promiscuity flag down to appropriate slaves
502 static int bond_set_promiscuity(struct bonding *bond, int inc)
504 struct list_head *iter;
507 if (USES_PRIMARY(bond->params.mode)) {
508 /* write lock already acquired */
509 if (bond->curr_active_slave) {
510 err = dev_set_promiscuity(bond->curr_active_slave->dev,
516 bond_for_each_slave(bond, slave, iter) {
517 err = dev_set_promiscuity(slave->dev, inc);
526 * Push the allmulti flag down to all slaves
528 static int bond_set_allmulti(struct bonding *bond, int inc)
530 struct list_head *iter;
533 if (USES_PRIMARY(bond->params.mode)) {
534 /* write lock already acquired */
535 if (bond->curr_active_slave) {
536 err = dev_set_allmulti(bond->curr_active_slave->dev,
542 bond_for_each_slave(bond, slave, iter) {
543 err = dev_set_allmulti(slave->dev, inc);
552 * Retrieve the list of registered multicast addresses for the bonding
553 * device and retransmit an IGMP JOIN request to the current active
556 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
558 struct bonding *bond = container_of(work, struct bonding,
561 if (!rtnl_trylock()) {
562 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
565 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
567 if (bond->igmp_retrans > 1) {
568 bond->igmp_retrans--;
569 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
574 /* Flush bond's hardware addresses from slave
576 static void bond_hw_addr_flush(struct net_device *bond_dev,
577 struct net_device *slave_dev)
579 struct bonding *bond = netdev_priv(bond_dev);
581 dev_uc_unsync(slave_dev, bond_dev);
582 dev_mc_unsync(slave_dev, bond_dev);
584 if (bond->params.mode == BOND_MODE_8023AD) {
585 /* del lacpdu mc addr from mc list */
586 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
588 dev_mc_del(slave_dev, lacpdu_multicast);
592 /*--------------------------- Active slave change ---------------------------*/
594 /* Update the hardware address list and promisc/allmulti for the new and
595 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
596 * slaves up date at all times; only the USES_PRIMARY modes need to call
597 * this function to swap these settings during a failover.
599 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
600 struct slave *old_active)
605 if (bond->dev->flags & IFF_PROMISC)
606 dev_set_promiscuity(old_active->dev, -1);
608 if (bond->dev->flags & IFF_ALLMULTI)
609 dev_set_allmulti(old_active->dev, -1);
611 bond_hw_addr_flush(bond->dev, old_active->dev);
615 /* FIXME: Signal errors upstream. */
616 if (bond->dev->flags & IFF_PROMISC)
617 dev_set_promiscuity(new_active->dev, 1);
619 if (bond->dev->flags & IFF_ALLMULTI)
620 dev_set_allmulti(new_active->dev, 1);
622 netif_addr_lock_bh(bond->dev);
623 dev_uc_sync(new_active->dev, bond->dev);
624 dev_mc_sync(new_active->dev, bond->dev);
625 netif_addr_unlock_bh(bond->dev);
630 * bond_set_dev_addr - clone slave's address to bond
631 * @bond_dev: bond net device
632 * @slave_dev: slave net device
634 * Should be called with RTNL held.
636 static void bond_set_dev_addr(struct net_device *bond_dev,
637 struct net_device *slave_dev)
639 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
640 bond_dev, slave_dev, slave_dev->addr_len);
641 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
642 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
643 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
647 * bond_do_fail_over_mac
649 * Perform special MAC address swapping for fail_over_mac settings
651 * Called with RTNL, curr_slave_lock for write_bh.
653 static void bond_do_fail_over_mac(struct bonding *bond,
654 struct slave *new_active,
655 struct slave *old_active)
656 __releases(&bond->curr_slave_lock)
657 __acquires(&bond->curr_slave_lock)
659 u8 tmp_mac[ETH_ALEN];
660 struct sockaddr saddr;
663 switch (bond->params.fail_over_mac) {
664 case BOND_FOM_ACTIVE:
666 write_unlock_bh(&bond->curr_slave_lock);
667 bond_set_dev_addr(bond->dev, new_active->dev);
668 write_lock_bh(&bond->curr_slave_lock);
671 case BOND_FOM_FOLLOW:
673 * if new_active && old_active, swap them
674 * if just old_active, do nothing (going to no active slave)
675 * if just new_active, set new_active to bond's MAC
680 write_unlock_bh(&bond->curr_slave_lock);
683 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
684 memcpy(saddr.sa_data, old_active->dev->dev_addr,
686 saddr.sa_family = new_active->dev->type;
688 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
689 saddr.sa_family = bond->dev->type;
692 rv = dev_set_mac_address(new_active->dev, &saddr);
694 pr_err("%s: Error %d setting MAC of slave %s\n",
695 bond->dev->name, -rv, new_active->dev->name);
702 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
703 saddr.sa_family = old_active->dev->type;
705 rv = dev_set_mac_address(old_active->dev, &saddr);
707 pr_err("%s: Error %d setting MAC of slave %s\n",
708 bond->dev->name, -rv, new_active->dev->name);
710 write_lock_bh(&bond->curr_slave_lock);
713 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
714 bond->dev->name, bond->params.fail_over_mac);
720 static bool bond_should_change_active(struct bonding *bond)
722 struct slave *prim = bond->primary_slave;
723 struct slave *curr = bond->curr_active_slave;
725 if (!prim || !curr || curr->link != BOND_LINK_UP)
727 if (bond->force_primary) {
728 bond->force_primary = false;
731 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
732 (prim->speed < curr->speed ||
733 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
735 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
741 * find_best_interface - select the best available slave to be the active one
742 * @bond: our bonding struct
744 static struct slave *bond_find_best_slave(struct bonding *bond)
746 struct slave *slave, *bestslave = NULL;
747 struct list_head *iter;
748 int mintime = bond->params.updelay;
750 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
751 bond_should_change_active(bond))
752 return bond->primary_slave;
754 bond_for_each_slave(bond, slave, iter) {
755 if (slave->link == BOND_LINK_UP)
757 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
758 slave->delay < mintime) {
759 mintime = slave->delay;
767 static bool bond_should_notify_peers(struct bonding *bond)
772 slave = rcu_dereference(bond->curr_active_slave);
775 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
776 bond->dev->name, slave ? slave->dev->name : "NULL");
778 if (!slave || !bond->send_peer_notif ||
779 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
786 * change_active_interface - change the active slave into the specified one
787 * @bond: our bonding struct
788 * @new: the new slave to make the active one
790 * Set the new slave to the bond's settings and unset them on the old
792 * Setting include flags, mc-list, promiscuity, allmulti, etc.
794 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
795 * because it is apparently the best available slave we have, even though its
796 * updelay hasn't timed out yet.
798 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
800 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
802 struct slave *old_active = bond->curr_active_slave;
804 if (old_active == new_active)
808 new_active->jiffies = jiffies;
810 if (new_active->link == BOND_LINK_BACK) {
811 if (USES_PRIMARY(bond->params.mode)) {
812 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
813 bond->dev->name, new_active->dev->name,
814 (bond->params.updelay - new_active->delay) * bond->params.miimon);
817 new_active->delay = 0;
818 new_active->link = BOND_LINK_UP;
820 if (bond->params.mode == BOND_MODE_8023AD)
821 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
823 if (bond_is_lb(bond))
824 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
826 if (USES_PRIMARY(bond->params.mode)) {
827 pr_info("%s: making interface %s the new active one.\n",
828 bond->dev->name, new_active->dev->name);
833 if (USES_PRIMARY(bond->params.mode))
834 bond_hw_addr_swap(bond, new_active, old_active);
836 if (bond_is_lb(bond)) {
837 bond_alb_handle_active_change(bond, new_active);
839 bond_set_slave_inactive_flags(old_active);
841 bond_set_slave_active_flags(new_active);
843 rcu_assign_pointer(bond->curr_active_slave, new_active);
846 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
848 bond_set_slave_inactive_flags(old_active);
851 bool should_notify_peers = false;
853 bond_set_slave_active_flags(new_active);
855 if (bond->params.fail_over_mac)
856 bond_do_fail_over_mac(bond, new_active,
859 if (netif_running(bond->dev)) {
860 bond->send_peer_notif =
861 bond->params.num_peer_notif;
862 should_notify_peers =
863 bond_should_notify_peers(bond);
866 write_unlock_bh(&bond->curr_slave_lock);
868 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
869 if (should_notify_peers)
870 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
873 write_lock_bh(&bond->curr_slave_lock);
877 /* resend IGMP joins since active slave has changed or
878 * all were sent on curr_active_slave.
879 * resend only if bond is brought up with the affected
880 * bonding modes and the retransmission is enabled */
881 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
882 ((USES_PRIMARY(bond->params.mode) && new_active) ||
883 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
884 bond->igmp_retrans = bond->params.resend_igmp;
885 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
890 * bond_select_active_slave - select a new active slave, if needed
891 * @bond: our bonding struct
893 * This functions should be called when one of the following occurs:
894 * - The old curr_active_slave has been released or lost its link.
895 * - The primary_slave has got its link back.
896 * - A slave has got its link back and there's no old curr_active_slave.
898 * Caller must hold curr_slave_lock for write_bh.
900 void bond_select_active_slave(struct bonding *bond)
902 struct slave *best_slave;
905 best_slave = bond_find_best_slave(bond);
906 if (best_slave != bond->curr_active_slave) {
907 bond_change_active_slave(bond, best_slave);
908 rv = bond_set_carrier(bond);
912 if (netif_carrier_ok(bond->dev)) {
913 pr_info("%s: first active interface up!\n",
916 pr_info("%s: now running without any active interface !\n",
922 #ifdef CONFIG_NET_POLL_CONTROLLER
923 static inline int slave_enable_netpoll(struct slave *slave)
928 np = kzalloc(sizeof(*np), GFP_ATOMIC);
933 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
942 static inline void slave_disable_netpoll(struct slave *slave)
944 struct netpoll *np = slave->np;
950 __netpoll_free_async(np);
952 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
954 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
956 if (!slave_dev->netdev_ops->ndo_poll_controller)
961 static void bond_poll_controller(struct net_device *bond_dev)
965 static void bond_netpoll_cleanup(struct net_device *bond_dev)
967 struct bonding *bond = netdev_priv(bond_dev);
968 struct list_head *iter;
971 bond_for_each_slave(bond, slave, iter)
972 if (IS_UP(slave->dev))
973 slave_disable_netpoll(slave);
976 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
978 struct bonding *bond = netdev_priv(dev);
979 struct list_head *iter;
983 bond_for_each_slave(bond, slave, iter) {
984 err = slave_enable_netpoll(slave);
986 bond_netpoll_cleanup(dev);
993 static inline int slave_enable_netpoll(struct slave *slave)
997 static inline void slave_disable_netpoll(struct slave *slave)
1000 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1005 /*---------------------------------- IOCTL ----------------------------------*/
1007 static netdev_features_t bond_fix_features(struct net_device *dev,
1008 netdev_features_t features)
1010 struct bonding *bond = netdev_priv(dev);
1011 struct list_head *iter;
1012 netdev_features_t mask;
1013 struct slave *slave;
1015 if (!bond_has_slaves(bond)) {
1016 /* Disable adding VLANs to empty bond. But why? --mq */
1017 features |= NETIF_F_VLAN_CHALLENGED;
1022 features &= ~NETIF_F_ONE_FOR_ALL;
1023 features |= NETIF_F_ALL_FOR_ALL;
1025 bond_for_each_slave(bond, slave, iter) {
1026 features = netdev_increment_features(features,
1027 slave->dev->features,
1030 features = netdev_add_tso_features(features, mask);
1035 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1036 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1037 NETIF_F_HIGHDMA | NETIF_F_LRO)
1039 static void bond_compute_features(struct bonding *bond)
1041 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1042 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1043 struct net_device *bond_dev = bond->dev;
1044 struct list_head *iter;
1045 struct slave *slave;
1046 unsigned short max_hard_header_len = ETH_HLEN;
1047 unsigned int gso_max_size = GSO_MAX_SIZE;
1048 u16 gso_max_segs = GSO_MAX_SEGS;
1050 if (!bond_has_slaves(bond))
1053 bond_for_each_slave(bond, slave, iter) {
1054 vlan_features = netdev_increment_features(vlan_features,
1055 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1057 dst_release_flag &= slave->dev->priv_flags;
1058 if (slave->dev->hard_header_len > max_hard_header_len)
1059 max_hard_header_len = slave->dev->hard_header_len;
1061 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1062 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1066 bond_dev->vlan_features = vlan_features;
1067 bond_dev->hard_header_len = max_hard_header_len;
1068 bond_dev->gso_max_segs = gso_max_segs;
1069 netif_set_gso_max_size(bond_dev, gso_max_size);
1071 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1072 bond_dev->priv_flags = flags | dst_release_flag;
1074 netdev_change_features(bond_dev);
1077 static void bond_setup_by_slave(struct net_device *bond_dev,
1078 struct net_device *slave_dev)
1080 bond_dev->header_ops = slave_dev->header_ops;
1082 bond_dev->type = slave_dev->type;
1083 bond_dev->hard_header_len = slave_dev->hard_header_len;
1084 bond_dev->addr_len = slave_dev->addr_len;
1086 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1087 slave_dev->addr_len);
1090 /* On bonding slaves other than the currently active slave, suppress
1091 * duplicates except for alb non-mcast/bcast.
