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 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_keys.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 /*---------------------------- Module parameters ----------------------------*/
87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
90 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
91 static int num_peer_notif = 1;
95 static int use_carrier = 1;
98 static char *primary_reselect;
99 static char *lacp_rate;
100 static int min_links;
101 static char *ad_select;
102 static char *xmit_hash_policy;
103 static int arp_interval;
104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
105 static char *arp_validate;
106 static char *arp_all_targets;
107 static char *fail_over_mac;
108 static int all_slaves_active;
109 static struct bond_params bonding_defaults;
110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
111 static int packets_per_slave = 1;
112 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
114 module_param(max_bonds, int, 0);
115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
116 module_param(tx_queues, int, 0);
117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
118 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
119 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
120 "failover event (alias of num_unsol_na)");
121 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
122 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
123 "failover event (alias of num_grat_arp)");
124 module_param(miimon, int, 0);
125 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
126 module_param(updelay, int, 0);
127 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
128 module_param(downdelay, int, 0);
129 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
131 module_param(use_carrier, int, 0);
132 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
133 "0 for off, 1 for on (default)");
134 module_param(mode, charp, 0);
135 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
136 "1 for active-backup, 2 for balance-xor, "
137 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
138 "6 for balance-alb");
139 module_param(primary, charp, 0);
140 MODULE_PARM_DESC(primary, "Primary network device to use");
141 module_param(primary_reselect, charp, 0);
142 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
144 "0 for always (default), "
145 "1 for only if speed of primary is "
147 "2 for only on active slave "
149 module_param(lacp_rate, charp, 0);
150 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
151 "0 for slow, 1 for fast");
152 module_param(ad_select, charp, 0);
153 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
154 "0 for stable (default), 1 for bandwidth, "
156 module_param(min_links, int, 0);
157 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
159 module_param(xmit_hash_policy, charp, 0);
160 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
161 "0 for layer 2 (default), 1 for layer 3+4, "
162 "2 for layer 2+3, 3 for encap layer 2+3, "
163 "4 for encap layer 3+4");
164 module_param(arp_interval, int, 0);
165 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
166 module_param_array(arp_ip_target, charp, NULL, 0);
167 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
168 module_param(arp_validate, charp, 0);
169 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
170 "0 for none (default), 1 for active, "
171 "2 for backup, 3 for all");
172 module_param(arp_all_targets, charp, 0);
173 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
174 module_param(fail_over_mac, charp, 0);
175 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
176 "the same MAC; 0 for none (default), "
177 "1 for active, 2 for follow");
178 module_param(all_slaves_active, int, 0);
179 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
180 "by setting active flag for all slaves; "
181 "0 for never (default), 1 for always.");
182 module_param(resend_igmp, int, 0);
183 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
185 module_param(packets_per_slave, int, 0);
186 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
187 "mode; 0 for a random slave, 1 packet per "
188 "slave (default), >1 packets per slave.");
189 module_param(lp_interval, uint, 0);
190 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
191 "the bonding driver sends learning packets to "
192 "each slaves peer switch. The default is 1.");
194 /*----------------------------- Global variables ----------------------------*/
196 #ifdef CONFIG_NET_POLL_CONTROLLER
197 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
200 int bond_net_id __read_mostly;
202 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
203 static int arp_ip_count;
204 static int bond_mode = BOND_MODE_ROUNDROBIN;
205 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
206 static int lacp_fast;
208 /*-------------------------- Forward declarations ---------------------------*/
210 static int bond_init(struct net_device *bond_dev);
211 static void bond_uninit(struct net_device *bond_dev);
212 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
213 struct rtnl_link_stats64 *stats);
214 static void bond_slave_arr_handler(struct work_struct *work);
216 /*---------------------------- General routines -----------------------------*/
218 const char *bond_mode_name(int mode)
220 static const char *names[] = {
221 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
222 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
223 [BOND_MODE_XOR] = "load balancing (xor)",
224 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
225 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
226 [BOND_MODE_TLB] = "transmit load balancing",
227 [BOND_MODE_ALB] = "adaptive load balancing",
230 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
236 /*---------------------------------- VLAN -----------------------------------*/
239 * bond_dev_queue_xmit - Prepare skb for xmit.
241 * @bond: bond device that got this skb for tx.
242 * @skb: hw accel VLAN tagged skb to transmit
243 * @slave_dev: slave that is supposed to xmit this skbuff
245 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
246 struct net_device *slave_dev)
248 skb->dev = slave_dev;
250 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
251 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
252 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
254 if (unlikely(netpoll_tx_running(bond->dev)))
255 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
260 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
261 * We don't protect the slave list iteration with a lock because:
262 * a. This operation is performed in IOCTL context,
263 * b. The operation is protected by the RTNL semaphore in the 8021q code,
264 * c. Holding a lock with BH disabled while directly calling a base driver
265 * entry point is generally a BAD idea.
267 * The design of synchronization/protection for this operation in the 8021q
268 * module is good for one or more VLAN devices over a single physical device
269 * and cannot be extended for a teaming solution like bonding, so there is a
270 * potential race condition here where a net device from the vlan group might
271 * be referenced (either by a base driver or the 8021q code) while it is being
272 * removed from the system. However, it turns out we're not making matters
273 * worse, and if it works for regular VLAN usage it will work here too.
277 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
278 * @bond_dev: bonding net device that got called
279 * @vid: vlan id being added
281 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
282 __be16 proto, u16 vid)
284 struct bonding *bond = netdev_priv(bond_dev);
285 struct slave *slave, *rollback_slave;
286 struct list_head *iter;
289 bond_for_each_slave(bond, slave, iter) {
290 res = vlan_vid_add(slave->dev, proto, vid);
298 /* unwind to the slave that failed */
299 bond_for_each_slave(bond, rollback_slave, iter) {
300 if (rollback_slave == slave)
303 vlan_vid_del(rollback_slave->dev, proto, vid);
310 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
311 * @bond_dev: bonding net device that got called
312 * @vid: vlan id being removed
314 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
315 __be16 proto, u16 vid)
317 struct bonding *bond = netdev_priv(bond_dev);
318 struct list_head *iter;
321 bond_for_each_slave(bond, slave, iter)
322 vlan_vid_del(slave->dev, proto, vid);
324 if (bond_is_lb(bond))
325 bond_alb_clear_vlan(bond, vid);
330 /*------------------------------- Link status -------------------------------*/
332 /* Set the carrier state for the master according to the state of its
333 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
334 * do special 802.3ad magic.
336 * Returns zero if carrier state does not change, nonzero if it does.
338 int bond_set_carrier(struct bonding *bond)
340 struct list_head *iter;
343 if (!bond_has_slaves(bond))
346 if (BOND_MODE(bond) == BOND_MODE_8023AD)
347 return bond_3ad_set_carrier(bond);
349 bond_for_each_slave(bond, slave, iter) {
350 if (slave->link == BOND_LINK_UP) {
351 if (!netif_carrier_ok(bond->dev)) {
352 netif_carrier_on(bond->dev);
360 if (netif_carrier_ok(bond->dev)) {
361 netif_carrier_off(bond->dev);
367 /* Get link speed and duplex from the slave's base driver
368 * using ethtool. If for some reason the call fails or the
369 * values are invalid, set speed and duplex to -1,
372 static void bond_update_speed_duplex(struct slave *slave)
374 struct net_device *slave_dev = slave->dev;
375 struct ethtool_cmd ecmd;
379 slave->speed = SPEED_UNKNOWN;
380 slave->duplex = DUPLEX_UNKNOWN;
382 res = __ethtool_get_settings(slave_dev, &ecmd);
386 slave_speed = ethtool_cmd_speed(&ecmd);
387 if (slave_speed == 0 || slave_speed == ((__u32) -1))
390 switch (ecmd.duplex) {
398 slave->speed = slave_speed;
399 slave->duplex = ecmd.duplex;
404 const char *bond_slave_link_status(s8 link)
420 /* if <dev> supports MII link status reporting, check its link status.
422 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
423 * depending upon the setting of the use_carrier parameter.
425 * Return either BMSR_LSTATUS, meaning that the link is up (or we
426 * can't tell and just pretend it is), or 0, meaning that the link is
429 * If reporting is non-zero, instead of faking link up, return -1 if
430 * both ETHTOOL and MII ioctls fail (meaning the device does not
431 * support them). If use_carrier is set, return whatever it says.
432 * It'd be nice if there was a good way to tell if a driver supports
433 * netif_carrier, but there really isn't.
435 static int bond_check_dev_link(struct bonding *bond,
436 struct net_device *slave_dev, int reporting)
438 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
439 int (*ioctl)(struct net_device *, struct ifreq *, int);
441 struct mii_ioctl_data *mii;
443 if (!reporting && !netif_running(slave_dev))
446 if (bond->params.use_carrier)
447 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
449 /* Try to get link status using Ethtool first. */
450 if (slave_dev->ethtool_ops->get_link)
451 return slave_dev->ethtool_ops->get_link(slave_dev) ?
454 /* Ethtool can't be used, fallback to MII ioctls. */
455 ioctl = slave_ops->ndo_do_ioctl;
457 /* TODO: set pointer to correct ioctl on a per team member
458 * bases to make this more efficient. that is, once
459 * we determine the correct ioctl, we will always
460 * call it and not the others for that team
464 /* We cannot assume that SIOCGMIIPHY will also read a
465 * register; not all network drivers (e.g., e100)
469 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
470 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
472 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
473 mii->reg_num = MII_BMSR;
474 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
475 return mii->val_out & BMSR_LSTATUS;
479 /* If reporting, report that either there's no dev->do_ioctl,
480 * or both SIOCGMIIREG and get_link failed (meaning that we
481 * cannot report link status). If not reporting, pretend
484 return reporting ? -1 : BMSR_LSTATUS;
487 /*----------------------------- Multicast list ------------------------------*/
489 /* Push the promiscuity flag down to appropriate slaves */
490 static int bond_set_promiscuity(struct bonding *bond, int inc)
492 struct list_head *iter;
495 if (bond_uses_primary(bond)) {
496 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
499 err = dev_set_promiscuity(curr_active->dev, inc);
503 bond_for_each_slave(bond, slave, iter) {
504 err = dev_set_promiscuity(slave->dev, inc);
512 /* Push the allmulti flag down to all slaves */
513 static int bond_set_allmulti(struct bonding *bond, int inc)
515 struct list_head *iter;
518 if (bond_uses_primary(bond)) {
519 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
522 err = dev_set_allmulti(curr_active->dev, inc);
526 bond_for_each_slave(bond, slave, iter) {
527 err = dev_set_allmulti(slave->dev, inc);
535 /* Retrieve the list of registered multicast addresses for the bonding
536 * device and retransmit an IGMP JOIN request to the current active
539 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
541 struct bonding *bond = container_of(work, struct bonding,
544 if (!rtnl_trylock()) {
545 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
548 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
550 if (bond->igmp_retrans > 1) {
551 bond->igmp_retrans--;
552 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
557 /* Flush bond's hardware addresses from slave */
558 static void bond_hw_addr_flush(struct net_device *bond_dev,
559 struct net_device *slave_dev)
561 struct bonding *bond = netdev_priv(bond_dev);
563 dev_uc_unsync(slave_dev, bond_dev);
564 dev_mc_unsync(slave_dev, bond_dev);
566 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
567 /* del lacpdu mc addr from mc list */
568 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
570 dev_mc_del(slave_dev, lacpdu_multicast);
574 /*--------------------------- Active slave change ---------------------------*/
576 /* Update the hardware address list and promisc/allmulti for the new and
577 * old active slaves (if any). Modes that are not using primary keep all
578 * slaves up date at all times; only the modes that use primary need to call
579 * this function to swap these settings during a failover.