1093 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1094 struct slave *slave,
1095 struct bonding *bond)
1097 if (bond_is_slave_inactive(slave)) {
1098 if (bond->params.mode == BOND_MODE_ALB &&
1099 skb->pkt_type != PACKET_BROADCAST &&
1100 skb->pkt_type != PACKET_MULTICAST)
1107 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1109 struct sk_buff *skb = *pskb;
1110 struct slave *slave;
1111 struct bonding *bond;
1112 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1114 int ret = RX_HANDLER_ANOTHER;
1116 skb = skb_share_check(skb, GFP_ATOMIC);
1118 return RX_HANDLER_CONSUMED;
1122 slave = bond_slave_get_rcu(skb->dev);
1125 if (bond->params.arp_interval)
1126 slave->dev->last_rx = jiffies;
1128 recv_probe = ACCESS_ONCE(bond->recv_probe);
1130 ret = recv_probe(skb, bond, slave);
1131 if (ret == RX_HANDLER_CONSUMED) {
1137 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1138 return RX_HANDLER_EXACT;
1141 skb->dev = bond->dev;
1143 if (bond->params.mode == BOND_MODE_ALB &&
1144 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1145 skb->pkt_type == PACKET_HOST) {
1147 if (unlikely(skb_cow_head(skb,
1148 skb->data - skb_mac_header(skb)))) {
1150 return RX_HANDLER_CONSUMED;
1152 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1158 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1159 struct net_device *slave_dev,
1160 struct slave *slave)
1164 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1167 slave_dev->flags |= IFF_SLAVE;
1168 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1172 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1173 struct net_device *slave_dev)
1175 netdev_upper_dev_unlink(slave_dev, bond_dev);
1176 slave_dev->flags &= ~IFF_SLAVE;
1177 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1180 /* enslave device <slave> to bond device <master> */
1181 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1183 struct bonding *bond = netdev_priv(bond_dev);
1184 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1185 struct slave *new_slave = NULL, *prev_slave;
1186 struct sockaddr addr;
1190 if (!bond->params.use_carrier &&
1191 slave_dev->ethtool_ops->get_link == NULL &&
1192 slave_ops->ndo_do_ioctl == NULL) {
1193 pr_warning("%s: Warning: no link monitoring support for %s\n",
1194 bond_dev->name, slave_dev->name);
1197 /* already enslaved */
1198 if (slave_dev->flags & IFF_SLAVE) {
1199 pr_debug("Error, Device was already enslaved\n");
1203 /* vlan challenged mutual exclusion */
1204 /* no need to lock since we're protected by rtnl_lock */
1205 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1206 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1207 if (vlan_uses_dev(bond_dev)) {
1208 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1209 bond_dev->name, slave_dev->name, bond_dev->name);
1212 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1213 bond_dev->name, slave_dev->name,
1214 slave_dev->name, bond_dev->name);
1217 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1221 * Old ifenslave binaries are no longer supported. These can
1222 * be identified with moderate accuracy by the state of the slave:
1223 * the current ifenslave will set the interface down prior to
1224 * enslaving it; the old ifenslave will not.
1226 if ((slave_dev->flags & IFF_UP)) {
1227 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1230 goto err_undo_flags;
1233 /* set bonding device ether type by slave - bonding netdevices are
1234 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1235 * there is a need to override some of the type dependent attribs/funcs.
1237 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1238 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1240 if (!bond_has_slaves(bond)) {
1241 if (bond_dev->type != slave_dev->type) {
1242 pr_debug("%s: change device type from %d to %d\n",
1244 bond_dev->type, slave_dev->type);
1246 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1248 res = notifier_to_errno(res);
1250 pr_err("%s: refused to change device type\n",
1253 goto err_undo_flags;
1256 /* Flush unicast and multicast addresses */
1257 dev_uc_flush(bond_dev);
1258 dev_mc_flush(bond_dev);
1260 if (slave_dev->type != ARPHRD_ETHER)
1261 bond_setup_by_slave(bond_dev, slave_dev);
1263 ether_setup(bond_dev);
1264 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1267 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1270 } else if (bond_dev->type != slave_dev->type) {
1271 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1273 slave_dev->type, bond_dev->type);
1275 goto err_undo_flags;
1278 if (slave_ops->ndo_set_mac_address == NULL) {
1279 if (!bond_has_slaves(bond)) {
1280 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1282 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1283 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1284 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",
1287 goto err_undo_flags;
1291 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1293 /* If this is the first slave, then we need to set the master's hardware
1294 * address to be the same as the slave's. */
1295 if (!bond_has_slaves(bond) &&
1296 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1297 bond_set_dev_addr(bond->dev, slave_dev);
1299 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1302 goto err_undo_flags;
1305 * Set the new_slave's queue_id to be zero. Queue ID mapping
1306 * is set via sysfs or module option if desired.
1308 new_slave->queue_id = 0;
1310 /* Save slave's original mtu and then set it to match the bond */
1311 new_slave->original_mtu = slave_dev->mtu;
1312 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1314 pr_debug("Error %d calling dev_set_mtu\n", res);
1319 * Save slave's original ("permanent") mac address for modes
1320 * that need it, and for restoring it upon release, and then
1321 * set it to the master's address
1323 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1325 if (!bond->params.fail_over_mac) {
1327 * Set slave to master's mac address. The application already
1328 * set the master's mac address to that of the first slave
1330 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1331 addr.sa_family = slave_dev->type;
1332 res = dev_set_mac_address(slave_dev, &addr);
1334 pr_debug("Error %d calling set_mac_address\n", res);
1335 goto err_restore_mtu;
1339 /* open the slave since the application closed it */
1340 res = dev_open(slave_dev);
1342 pr_debug("Opening slave %s failed\n", slave_dev->name);
1343 goto err_restore_mac;
1346 new_slave->bond = bond;
1347 new_slave->dev = slave_dev;
1348 slave_dev->priv_flags |= IFF_BONDING;
1350 if (bond_is_lb(bond)) {
1351 /* bond_alb_init_slave() must be called before all other stages since
1352 * it might fail and we do not want to have to undo everything
1354 res = bond_alb_init_slave(bond, new_slave);
1359 /* If the mode USES_PRIMARY, then the following is handled by
1360 * bond_change_active_slave().
1362 if (!USES_PRIMARY(bond->params.mode)) {
1363 /* set promiscuity level to new slave */
1364 if (bond_dev->flags & IFF_PROMISC) {
1365 res = dev_set_promiscuity(slave_dev, 1);
1370 /* set allmulti level to new slave */
1371 if (bond_dev->flags & IFF_ALLMULTI) {
1372 res = dev_set_allmulti(slave_dev, 1);
1377 netif_addr_lock_bh(bond_dev);
1379 dev_mc_sync_multiple(slave_dev, bond_dev);
1380 dev_uc_sync_multiple(slave_dev, bond_dev);
1382 netif_addr_unlock_bh(bond_dev);
1385 if (bond->params.mode == BOND_MODE_8023AD) {
1386 /* add lacpdu mc addr to mc list */
1387 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1389 dev_mc_add(slave_dev, lacpdu_multicast);
1392 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1394 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1395 bond_dev->name, slave_dev->name);
1399 prev_slave = bond_last_slave(bond);
1401 new_slave->delay = 0;
1402 new_slave->link_failure_count = 0;
1404 bond_update_speed_duplex(new_slave);
1406 new_slave->last_arp_rx = jiffies -
1407 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1408 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1409 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1411 if (bond->params.miimon && !bond->params.use_carrier) {
1412 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1414 if ((link_reporting == -1) && !bond->params.arp_interval) {
1416 * miimon is set but a bonded network driver
1417 * does not support ETHTOOL/MII and
1418 * arp_interval is not set. Note: if
1419 * use_carrier is enabled, we will never go
1420 * here (because netif_carrier is always
1421 * supported); thus, we don't need to change
1422 * the messages for netif_carrier.
1424 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",
1425 bond_dev->name, slave_dev->name);
1426 } else if (link_reporting == -1) {
1427 /* unable get link status using mii/ethtool */
1428 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",
1429 bond_dev->name, slave_dev->name);
1433 /* check for initial state */
1434 if (bond->params.miimon) {
1435 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1436 if (bond->params.updelay) {
1437 new_slave->link = BOND_LINK_BACK;
1438 new_slave->delay = bond->params.updelay;
1440 new_slave->link = BOND_LINK_UP;
1443 new_slave->link = BOND_LINK_DOWN;
1445 } else if (bond->params.arp_interval) {
1446 new_slave->link = (netif_carrier_ok(slave_dev) ?
1447 BOND_LINK_UP : BOND_LINK_DOWN);
1449 new_slave->link = BOND_LINK_UP;
1452 if (new_slave->link != BOND_LINK_DOWN)
1453 new_slave->jiffies = jiffies;
1454 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1455 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1456 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1458 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1459 /* if there is a primary slave, remember it */
1460 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1461 bond->primary_slave = new_slave;
1462 bond->force_primary = true;
1466 switch (bond->params.mode) {
1467 case BOND_MODE_ACTIVEBACKUP:
1468 bond_set_slave_inactive_flags(new_slave);
1470 case BOND_MODE_8023AD:
1471 /* in 802.3ad mode, the internal mechanism
1472 * will activate the slaves in the selected
1475 bond_set_slave_inactive_flags(new_slave);
1476 /* if this is the first slave */
1478 SLAVE_AD_INFO(new_slave).id = 1;
1479 /* Initialize AD with the number of times that the AD timer is called in 1 second
1480 * can be called only after the mac address of the bond is set
1482 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1484 SLAVE_AD_INFO(new_slave).id =
1485 SLAVE_AD_INFO(prev_slave).id + 1;
1488 bond_3ad_bind_slave(new_slave);
1492 bond_set_active_slave(new_slave);
1493 bond_set_slave_inactive_flags(new_slave);
1496 pr_debug("This slave is always active in trunk mode\n");
1498 /* always active in trunk mode */
1499 bond_set_active_slave(new_slave);
1501 /* In trunking mode there is little meaning to curr_active_slave
1502 * anyway (it holds no special properties of the bond device),
1503 * so we can change it without calling change_active_interface()
1505 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1506 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1509 } /* switch(bond_mode) */
1511 #ifdef CONFIG_NET_POLL_CONTROLLER
1512 slave_dev->npinfo = bond->dev->npinfo;
1513 if (slave_dev->npinfo) {
1514 if (slave_enable_netpoll(new_slave)) {
1515 read_unlock(&bond->lock);
1516 pr_info("Error, %s: master_dev is using netpoll, "
1517 "but new slave device does not support netpoll.\n",
1525 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1528 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1532 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1534 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1535 goto err_unregister;
1538 res = bond_sysfs_slave_add(new_slave);
1540 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1541 goto err_upper_unlink;
1545 bond_compute_features(bond);
1546 bond_set_carrier(bond);
1548 if (USES_PRIMARY(bond->params.mode)) {
1549 write_lock_bh(&bond->curr_slave_lock);
1550 bond_select_active_slave(bond);
1551 write_unlock_bh(&bond->curr_slave_lock);
1554 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1555 bond_dev->name, slave_dev->name,
1556 bond_is_active_slave(new_slave) ? "n active" : " backup",
1557 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1559 /* enslave is successful */
1562 /* Undo stages on error */
1564 bond_upper_dev_unlink(bond_dev, slave_dev);
1567 netdev_rx_handler_unregister(slave_dev);
1570 if (!USES_PRIMARY(bond->params.mode))
1571 bond_hw_addr_flush(bond_dev, slave_dev);
1573 vlan_vids_del_by_dev(slave_dev, bond_dev);
1574 if (bond->primary_slave == new_slave)
1575 bond->primary_slave = NULL;
1576 if (bond->curr_active_slave == new_slave) {
1577 write_lock_bh(&bond->curr_slave_lock);
1578 bond_change_active_slave(bond, NULL);
1579 bond_select_active_slave(bond);
1580 write_unlock_bh(&bond->curr_slave_lock);
1582 slave_disable_netpoll(new_slave);
1585 slave_dev->priv_flags &= ~IFF_BONDING;
1586 dev_close(slave_dev);
1589 if (!bond->params.fail_over_mac) {
1590 /* XXX TODO - fom follow mode needs to change master's
1591 * MAC if this slave's MAC is in use by the bond, or at
1592 * least print a warning.
1594 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1595 addr.sa_family = slave_dev->type;
1596 dev_set_mac_address(slave_dev, &addr);
1600 dev_set_mtu(slave_dev, new_slave->original_mtu);
1606 /* Enslave of first slave has failed and we need to fix master's mac */
1607 if (!bond_has_slaves(bond) &&
1608 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1609 eth_hw_addr_random(bond_dev);
1615 * Try to release the slave device <slave> from the bond device <master>
1616 * It is legal to access curr_active_slave without a lock because all the function
1617 * is write-locked. If "all" is true it means that the function is being called
1618 * while destroying a bond interface and all slaves are being released.
1620 * The rules for slave state should be:
1621 * for Active/Backup:
1622 * Active stays on all backups go down
1623 * for Bonded connections:
1624 * The first up interface should be left on and all others downed.
1626 static int __bond_release_one(struct net_device *bond_dev,
1627 struct net_device *slave_dev,
1630 struct bonding *bond = netdev_priv(bond_dev);
1631 struct slave *slave, *oldcurrent;
1632 struct sockaddr addr;
1633 int old_flags = bond_dev->flags;
1634 netdev_features_t old_features = bond_dev->features;
1636 /* slave is not a slave or master is not master of this slave */
1637 if (!(slave_dev->flags & IFF_SLAVE) ||
1638 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1639 pr_err("%s: Error: cannot release %s.\n",
1640 bond_dev->name, slave_dev->name);
1646 slave = bond_get_slave_by_dev(bond, slave_dev);
1648 /* not a slave of this bond */
1649 pr_info("%s: %s not enslaved\n",
1650 bond_dev->name, slave_dev->name);
1651 unblock_netpoll_tx();
1655 /* release the slave from its bond */
1658 bond_sysfs_slave_del(slave);
1660 bond_upper_dev_unlink(bond_dev, slave_dev);
1661 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1662 * for this slave anymore.
1664 netdev_rx_handler_unregister(slave_dev);
1665 write_lock_bh(&bond->lock);
1667 /* Inform AD package of unbinding of slave. */
1668 if (bond->params.mode == BOND_MODE_8023AD)
1669 bond_3ad_unbind_slave(slave);
1671 write_unlock_bh(&bond->lock);
1673 pr_info("%s: releasing %s interface %s\n",
1675 bond_is_active_slave(slave) ? "active" : "backup",
1678 oldcurrent = bond->curr_active_slave;
1680 bond->current_arp_slave = NULL;
1682 if (!all && !bond->params.fail_over_mac) {
1683 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1684 bond_has_slaves(bond))
1685 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",
1686 bond_dev->name, slave_dev->name,
1688 bond_dev->name, slave_dev->name);
1691 if (bond->primary_slave == slave)
1692 bond->primary_slave = NULL;
1694 if (oldcurrent == slave) {
1695 write_lock_bh(&bond->curr_slave_lock);
1696 bond_change_active_slave(bond, NULL);
1697 write_unlock_bh(&bond->curr_slave_lock);
1700 if (bond_is_lb(bond)) {
1701 /* Must be called only after the slave has been
1702 * detached from the list and the curr_active_slave
1703 * has been cleared (if our_slave == old_current),
1704 * but before a new active slave is selected.
1706 bond_alb_deinit_slave(bond, slave);
1710 rcu_assign_pointer(bond->curr_active_slave, NULL);
1711 } else if (oldcurrent == slave) {
1713 * Note that we hold RTNL over this sequence, so there
1714 * is no concern that another slave add/remove event
1717 write_lock_bh(&bond->curr_slave_lock);
1719 bond_select_active_slave(bond);
1721 write_unlock_bh(&bond->curr_slave_lock);
1724 if (!bond_has_slaves(bond)) {
1725 bond_set_carrier(bond);
1726 eth_hw_addr_random(bond_dev);
1728 if (vlan_uses_dev(bond_dev)) {
1729 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1730 bond_dev->name, bond_dev->name);
1731 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1736 unblock_netpoll_tx();
1739 if (!bond_has_slaves(bond)) {
1740 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1741 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1744 bond_compute_features(bond);
1745 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1746 (old_features & NETIF_F_VLAN_CHALLENGED))
1747 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1748 bond_dev->name, slave_dev->name, bond_dev->name);
1750 /* must do this from outside any spinlocks */
1751 vlan_vids_del_by_dev(slave_dev, bond_dev);
1753 /* If the mode USES_PRIMARY, then this cases was handled above by
1754 * bond_change_active_slave(..., NULL)
1756 if (!USES_PRIMARY(bond->params.mode)) {
1757 /* unset promiscuity level from slave
1758 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1759 * of the IFF_PROMISC flag in the bond_dev, but we need the
1760 * value of that flag before that change, as that was the value
1761 * when this slave was attached, so we cache at the start of the
1762 * function and use it here. Same goes for ALLMULTI below
1764 if (old_flags & IFF_PROMISC)
1765 dev_set_promiscuity(slave_dev, -1);
1767 /* unset allmulti level from slave */
1768 if (old_flags & IFF_ALLMULTI)
1769 dev_set_allmulti(slave_dev, -1);
1771 bond_hw_addr_flush(bond_dev, slave_dev);
1774 slave_disable_netpoll(slave);
1776 /* close slave before restoring its mac address */
1777 dev_close(slave_dev);
1779 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1780 /* restore original ("permanent") mac address */
1781 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1782 addr.sa_family = slave_dev->type;
1783 dev_set_mac_address(slave_dev, &addr);
1786 dev_set_mtu(slave_dev, slave->original_mtu);
1788 slave_dev->priv_flags &= ~IFF_BONDING;
1792 return 0; /* deletion OK */
1795 /* A wrapper used because of ndo_del_link */
1796 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1798 return __bond_release_one(bond_dev, slave_dev, false);
1802 * First release a slave and then destroy the bond if no more slaves are left.
1803 * Must be under rtnl_lock when this function is called.
1805 static int bond_release_and_destroy(struct net_device *bond_dev,
1806 struct net_device *slave_dev)
1808 struct bonding *bond = netdev_priv(bond_dev);
1811 ret = bond_release(bond_dev, slave_dev);
1812 if (ret == 0 && !bond_has_slaves(bond)) {
1813 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1814 pr_info("%s: destroying bond %s.\n",
1815 bond_dev->name, bond_dev->name);
1816 unregister_netdevice(bond_dev);
1821 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1823 struct bonding *bond = netdev_priv(bond_dev);
1825 info->bond_mode = bond->params.mode;
1826 info->miimon = bond->params.miimon;
1828 read_lock(&bond->lock);
1829 info->num_slaves = bond->slave_cnt;
1830 read_unlock(&bond->lock);
1835 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1837 struct bonding *bond = netdev_priv(bond_dev);
1838 struct list_head *iter;
1839 int i = 0, res = -ENODEV;
1840 struct slave *slave;
1842 read_lock(&bond->lock);
1843 bond_for_each_slave(bond, slave, iter) {
1844 if (i++ == (int)info->slave_id) {
1846 strcpy(info->slave_name, slave->dev->name);
1847 info->link = slave->link;
1848 info->state = bond_slave_state(slave);
1849 info->link_failure_count = slave->link_failure_count;
1853 read_unlock(&bond->lock);
1858 /*-------------------------------- Monitoring -------------------------------*/
1861 static int bond_miimon_inspect(struct bonding *bond)
1863 int link_state, commit = 0;
1864 struct list_head *iter;
1865 struct slave *slave;
1866 bool ignore_updelay;
1868 ignore_updelay = !bond->curr_active_slave ? true : false;
1870 bond_for_each_slave_rcu(bond, slave, iter) {
1871 slave->new_link = BOND_LINK_NOCHANGE;
1873 link_state = bond_check_dev_link(bond, slave->dev, 0);
1875 switch (slave->link) {
1880 slave->link = BOND_LINK_FAIL;
1881 slave->delay = bond->params.downdelay;
1883 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1885 (bond->params.mode ==
1886 BOND_MODE_ACTIVEBACKUP) ?
1887 (bond_is_active_slave(slave) ?
1888 "active " : "backup ") : "",
1890 bond->params.downdelay * bond->params.miimon);
1893 case BOND_LINK_FAIL:
1896 * recovered before downdelay expired
1898 slave->link = BOND_LINK_UP;
1899 slave->jiffies = jiffies;
1900 pr_info("%s: link status up again after %d ms for interface %s.\n",
1902 (bond->params.downdelay - slave->delay) *
1903 bond->params.miimon,
1908 if (slave->delay <= 0) {
1909 slave->new_link = BOND_LINK_DOWN;
1917 case BOND_LINK_DOWN:
1921 slave->link = BOND_LINK_BACK;
1922 slave->delay = bond->params.updelay;
1925 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1926 bond->dev->name, slave->dev->name,
1927 ignore_updelay ? 0 :
1928 bond->params.updelay *
1929 bond->params.miimon);
1932 case BOND_LINK_BACK:
1934 slave->link = BOND_LINK_DOWN;
1935 pr_info("%s: link status down again after %d ms for interface %s.\n",
1937 (bond->params.updelay - slave->delay) *
1938 bond->params.miimon,
1947 if (slave->delay <= 0) {
1948 slave->new_link = BOND_LINK_UP;
1950 ignore_updelay = false;
1962 static void bond_miimon_commit(struct bonding *bond)
1964 struct list_head *iter;
1965 struct slave *slave;
1967 bond_for_each_slave(bond, slave, iter) {
1968 switch (slave->new_link) {
1969 case BOND_LINK_NOCHANGE:
1973 slave->link = BOND_LINK_UP;
1974 slave->jiffies = jiffies;
1976 if (bond->params.mode == BOND_MODE_8023AD) {
1977 /* prevent it from being the active one */
1978 bond_set_backup_slave(slave);
1979 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1980 /* make it immediately active */
1981 bond_set_active_slave(slave);
1982 } else if (slave != bond->primary_slave) {
1983 /* prevent it from being the active one */
1984 bond_set_backup_slave(slave);
1987 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
1988 bond->dev->name, slave->dev->name,
1989 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1990 slave->duplex ? "full" : "half");
1992 /* notify ad that the link status has changed */
1993 if (bond->params.mode == BOND_MODE_8023AD)
1994 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
1996 if (bond_is_lb(bond))
1997 bond_alb_handle_link_change(bond, slave,
2000 if (!bond->curr_active_slave ||
2001 (slave == bond->primary_slave))
2006 case BOND_LINK_DOWN:
2007 if (slave->link_failure_count < UINT_MAX)
2008 slave->link_failure_count++;
2010 slave->link = BOND_LINK_DOWN;
2012 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2013 bond->params.mode == BOND_MODE_8023AD)
2014 bond_set_slave_inactive_flags(slave);
2016 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2017 bond->dev->name, slave->dev->name);
2019 if (bond->params.mode == BOND_MODE_8023AD)
2020 bond_3ad_handle_link_change(slave,
2023 if (bond_is_lb(bond))
2024 bond_alb_handle_link_change(bond, slave,
2027 if (slave == bond->curr_active_slave)
2033 pr_err("%s: invalid new link %d on slave %s\n",
2034 bond->dev->name, slave->new_link,
2036 slave->new_link = BOND_LINK_NOCHANGE;
2044 write_lock_bh(&bond->curr_slave_lock);
2045 bond_select_active_slave(bond);
2046 write_unlock_bh(&bond->curr_slave_lock);
2047 unblock_netpoll_tx();
2050 bond_set_carrier(bond);
2056 * Really a wrapper that splits the mii monitor into two phases: an
2057 * inspection, then (if inspection indicates something needs to be done)
2058 * an acquisition of appropriate locks followed by a commit phase to
2059 * implement whatever link state changes are indicated.