581 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
582 struct slave *old_active)
585 if (bond->dev->flags & IFF_PROMISC)
586 dev_set_promiscuity(old_active->dev, -1);
588 if (bond->dev->flags & IFF_ALLMULTI)
589 dev_set_allmulti(old_active->dev, -1);
591 bond_hw_addr_flush(bond->dev, old_active->dev);
595 /* FIXME: Signal errors upstream. */
596 if (bond->dev->flags & IFF_PROMISC)
597 dev_set_promiscuity(new_active->dev, 1);
599 if (bond->dev->flags & IFF_ALLMULTI)
600 dev_set_allmulti(new_active->dev, 1);
602 netif_addr_lock_bh(bond->dev);
603 dev_uc_sync(new_active->dev, bond->dev);
604 dev_mc_sync(new_active->dev, bond->dev);
605 netif_addr_unlock_bh(bond->dev);
610 * bond_set_dev_addr - clone slave's address to bond
611 * @bond_dev: bond net device
612 * @slave_dev: slave net device
614 * Should be called with RTNL held.
616 static void bond_set_dev_addr(struct net_device *bond_dev,
617 struct net_device *slave_dev)
619 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
620 bond_dev, slave_dev, slave_dev->addr_len);
621 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
622 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
623 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
626 /* bond_do_fail_over_mac
628 * Perform special MAC address swapping for fail_over_mac settings
632 static void bond_do_fail_over_mac(struct bonding *bond,
633 struct slave *new_active,
634 struct slave *old_active)
636 u8 tmp_mac[ETH_ALEN];
637 struct sockaddr saddr;
640 switch (bond->params.fail_over_mac) {
641 case BOND_FOM_ACTIVE:
643 bond_set_dev_addr(bond->dev, new_active->dev);
645 case BOND_FOM_FOLLOW:
646 /* if new_active && old_active, swap them
647 * if just old_active, do nothing (going to no active slave)
648 * if just new_active, set new_active to bond's MAC
654 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
655 ether_addr_copy(saddr.sa_data,
656 old_active->dev->dev_addr);
657 saddr.sa_family = new_active->dev->type;
659 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
660 saddr.sa_family = bond->dev->type;
663 rv = dev_set_mac_address(new_active->dev, &saddr);
665 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
666 -rv, new_active->dev->name);
673 ether_addr_copy(saddr.sa_data, tmp_mac);
674 saddr.sa_family = old_active->dev->type;
676 rv = dev_set_mac_address(old_active->dev, &saddr);
678 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
679 -rv, new_active->dev->name);
683 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
684 bond->params.fail_over_mac);
690 static bool bond_should_change_active(struct bonding *bond)
692 struct slave *prim = rtnl_dereference(bond->primary_slave);
693 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
695 if (!prim || !curr || curr->link != BOND_LINK_UP)
697 if (bond->force_primary) {
698 bond->force_primary = false;
701 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
702 (prim->speed < curr->speed ||
703 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
705 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
711 * find_best_interface - select the best available slave to be the active one
712 * @bond: our bonding struct
714 static struct slave *bond_find_best_slave(struct bonding *bond)
716 struct slave *slave, *bestslave = NULL, *primary;
717 struct list_head *iter;
718 int mintime = bond->params.updelay;
720 primary = rtnl_dereference(bond->primary_slave);
721 if (primary && primary->link == BOND_LINK_UP &&
722 bond_should_change_active(bond))
725 bond_for_each_slave(bond, slave, iter) {
726 if (slave->link == BOND_LINK_UP)
728 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
729 slave->delay < mintime) {
730 mintime = slave->delay;
738 static bool bond_should_notify_peers(struct bonding *bond)
743 slave = rcu_dereference(bond->curr_active_slave);
746 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
747 slave ? slave->dev->name : "NULL");
749 if (!slave || !bond->send_peer_notif ||
750 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
757 * change_active_interface - change the active slave into the specified one
758 * @bond: our bonding struct
759 * @new: the new slave to make the active one
761 * Set the new slave to the bond's settings and unset them on the old
763 * Setting include flags, mc-list, promiscuity, allmulti, etc.
765 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
766 * because it is apparently the best available slave we have, even though its
767 * updelay hasn't timed out yet.
769 * Caller must hold RTNL.
771 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
773 struct slave *old_active;
777 old_active = rtnl_dereference(bond->curr_active_slave);
779 if (old_active == new_active)
783 new_active->last_link_up = jiffies;
785 if (new_active->link == BOND_LINK_BACK) {
786 if (bond_uses_primary(bond)) {
787 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
788 new_active->dev->name,
789 (bond->params.updelay - new_active->delay) * bond->params.miimon);
792 new_active->delay = 0;
793 bond_set_slave_link_state(new_active, BOND_LINK_UP);
795 if (BOND_MODE(bond) == BOND_MODE_8023AD)
796 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
798 if (bond_is_lb(bond))
799 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
801 if (bond_uses_primary(bond)) {
802 netdev_info(bond->dev, "making interface %s the new active one\n",
803 new_active->dev->name);
808 if (bond_uses_primary(bond))
809 bond_hw_addr_swap(bond, new_active, old_active);
811 if (bond_is_lb(bond)) {
812 bond_alb_handle_active_change(bond, new_active);
814 bond_set_slave_inactive_flags(old_active,
815 BOND_SLAVE_NOTIFY_NOW);
817 bond_set_slave_active_flags(new_active,
818 BOND_SLAVE_NOTIFY_NOW);
820 rcu_assign_pointer(bond->curr_active_slave, new_active);
823 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
825 bond_set_slave_inactive_flags(old_active,
826 BOND_SLAVE_NOTIFY_NOW);
829 bool should_notify_peers = false;
831 bond_set_slave_active_flags(new_active,
832 BOND_SLAVE_NOTIFY_NOW);
834 if (bond->params.fail_over_mac)
835 bond_do_fail_over_mac(bond, new_active,
838 if (netif_running(bond->dev)) {
839 bond->send_peer_notif =
840 bond->params.num_peer_notif;
841 should_notify_peers =
842 bond_should_notify_peers(bond);
845 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
846 if (should_notify_peers)
847 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
852 /* resend IGMP joins since active slave has changed or
853 * all were sent on curr_active_slave.
854 * resend only if bond is brought up with the affected
855 * bonding modes and the retransmission is enabled
857 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
858 ((bond_uses_primary(bond) && new_active) ||
859 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
860 bond->igmp_retrans = bond->params.resend_igmp;
861 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
866 * bond_select_active_slave - select a new active slave, if needed
867 * @bond: our bonding struct
869 * This functions should be called when one of the following occurs:
870 * - The old curr_active_slave has been released or lost its link.
871 * - The primary_slave has got its link back.
872 * - A slave has got its link back and there's no old curr_active_slave.
874 * Caller must hold RTNL.
876 void bond_select_active_slave(struct bonding *bond)
878 struct slave *best_slave;
883 best_slave = bond_find_best_slave(bond);
884 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
885 bond_change_active_slave(bond, best_slave);
886 rv = bond_set_carrier(bond);
890 if (netif_carrier_ok(bond->dev)) {
891 netdev_info(bond->dev, "first active interface up!\n");
893 netdev_info(bond->dev, "now running without any active interface!\n");
898 #ifdef CONFIG_NET_POLL_CONTROLLER
899 static inline int slave_enable_netpoll(struct slave *slave)
904 np = kzalloc(sizeof(*np), GFP_KERNEL);
909 err = __netpoll_setup(np, slave->dev);
918 static inline void slave_disable_netpoll(struct slave *slave)
920 struct netpoll *np = slave->np;
926 __netpoll_free_async(np);
929 static void bond_poll_controller(struct net_device *bond_dev)
933 static void bond_netpoll_cleanup(struct net_device *bond_dev)
935 struct bonding *bond = netdev_priv(bond_dev);
936 struct list_head *iter;
939 bond_for_each_slave(bond, slave, iter)
940 if (bond_slave_is_up(slave))
941 slave_disable_netpoll(slave);
944 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
946 struct bonding *bond = netdev_priv(dev);
947 struct list_head *iter;
951 bond_for_each_slave(bond, slave, iter) {
952 err = slave_enable_netpoll(slave);
954 bond_netpoll_cleanup(dev);
961 static inline int slave_enable_netpoll(struct slave *slave)
965 static inline void slave_disable_netpoll(struct slave *slave)
968 static void bond_netpoll_cleanup(struct net_device *bond_dev)
973 /*---------------------------------- IOCTL ----------------------------------*/
975 static netdev_features_t bond_fix_features(struct net_device *dev,
976 netdev_features_t features)
978 struct bonding *bond = netdev_priv(dev);
979 struct list_head *iter;
980 netdev_features_t mask;
983 /* If any slave has the offload feature flag set,
984 * set the offload flag on the bond.
986 mask = features | NETIF_F_HW_SWITCH_OFFLOAD;
988 features &= ~NETIF_F_ONE_FOR_ALL;
989 features |= NETIF_F_ALL_FOR_ALL;
991 bond_for_each_slave(bond, slave, iter) {
992 features = netdev_increment_features(features,
993 slave->dev->features,
996 features = netdev_add_tso_features(features, mask);
1001 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1002 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1003 NETIF_F_HIGHDMA | NETIF_F_LRO)
1005 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1008 static void bond_compute_features(struct bonding *bond)
1010 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1011 IFF_XMIT_DST_RELEASE_PERM;
1012 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1013 netdev_features_t enc_features = BOND_ENC_FEATURES;
1014 struct net_device *bond_dev = bond->dev;
1015 struct list_head *iter;
1016 struct slave *slave;
1017 unsigned short max_hard_header_len = ETH_HLEN;
1018 unsigned int gso_max_size = GSO_MAX_SIZE;
1019 u16 gso_max_segs = GSO_MAX_SEGS;
1021 if (!bond_has_slaves(bond))
1023 vlan_features &= NETIF_F_ALL_FOR_ALL;
1025 bond_for_each_slave(bond, slave, iter) {
1026 vlan_features = netdev_increment_features(vlan_features,
1027 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1029 enc_features = netdev_increment_features(enc_features,
1030 slave->dev->hw_enc_features,
1032 dst_release_flag &= slave->dev->priv_flags;
1033 if (slave->dev->hard_header_len > max_hard_header_len)
1034 max_hard_header_len = slave->dev->hard_header_len;
1036 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1037 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1041 bond_dev->vlan_features = vlan_features;
1042 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1043 bond_dev->hard_header_len = max_hard_header_len;
1044 bond_dev->gso_max_segs = gso_max_segs;
1045 netif_set_gso_max_size(bond_dev, gso_max_size);
1047 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1048 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1049 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1050 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1052 netdev_change_features(bond_dev);
1055 static void bond_setup_by_slave(struct net_device *bond_dev,
1056 struct net_device *slave_dev)
1058 bond_dev->header_ops = slave_dev->header_ops;
1060 bond_dev->type = slave_dev->type;
1061 bond_dev->hard_header_len = slave_dev->hard_header_len;
1062 bond_dev->addr_len = slave_dev->addr_len;
1064 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1065 slave_dev->addr_len);
1068 /* On bonding slaves other than the currently active slave, suppress
1069 * duplicates except for alb non-mcast/bcast.
1071 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1072 struct slave *slave,
1073 struct bonding *bond)
1075 if (bond_is_slave_inactive(slave)) {
1076 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1077 skb->pkt_type != PACKET_BROADCAST &&
1078 skb->pkt_type != PACKET_MULTICAST)
1085 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1087 struct sk_buff *skb = *pskb;
1088 struct slave *slave;
1089 struct bonding *bond;
1090 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1092 int ret = RX_HANDLER_ANOTHER;
1094 skb = skb_share_check(skb, GFP_ATOMIC);
1096 return RX_HANDLER_CONSUMED;
1100 slave = bond_slave_get_rcu(skb->dev);
1103 recv_probe = ACCESS_ONCE(bond->recv_probe);
1105 ret = recv_probe(skb, bond, slave);
1106 if (ret == RX_HANDLER_CONSUMED) {
1112 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1113 return RX_HANDLER_EXACT;
1116 skb->dev = bond->dev;
1118 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1119 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1120 skb->pkt_type == PACKET_HOST) {
1122 if (unlikely(skb_cow_head(skb,
1123 skb->data - skb_mac_header(skb)))) {
1125 return RX_HANDLER_CONSUMED;
1127 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1133 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1134 struct net_device *slave_dev,
1135 struct slave *slave)
1139 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1142 slave_dev->flags |= IFF_SLAVE;
1143 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1147 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1148 struct net_device *slave_dev)
1150 netdev_upper_dev_unlink(slave_dev, bond_dev);
1151 slave_dev->flags &= ~IFF_SLAVE;
1152 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1155 static struct slave *bond_alloc_slave(struct bonding *bond)
1157 struct slave *slave = NULL;
1159 slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1163 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1164 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1166 if (!SLAVE_AD_INFO(slave)) {
1174 static void bond_free_slave(struct slave *slave)
1176 struct bonding *bond = bond_get_bond_by_slave(slave);
1178 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1179 kfree(SLAVE_AD_INFO(slave));
1184 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1186 info->bond_mode = BOND_MODE(bond);
1187 info->miimon = bond->params.miimon;
1188 info->num_slaves = bond->slave_cnt;
1191 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1193 strcpy(info->slave_name, slave->dev->name);
1194 info->link = slave->link;
1195 info->state = bond_slave_state(slave);
1196 info->link_failure_count = slave->link_failure_count;
1199 static void bond_netdev_notify(struct net_device *dev,
1200 struct netdev_bonding_info *info)
1203 netdev_bonding_info_change(dev, info);
1207 static void bond_netdev_notify_work(struct work_struct *_work)
1209 struct netdev_notify_work *w =
1210 container_of(_work, struct netdev_notify_work, work.work);
1212 bond_netdev_notify(w->dev, &w->bonding_info);
1217 void bond_queue_slave_event(struct slave *slave)
1219 struct bonding *bond = slave->bond;
1220 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1225 dev_hold(slave->dev);
1226 nnw->dev = slave->dev;
1227 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1228 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1229 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1231 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1234 /* enslave device <slave> to bond device <master> */
1235 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1237 struct bonding *bond = netdev_priv(bond_dev);
1238 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1239 struct slave *new_slave = NULL, *prev_slave;
1240 struct sockaddr addr;
1244 if (!bond->params.use_carrier &&
1245 slave_dev->ethtool_ops->get_link == NULL &&
1246 slave_ops->ndo_do_ioctl == NULL) {
1247 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1251 /* already enslaved */
1252 if (slave_dev->flags & IFF_SLAVE) {
1253 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1257 if (bond_dev == slave_dev) {
1258 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1262 /* vlan challenged mutual exclusion */
1263 /* no need to lock since we're protected by rtnl_lock */
1264 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1265 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1267 if (vlan_uses_dev(bond_dev)) {
1268 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1269 slave_dev->name, bond_dev->name);
1272 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1273 slave_dev->name, slave_dev->name,
1277 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1281 /* Old ifenslave binaries are no longer supported. These can
1282 * be identified with moderate accuracy by the state of the slave:
1283 * the current ifenslave will set the interface down prior to
1284 * enslaving it; the old ifenslave will not.
1286 if ((slave_dev->flags & IFF_UP)) {
1287 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1290 goto err_undo_flags;
1293 /* set bonding device ether type by slave - bonding netdevices are
1294 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1295 * there is a need to override some of the type dependent attribs/funcs.
1297 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1298 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1300 if (!bond_has_slaves(bond)) {
1301 if (bond_dev->type != slave_dev->type) {
1302 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1303 bond_dev->type, slave_dev->type);
1305 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1307 res = notifier_to_errno(res);
1309 netdev_err(bond_dev, "refused to change device type\n");
1311 goto err_undo_flags;
1314 /* Flush unicast and multicast addresses */
1315 dev_uc_flush(bond_dev);
1316 dev_mc_flush(bond_dev);
1318 if (slave_dev->type != ARPHRD_ETHER)
1319 bond_setup_by_slave(bond_dev, slave_dev);
1321 ether_setup(bond_dev);
1322 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1325 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1328 } else if (bond_dev->type != slave_dev->type) {
1329 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1330 slave_dev->name, slave_dev->type, bond_dev->type);
1332 goto err_undo_flags;
1335 if (slave_ops->ndo_set_mac_address == NULL) {
1336 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1337 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1338 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1339 if (!bond_has_slaves(bond)) {
1340 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1341 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1343 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1345 goto err_undo_flags;
1350 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1352 /* If this is the first slave, then we need to set the master's hardware
1353 * address to be the same as the slave's.
1355 if (!bond_has_slaves(bond) &&
1356 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1357 bond_set_dev_addr(bond->dev, slave_dev);
1359 new_slave = bond_alloc_slave(bond);
1362 goto err_undo_flags;
1365 new_slave->bond = bond;
1366 new_slave->dev = slave_dev;
1367 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1368 * is set via sysfs or module option if desired.
1370 new_slave->queue_id = 0;
1372 /* Save slave's original mtu and then set it to match the bond */
1373 new_slave->original_mtu = slave_dev->mtu;
1374 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1376 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1380 /* Save slave's original ("permanent") mac address for modes
1381 * that need it, and for restoring it upon release, and then
1382 * set it to the master's address
1384 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1386 if (!bond->params.fail_over_mac ||
1387 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1388 /* Set slave to master's mac address. The application already
1389 * set the master's mac address to that of the first slave
1391 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1392 addr.sa_family = slave_dev->type;
1393 res = dev_set_mac_address(slave_dev, &addr);
1395 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1396 goto err_restore_mtu;
1400 /* open the slave since the application closed it */
1401 res = dev_open(slave_dev);
1403 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1404 goto err_restore_mac;
1407 slave_dev->priv_flags |= IFF_BONDING;
1408 /* initialize slave stats */
1409 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1411 if (bond_is_lb(bond)) {
1412 /* bond_alb_init_slave() must be called before all other stages since
1413 * it might fail and we do not want to have to undo everything
1415 res = bond_alb_init_slave(bond, new_slave);
1420 /* If the mode uses primary, then the following is handled by
1421 * bond_change_active_slave().
1423 if (!bond_uses_primary(bond)) {
1424 /* set promiscuity level to new slave */
1425 if (bond_dev->flags & IFF_PROMISC) {
1426 res = dev_set_promiscuity(slave_dev, 1);
1431 /* set allmulti level to new slave */
1432 if (bond_dev->flags & IFF_ALLMULTI) {
1433 res = dev_set_allmulti(slave_dev, 1);
1438 netif_addr_lock_bh(bond_dev);
1440 dev_mc_sync_multiple(slave_dev, bond_dev);
1441 dev_uc_sync_multiple(slave_dev, bond_dev);
1443 netif_addr_unlock_bh(bond_dev);
1446 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1447 /* add lacpdu mc addr to mc list */
1448 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1450 dev_mc_add(slave_dev, lacpdu_multicast);
1453 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1455 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1460 prev_slave = bond_last_slave(bond);
1462 new_slave->delay = 0;
1463 new_slave->link_failure_count = 0;
1465 bond_update_speed_duplex(new_slave);
1467 new_slave->last_rx = jiffies -
1468 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1469 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1470 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1472 if (bond->params.miimon && !bond->params.use_carrier) {
1473 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1475 if ((link_reporting == -1) && !bond->params.arp_interval) {
1476 /* miimon is set but a bonded network driver
1477 * does not support ETHTOOL/MII and
1478 * arp_interval is not set. Note: if
1479 * use_carrier is enabled, we will never go
1480 * here (because netif_carrier is always
1481 * supported); thus, we don't need to change
1482 * the messages for netif_carrier.
1484 netdev_warn(bond_dev, "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",
1486 } else if (link_reporting == -1) {
1487 /* unable get link status using mii/ethtool */
1488 netdev_warn(bond_dev, "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",
1493 /* check for initial state */
1494 if (bond->params.miimon) {
1495 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1496 if (bond->params.updelay) {
1497 bond_set_slave_link_state(new_slave,
1499 new_slave->delay = bond->params.updelay;
1501 bond_set_slave_link_state(new_slave,
1505 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1507 } else if (bond->params.arp_interval) {
1508 bond_set_slave_link_state(new_slave,
1509 (netif_carrier_ok(slave_dev) ?