2061 static void bond_mii_monitor(struct work_struct *work)
2063 struct bonding *bond = container_of(work, struct bonding,
2065 bool should_notify_peers = false;
2066 unsigned long delay;
2068 delay = msecs_to_jiffies(bond->params.miimon);
2070 if (!bond_has_slaves(bond))
2075 should_notify_peers = bond_should_notify_peers(bond);
2077 if (bond_miimon_inspect(bond)) {
2080 /* Race avoidance with bond_close cancel of workqueue */
2081 if (!rtnl_trylock()) {
2083 should_notify_peers = false;
2087 bond_miimon_commit(bond);
2089 rtnl_unlock(); /* might sleep, hold no other locks */
2094 if (bond->params.miimon)
2095 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2097 if (should_notify_peers) {
2098 if (!rtnl_trylock())
2100 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2105 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2107 struct net_device *upper;
2108 struct list_head *iter;
2111 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2115 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2116 if (ip == bond_confirm_addr(upper, 0, ip)) {
2127 * We go to the (large) trouble of VLAN tagging ARP frames because
2128 * switches in VLAN mode (especially if ports are configured as
2129 * "native" to a VLAN) might not pass non-tagged frames.
2131 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2133 struct sk_buff *skb;
2135 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2136 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2138 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2139 NULL, slave_dev->dev_addr, NULL);
2142 pr_err("ARP packet allocation failed\n");
2146 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2148 pr_err("failed to insert VLAN tag\n");
2156 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2158 struct net_device *upper, *vlan_upper;
2159 struct list_head *iter, *vlan_iter;
2161 __be32 *targets = bond->params.arp_targets, addr;
2164 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2165 pr_debug("basa: target %pI4\n", &targets[i]);
2167 /* Find out through which dev should the packet go */
2168 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2171 pr_debug("%s: no route to arp_ip_target %pI4\n",
2172 bond->dev->name, &targets[i]);
2178 /* bond device itself */
2179 if (rt->dst.dev == bond->dev)
2183 /* first we search only for vlan devices. for every vlan
2184 * found we verify its upper dev list, searching for the
2185 * rt->dst.dev. If found we save the tag of the vlan and
2186 * proceed to send the packet.
2190 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2192 if (!is_vlan_dev(vlan_upper))
2194 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2196 if (upper == rt->dst.dev) {
2197 vlan_id = vlan_dev_vlan_id(vlan_upper);
2204 /* if the device we're looking for is not on top of any of
2205 * our upper vlans, then just search for any dev that
2206 * matches, and in case it's a vlan - save the id
2208 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2209 if (upper == rt->dst.dev) {
2210 /* if it's a vlan - get its VID */
2211 if (is_vlan_dev(upper))
2212 vlan_id = vlan_dev_vlan_id(upper);
2220 /* Not our device - skip */
2221 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2222 bond->dev->name, &targets[i],
2223 rt->dst.dev ? rt->dst.dev->name : "NULL");
2229 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2231 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2236 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2240 if (!sip || !bond_has_this_ip(bond, tip)) {
2241 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2245 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2247 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2250 slave->last_arp_rx = jiffies;
2251 slave->target_last_arp_rx[i] = jiffies;
2254 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2255 struct slave *slave)
2257 struct arphdr *arp = (struct arphdr *)skb->data;
2258 unsigned char *arp_ptr;
2262 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2263 return RX_HANDLER_ANOTHER;
2265 read_lock(&bond->lock);
2267 if (!slave_do_arp_validate(bond, slave))
2270 alen = arp_hdr_len(bond->dev);
2272 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2273 bond->dev->name, skb->dev->name);
2275 if (alen > skb_headlen(skb)) {
2276 arp = kmalloc(alen, GFP_ATOMIC);
2279 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2283 if (arp->ar_hln != bond->dev->addr_len ||
2284 skb->pkt_type == PACKET_OTHERHOST ||
2285 skb->pkt_type == PACKET_LOOPBACK ||
2286 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2287 arp->ar_pro != htons(ETH_P_IP) ||
2291 arp_ptr = (unsigned char *)(arp + 1);
2292 arp_ptr += bond->dev->addr_len;
2293 memcpy(&sip, arp_ptr, 4);
2294 arp_ptr += 4 + bond->dev->addr_len;
2295 memcpy(&tip, arp_ptr, 4);
2297 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2298 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2299 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2303 * Backup slaves won't see the ARP reply, but do come through
2304 * here for each ARP probe (so we swap the sip/tip to validate
2305 * the probe). In a "redundant switch, common router" type of
2306 * configuration, the ARP probe will (hopefully) travel from
2307 * the active, through one switch, the router, then the other
2308 * switch before reaching the backup.
2310 * We 'trust' the arp requests if there is an active slave and
2311 * it received valid arp reply(s) after it became active. This
2312 * is done to avoid endless looping when we can't reach the
2313 * arp_ip_target and fool ourselves with our own arp requests.
2315 if (bond_is_active_slave(slave))
2316 bond_validate_arp(bond, slave, sip, tip);
2317 else if (bond->curr_active_slave &&
2318 time_after(slave_last_rx(bond, bond->curr_active_slave),
2319 bond->curr_active_slave->jiffies))
2320 bond_validate_arp(bond, slave, tip, sip);
2323 read_unlock(&bond->lock);
2324 if (arp != (struct arphdr *)skb->data)
2326 return RX_HANDLER_ANOTHER;
2329 /* function to verify if we're in the arp_interval timeslice, returns true if
2330 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2331 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2333 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2336 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2338 return time_in_range(jiffies,
2339 last_act - delta_in_ticks,
2340 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2344 * this function is called regularly to monitor each slave's link
2345 * ensuring that traffic is being sent and received when arp monitoring
2346 * is used in load-balancing mode. if the adapter has been dormant, then an
2347 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2348 * arp monitoring in active backup mode.
2350 static void bond_loadbalance_arp_mon(struct work_struct *work)
2352 struct bonding *bond = container_of(work, struct bonding,
2354 struct slave *slave, *oldcurrent;
2355 struct list_head *iter;
2356 int do_failover = 0;
2358 if (!bond_has_slaves(bond))
2363 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2364 /* see if any of the previous devices are up now (i.e. they have
2365 * xmt and rcv traffic). the curr_active_slave does not come into
2366 * the picture unless it is null. also, slave->jiffies is not needed
2367 * here because we send an arp on each slave and give a slave as
2368 * long as it needs to get the tx/rx within the delta.
2369 * TODO: what about up/down delay in arp mode? it wasn't here before
2372 bond_for_each_slave_rcu(bond, slave, iter) {
2373 unsigned long trans_start = dev_trans_start(slave->dev);
2375 if (slave->link != BOND_LINK_UP) {
2376 if (bond_time_in_interval(bond, trans_start, 1) &&
2377 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2379 slave->link = BOND_LINK_UP;
2380 bond_set_active_slave(slave);
2382 /* primary_slave has no meaning in round-robin
2383 * mode. the window of a slave being up and
2384 * curr_active_slave being null after enslaving
2388 pr_info("%s: link status definitely up for interface %s, ",
2393 pr_info("%s: interface %s is now up\n",
2399 /* slave->link == BOND_LINK_UP */
2401 /* not all switches will respond to an arp request
2402 * when the source ip is 0, so don't take the link down
2403 * if we don't know our ip yet
2405 if (!bond_time_in_interval(bond, trans_start, 2) ||
2406 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2408 slave->link = BOND_LINK_DOWN;
2409 bond_set_backup_slave(slave);
2411 if (slave->link_failure_count < UINT_MAX)
2412 slave->link_failure_count++;
2414 pr_info("%s: interface %s is now down.\n",
2418 if (slave == oldcurrent)
2423 /* note: if switch is in round-robin mode, all links
2424 * must tx arp to ensure all links rx an arp - otherwise
2425 * links may oscillate or not come up at all; if switch is
2426 * in something like xor mode, there is nothing we can
2427 * do - all replies will be rx'ed on same link causing slaves
2428 * to be unstable during low/no traffic periods
2430 if (IS_UP(slave->dev))
2431 bond_arp_send_all(bond, slave);
2437 /* the bond_select_active_slave must hold RTNL
2438 * and curr_slave_lock for write.
2440 if (!rtnl_trylock())
2443 write_lock_bh(&bond->curr_slave_lock);
2445 bond_select_active_slave(bond);
2447 write_unlock_bh(&bond->curr_slave_lock);
2448 unblock_netpoll_tx();
2453 if (bond->params.arp_interval)
2454 queue_delayed_work(bond->wq, &bond->arp_work,
2455 msecs_to_jiffies(bond->params.arp_interval));
2459 * Called to inspect slaves for active-backup mode ARP monitor link state
2460 * changes. Sets new_link in slaves to specify what action should take
2461 * place for the slave. Returns 0 if no changes are found, >0 if changes
2462 * to link states must be committed.
2464 * Called with rcu_read_lock hold.
2466 static int bond_ab_arp_inspect(struct bonding *bond)
2468 unsigned long trans_start, last_rx;
2469 struct list_head *iter;
2470 struct slave *slave;
2473 bond_for_each_slave_rcu(bond, slave, iter) {
2474 slave->new_link = BOND_LINK_NOCHANGE;
2475 last_rx = slave_last_rx(bond, slave);
2477 if (slave->link != BOND_LINK_UP) {
2478 if (bond_time_in_interval(bond, last_rx, 1)) {
2479 slave->new_link = BOND_LINK_UP;
2486 * Give slaves 2*delta after being enslaved or made
2487 * active. This avoids bouncing, as the last receive
2488 * times need a full ARP monitor cycle to be updated.
2490 if (bond_time_in_interval(bond, slave->jiffies, 2))
2494 * Backup slave is down if:
2495 * - No current_arp_slave AND
2496 * - more than 3*delta since last receive AND
2497 * - the bond has an IP address
2499 * Note: a non-null current_arp_slave indicates
2500 * the curr_active_slave went down and we are
2501 * searching for a new one; under this condition
2502 * we only take the curr_active_slave down - this
2503 * gives each slave a chance to tx/rx traffic
2504 * before being taken out
2506 if (!bond_is_active_slave(slave) &&
2507 !bond->current_arp_slave &&
2508 !bond_time_in_interval(bond, last_rx, 3)) {
2509 slave->new_link = BOND_LINK_DOWN;
2514 * Active slave is down if:
2515 * - more than 2*delta since transmitting OR
2516 * - (more than 2*delta since receive AND
2517 * the bond has an IP address)
2519 trans_start = dev_trans_start(slave->dev);
2520 if (bond_is_active_slave(slave) &&
2521 (!bond_time_in_interval(bond, trans_start, 2) ||
2522 !bond_time_in_interval(bond, last_rx, 2))) {
2523 slave->new_link = BOND_LINK_DOWN;
2532 * Called to commit link state changes noted by inspection step of
2533 * active-backup mode ARP monitor.