1510 BOND_LINK_UP : BOND_LINK_DOWN));
1512 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1515 if (new_slave->link != BOND_LINK_DOWN)
1516 new_slave->last_link_up = jiffies;
1517 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1518 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1519 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1521 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1522 /* if there is a primary slave, remember it */
1523 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1524 rcu_assign_pointer(bond->primary_slave, new_slave);
1525 bond->force_primary = true;
1529 switch (BOND_MODE(bond)) {
1530 case BOND_MODE_ACTIVEBACKUP:
1531 bond_set_slave_inactive_flags(new_slave,
1532 BOND_SLAVE_NOTIFY_NOW);
1534 case BOND_MODE_8023AD:
1535 /* in 802.3ad mode, the internal mechanism
1536 * will activate the slaves in the selected
1539 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1540 /* if this is the first slave */
1542 SLAVE_AD_INFO(new_slave)->id = 1;
1543 /* Initialize AD with the number of times that the AD timer is called in 1 second
1544 * can be called only after the mac address of the bond is set
1546 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1548 SLAVE_AD_INFO(new_slave)->id =
1549 SLAVE_AD_INFO(prev_slave)->id + 1;
1552 bond_3ad_bind_slave(new_slave);
1556 bond_set_active_slave(new_slave);
1557 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1560 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1562 /* always active in trunk mode */
1563 bond_set_active_slave(new_slave);
1565 /* In trunking mode there is little meaning to curr_active_slave
1566 * anyway (it holds no special properties of the bond device),
1567 * so we can change it without calling change_active_interface()
1569 if (!rcu_access_pointer(bond->curr_active_slave) &&
1570 new_slave->link == BOND_LINK_UP)
1571 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1574 } /* switch(bond_mode) */
1576 #ifdef CONFIG_NET_POLL_CONTROLLER
1577 slave_dev->npinfo = bond->dev->npinfo;
1578 if (slave_dev->npinfo) {
1579 if (slave_enable_netpoll(new_slave)) {
1580 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1587 if (!(bond_dev->features & NETIF_F_LRO))
1588 dev_disable_lro(slave_dev);
1590 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1593 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1597 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1599 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1600 goto err_unregister;
1603 res = bond_sysfs_slave_add(new_slave);
1605 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1606 goto err_upper_unlink;
1610 bond_compute_features(bond);
1611 bond_set_carrier(bond);
1613 if (bond_uses_primary(bond)) {
1615 bond_select_active_slave(bond);
1616 unblock_netpoll_tx();
1619 if (bond_mode_uses_xmit_hash(bond))
1620 bond_update_slave_arr(bond, NULL);
1622 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1624 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1625 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1627 /* enslave is successful */
1628 bond_queue_slave_event(new_slave);
1631 /* Undo stages on error */
1633 bond_upper_dev_unlink(bond_dev, slave_dev);
1636 netdev_rx_handler_unregister(slave_dev);
1639 if (!bond_uses_primary(bond))
1640 bond_hw_addr_flush(bond_dev, slave_dev);
1642 vlan_vids_del_by_dev(slave_dev, bond_dev);
1643 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1644 RCU_INIT_POINTER(bond->primary_slave, NULL);
1645 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1647 bond_change_active_slave(bond, NULL);
1648 bond_select_active_slave(bond);
1649 unblock_netpoll_tx();
1651 /* either primary_slave or curr_active_slave might've changed */
1653 slave_disable_netpoll(new_slave);
1656 slave_dev->priv_flags &= ~IFF_BONDING;
1657 dev_close(slave_dev);
1660 if (!bond->params.fail_over_mac ||
1661 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1662 /* XXX TODO - fom follow mode needs to change master's
1663 * MAC if this slave's MAC is in use by the bond, or at
1664 * least print a warning.
1666 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1667 addr.sa_family = slave_dev->type;
1668 dev_set_mac_address(slave_dev, &addr);
1672 dev_set_mtu(slave_dev, new_slave->original_mtu);
1675 bond_free_slave(new_slave);
1678 /* Enslave of first slave has failed and we need to fix master's mac */
1679 if (!bond_has_slaves(bond) &&
1680 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1681 eth_hw_addr_random(bond_dev);
1686 /* Try to release the slave device <slave> from the bond device <master>
1687 * It is legal to access curr_active_slave without a lock because all the function
1688 * is RTNL-locked. If "all" is true it means that the function is being called
1689 * while destroying a bond interface and all slaves are being released.
1691 * The rules for slave state should be:
1692 * for Active/Backup:
1693 * Active stays on all backups go down
1694 * for Bonded connections:
1695 * The first up interface should be left on and all others downed.
1697 static int __bond_release_one(struct net_device *bond_dev,
1698 struct net_device *slave_dev,
1701 struct bonding *bond = netdev_priv(bond_dev);
1702 struct slave *slave, *oldcurrent;
1703 struct sockaddr addr;
1704 int old_flags = bond_dev->flags;
1705 netdev_features_t old_features = bond_dev->features;
1707 /* slave is not a slave or master is not master of this slave */
1708 if (!(slave_dev->flags & IFF_SLAVE) ||
1709 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1710 netdev_dbg(bond_dev, "cannot release %s\n",
1717 slave = bond_get_slave_by_dev(bond, slave_dev);
1719 /* not a slave of this bond */
1720 netdev_info(bond_dev, "%s not enslaved\n",
1722 unblock_netpoll_tx();
1726 bond_sysfs_slave_del(slave);
1728 /* recompute stats just before removing the slave */
1729 bond_get_stats(bond->dev, &bond->bond_stats);
1731 bond_upper_dev_unlink(bond_dev, slave_dev);
1732 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1733 * for this slave anymore.
1735 netdev_rx_handler_unregister(slave_dev);
1737 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1738 bond_3ad_unbind_slave(slave);
1740 if (bond_mode_uses_xmit_hash(bond))
1741 bond_update_slave_arr(bond, slave);
1743 netdev_info(bond_dev, "Releasing %s interface %s\n",
1744 bond_is_active_slave(slave) ? "active" : "backup",
1747 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1749 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1751 if (!all && (!bond->params.fail_over_mac ||
1752 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1753 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1754 bond_has_slaves(bond))
1755 netdev_warn(bond_dev, "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",
1756 slave_dev->name, slave->perm_hwaddr,
1757 bond_dev->name, slave_dev->name);
1760 if (rtnl_dereference(bond->primary_slave) == slave)
1761 RCU_INIT_POINTER(bond->primary_slave, NULL);
1763 if (oldcurrent == slave)
1764 bond_change_active_slave(bond, NULL);
1766 if (bond_is_lb(bond)) {
1767 /* Must be called only after the slave has been
1768 * detached from the list and the curr_active_slave
1769 * has been cleared (if our_slave == old_current),
1770 * but before a new active slave is selected.
1772 bond_alb_deinit_slave(bond, slave);
1776 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1777 } else if (oldcurrent == slave) {
1778 /* Note that we hold RTNL over this sequence, so there
1779 * is no concern that another slave add/remove event
1782 bond_select_active_slave(bond);
1785 if (!bond_has_slaves(bond)) {
1786 bond_set_carrier(bond);
1787 eth_hw_addr_random(bond_dev);
1790 unblock_netpoll_tx();
1794 if (!bond_has_slaves(bond)) {
1795 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1796 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1799 bond_compute_features(bond);
1800 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1801 (old_features & NETIF_F_VLAN_CHALLENGED))
1802 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1803 slave_dev->name, bond_dev->name);
1805 vlan_vids_del_by_dev(slave_dev, bond_dev);
1807 /* If the mode uses primary, then this case was handled above by
1808 * bond_change_active_slave(..., NULL)
1810 if (!bond_uses_primary(bond)) {
1811 /* unset promiscuity level from slave
1812 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1813 * of the IFF_PROMISC flag in the bond_dev, but we need the
1814 * value of that flag before that change, as that was the value
1815 * when this slave was attached, so we cache at the start of the
1816 * function and use it here. Same goes for ALLMULTI below
1818 if (old_flags & IFF_PROMISC)
1819 dev_set_promiscuity(slave_dev, -1);
1821 /* unset allmulti level from slave */
1822 if (old_flags & IFF_ALLMULTI)
1823 dev_set_allmulti(slave_dev, -1);
1825 bond_hw_addr_flush(bond_dev, slave_dev);
1828 slave_disable_netpoll(slave);
1830 /* close slave before restoring its mac address */
1831 dev_close(slave_dev);
1833 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1834 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1835 /* restore original ("permanent") mac address */
1836 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1837 addr.sa_family = slave_dev->type;
1838 dev_set_mac_address(slave_dev, &addr);
1841 dev_set_mtu(slave_dev, slave->original_mtu);
1843 slave_dev->priv_flags &= ~IFF_BONDING;
1845 bond_free_slave(slave);
1850 /* A wrapper used because of ndo_del_link */
1851 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1853 return __bond_release_one(bond_dev, slave_dev, false);
1856 /* First release a slave and then destroy the bond if no more slaves are left.
1857 * Must be under rtnl_lock when this function is called.
1859 static int bond_release_and_destroy(struct net_device *bond_dev,
1860 struct net_device *slave_dev)
1862 struct bonding *bond = netdev_priv(bond_dev);
1865 ret = bond_release(bond_dev, slave_dev);
1866 if (ret == 0 && !bond_has_slaves(bond)) {
1867 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1868 netdev_info(bond_dev, "Destroying bond %s\n",
1870 unregister_netdevice(bond_dev);
1875 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1877 struct bonding *bond = netdev_priv(bond_dev);
1878 bond_fill_ifbond(bond, info);
1882 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1884 struct bonding *bond = netdev_priv(bond_dev);
1885 struct list_head *iter;
1886 int i = 0, res = -ENODEV;
1887 struct slave *slave;
1889 bond_for_each_slave(bond, slave, iter) {
1890 if (i++ == (int)info->slave_id) {
1892 bond_fill_ifslave(slave, info);
1900 /*-------------------------------- Monitoring -------------------------------*/
1902 /* called with rcu_read_lock() */
1903 static int bond_miimon_inspect(struct bonding *bond)
1905 int link_state, commit = 0;
1906 struct list_head *iter;
1907 struct slave *slave;
1908 bool ignore_updelay;
1910 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1912 bond_for_each_slave_rcu(bond, slave, iter) {
1913 slave->new_link = BOND_LINK_NOCHANGE;
1915 link_state = bond_check_dev_link(bond, slave->dev, 0);
1917 switch (slave->link) {
1922 bond_set_slave_link_state(slave, BOND_LINK_FAIL);
1923 slave->delay = bond->params.downdelay;
1925 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1927 BOND_MODE_ACTIVEBACKUP) ?
1928 (bond_is_active_slave(slave) ?
1929 "active " : "backup ") : "",
1931 bond->params.downdelay * bond->params.miimon);
1934 case BOND_LINK_FAIL:
1936 /* recovered before downdelay expired */
1937 bond_set_slave_link_state(slave, BOND_LINK_UP);
1938 slave->last_link_up = jiffies;
1939 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1940 (bond->params.downdelay - slave->delay) *
1941 bond->params.miimon,
1946 if (slave->delay <= 0) {
1947 slave->new_link = BOND_LINK_DOWN;
1955 case BOND_LINK_DOWN:
1959 bond_set_slave_link_state(slave, BOND_LINK_BACK);
1960 slave->delay = bond->params.updelay;
1963 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1965 ignore_updelay ? 0 :
1966 bond->params.updelay *
1967 bond->params.miimon);
1970 case BOND_LINK_BACK:
1972 bond_set_slave_link_state(slave,
1974 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
1975 (bond->params.updelay - slave->delay) *
1976 bond->params.miimon,
1985 if (slave->delay <= 0) {
1986 slave->new_link = BOND_LINK_UP;
1988 ignore_updelay = false;
2000 static void bond_miimon_commit(struct bonding *bond)
2002 struct list_head *iter;
2003 struct slave *slave, *primary;
2005 bond_for_each_slave(bond, slave, iter) {
2006 switch (slave->new_link) {
2007 case BOND_LINK_NOCHANGE:
2011 bond_set_slave_link_state(slave, BOND_LINK_UP);
2012 slave->last_link_up = jiffies;
2014 primary = rtnl_dereference(bond->primary_slave);
2015 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2016 /* prevent it from being the active one */
2017 bond_set_backup_slave(slave);
2018 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2019 /* make it immediately active */
2020 bond_set_active_slave(slave);
2021 } else if (slave != primary) {
2022 /* prevent it from being the active one */
2023 bond_set_backup_slave(slave);
2026 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2028 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2029 slave->duplex ? "full" : "half");
2031 /* notify ad that the link status has changed */
2032 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2033 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2035 if (bond_is_lb(bond))
2036 bond_alb_handle_link_change(bond, slave,
2039 if (BOND_MODE(bond) == BOND_MODE_XOR)
2040 bond_update_slave_arr(bond, NULL);
2042 if (!bond->curr_active_slave || slave == primary)
2047 case BOND_LINK_DOWN:
2048 if (slave->link_failure_count < UINT_MAX)
2049 slave->link_failure_count++;
2051 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2053 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2054 BOND_MODE(bond) == BOND_MODE_8023AD)
2055 bond_set_slave_inactive_flags(slave,
2056 BOND_SLAVE_NOTIFY_NOW);
2058 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2061 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2062 bond_3ad_handle_link_change(slave,
2065 if (bond_is_lb(bond))
2066 bond_alb_handle_link_change(bond, slave,
2069 if (BOND_MODE(bond) == BOND_MODE_XOR)
2070 bond_update_slave_arr(bond, NULL);
2072 if (slave == rcu_access_pointer(bond->curr_active_slave))
2078 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2079 slave->new_link, slave->dev->name);
2080 slave->new_link = BOND_LINK_NOCHANGE;
2087 bond_select_active_slave(bond);
2088 unblock_netpoll_tx();
2091 bond_set_carrier(bond);
2096 * Really a wrapper that splits the mii monitor into two phases: an
2097 * inspection, then (if inspection indicates something needs to be done)
2098 * an acquisition of appropriate locks followed by a commit phase to
2099 * implement whatever link state changes are indicated.
2101 static void bond_mii_monitor(struct work_struct *work)
2103 struct bonding *bond = container_of(work, struct bonding,
2105 bool should_notify_peers = false;
2106 unsigned long delay;
2108 delay = msecs_to_jiffies(bond->params.miimon);
2110 if (!bond_has_slaves(bond))
2115 should_notify_peers = bond_should_notify_peers(bond);
2117 if (bond_miimon_inspect(bond)) {
2120 /* Race avoidance with bond_close cancel of workqueue */
2121 if (!rtnl_trylock()) {
2123 should_notify_peers = false;
2127 bond_miimon_commit(bond);
2129 rtnl_unlock(); /* might sleep, hold no other locks */
2134 if (bond->params.miimon)
2135 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2137 if (should_notify_peers) {
2138 if (!rtnl_trylock())
2140 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2145 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2147 struct net_device *upper;
2148 struct list_head *iter;
2151 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2155 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2156 if (ip == bond_confirm_addr(upper, 0, ip)) {
2166 /* We go to the (large) trouble of VLAN tagging ARP frames because
2167 * switches in VLAN mode (especially if ports are configured as
2168 * "native" to a VLAN) might not pass non-tagged frames.
2170 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2171 __be32 dest_ip, __be32 src_ip,
2172 struct bond_vlan_tag *tags)
2174 struct sk_buff *skb;
2175 struct bond_vlan_tag *outer_tag = tags;
2177 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2178 arp_op, slave_dev->name, &dest_ip, &src_ip);
2180 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2181 NULL, slave_dev->dev_addr, NULL);
2184 net_err_ratelimited("ARP packet allocation failed\n");
2188 if (!tags || tags->vlan_proto == VLAN_N_VID)
2193 /* Go through all the tags backwards and add them to the packet */
2194 while (tags->vlan_proto != VLAN_N_VID) {
2195 if (!tags->vlan_id) {
2200 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2201 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2202 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2205 net_err_ratelimited("failed to insert inner VLAN tag\n");
2211 /* Set the outer tag */
2212 if (outer_tag->vlan_id) {
2213 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2214 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2215 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2216 outer_tag->vlan_id);
2223 /* Validate the device path between the @start_dev and the @end_dev.
2224 * The path is valid if the @end_dev is reachable through device
2226 * When the path is validated, collect any vlan information in the
2229 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2230 struct net_device *end_dev,
2233 struct bond_vlan_tag *tags;
2234 struct net_device *upper;
2235 struct list_head *iter;
2237 if (start_dev == end_dev) {
2238 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2240 return ERR_PTR(-ENOMEM);
2241 tags[level].vlan_proto = VLAN_N_VID;
2245 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2246 tags = bond_verify_device_path(upper, end_dev, level + 1);
2247 if (IS_ERR_OR_NULL(tags)) {
2252 if (is_vlan_dev(upper)) {
2253 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2254 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2263 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2266 struct bond_vlan_tag *tags;
2267 __be32 *targets = bond->params.arp_targets, addr;
2270 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2271 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2274 /* Find out through which dev should the packet go */
2275 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2278 /* there's no route to target - try to send arp
2279 * probe to generate any traffic (arp_validate=0)
2281 if (bond->params.arp_validate)
2282 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2285 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2290 /* bond device itself */
2291 if (rt->dst.dev == bond->dev)
2295 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2298 if (!IS_ERR_OR_NULL(tags))
2301 /* Not our device - skip */
2302 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2303 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2309 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2311 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2317 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2321 if (!sip || !bond_has_this_ip(bond, tip)) {
2322 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2327 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2329 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2333 slave->last_rx = jiffies;
2334 slave->target_last_arp_rx[i] = jiffies;
2337 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2338 struct slave *slave)
2340 struct arphdr *arp = (struct arphdr *)skb->data;
2341 struct slave *curr_active_slave;
2342 unsigned char *arp_ptr;
2344 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2346 if (!slave_do_arp_validate(bond, slave)) {
2347 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2348 !slave_do_arp_validate_only(bond))
2349 slave->last_rx = jiffies;
2350 return RX_HANDLER_ANOTHER;
2351 } else if (!is_arp) {
2352 return RX_HANDLER_ANOTHER;
2355 alen = arp_hdr_len(bond->dev);
2357 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2360 if (alen > skb_headlen(skb)) {
2361 arp = kmalloc(alen, GFP_ATOMIC);
2364 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2368 if (arp->ar_hln != bond->dev->addr_len ||
2369 skb->pkt_type == PACKET_OTHERHOST ||
2370 skb->pkt_type == PACKET_LOOPBACK ||
2371 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2372 arp->ar_pro != htons(ETH_P_IP) ||
2376 arp_ptr = (unsigned char *)(arp + 1);
2377 arp_ptr += bond->dev->addr_len;
2378 memcpy(&sip, arp_ptr, 4);
2379 arp_ptr += 4 + bond->dev->addr_len;
2380 memcpy(&tip, arp_ptr, 4);
2382 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2383 slave->dev->name, bond_slave_state(slave),
2384 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2387 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2389 /* Backup slaves won't see the ARP reply, but do come through
2390 * here for each ARP probe (so we swap the sip/tip to validate
2391 * the probe). In a "redundant switch, common router" type of
2392 * configuration, the ARP probe will (hopefully) travel from
2393 * the active, through one switch, the router, then the other
2394 * switch before reaching the backup.
2396 * We 'trust' the arp requests if there is an active slave and
2397 * it received valid arp reply(s) after it became active. This
2398 * is done to avoid endless looping when we can't reach the
2399 * arp_ip_target and fool ourselves with our own arp requests.
2402 if (bond_is_active_slave(slave))
2403 bond_validate_arp(bond, slave, sip, tip);
2404 else if (curr_active_slave &&
2405 time_after(slave_last_rx(bond, curr_active_slave),
2406 curr_active_slave->last_link_up))
2407 bond_validate_arp(bond, slave, tip, sip);
2410 if (arp != (struct arphdr *)skb->data)
2412 return RX_HANDLER_ANOTHER;
2415 /* function to verify if we're in the arp_interval timeslice, returns true if
2416 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2417 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2419 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2422 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2424 return time_in_range(jiffies,
2425 last_act - delta_in_ticks,
2426 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2429 /* This function is called regularly to monitor each slave's link
2430 * ensuring that traffic is being sent and received when arp monitoring
2431 * is used in load-balancing mode. if the adapter has been dormant, then an
2432 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2433 * arp monitoring in active backup mode.
2435 static void bond_loadbalance_arp_mon(struct work_struct *work)
2437 struct bonding *bond = container_of(work, struct bonding,
2439 struct slave *slave, *oldcurrent;
2440 struct list_head *iter;
2441 int do_failover = 0, slave_state_changed = 0;
2443 if (!bond_has_slaves(bond))
2448 oldcurrent = rcu_dereference(bond->curr_active_slave);
2449 /* see if any of the previous devices are up now (i.e. they have
2450 * xmt and rcv traffic). the curr_active_slave does not come into
2451 * the picture unless it is null. also, slave->last_link_up is not
2452 * needed here because we send an arp on each slave and give a slave
2453 * as long as it needs to get the tx/rx within the delta.
2454 * TODO: what about up/down delay in arp mode? it wasn't here before
2457 bond_for_each_slave_rcu(bond, slave, iter) {
2458 unsigned long trans_start = dev_trans_start(slave->dev);
2460 if (slave->link != BOND_LINK_UP) {
2461 if (bond_time_in_interval(bond, trans_start, 1) &&
2462 bond_time_in_interval(bond, slave->last_rx, 1)) {
2464 slave->link = BOND_LINK_UP;
2465 slave_state_changed = 1;
2467 /* primary_slave has no meaning in round-robin
2468 * mode. the window of a slave being up and
2469 * curr_active_slave being null after enslaving
2473 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2477 netdev_info(bond->dev, "interface %s is now up\n",
2482 /* slave->link == BOND_LINK_UP */
2484 /* not all switches will respond to an arp request
2485 * when the source ip is 0, so don't take the link down
2486 * if we don't know our ip yet
2488 if (!bond_time_in_interval(bond, trans_start, 2) ||
2489 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2491 slave->link = BOND_LINK_DOWN;
2492 slave_state_changed = 1;
2494 if (slave->link_failure_count < UINT_MAX)
2495 slave->link_failure_count++;
2497 netdev_info(bond->dev, "interface %s is now down\n",
2500 if (slave == oldcurrent)
2505 /* note: if switch is in round-robin mode, all links
2506 * must tx arp to ensure all links rx an arp - otherwise
2507 * links may oscillate or not come up at all; if switch is
2508 * in something like xor mode, there is nothing we can
2509 * do - all replies will be rx'ed on same link causing slaves
2510 * to be unstable during low/no traffic periods
2512 if (bond_slave_is_up(slave))
2513 bond_arp_send_all(bond, slave);
2518 if (do_failover || slave_state_changed) {
2519 if (!rtnl_trylock())
2522 if (slave_state_changed) {
2523 bond_slave_state_change(bond);
2524 if (BOND_MODE(bond) == BOND_MODE_XOR)
2525 bond_update_slave_arr(bond, NULL);
2529 bond_select_active_slave(bond);
2530 unblock_netpoll_tx();
2536 if (bond->params.arp_interval)
2537 queue_delayed_work(bond->wq, &bond->arp_work,
2538 msecs_to_jiffies(bond->params.arp_interval));
2541 /* Called to inspect slaves for active-backup mode ARP monitor link state
2542 * changes. Sets new_link in slaves to specify what action should take
2543 * place for the slave. Returns 0 if no changes are found, >0 if changes
2544 * to link states must be committed.