2535 * Called with RTNL hold.
2537 static void bond_ab_arp_commit(struct bonding *bond)
2539 unsigned long trans_start;
2540 struct list_head *iter;
2541 struct slave *slave;
2543 bond_for_each_slave(bond, slave, iter) {
2544 switch (slave->new_link) {
2545 case BOND_LINK_NOCHANGE:
2549 trans_start = dev_trans_start(slave->dev);
2550 if (bond->curr_active_slave != slave ||
2551 (!bond->curr_active_slave &&
2552 bond_time_in_interval(bond, trans_start, 1))) {
2553 slave->link = BOND_LINK_UP;
2554 if (bond->current_arp_slave) {
2555 bond_set_slave_inactive_flags(
2556 bond->current_arp_slave);
2557 bond->current_arp_slave = NULL;
2560 pr_info("%s: link status definitely up for interface %s.\n",
2561 bond->dev->name, slave->dev->name);
2563 if (!bond->curr_active_slave ||
2564 (slave == bond->primary_slave))
2571 case BOND_LINK_DOWN:
2572 if (slave->link_failure_count < UINT_MAX)
2573 slave->link_failure_count++;
2575 slave->link = BOND_LINK_DOWN;
2576 bond_set_slave_inactive_flags(slave);
2578 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2579 bond->dev->name, slave->dev->name);
2581 if (slave == bond->curr_active_slave) {
2582 bond->current_arp_slave = NULL;
2589 pr_err("%s: impossible: new_link %d on slave %s\n",
2590 bond->dev->name, slave->new_link,
2598 write_lock_bh(&bond->curr_slave_lock);
2599 bond_select_active_slave(bond);
2600 write_unlock_bh(&bond->curr_slave_lock);
2601 unblock_netpoll_tx();
2604 bond_set_carrier(bond);
2608 * Send ARP probes for active-backup mode ARP monitor.
2610 * Called with rcu_read_lock hold.
2612 static void bond_ab_arp_probe(struct bonding *bond)
2614 struct slave *slave, *before = NULL, *new_slave = NULL,
2615 *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2616 struct list_head *iter;
2619 read_lock(&bond->curr_slave_lock);
2621 if (curr_arp_slave && bond->curr_active_slave)
2622 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2623 curr_arp_slave->dev->name,
2624 bond->curr_active_slave->dev->name);
2626 if (bond->curr_active_slave) {
2627 bond_arp_send_all(bond, bond->curr_active_slave);
2628 read_unlock(&bond->curr_slave_lock);
2632 read_unlock(&bond->curr_slave_lock);
2634 /* if we don't have a curr_active_slave, search for the next available
2635 * backup slave from the current_arp_slave and make it the candidate
2636 * for becoming the curr_active_slave
2639 if (!curr_arp_slave) {
2640 curr_arp_slave = bond_first_slave_rcu(bond);
2641 if (!curr_arp_slave)
2645 bond_set_slave_inactive_flags(curr_arp_slave);
2647 bond_for_each_slave_rcu(bond, slave, iter) {
2648 if (!found && !before && IS_UP(slave->dev))
2651 if (found && !new_slave && IS_UP(slave->dev))
2653 /* if the link state is up at this point, we
2654 * mark it down - this can happen if we have
2655 * simultaneous link failures and
2656 * reselect_active_interface doesn't make this
2657 * one the current slave so it is still marked
2658 * up when it is actually down
2660 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2661 slave->link = BOND_LINK_DOWN;
2662 if (slave->link_failure_count < UINT_MAX)
2663 slave->link_failure_count++;
2665 bond_set_slave_inactive_flags(slave);
2667 pr_info("%s: backup interface %s is now down.\n",
2668 bond->dev->name, slave->dev->name);
2670 if (slave == curr_arp_slave)
2674 if (!new_slave && before)
2680 new_slave->link = BOND_LINK_BACK;
2681 bond_set_slave_active_flags(new_slave);
2682 bond_arp_send_all(bond, new_slave);
2683 new_slave->jiffies = jiffies;
2684 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2687 static void bond_activebackup_arp_mon(struct work_struct *work)
2689 struct bonding *bond = container_of(work, struct bonding,
2691 bool should_notify_peers = false;
2694 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2696 if (!bond_has_slaves(bond))
2701 should_notify_peers = bond_should_notify_peers(bond);
2703 if (bond_ab_arp_inspect(bond)) {
2706 /* Race avoidance with bond_close flush of workqueue */
2707 if (!rtnl_trylock()) {
2709 should_notify_peers = false;
2713 bond_ab_arp_commit(bond);
2719 bond_ab_arp_probe(bond);
2723 if (bond->params.arp_interval)
2724 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2726 if (should_notify_peers) {
2727 if (!rtnl_trylock())
2729 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2734 /*-------------------------- netdev event handling --------------------------*/
2737 * Change device name
2739 static int bond_event_changename(struct bonding *bond)
2741 bond_remove_proc_entry(bond);
2742 bond_create_proc_entry(bond);
2744 bond_debug_reregister(bond);
2749 static int bond_master_netdev_event(unsigned long event,
2750 struct net_device *bond_dev)
2752 struct bonding *event_bond = netdev_priv(bond_dev);
2755 case NETDEV_CHANGENAME:
2756 return bond_event_changename(event_bond);
2757 case NETDEV_UNREGISTER:
2758 bond_remove_proc_entry(event_bond);
2760 case NETDEV_REGISTER:
2761 bond_create_proc_entry(event_bond);
2763 case NETDEV_NOTIFY_PEERS:
2764 if (event_bond->send_peer_notif)
2765 event_bond->send_peer_notif--;
2774 static int bond_slave_netdev_event(unsigned long event,
2775 struct net_device *slave_dev)
2777 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2778 struct bonding *bond;
2779 struct net_device *bond_dev;
2783 /* A netdev event can be generated while enslaving a device
2784 * before netdev_rx_handler_register is called in which case
2785 * slave will be NULL
2789 bond_dev = slave->bond->dev;
2793 case NETDEV_UNREGISTER:
2794 if (bond_dev->type != ARPHRD_ETHER)
2795 bond_release_and_destroy(bond_dev, slave_dev);
2797 bond_release(bond_dev, slave_dev);
2801 old_speed = slave->speed;
2802 old_duplex = slave->duplex;
2804 bond_update_speed_duplex(slave);
2806 if (bond->params.mode == BOND_MODE_8023AD) {
2807 if (old_speed != slave->speed)
2808 bond_3ad_adapter_speed_changed(slave);
2809 if (old_duplex != slave->duplex)
2810 bond_3ad_adapter_duplex_changed(slave);
2815 * ... Or is it this?
2818 case NETDEV_CHANGEMTU:
2820 * TODO: Should slaves be allowed to
2821 * independently alter their MTU? For
2822 * an active-backup bond, slaves need
2823 * not be the same type of device, so
2824 * MTUs may vary. For other modes,
2825 * slaves arguably should have the
2826 * same MTUs. To do this, we'd need to
2827 * take over the slave's change_mtu
2828 * function for the duration of their
2832 case NETDEV_CHANGENAME:
2833 /* we don't care if we don't have primary set */
2834 if (!USES_PRIMARY(bond->params.mode) ||
2835 !bond->params.primary[0])
2838 if (slave == bond->primary_slave) {
2839 /* slave's name changed - he's no longer primary */
2840 bond->primary_slave = NULL;
2841 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2842 /* we have a new primary slave */
2843 bond->primary_slave = slave;
2844 } else { /* we didn't change primary - exit */
2848 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2849 bond->dev->name, bond->primary_slave ? slave_dev->name :
2851 write_lock_bh(&bond->curr_slave_lock);
2852 bond_select_active_slave(bond);
2853 write_unlock_bh(&bond->curr_slave_lock);
2855 case NETDEV_FEAT_CHANGE:
2856 bond_compute_features(bond);
2858 case NETDEV_RESEND_IGMP:
2859 /* Propagate to master device */
2860 call_netdevice_notifiers(event, slave->bond->dev);
2870 * bond_netdev_event: handle netdev notifier chain events.
2872 * This function receives events for the netdev chain. The caller (an
2873 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2874 * locks for us to safely manipulate the slave devices (RTNL lock,
2877 static int bond_netdev_event(struct notifier_block *this,
2878 unsigned long event, void *ptr)
2880 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2882 pr_debug("event_dev: %s, event: %lx\n",
2883 event_dev ? event_dev->name : "None",
2886 if (!(event_dev->priv_flags & IFF_BONDING))
2889 if (event_dev->flags & IFF_MASTER) {
2890 pr_debug("IFF_MASTER\n");
2891 return bond_master_netdev_event(event, event_dev);
2894 if (event_dev->flags & IFF_SLAVE) {
2895 pr_debug("IFF_SLAVE\n");
2896 return bond_slave_netdev_event(event, event_dev);
2902 static struct notifier_block bond_netdev_notifier = {
2903 .notifier_call = bond_netdev_event,
2906 /*---------------------------- Hashing Policies -----------------------------*/
2908 /* L2 hash helper */
2909 static inline u32 bond_eth_hash(struct sk_buff *skb)
2911 struct ethhdr *data = (struct ethhdr *)skb->data;
2913 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2914 return data->h_dest[5] ^ data->h_source[5];
2919 /* Extract the appropriate headers based on bond's xmit policy */
2920 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2921 struct flow_keys *fk)
2923 const struct ipv6hdr *iph6;
2924 const struct iphdr *iph;
2925 int noff, proto = -1;
2927 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2928 return skb_flow_dissect(skb, fk);
2931 noff = skb_network_offset(skb);
2932 if (skb->protocol == htons(ETH_P_IP)) {
2933 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2936 fk->src = iph->saddr;
2937 fk->dst = iph->daddr;
2938 noff += iph->ihl << 2;
2939 if (!ip_is_fragment(iph))
2940 proto = iph->protocol;
2941 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2942 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2944 iph6 = ipv6_hdr(skb);
2945 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2946 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2947 noff += sizeof(*iph6);
2948 proto = iph6->nexthdr;
2952 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2953 fk->ports = skb_flow_get_ports(skb, noff, proto);
2959 * bond_xmit_hash - generate a hash value based on the xmit policy
2960 * @bond: bonding device
2961 * @skb: buffer to use for headers
2962 * @count: modulo value
2964 * This function will extract the necessary headers from the skb buffer and use
2965 * them to generate a hash based on the xmit_policy set in the bonding device
2966 * which will be reduced modulo count before returning.