2546 * Called with rcu_read_lock held.
2548 static int bond_ab_arp_inspect(struct bonding *bond)
2550 unsigned long trans_start, last_rx;
2551 struct list_head *iter;
2552 struct slave *slave;
2555 bond_for_each_slave_rcu(bond, slave, iter) {
2556 slave->new_link = BOND_LINK_NOCHANGE;
2557 last_rx = slave_last_rx(bond, slave);
2559 if (slave->link != BOND_LINK_UP) {
2560 if (bond_time_in_interval(bond, last_rx, 1)) {
2561 slave->new_link = BOND_LINK_UP;
2567 /* Give slaves 2*delta after being enslaved or made
2568 * active. This avoids bouncing, as the last receive
2569 * times need a full ARP monitor cycle to be updated.
2571 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2574 /* Backup slave is down if:
2575 * - No current_arp_slave AND
2576 * - more than 3*delta since last receive AND
2577 * - the bond has an IP address
2579 * Note: a non-null current_arp_slave indicates
2580 * the curr_active_slave went down and we are
2581 * searching for a new one; under this condition
2582 * we only take the curr_active_slave down - this
2583 * gives each slave a chance to tx/rx traffic
2584 * before being taken out
2586 if (!bond_is_active_slave(slave) &&
2587 !rcu_access_pointer(bond->current_arp_slave) &&
2588 !bond_time_in_interval(bond, last_rx, 3)) {
2589 slave->new_link = BOND_LINK_DOWN;
2593 /* Active slave is down if:
2594 * - more than 2*delta since transmitting OR
2595 * - (more than 2*delta since receive AND
2596 * the bond has an IP address)
2598 trans_start = dev_trans_start(slave->dev);
2599 if (bond_is_active_slave(slave) &&
2600 (!bond_time_in_interval(bond, trans_start, 2) ||
2601 !bond_time_in_interval(bond, last_rx, 2))) {
2602 slave->new_link = BOND_LINK_DOWN;
2610 /* Called to commit link state changes noted by inspection step of
2611 * active-backup mode ARP monitor.
2613 * Called with RTNL hold.
2615 static void bond_ab_arp_commit(struct bonding *bond)
2617 unsigned long trans_start;
2618 struct list_head *iter;
2619 struct slave *slave;
2621 bond_for_each_slave(bond, slave, iter) {
2622 switch (slave->new_link) {
2623 case BOND_LINK_NOCHANGE:
2627 trans_start = dev_trans_start(slave->dev);
2628 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2629 (!rtnl_dereference(bond->curr_active_slave) &&
2630 bond_time_in_interval(bond, trans_start, 1))) {
2631 struct slave *current_arp_slave;
2633 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2634 bond_set_slave_link_state(slave, BOND_LINK_UP);
2635 if (current_arp_slave) {
2636 bond_set_slave_inactive_flags(
2638 BOND_SLAVE_NOTIFY_NOW);
2639 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2642 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2645 if (!rtnl_dereference(bond->curr_active_slave) ||
2646 slave == rtnl_dereference(bond->primary_slave))
2653 case BOND_LINK_DOWN:
2654 if (slave->link_failure_count < UINT_MAX)
2655 slave->link_failure_count++;
2657 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2658 bond_set_slave_inactive_flags(slave,
2659 BOND_SLAVE_NOTIFY_NOW);
2661 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2664 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2665 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2672 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2673 slave->new_link, slave->dev->name);
2679 bond_select_active_slave(bond);
2680 unblock_netpoll_tx();
2683 bond_set_carrier(bond);
2686 /* Send ARP probes for active-backup mode ARP monitor.
2688 * Called with rcu_read_lock held.
2690 static bool bond_ab_arp_probe(struct bonding *bond)
2692 struct slave *slave, *before = NULL, *new_slave = NULL,
2693 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2694 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2695 struct list_head *iter;
2697 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2699 if (curr_arp_slave && curr_active_slave)
2700 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2701 curr_arp_slave->dev->name,
2702 curr_active_slave->dev->name);
2704 if (curr_active_slave) {
2705 bond_arp_send_all(bond, curr_active_slave);
2706 return should_notify_rtnl;
2709 /* if we don't have a curr_active_slave, search for the next available
2710 * backup slave from the current_arp_slave and make it the candidate
2711 * for becoming the curr_active_slave
2714 if (!curr_arp_slave) {
2715 curr_arp_slave = bond_first_slave_rcu(bond);
2716 if (!curr_arp_slave)
2717 return should_notify_rtnl;
2720 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2722 bond_for_each_slave_rcu(bond, slave, iter) {
2723 if (!found && !before && bond_slave_is_up(slave))
2726 if (found && !new_slave && bond_slave_is_up(slave))
2728 /* if the link state is up at this point, we
2729 * mark it down - this can happen if we have
2730 * simultaneous link failures and
2731 * reselect_active_interface doesn't make this
2732 * one the current slave so it is still marked
2733 * up when it is actually down
2735 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2736 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2737 if (slave->link_failure_count < UINT_MAX)
2738 slave->link_failure_count++;
2740 bond_set_slave_inactive_flags(slave,
2741 BOND_SLAVE_NOTIFY_LATER);
2743 netdev_info(bond->dev, "backup interface %s is now down\n",
2746 if (slave == curr_arp_slave)
2750 if (!new_slave && before)
2756 bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2757 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2758 bond_arp_send_all(bond, new_slave);
2759 new_slave->last_link_up = jiffies;
2760 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2763 bond_for_each_slave_rcu(bond, slave, iter) {
2764 if (slave->should_notify) {
2765 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2769 return should_notify_rtnl;
2772 static void bond_activebackup_arp_mon(struct work_struct *work)
2774 struct bonding *bond = container_of(work, struct bonding,
2776 bool should_notify_peers = false;
2777 bool should_notify_rtnl = false;
2780 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2782 if (!bond_has_slaves(bond))
2787 should_notify_peers = bond_should_notify_peers(bond);
2789 if (bond_ab_arp_inspect(bond)) {
2792 /* Race avoidance with bond_close flush of workqueue */
2793 if (!rtnl_trylock()) {
2795 should_notify_peers = false;
2799 bond_ab_arp_commit(bond);
2805 should_notify_rtnl = bond_ab_arp_probe(bond);
2809 if (bond->params.arp_interval)
2810 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2812 if (should_notify_peers || should_notify_rtnl) {
2813 if (!rtnl_trylock())
2816 if (should_notify_peers)
2817 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2819 if (should_notify_rtnl)
2820 bond_slave_state_notify(bond);
2826 /*-------------------------- netdev event handling --------------------------*/
2828 /* Change device name */
2829 static int bond_event_changename(struct bonding *bond)
2831 bond_remove_proc_entry(bond);
2832 bond_create_proc_entry(bond);
2834 bond_debug_reregister(bond);
2839 static int bond_master_netdev_event(unsigned long event,
2840 struct net_device *bond_dev)
2842 struct bonding *event_bond = netdev_priv(bond_dev);
2845 case NETDEV_CHANGENAME:
2846 return bond_event_changename(event_bond);
2847 case NETDEV_UNREGISTER:
2848 bond_remove_proc_entry(event_bond);
2850 case NETDEV_REGISTER:
2851 bond_create_proc_entry(event_bond);
2853 case NETDEV_NOTIFY_PEERS:
2854 if (event_bond->send_peer_notif)
2855 event_bond->send_peer_notif--;
2864 static int bond_slave_netdev_event(unsigned long event,
2865 struct net_device *slave_dev)
2867 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2868 struct bonding *bond;
2869 struct net_device *bond_dev;
2873 /* A netdev event can be generated while enslaving a device
2874 * before netdev_rx_handler_register is called in which case
2875 * slave will be NULL
2879 bond_dev = slave->bond->dev;
2881 primary = rtnl_dereference(bond->primary_slave);
2884 case NETDEV_UNREGISTER:
2885 if (bond_dev->type != ARPHRD_ETHER)
2886 bond_release_and_destroy(bond_dev, slave_dev);
2888 bond_release(bond_dev, slave_dev);
2892 old_speed = slave->speed;
2893 old_duplex = slave->duplex;
2895 bond_update_speed_duplex(slave);
2897 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2898 if (old_speed != slave->speed)
2899 bond_3ad_adapter_speed_changed(slave);
2900 if (old_duplex != slave->duplex)
2901 bond_3ad_adapter_duplex_changed(slave);
2903 /* Refresh slave-array if applicable!
2904 * If the setup does not use miimon or arpmon (mode-specific!),
2905 * then these events will not cause the slave-array to be
2906 * refreshed. This will cause xmit to use a slave that is not
2907 * usable. Avoid such situation by refeshing the array at these
2908 * events. If these (miimon/arpmon) parameters are configured
2909 * then array gets refreshed twice and that should be fine!
2911 if (bond_mode_uses_xmit_hash(bond))
2912 bond_update_slave_arr(bond, NULL);
2915 if (bond_mode_uses_xmit_hash(bond))
2916 bond_update_slave_arr(bond, NULL);
2918 case NETDEV_CHANGEMTU:
2919 /* TODO: Should slaves be allowed to
2920 * independently alter their MTU? For
2921 * an active-backup bond, slaves need
2922 * not be the same type of device, so
2923 * MTUs may vary. For other modes,
2924 * slaves arguably should have the
2925 * same MTUs. To do this, we'd need to
2926 * take over the slave's change_mtu
2927 * function for the duration of their
2931 case NETDEV_CHANGENAME:
2932 /* we don't care if we don't have primary set */
2933 if (!bond_uses_primary(bond) ||
2934 !bond->params.primary[0])
2937 if (slave == primary) {
2938 /* slave's name changed - he's no longer primary */
2939 RCU_INIT_POINTER(bond->primary_slave, NULL);
2940 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2941 /* we have a new primary slave */
2942 rcu_assign_pointer(bond->primary_slave, slave);
2943 } else { /* we didn't change primary - exit */
2947 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2948 primary ? slave_dev->name : "none");
2951 bond_select_active_slave(bond);
2952 unblock_netpoll_tx();
2954 case NETDEV_FEAT_CHANGE:
2955 bond_compute_features(bond);
2957 case NETDEV_RESEND_IGMP:
2958 /* Propagate to master device */
2959 call_netdevice_notifiers(event, slave->bond->dev);
2968 /* bond_netdev_event: handle netdev notifier chain events.
2970 * This function receives events for the netdev chain. The caller (an
2971 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2972 * locks for us to safely manipulate the slave devices (RTNL lock,
2975 static int bond_netdev_event(struct notifier_block *this,
2976 unsigned long event, void *ptr)
2978 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2980 netdev_dbg(event_dev, "event: %lx\n", event);
2982 if (!(event_dev->priv_flags & IFF_BONDING))
2985 if (event_dev->flags & IFF_MASTER) {
2986 netdev_dbg(event_dev, "IFF_MASTER\n");
2987 return bond_master_netdev_event(event, event_dev);
2990 if (event_dev->flags & IFF_SLAVE) {
2991 netdev_dbg(event_dev, "IFF_SLAVE\n");
2992 return bond_slave_netdev_event(event, event_dev);
2998 static struct notifier_block bond_netdev_notifier = {
2999 .notifier_call = bond_netdev_event,
3002 /*---------------------------- Hashing Policies -----------------------------*/
3004 /* L2 hash helper */
3005 static inline u32 bond_eth_hash(struct sk_buff *skb)
3007 struct ethhdr *ep, hdr_tmp;
3009 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3011 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3015 /* Extract the appropriate headers based on bond's xmit policy */
3016 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3017 struct flow_keys *fk)
3019 const struct ipv6hdr *iph6;
3020 const struct iphdr *iph;
3021 int noff, proto = -1;
3023 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3024 return skb_flow_dissect(skb, fk);
3027 noff = skb_network_offset(skb);
3028 if (skb->protocol == htons(ETH_P_IP)) {
3029 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3032 fk->src = iph->saddr;
3033 fk->dst = iph->daddr;
3034 noff += iph->ihl << 2;
3035 if (!ip_is_fragment(iph))
3036 proto = iph->protocol;
3037 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3038 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3040 iph6 = ipv6_hdr(skb);
3041 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3042 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3043 noff += sizeof(*iph6);
3044 proto = iph6->nexthdr;
3048 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3049 fk->ports = skb_flow_get_ports(skb, noff, proto);
3055 * bond_xmit_hash - generate a hash value based on the xmit policy
3056 * @bond: bonding device
3057 * @skb: buffer to use for headers
3059 * This function will extract the necessary headers from the skb buffer and use
3060 * them to generate a hash based on the xmit_policy set in the bonding device
3062 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3064 struct flow_keys flow;
3067 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3068 !bond_flow_dissect(bond, skb, &flow))
3069 return bond_eth_hash(skb);
3071 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3072 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3073 hash = bond_eth_hash(skb);
3075 hash = (__force u32)flow.ports;
3076 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3077 hash ^= (hash >> 16);
3078 hash ^= (hash >> 8);
3083 /*-------------------------- Device entry points ----------------------------*/
3085 static void bond_work_init_all(struct bonding *bond)
3087 INIT_DELAYED_WORK(&bond->mcast_work,
3088 bond_resend_igmp_join_requests_delayed);
3089 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3090 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3091 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3092 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3094 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3095 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3096 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3099 static void bond_work_cancel_all(struct bonding *bond)
3101 cancel_delayed_work_sync(&bond->mii_work);
3102 cancel_delayed_work_sync(&bond->arp_work);
3103 cancel_delayed_work_sync(&bond->alb_work);
3104 cancel_delayed_work_sync(&bond->ad_work);
3105 cancel_delayed_work_sync(&bond->mcast_work);
3106 cancel_delayed_work_sync(&bond->slave_arr_work);
3109 static int bond_open(struct net_device *bond_dev)
3111 struct bonding *bond = netdev_priv(bond_dev);
3112 struct list_head *iter;
3113 struct slave *slave;
3115 /* reset slave->backup and slave->inactive */
3116 if (bond_has_slaves(bond)) {
3117 bond_for_each_slave(bond, slave, iter) {
3118 if (bond_uses_primary(bond) &&
3119 slave != rcu_access_pointer(bond->curr_active_slave)) {
3120 bond_set_slave_inactive_flags(slave,
3121 BOND_SLAVE_NOTIFY_NOW);
3122 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3123 bond_set_slave_active_flags(slave,
3124 BOND_SLAVE_NOTIFY_NOW);
3129 bond_work_init_all(bond);
3131 if (bond_is_lb(bond)) {
3132 /* bond_alb_initialize must be called before the timer
3135 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3137 if (bond->params.tlb_dynamic_lb)
3138 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3141 if (bond->params.miimon) /* link check interval, in milliseconds. */
3142 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3144 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3145 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3146 bond->recv_probe = bond_arp_rcv;
3149 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3150 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3151 /* register to receive LACPDUs */
3152 bond->recv_probe = bond_3ad_lacpdu_recv;
3153 bond_3ad_initiate_agg_selection(bond, 1);
3156 if (bond_mode_uses_xmit_hash(bond))
3157 bond_update_slave_arr(bond, NULL);
3162 static int bond_close(struct net_device *bond_dev)
3164 struct bonding *bond = netdev_priv(bond_dev);
3166 bond_work_cancel_all(bond);
3167 bond->send_peer_notif = 0;
3168 if (bond_is_lb(bond))
3169 bond_alb_deinitialize(bond);
3170 bond->recv_probe = NULL;
3175 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3176 struct rtnl_link_stats64 *stats)
3178 struct bonding *bond = netdev_priv(bond_dev);
3179 struct rtnl_link_stats64 temp;
3180 struct list_head *iter;
3181 struct slave *slave;
3183 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3185 bond_for_each_slave(bond, slave, iter) {
3186 const struct rtnl_link_stats64 *sstats =
3187 dev_get_stats(slave->dev, &temp);
3188 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3190 stats->rx_packets += sstats->rx_packets - pstats->rx_packets;
3191 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3192 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3193 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3195 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3196 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3197 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3198 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3200 stats->multicast += sstats->multicast - pstats->multicast;
3201 stats->collisions += sstats->collisions - pstats->collisions;
3203 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3204 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3205 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3206 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3207 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3208 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3210 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3211 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3212 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3213 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3214 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3216 /* save off the slave stats for the next run */
3217 memcpy(pstats, sstats, sizeof(*sstats));
3219 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3224 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3226 struct bonding *bond = netdev_priv(bond_dev);
3227 struct net_device *slave_dev = NULL;
3228 struct ifbond k_binfo;
3229 struct ifbond __user *u_binfo = NULL;
3230 struct ifslave k_sinfo;
3231 struct ifslave __user *u_sinfo = NULL;
3232 struct mii_ioctl_data *mii = NULL;
3233 struct bond_opt_value newval;
3237 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3248 /* We do this again just in case we were called by SIOCGMIIREG
3249 * instead of SIOCGMIIPHY.
3255 if (mii->reg_num == 1) {
3257 if (netif_carrier_ok(bond->dev))
3258 mii->val_out = BMSR_LSTATUS;
3262 case BOND_INFO_QUERY_OLD:
3263 case SIOCBONDINFOQUERY:
3264 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3266 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3269 res = bond_info_query(bond_dev, &k_binfo);
3271 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3275 case BOND_SLAVE_INFO_QUERY_OLD:
3276 case SIOCBONDSLAVEINFOQUERY:
3277 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3279 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3282 res = bond_slave_info_query(bond_dev, &k_sinfo);
3284 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3292 net = dev_net(bond_dev);
3294 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3297 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3299 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3304 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3306 case BOND_ENSLAVE_OLD:
3307 case SIOCBONDENSLAVE:
3308 res = bond_enslave(bond_dev, slave_dev);
3310 case BOND_RELEASE_OLD:
3311 case SIOCBONDRELEASE:
3312 res = bond_release(bond_dev, slave_dev);
3314 case BOND_SETHWADDR_OLD:
3315 case SIOCBONDSETHWADDR:
3316 bond_set_dev_addr(bond_dev, slave_dev);
3319 case BOND_CHANGE_ACTIVE_OLD:
3320 case SIOCBONDCHANGEACTIVE:
3321 bond_opt_initstr(&newval, slave_dev->name);
3322 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3331 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3333 struct bonding *bond = netdev_priv(bond_dev);
3335 if (change & IFF_PROMISC)
3336 bond_set_promiscuity(bond,
3337 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3339 if (change & IFF_ALLMULTI)
3340 bond_set_allmulti(bond,
3341 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3344 static void bond_set_rx_mode(struct net_device *bond_dev)
3346 struct bonding *bond = netdev_priv(bond_dev);
3347 struct list_head *iter;
3348 struct slave *slave;
3351 if (bond_uses_primary(bond)) {
3352 slave = rcu_dereference(bond->curr_active_slave);
3354 dev_uc_sync(slave->dev, bond_dev);
3355 dev_mc_sync(slave->dev, bond_dev);
3358 bond_for_each_slave_rcu(bond, slave, iter) {
3359 dev_uc_sync_multiple(slave->dev, bond_dev);
3360 dev_mc_sync_multiple(slave->dev, bond_dev);
3366 static int bond_neigh_init(struct neighbour *n)
3368 struct bonding *bond = netdev_priv(n->dev);
3369 const struct net_device_ops *slave_ops;
3370 struct neigh_parms parms;
3371 struct slave *slave;
3374 slave = bond_first_slave(bond);
3377 slave_ops = slave->dev->netdev_ops;
3378 if (!slave_ops->ndo_neigh_setup)
3381 parms.neigh_setup = NULL;
3382 parms.neigh_cleanup = NULL;
3383 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3387 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3388 * after the last slave has been detached. Assumes that all slaves
3389 * utilize the same neigh_cleanup (true at this writing as only user
3392 n->parms->neigh_cleanup = parms.neigh_cleanup;
3394 if (!parms.neigh_setup)
3397 return parms.neigh_setup(n);
3400 /* The bonding ndo_neigh_setup is called at init time beofre any
3401 * slave exists. So we must declare proxy setup function which will
3402 * be used at run time to resolve the actual slave neigh param setup.
3404 * It's also called by master devices (such as vlans) to setup their
3405 * underlying devices. In that case - do nothing, we're already set up from
3408 static int bond_neigh_setup(struct net_device *dev,
3409 struct neigh_parms *parms)
3411 /* modify only our neigh_parms */
3412 if (parms->dev == dev)
3413 parms->neigh_setup = bond_neigh_init;
3418 /* Change the MTU of all of a master's slaves to match the master */
3419 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3421 struct bonding *bond = netdev_priv(bond_dev);
3422 struct slave *slave, *rollback_slave;
3423 struct list_head *iter;
3426 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3428 bond_for_each_slave(bond, slave, iter) {
3429 netdev_dbg(bond_dev, "s %p c_m %p\n",
3430 slave, slave->dev->netdev_ops->ndo_change_mtu);
3432 res = dev_set_mtu(slave->dev, new_mtu);
3435 /* If we failed to set the slave's mtu to the new value
3436 * we must abort the operation even in ACTIVE_BACKUP
3437 * mode, because if we allow the backup slaves to have
3438 * different mtu values than the active slave we'll
3439 * need to change their mtu when doing a failover. That
3440 * means changing their mtu from timer context, which
3441 * is probably not a good idea.