2968 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
2970 struct flow_keys flow;
2973 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2974 !bond_flow_dissect(bond, skb, &flow))
2975 return bond_eth_hash(skb) % count;
2977 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
2978 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
2979 hash = bond_eth_hash(skb);
2981 hash = (__force u32)flow.ports;
2982 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
2983 hash ^= (hash >> 16);
2984 hash ^= (hash >> 8);
2986 return hash % count;
2989 /*-------------------------- Device entry points ----------------------------*/
2991 static void bond_work_init_all(struct bonding *bond)
2993 INIT_DELAYED_WORK(&bond->mcast_work,
2994 bond_resend_igmp_join_requests_delayed);
2995 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
2996 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
2997 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2998 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3000 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3001 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3004 static void bond_work_cancel_all(struct bonding *bond)
3006 cancel_delayed_work_sync(&bond->mii_work);
3007 cancel_delayed_work_sync(&bond->arp_work);
3008 cancel_delayed_work_sync(&bond->alb_work);
3009 cancel_delayed_work_sync(&bond->ad_work);
3010 cancel_delayed_work_sync(&bond->mcast_work);
3013 static int bond_open(struct net_device *bond_dev)
3015 struct bonding *bond = netdev_priv(bond_dev);
3016 struct list_head *iter;
3017 struct slave *slave;
3019 /* reset slave->backup and slave->inactive */
3020 read_lock(&bond->lock);
3021 if (bond_has_slaves(bond)) {
3022 read_lock(&bond->curr_slave_lock);
3023 bond_for_each_slave(bond, slave, iter) {
3024 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3025 && (slave != bond->curr_active_slave)) {
3026 bond_set_slave_inactive_flags(slave);
3028 bond_set_slave_active_flags(slave);
3031 read_unlock(&bond->curr_slave_lock);
3033 read_unlock(&bond->lock);
3035 bond_work_init_all(bond);
3037 if (bond_is_lb(bond)) {
3038 /* bond_alb_initialize must be called before the timer
3041 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3043 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3046 if (bond->params.miimon) /* link check interval, in milliseconds. */
3047 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3049 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3050 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3051 if (bond->params.arp_validate)
3052 bond->recv_probe = bond_arp_rcv;
3055 if (bond->params.mode == BOND_MODE_8023AD) {
3056 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3057 /* register to receive LACPDUs */
3058 bond->recv_probe = bond_3ad_lacpdu_recv;
3059 bond_3ad_initiate_agg_selection(bond, 1);
3065 static int bond_close(struct net_device *bond_dev)
3067 struct bonding *bond = netdev_priv(bond_dev);
3069 bond_work_cancel_all(bond);
3070 bond->send_peer_notif = 0;
3071 if (bond_is_lb(bond))
3072 bond_alb_deinitialize(bond);
3073 bond->recv_probe = NULL;
3078 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3079 struct rtnl_link_stats64 *stats)
3081 struct bonding *bond = netdev_priv(bond_dev);
3082 struct rtnl_link_stats64 temp;
3083 struct list_head *iter;
3084 struct slave *slave;
3086 memset(stats, 0, sizeof(*stats));
3088 read_lock_bh(&bond->lock);
3089 bond_for_each_slave(bond, slave, iter) {
3090 const struct rtnl_link_stats64 *sstats =
3091 dev_get_stats(slave->dev, &temp);
3093 stats->rx_packets += sstats->rx_packets;
3094 stats->rx_bytes += sstats->rx_bytes;
3095 stats->rx_errors += sstats->rx_errors;
3096 stats->rx_dropped += sstats->rx_dropped;
3098 stats->tx_packets += sstats->tx_packets;
3099 stats->tx_bytes += sstats->tx_bytes;
3100 stats->tx_errors += sstats->tx_errors;
3101 stats->tx_dropped += sstats->tx_dropped;
3103 stats->multicast += sstats->multicast;
3104 stats->collisions += sstats->collisions;
3106 stats->rx_length_errors += sstats->rx_length_errors;
3107 stats->rx_over_errors += sstats->rx_over_errors;
3108 stats->rx_crc_errors += sstats->rx_crc_errors;
3109 stats->rx_frame_errors += sstats->rx_frame_errors;
3110 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3111 stats->rx_missed_errors += sstats->rx_missed_errors;
3113 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3114 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3115 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3116 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3117 stats->tx_window_errors += sstats->tx_window_errors;
3119 read_unlock_bh(&bond->lock);
3124 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3126 struct bonding *bond = netdev_priv(bond_dev);
3127 struct net_device *slave_dev = NULL;
3128 struct ifbond k_binfo;
3129 struct ifbond __user *u_binfo = NULL;
3130 struct ifslave k_sinfo;
3131 struct ifslave __user *u_sinfo = NULL;
3132 struct mii_ioctl_data *mii = NULL;
3136 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3148 * We do this again just in case we were called by SIOCGMIIREG
3149 * instead of SIOCGMIIPHY.
3156 if (mii->reg_num == 1) {
3158 read_lock(&bond->lock);
3159 read_lock(&bond->curr_slave_lock);
3160 if (netif_carrier_ok(bond->dev))
3161 mii->val_out = BMSR_LSTATUS;
3163 read_unlock(&bond->curr_slave_lock);
3164 read_unlock(&bond->lock);
3168 case BOND_INFO_QUERY_OLD:
3169 case SIOCBONDINFOQUERY:
3170 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3172 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3175 res = bond_info_query(bond_dev, &k_binfo);
3177 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3181 case BOND_SLAVE_INFO_QUERY_OLD:
3182 case SIOCBONDSLAVEINFOQUERY:
3183 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3185 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3188 res = bond_slave_info_query(bond_dev, &k_sinfo);
3190 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3199 net = dev_net(bond_dev);
3201 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3204 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3206 pr_debug("slave_dev=%p:\n", slave_dev);
3211 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3213 case BOND_ENSLAVE_OLD:
3214 case SIOCBONDENSLAVE:
3215 res = bond_enslave(bond_dev, slave_dev);
3217 case BOND_RELEASE_OLD:
3218 case SIOCBONDRELEASE:
3219 res = bond_release(bond_dev, slave_dev);
3221 case BOND_SETHWADDR_OLD:
3222 case SIOCBONDSETHWADDR:
3223 bond_set_dev_addr(bond_dev, slave_dev);
3226 case BOND_CHANGE_ACTIVE_OLD:
3227 case SIOCBONDCHANGEACTIVE:
3228 res = bond_option_active_slave_set(bond, slave_dev);
3237 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3239 struct bonding *bond = netdev_priv(bond_dev);
3241 if (change & IFF_PROMISC)
3242 bond_set_promiscuity(bond,
3243 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3245 if (change & IFF_ALLMULTI)
3246 bond_set_allmulti(bond,
3247 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3250 static void bond_set_rx_mode(struct net_device *bond_dev)
3252 struct bonding *bond = netdev_priv(bond_dev);
3253 struct list_head *iter;
3254 struct slave *slave;
3258 if (USES_PRIMARY(bond->params.mode)) {
3259 slave = rcu_dereference(bond->curr_active_slave);
3261 dev_uc_sync(slave->dev, bond_dev);
3262 dev_mc_sync(slave->dev, bond_dev);
3265 bond_for_each_slave_rcu(bond, slave, iter) {
3266 dev_uc_sync_multiple(slave->dev, bond_dev);
3267 dev_mc_sync_multiple(slave->dev, bond_dev);
3273 static int bond_neigh_init(struct neighbour *n)
3275 struct bonding *bond = netdev_priv(n->dev);
3276 const struct net_device_ops *slave_ops;
3277 struct neigh_parms parms;
3278 struct slave *slave;
3281 slave = bond_first_slave(bond);
3284 slave_ops = slave->dev->netdev_ops;
3285 if (!slave_ops->ndo_neigh_setup)
3288 parms.neigh_setup = NULL;
3289 parms.neigh_cleanup = NULL;
3290 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3295 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3296 * after the last slave has been detached. Assumes that all slaves
3297 * utilize the same neigh_cleanup (true at this writing as only user
3300 n->parms->neigh_cleanup = parms.neigh_cleanup;
3302 if (!parms.neigh_setup)
3305 return parms.neigh_setup(n);
3309 * The bonding ndo_neigh_setup is called at init time beofre any
3310 * slave exists. So we must declare proxy setup function which will
3311 * be used at run time to resolve the actual slave neigh param setup.
3313 * It's also called by master devices (such as vlans) to setup their
3314 * underlying devices. In that case - do nothing, we're already set up from
3317 static int bond_neigh_setup(struct net_device *dev,
3318 struct neigh_parms *parms)
3320 /* modify only our neigh_parms */
3321 if (parms->dev == dev)
3322 parms->neigh_setup = bond_neigh_init;
3328 * Change the MTU of all of a master's slaves to match the master
3330 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3332 struct bonding *bond = netdev_priv(bond_dev);
3333 struct slave *slave, *rollback_slave;
3334 struct list_head *iter;
3337 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3338 (bond_dev ? bond_dev->name : "None"), new_mtu);
3340 /* Can't hold bond->lock with bh disabled here since
3341 * some base drivers panic. On the other hand we can't
3342 * hold bond->lock without bh disabled because we'll
3343 * deadlock. The only solution is to rely on the fact
3344 * that we're under rtnl_lock here, and the slaves
3345 * list won't change. This doesn't solve the problem
3346 * of setting the slave's MTU while it is
3347 * transmitting, but the assumption is that the base
3348 * driver can handle that.
3350 * TODO: figure out a way to safely iterate the slaves
3351 * list, but without holding a lock around the actual
3352 * call to the base driver.
3355 bond_for_each_slave(bond, slave, iter) {
3356 pr_debug("s %p c_m %p\n",
3358 slave->dev->netdev_ops->ndo_change_mtu);
3360 res = dev_set_mtu(slave->dev, new_mtu);
3363 /* If we failed to set the slave's mtu to the new value
3364 * we must abort the operation even in ACTIVE_BACKUP
3365 * mode, because if we allow the backup slaves to have
3366 * different mtu values than the active slave we'll
3367 * need to change their mtu when doing a failover. That
3368 * means changing their mtu from timer context, which
3369 * is probably not a good idea.
3371 pr_debug("err %d %s\n", res, slave->dev->name);
3376 bond_dev->mtu = new_mtu;
3381 /* unwind from head to the slave that failed */
3382 bond_for_each_slave(bond, rollback_slave, iter) {
3385 if (rollback_slave == slave)
3388 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3390 pr_debug("unwind err %d dev %s\n",
3391 tmp_res, rollback_slave->dev->name);
3401 * Note that many devices must be down to change the HW address, and
3402 * downing the master releases all slaves. We can make bonds full of
3403 * bonding devices to test this, however.
3405 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3407 struct bonding *bond = netdev_priv(bond_dev);
3408 struct slave *slave, *rollback_slave;
3409 struct sockaddr *sa = addr, tmp_sa;
3410 struct list_head *iter;
3413 if (bond->params.mode == BOND_MODE_ALB)
3414 return bond_alb_set_mac_address(bond_dev, addr);
3417 pr_debug("bond=%p, name=%s\n",
3418 bond, bond_dev ? bond_dev->name : "None");
3420 /* If fail_over_mac is enabled, do nothing and return success.
3421 * Returning an error causes ifenslave to fail.
3423 if (bond->params.fail_over_mac)
3426 if (!is_valid_ether_addr(sa->sa_data))
3427 return -EADDRNOTAVAIL;
3429 /* Can't hold bond->lock with bh disabled here since
3430 * some base drivers panic. On the other hand we can't
3431 * hold bond->lock without bh disabled because we'll
3432 * deadlock. The only solution is to rely on the fact
3433 * that we're under rtnl_lock here, and the slaves
3434 * list won't change. This doesn't solve the problem
3435 * of setting the slave's hw address while it is
3436 * transmitting, but the assumption is that the base
3437 * driver can handle that.
3439 * TODO: figure out a way to safely iterate the slaves
3440 * list, but without holding a lock around the actual
3441 * call to the base driver.
3444 bond_for_each_slave(bond, slave, iter) {
3445 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3446 pr_debug("slave %p %s\n", slave, slave->dev->name);
3448 if (slave_ops->ndo_set_mac_address == NULL) {
3450 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3454 res = dev_set_mac_address(slave->dev, addr);
3456 /* TODO: consider downing the slave
3458 * User should expect communications
3459 * breakage anyway until ARP finish
3462 pr_debug("err %d %s\n", res, slave->dev->name);
3468 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3472 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3473 tmp_sa.sa_family = bond_dev->type;
3475 /* unwind from head to the slave that failed */
3476 bond_for_each_slave(bond, rollback_slave, iter) {
3479 if (rollback_slave == slave)
3482 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3484 pr_debug("unwind err %d dev %s\n",
3485 tmp_res, rollback_slave->dev->name);
3493 * bond_xmit_slave_id - transmit skb through slave with slave_id
3494 * @bond: bonding device that is transmitting
3495 * @skb: buffer to transmit
3496 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3498 * This function tries to transmit through slave with slave_id but in case
3499 * it fails, it tries to find the first available slave for transmission.
3500 * The skb is consumed in all cases, thus the function is void.
3502 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3504 struct list_head *iter;
3505 struct slave *slave;
3508 /* Here we start from the slave with slave_id */
3509 bond_for_each_slave_rcu(bond, slave, iter) {
3511 if (slave_can_tx(slave)) {
3512 bond_dev_queue_xmit(bond, skb, slave->dev);
3518 /* Here we start from the first slave up to slave_id */
3520 bond_for_each_slave_rcu(bond, slave, iter) {
3523 if (slave_can_tx(slave)) {
3524 bond_dev_queue_xmit(bond, skb, slave->dev);
3528 /* no slave that can tx has been found */
3533 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3534 * @bond: bonding device to use
3536 * Based on the value of the bonding device's packets_per_slave parameter
3537 * this function generates a slave id, which is usually used as the next
3538 * slave to transmit through.
3540 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3543 struct reciprocal_value reciprocal_packets_per_slave;
3544 int packets_per_slave = bond->params.packets_per_slave;
3546 switch (packets_per_slave) {
3548 slave_id = prandom_u32();
3551 slave_id = bond->rr_tx_counter;
3554 reciprocal_packets_per_slave =
3555 bond->params.reciprocal_packets_per_slave;
3556 slave_id = reciprocal_divide(bond->rr_tx_counter,
3557 reciprocal_packets_per_slave);
3560 bond->rr_tx_counter++;
3565 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3567 struct bonding *bond = netdev_priv(bond_dev);
3568 struct iphdr *iph = ip_hdr(skb);
3569 struct slave *slave;
3572 /* Start with the curr_active_slave that joined the bond as the
3573 * default for sending IGMP traffic. For failover purposes one
3574 * needs to maintain some consistency for the interface that will
3575 * send the join/membership reports. The curr_active_slave found
3576 * will send all of this type of traffic.
3578 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3579 slave = rcu_dereference(bond->curr_active_slave);
3580 if (slave && slave_can_tx(slave))
3581 bond_dev_queue_xmit(bond, skb, slave->dev);
3583 bond_xmit_slave_id(bond, skb, 0);
3585 slave_id = bond_rr_gen_slave_id(bond);
3586 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3589 return NETDEV_TX_OK;
3593 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3594 * the bond has a usable interface.
3596 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3598 struct bonding *bond = netdev_priv(bond_dev);
3599 struct slave *slave;
3601 slave = rcu_dereference(bond->curr_active_slave);
3603 bond_dev_queue_xmit(bond, skb, slave->dev);
3607 return NETDEV_TX_OK;
3610 /* In bond_xmit_xor() , we determine the output device by using a pre-
3611 * determined xmit_hash_policy(), If the selected device is not enabled,
3612 * find the next active slave.
3614 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3616 struct bonding *bond = netdev_priv(bond_dev);
3618 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3620 return NETDEV_TX_OK;
3623 /* in broadcast mode, we send everything to all usable interfaces. */
3624 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3626 struct bonding *bond = netdev_priv(bond_dev);
3627 struct slave *slave = NULL;
3628 struct list_head *iter;
3630 bond_for_each_slave_rcu(bond, slave, iter) {
3631 if (bond_is_last_slave(bond, slave))
3633 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3634 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3637 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3641 /* bond_dev_queue_xmit always returns 0 */
3642 bond_dev_queue_xmit(bond, skb2, slave->dev);
3645 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3646 bond_dev_queue_xmit(bond, skb, slave->dev);
3650 return NETDEV_TX_OK;
3653 /*------------------------- Device initialization ---------------------------*/
3656 * Lookup the slave that corresponds to a qid
3658 static inline int bond_slave_override(struct bonding *bond,
3659 struct sk_buff *skb)
3661 struct slave *slave = NULL;
3662 struct list_head *iter;
3664 if (!skb->queue_mapping)
3667 /* Find out if any slaves have the same mapping as this skb. */
3668 bond_for_each_slave_rcu(bond, slave, iter) {
3669 if (slave->queue_id == skb->queue_mapping) {
3670 if (slave_can_tx(slave)) {
3671 bond_dev_queue_xmit(bond, skb, slave->dev);
3674 /* If the slave isn't UP, use default transmit policy. */
3683 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3687 * This helper function exists to help dev_pick_tx get the correct
3688 * destination queue. Using a helper function skips a call to
3689 * skb_tx_hash and will put the skbs in the queue we expect on their
3690 * way down to the bonding driver.
3692 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3695 * Save the original txq to restore before passing to the driver
3697 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3699 if (unlikely(txq >= dev->real_num_tx_queues)) {
3701 txq -= dev->real_num_tx_queues;
3702 } while (txq >= dev->real_num_tx_queues);
3707 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3709 struct bonding *bond = netdev_priv(dev);
3711 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3712 if (!bond_slave_override(bond, skb))
3713 return NETDEV_TX_OK;
3716 switch (bond->params.mode) {
3717 case BOND_MODE_ROUNDROBIN:
3718 return bond_xmit_roundrobin(skb, dev);
3719 case BOND_MODE_ACTIVEBACKUP:
3720 return bond_xmit_activebackup(skb, dev);
3722 return bond_xmit_xor(skb, dev);
3723 case BOND_MODE_BROADCAST:
3724 return bond_xmit_broadcast(skb, dev);
3725 case BOND_MODE_8023AD:
3726 return bond_3ad_xmit_xor(skb, dev);
3729 return bond_alb_xmit(skb, dev);
3731 /* Should never happen, mode already checked */
3732 pr_err("%s: Error: Unknown bonding mode %d\n",
3733 dev->name, bond->params.mode);
3736 return NETDEV_TX_OK;
3740 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3742 struct bonding *bond = netdev_priv(dev);
3743 netdev_tx_t ret = NETDEV_TX_OK;
3746 * If we risk deadlock from transmitting this in the
3747 * netpoll path, tell netpoll to queue the frame for later tx
3749 if (is_netpoll_tx_blocked(dev))
3750 return NETDEV_TX_BUSY;
3753 if (bond_has_slaves(bond))
3754 ret = __bond_start_xmit(skb, dev);
3762 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3763 struct ethtool_cmd *ecmd)
3765 struct bonding *bond = netdev_priv(bond_dev);
3766 unsigned long speed = 0;
3767 struct list_head *iter;
3768 struct slave *slave;
3770 ecmd->duplex = DUPLEX_UNKNOWN;
3771 ecmd->port = PORT_OTHER;
3773 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3774 * do not need to check mode. Though link speed might not represent
3775 * the true receive or transmit bandwidth (not all modes are symmetric)
3776 * this is an accurate maximum.
3778 read_lock(&bond->lock);
3779 bond_for_each_slave(bond, slave, iter) {
3780 if (SLAVE_IS_OK(slave)) {
3781 if (slave->speed != SPEED_UNKNOWN)
3782 speed += slave->speed;
3783 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3784 slave->duplex != DUPLEX_UNKNOWN)
3785 ecmd->duplex = slave->duplex;
3788 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3789 read_unlock(&bond->lock);
3794 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3795 struct ethtool_drvinfo *drvinfo)
3797 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3798 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3799 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3803 static const struct ethtool_ops bond_ethtool_ops = {
3804 .get_drvinfo = bond_ethtool_get_drvinfo,
3805 .get_settings = bond_ethtool_get_settings,
3806 .get_link = ethtool_op_get_link,
3809 static const struct net_device_ops bond_netdev_ops = {
3810 .ndo_init = bond_init,
3811 .ndo_uninit = bond_uninit,
3812 .ndo_open = bond_open,
3813 .ndo_stop = bond_close,
3814 .ndo_start_xmit = bond_start_xmit,
3815 .ndo_select_queue = bond_select_queue,
3816 .ndo_get_stats64 = bond_get_stats,
3817 .ndo_do_ioctl = bond_do_ioctl,
3818 .ndo_change_rx_flags = bond_change_rx_flags,
3819 .ndo_set_rx_mode = bond_set_rx_mode,
3820 .ndo_change_mtu = bond_change_mtu,
3821 .ndo_set_mac_address = bond_set_mac_address,
3822 .ndo_neigh_setup = bond_neigh_setup,
3823 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3824 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3825 #ifdef CONFIG_NET_POLL_CONTROLLER
3826 .ndo_netpoll_setup = bond_netpoll_setup,
3827 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3828 .ndo_poll_controller = bond_poll_controller,
3830 .ndo_add_slave = bond_enslave,
3831 .ndo_del_slave = bond_release,
3832 .ndo_get_slave = bond_get_slave,
3833 .ndo_fix_features = bond_fix_features,
3836 static const struct device_type bond_type = {
3840 static void bond_destructor(struct net_device *bond_dev)
3842 struct bonding *bond = netdev_priv(bond_dev);
3844 destroy_workqueue(bond->wq);
3845 free_netdev(bond_dev);
3848 void bond_setup(struct net_device *bond_dev)
3850 struct bonding *bond = netdev_priv(bond_dev);
3852 /* initialize rwlocks */
3853 rwlock_init(&bond->lock);
3854 rwlock_init(&bond->curr_slave_lock);
3855 bond->params = bonding_defaults;
3857 /* Initialize pointers */
3858 bond->dev = bond_dev;
3860 /* Initialize the device entry points */
3861 ether_setup(bond_dev);
3862 bond_dev->netdev_ops = &bond_netdev_ops;
3863 bond_dev->ethtool_ops = &bond_ethtool_ops;
3865 bond_dev->destructor = bond_destructor;
3867 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3869 /* Initialize the device options */
3870 bond_dev->tx_queue_len = 0;
3871 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3872 bond_dev->priv_flags |= IFF_BONDING;
3873 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3875 /* At first, we block adding VLANs. That's the only way to
3876 * prevent problems that occur when adding VLANs over an
3877 * empty bond. The block will be removed once non-challenged
3878 * slaves are enslaved.
3880 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3882 /* don't acquire bond device's netif_tx_lock when
3884 bond_dev->features |= NETIF_F_LLTX;
3886 /* By default, we declare the bond to be fully
3887 * VLAN hardware accelerated capable. Special
3888 * care is taken in the various xmit functions
3889 * when there are slaves that are not hw accel
3893 bond_dev->hw_features = BOND_VLAN_FEATURES |
3894 NETIF_F_HW_VLAN_CTAG_TX |
3895 NETIF_F_HW_VLAN_CTAG_RX |
3896 NETIF_F_HW_VLAN_CTAG_FILTER;
3898 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3899 bond_dev->features |= bond_dev->hw_features;
3903 * Destroy a bonding device.
3904 * Must be under rtnl_lock when this function is called.
3906 static void bond_uninit(struct net_device *bond_dev)
3908 struct bonding *bond = netdev_priv(bond_dev);
3909 struct list_head *iter;
3910 struct slave *slave;
3912 bond_netpoll_cleanup(bond_dev);
3914 /* Release the bonded slaves */
3915 bond_for_each_slave(bond, slave, iter)
3916 __bond_release_one(bond_dev, slave->dev, true);
3917 pr_info("%s: released all slaves\n", bond_dev->name);
3919 list_del(&bond->bond_list);
3921 bond_debug_unregister(bond);
3924 /*------------------------- Module initialization ---------------------------*/
3926 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3930 for (i = 0; tbl[i].modename; i++)
3931 if (mode == tbl[i].mode)
3937 static int bond_parm_tbl_lookup_name(const char *modename,
3938 const struct bond_parm_tbl *tbl)
3942 for (i = 0; tbl[i].modename; i++)
3943 if (strcmp(modename, tbl[i].modename) == 0)
3950 * Convert string input module parms. Accept either the
3951 * number of the mode or its string name. A bit complicated because
3952 * some mode names are substrings of other names, and calls from sysfs
3953 * may have whitespace in the name (trailing newlines, for example).
3955 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3958 char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
3960 for (p = (char *)buf; *p; p++)
3961 if (!(isdigit(*p) || isspace(*p)))
3964 if (*p && sscanf(buf, "%20s", modestr) != 0)
3965 return bond_parm_tbl_lookup_name(modestr, tbl);
3966 else if (sscanf(buf, "%d", &modeint) != 0)
3967 return bond_parm_tbl_lookup(modeint, tbl);
3972 static int bond_check_params(struct bond_params *params)
3974 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3975 struct bond_opt_value newval, *valptr;
3976 int arp_all_targets_value;
3979 * Convert string parameters.
3982 bond_opt_initstr(&newval, mode);
3983 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
3985 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
3988 bond_mode = valptr->value;
3991 if (xmit_hash_policy) {
3992 if ((bond_mode != BOND_MODE_XOR) &&
3993 (bond_mode != BOND_MODE_8023AD)) {
3994 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3995 bond_mode_name(bond_mode));
3997 bond_opt_initstr(&newval, xmit_hash_policy);
3998 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4001 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4005 xmit_hashtype = valptr->value;
4010 if (bond_mode != BOND_MODE_8023AD) {
4011 pr_info("lacp_rate param is irrelevant in mode %s\n",
4012 bond_mode_name(bond_mode));
4014 bond_opt_initstr(&newval, lacp_rate);
4015 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4018 pr_err("Error: Invalid lacp rate \"%s\"\n",
4022 lacp_fast = valptr->value;
4027 bond_opt_initstr(&newval, lacp_rate);
4028 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4031 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4034 params->ad_select = valptr->value;
4035 if (bond_mode != BOND_MODE_8023AD)
4036 pr_warning("ad_select param only affects 802.3ad mode\n");
4038 params->ad_select = BOND_AD_STABLE;
4041 if (max_bonds < 0) {
4042 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4043 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4044 max_bonds = BOND_DEFAULT_MAX_BONDS;
4048 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4054 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4059 if (downdelay < 0) {
4060 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4061 downdelay, INT_MAX);
4065 if ((use_carrier != 0) && (use_carrier != 1)) {
4066 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4071 if (num_peer_notif < 0 || num_peer_notif > 255) {
4072 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4077 /* reset values for 802.3ad/TLB/ALB */
4078 if (BOND_NO_USES_ARP(bond_mode)) {
4080 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");
4081 pr_warning("Forcing miimon to 100msec\n");
4082 miimon = BOND_DEFAULT_MIIMON;
4086 if (tx_queues < 1 || tx_queues > 255) {
4087 pr_warning("Warning: tx_queues (%d) should be between "
4088 "1 and 255, resetting to %d\n",
4089 tx_queues, BOND_DEFAULT_TX_QUEUES);
4090 tx_queues = BOND_DEFAULT_TX_QUEUES;
4093 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4094 pr_warning("Warning: all_slaves_active module parameter (%d), "
4095 "not of valid value (0/1), so it was set to "
4096 "0\n", all_slaves_active);
4097 all_slaves_active = 0;
4100 if (resend_igmp < 0 || resend_igmp > 255) {
4101 pr_warning("Warning: resend_igmp (%d) should be between "
4102 "0 and 255, resetting to %d\n",
4103 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4104 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4107 bond_opt_initval(&newval, packets_per_slave);
4108 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4109 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4110 packets_per_slave, USHRT_MAX);
4111 packets_per_slave = 1;
4114 if (bond_mode == BOND_MODE_ALB) {
4115 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",
4120 if (updelay || downdelay) {
4121 /* just warn the user the up/down delay will have
4122 * no effect since miimon is zero...
4124 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",
4125 updelay, downdelay);
4128 /* don't allow arp monitoring */
4130 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4131 miimon, arp_interval);
4135 if ((updelay % miimon) != 0) {
4136 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4138 (updelay / miimon) * miimon);
4143 if ((downdelay % miimon) != 0) {
4144 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4146 (downdelay / miimon) * miimon);
4149 downdelay /= miimon;
4152 if (arp_interval < 0) {
4153 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to 0\n",
4154 arp_interval, INT_MAX);
4158 for (arp_ip_count = 0, i = 0;
4159 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4160 /* not complete check, but should be good enough to
4163 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4164 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4165 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4169 if (bond_get_targets_ip(arp_target, ip) == -1)
4170 arp_target[arp_ip_count++] = ip;
4172 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4177 if (arp_interval && !arp_ip_count) {
4178 /* don't allow arping if no arp_ip_target given... */
4179 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4185 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4186 pr_err("arp_validate only supported in active-backup mode\n");
4189 if (!arp_interval) {
4190 pr_err("arp_validate requires arp_interval\n");
4194 bond_opt_initstr(&newval, arp_validate);
4195 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4198 pr_err("Error: invalid arp_validate \"%s\"\n",
4202 arp_validate_value = valptr->value;
4204 arp_validate_value = 0;
4207 arp_all_targets_value = 0;
4208 if (arp_all_targets) {
4209 bond_opt_initstr(&newval, arp_all_targets);
4210 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4213 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4215 arp_all_targets_value = 0;
4217 arp_all_targets_value = valptr->value;
4222 pr_info("MII link monitoring set to %d ms\n", miimon);
4223 } else if (arp_interval) {
4224 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4225 arp_validate_value);
4226 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4227 arp_interval, valptr->string, arp_ip_count);
4229 for (i = 0; i < arp_ip_count; i++)
4230 pr_info(" %s", arp_ip_target[i]);
4234 } else if (max_bonds) {
4235 /* miimon and arp_interval not set, we need one so things
4236 * work as expected, see bonding.txt for details
4238 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");
4241 if (primary && !USES_PRIMARY(bond_mode)) {
4242 /* currently, using a primary only makes sense
4243 * in active backup, TLB or ALB modes
4245 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4246 primary, bond_mode_name(bond_mode));
4250 if (primary && primary_reselect) {
4251 primary_reselect_value = bond_parse_parm(primary_reselect,
4253 if (primary_reselect_value == -1) {
4254 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4256 NULL ? "NULL" : primary_reselect);
4260 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4263 if (fail_over_mac) {
4264 bond_opt_initstr(&newval, fail_over_mac);
4265 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4268 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4272 fail_over_mac_value = valptr->value;
4273 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4274 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4276 fail_over_mac_value = BOND_FOM_NONE;
4279 if (lp_interval == 0) {
4280 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4281 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4282 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4285 /* fill params struct with the proper values */
4286 params->mode = bond_mode;
4287 params->xmit_policy = xmit_hashtype;
4288 params->miimon = miimon;
4289 params->num_peer_notif = num_peer_notif;
4290 params->arp_interval = arp_interval;
4291 params->arp_validate = arp_validate_value;
4292 params->arp_all_targets = arp_all_targets_value;
4293 params->updelay = updelay;
4294 params->downdelay = downdelay;
4295 params->use_carrier = use_carrier;
4296 params->lacp_fast = lacp_fast;
4297 params->primary[0] = 0;
4298 params->primary_reselect = primary_reselect_value;
4299 params->fail_over_mac = fail_over_mac_value;
4300 params->tx_queues = tx_queues;
4301 params->all_slaves_active = all_slaves_active;
4302 params->resend_igmp = resend_igmp;
4303 params->min_links = min_links;
4304 params->lp_interval = lp_interval;
4305 params->packets_per_slave = packets_per_slave;
4306 if (packets_per_slave > 0) {
4307 params->reciprocal_packets_per_slave =
4308 reciprocal_value(packets_per_slave);
4310 /* reciprocal_packets_per_slave is unused if
4311 * packets_per_slave is 0 or 1, just initialize it
4313 params->reciprocal_packets_per_slave =
4314 (struct reciprocal_value) { 0 };
4318 strncpy(params->primary, primary, IFNAMSIZ);
4319 params->primary[IFNAMSIZ - 1] = 0;
4322 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4327 static struct lock_class_key bonding_netdev_xmit_lock_key;
4328 static struct lock_class_key bonding_netdev_addr_lock_key;
4329 static struct lock_class_key bonding_tx_busylock_key;
4331 static void bond_set_lockdep_class_one(struct net_device *dev,
4332 struct netdev_queue *txq,
4335 lockdep_set_class(&txq->_xmit_lock,
4336 &bonding_netdev_xmit_lock_key);
4339 static void bond_set_lockdep_class(struct net_device *dev)
4341 lockdep_set_class(&dev->addr_list_lock,
4342 &bonding_netdev_addr_lock_key);
4343 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4344 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4348 * Called from registration process
4350 static int bond_init(struct net_device *bond_dev)
4352 struct bonding *bond = netdev_priv(bond_dev);
4353 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4354 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4356 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4359 * Initialize locks that may be required during
4360 * en/deslave operations. All of the bond_open work
4361 * (of which this is part) should really be moved to
4362 * a phase prior to dev_open
4364 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4365 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4367 bond->wq = create_singlethread_workqueue(bond_dev->name);
4371 bond_set_lockdep_class(bond_dev);
4373 list_add_tail(&bond->bond_list, &bn->dev_list);
4375 bond_prepare_sysfs_group(bond);
4377 bond_debug_register(bond);
4379 /* Ensure valid dev_addr */
4380 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4381 bond_dev->addr_assign_type == NET_ADDR_PERM)
4382 eth_hw_addr_random(bond_dev);
4387 unsigned int bond_get_num_tx_queues(void)
4392 /* Create a new bond based on the specified name and bonding parameters.
4393 * If name is NULL, obtain a suitable "bond%d" name for us.
4394 * Caller must NOT hold rtnl_lock; we need to release it here before we
4395 * set up our sysfs entries.
4397 int bond_create(struct net *net, const char *name)
4399 struct net_device *bond_dev;
4404 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4405 name ? name : "bond%d",
4406 bond_setup, tx_queues);
4408 pr_err("%s: eek! can't alloc netdev!\n", name);
4413 dev_net_set(bond_dev, net);
4414 bond_dev->rtnl_link_ops = &bond_link_ops;
4416 res = register_netdevice(bond_dev);
4418 netif_carrier_off(bond_dev);
4422 bond_destructor(bond_dev);
4426 static int __net_init bond_net_init(struct net *net)
4428 struct bond_net *bn = net_generic(net, bond_net_id);
4431 INIT_LIST_HEAD(&bn->dev_list);
4433 bond_create_proc_dir(bn);
4434 bond_create_sysfs(bn);
4439 static void __net_exit bond_net_exit(struct net *net)
4441 struct bond_net *bn = net_generic(net, bond_net_id);
4442 struct bonding *bond, *tmp_bond;
4445 bond_destroy_sysfs(bn);
4446 bond_destroy_proc_dir(bn);
4448 /* Kill off any bonds created after unregistering bond rtnl ops */
4450 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4451 unregister_netdevice_queue(bond->dev, &list);
4452 unregister_netdevice_many(&list);
4456 static struct pernet_operations bond_net_ops = {
4457 .init = bond_net_init,
4458 .exit = bond_net_exit,
4460 .size = sizeof(struct bond_net),
4463 static int __init bonding_init(void)
4468 pr_info("%s", bond_version);
4470 res = bond_check_params(&bonding_defaults);
4474 res = register_pernet_subsys(&bond_net_ops);
4478 res = bond_netlink_init();
4482 bond_create_debugfs();
4484 for (i = 0; i < max_bonds; i++) {
4485 res = bond_create(&init_net, NULL);
4490 register_netdevice_notifier(&bond_netdev_notifier);
4494 bond_netlink_fini();
4496 unregister_pernet_subsys(&bond_net_ops);
4501 static void __exit bonding_exit(void)
4503 unregister_netdevice_notifier(&bond_netdev_notifier);
4505 bond_destroy_debugfs();
4507 bond_netlink_fini();
4508 unregister_pernet_subsys(&bond_net_ops);
4510 #ifdef CONFIG_NET_POLL_CONTROLLER
4512 * Make sure we don't have an imbalance on our netpoll blocking
4514 WARN_ON(atomic_read(&netpoll_block_tx));
4518 module_init(bonding_init);
4519 module_exit(bonding_exit);
4520 MODULE_LICENSE("GPL");
4521 MODULE_VERSION(DRV_VERSION);
4522 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4523 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");