3443 netdev_dbg(bond_dev, "err %d %s\n", res,
3449 bond_dev->mtu = new_mtu;
3454 /* unwind from head to the slave that failed */
3455 bond_for_each_slave(bond, rollback_slave, iter) {
3458 if (rollback_slave == slave)
3461 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3463 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3464 tmp_res, rollback_slave->dev->name);
3471 /* Change HW address
3473 * Note that many devices must be down to change the HW address, and
3474 * downing the master releases all slaves. We can make bonds full of
3475 * bonding devices to test this, however.
3477 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3479 struct bonding *bond = netdev_priv(bond_dev);
3480 struct slave *slave, *rollback_slave;
3481 struct sockaddr *sa = addr, tmp_sa;
3482 struct list_head *iter;
3485 if (BOND_MODE(bond) == BOND_MODE_ALB)
3486 return bond_alb_set_mac_address(bond_dev, addr);
3489 netdev_dbg(bond_dev, "bond=%p\n", bond);
3491 /* If fail_over_mac is enabled, do nothing and return success.
3492 * Returning an error causes ifenslave to fail.
3494 if (bond->params.fail_over_mac &&
3495 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3498 if (!is_valid_ether_addr(sa->sa_data))
3499 return -EADDRNOTAVAIL;
3501 bond_for_each_slave(bond, slave, iter) {
3502 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3503 res = dev_set_mac_address(slave->dev, addr);
3505 /* TODO: consider downing the slave
3507 * User should expect communications
3508 * breakage anyway until ARP finish
3511 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3517 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3521 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3522 tmp_sa.sa_family = bond_dev->type;
3524 /* unwind from head to the slave that failed */
3525 bond_for_each_slave(bond, rollback_slave, iter) {
3528 if (rollback_slave == slave)
3531 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3533 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3534 tmp_res, rollback_slave->dev->name);
3542 * bond_xmit_slave_id - transmit skb through slave with slave_id
3543 * @bond: bonding device that is transmitting
3544 * @skb: buffer to transmit
3545 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3547 * This function tries to transmit through slave with slave_id but in case
3548 * it fails, it tries to find the first available slave for transmission.
3549 * The skb is consumed in all cases, thus the function is void.
3551 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3553 struct list_head *iter;
3554 struct slave *slave;
3557 /* Here we start from the slave with slave_id */
3558 bond_for_each_slave_rcu(bond, slave, iter) {
3560 if (bond_slave_can_tx(slave)) {
3561 bond_dev_queue_xmit(bond, skb, slave->dev);
3567 /* Here we start from the first slave up to slave_id */
3569 bond_for_each_slave_rcu(bond, slave, iter) {
3572 if (bond_slave_can_tx(slave)) {
3573 bond_dev_queue_xmit(bond, skb, slave->dev);
3577 /* no slave that can tx has been found */
3578 bond_tx_drop(bond->dev, skb);
3582 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3583 * @bond: bonding device to use
3585 * Based on the value of the bonding device's packets_per_slave parameter
3586 * this function generates a slave id, which is usually used as the next
3587 * slave to transmit through.
3589 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3592 struct reciprocal_value reciprocal_packets_per_slave;
3593 int packets_per_slave = bond->params.packets_per_slave;
3595 switch (packets_per_slave) {
3597 slave_id = prandom_u32();
3600 slave_id = bond->rr_tx_counter;
3603 reciprocal_packets_per_slave =
3604 bond->params.reciprocal_packets_per_slave;
3605 slave_id = reciprocal_divide(bond->rr_tx_counter,
3606 reciprocal_packets_per_slave);
3609 bond->rr_tx_counter++;
3614 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3616 struct bonding *bond = netdev_priv(bond_dev);
3617 struct iphdr *iph = ip_hdr(skb);
3618 struct slave *slave;
3621 /* Start with the curr_active_slave that joined the bond as the
3622 * default for sending IGMP traffic. For failover purposes one
3623 * needs to maintain some consistency for the interface that will
3624 * send the join/membership reports. The curr_active_slave found
3625 * will send all of this type of traffic.
3627 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3628 slave = rcu_dereference(bond->curr_active_slave);
3630 bond_dev_queue_xmit(bond, skb, slave->dev);
3632 bond_xmit_slave_id(bond, skb, 0);
3634 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3636 if (likely(slave_cnt)) {
3637 slave_id = bond_rr_gen_slave_id(bond);
3638 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3640 bond_tx_drop(bond_dev, skb);
3644 return NETDEV_TX_OK;
3647 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3648 * the bond has a usable interface.
3650 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3652 struct bonding *bond = netdev_priv(bond_dev);
3653 struct slave *slave;
3655 slave = rcu_dereference(bond->curr_active_slave);
3657 bond_dev_queue_xmit(bond, skb, slave->dev);
3659 bond_tx_drop(bond_dev, skb);
3661 return NETDEV_TX_OK;
3664 /* Use this to update slave_array when (a) it's not appropriate to update
3665 * slave_array right away (note that update_slave_array() may sleep)
3666 * and / or (b) RTNL is not held.
3668 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3670 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3673 /* Slave array work handler. Holds only RTNL */
3674 static void bond_slave_arr_handler(struct work_struct *work)
3676 struct bonding *bond = container_of(work, struct bonding,
3677 slave_arr_work.work);
3680 if (!rtnl_trylock())
3683 ret = bond_update_slave_arr(bond, NULL);
3686 pr_warn_ratelimited("Failed to update slave array from WT\n");
3692 bond_slave_arr_work_rearm(bond, 1);
3695 /* Build the usable slaves array in control path for modes that use xmit-hash
3696 * to determine the slave interface -
3697 * (a) BOND_MODE_8023AD
3699 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3701 * The caller is expected to hold RTNL only and NO other lock!
3703 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3705 struct slave *slave;
3706 struct list_head *iter;
3707 struct bond_up_slave *new_arr, *old_arr;
3712 #ifdef CONFIG_LOCKDEP
3713 WARN_ON(lockdep_is_held(&bond->mode_lock));
3716 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3720 pr_err("Failed to build slave-array.\n");
3723 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3724 struct ad_info ad_info;
3726 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3727 pr_debug("bond_3ad_get_active_agg_info failed\n");
3728 kfree_rcu(new_arr, rcu);
3729 /* No active aggragator means it's not safe to use
3730 * the previous array.
3732 old_arr = rtnl_dereference(bond->slave_arr);
3734 RCU_INIT_POINTER(bond->slave_arr, NULL);
3735 kfree_rcu(old_arr, rcu);
3739 slaves_in_agg = ad_info.ports;
3740 agg_id = ad_info.aggregator_id;
3742 bond_for_each_slave(bond, slave, iter) {
3743 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3744 struct aggregator *agg;
3746 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3747 if (!agg || agg->aggregator_identifier != agg_id)
3750 if (!bond_slave_can_tx(slave))
3752 if (skipslave == slave)
3754 new_arr->arr[new_arr->count++] = slave;
3757 old_arr = rtnl_dereference(bond->slave_arr);
3758 rcu_assign_pointer(bond->slave_arr, new_arr);
3760 kfree_rcu(old_arr, rcu);
3762 if (ret != 0 && skipslave) {
3765 /* Rare situation where caller has asked to skip a specific
3766 * slave but allocation failed (most likely!). BTW this is
3767 * only possible when the call is initiated from
3768 * __bond_release_one(). In this situation; overwrite the
3769 * skipslave entry in the array with the last entry from the
3770 * array to avoid a situation where the xmit path may choose
3771 * this to-be-skipped slave to send a packet out.
3773 old_arr = rtnl_dereference(bond->slave_arr);
3774 for (idx = 0; idx < old_arr->count; idx++) {
3775 if (skipslave == old_arr->arr[idx]) {
3777 old_arr->arr[old_arr->count-1];
3786 /* Use this Xmit function for 3AD as well as XOR modes. The current
3787 * usable slave array is formed in the control path. The xmit function
3788 * just calculates hash and sends the packet out.
3790 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3792 struct bonding *bond = netdev_priv(dev);
3793 struct slave *slave;
3794 struct bond_up_slave *slaves;
3797 slaves = rcu_dereference(bond->slave_arr);
3798 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3799 if (likely(count)) {
3800 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3801 bond_dev_queue_xmit(bond, skb, slave->dev);
3803 bond_tx_drop(dev, skb);
3806 return NETDEV_TX_OK;
3809 /* in broadcast mode, we send everything to all usable interfaces. */
3810 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3812 struct bonding *bond = netdev_priv(bond_dev);
3813 struct slave *slave = NULL;
3814 struct list_head *iter;
3816 bond_for_each_slave_rcu(bond, slave, iter) {
3817 if (bond_is_last_slave(bond, slave))
3819 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3820 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3823 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3824 bond_dev->name, __func__);
3827 bond_dev_queue_xmit(bond, skb2, slave->dev);
3830 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3831 bond_dev_queue_xmit(bond, skb, slave->dev);
3833 bond_tx_drop(bond_dev, skb);
3835 return NETDEV_TX_OK;
3838 /*------------------------- Device initialization ---------------------------*/
3840 /* Lookup the slave that corresponds to a qid */
3841 static inline int bond_slave_override(struct bonding *bond,
3842 struct sk_buff *skb)
3844 struct slave *slave = NULL;
3845 struct list_head *iter;
3847 if (!skb->queue_mapping)
3850 /* Find out if any slaves have the same mapping as this skb. */
3851 bond_for_each_slave_rcu(bond, slave, iter) {
3852 if (slave->queue_id == skb->queue_mapping) {
3853 if (bond_slave_is_up(slave) &&
3854 slave->link == BOND_LINK_UP) {
3855 bond_dev_queue_xmit(bond, skb, slave->dev);
3858 /* If the slave isn't UP, use default transmit policy. */
3867 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3868 void *accel_priv, select_queue_fallback_t fallback)
3870 /* This helper function exists to help dev_pick_tx get the correct
3871 * destination queue. Using a helper function skips a call to
3872 * skb_tx_hash and will put the skbs in the queue we expect on their
3873 * way down to the bonding driver.
3875 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3877 /* Save the original txq to restore before passing to the driver */
3878 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3880 if (unlikely(txq >= dev->real_num_tx_queues)) {
3882 txq -= dev->real_num_tx_queues;
3883 } while (txq >= dev->real_num_tx_queues);
3888 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3890 struct bonding *bond = netdev_priv(dev);
3892 if (bond_should_override_tx_queue(bond) &&
3893 !bond_slave_override(bond, skb))
3894 return NETDEV_TX_OK;
3896 switch (BOND_MODE(bond)) {
3897 case BOND_MODE_ROUNDROBIN:
3898 return bond_xmit_roundrobin(skb, dev);
3899 case BOND_MODE_ACTIVEBACKUP:
3900 return bond_xmit_activebackup(skb, dev);
3901 case BOND_MODE_8023AD:
3903 return bond_3ad_xor_xmit(skb, dev);
3904 case BOND_MODE_BROADCAST:
3905 return bond_xmit_broadcast(skb, dev);
3907 return bond_alb_xmit(skb, dev);
3909 return bond_tlb_xmit(skb, dev);
3911 /* Should never happen, mode already checked */
3912 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3914 bond_tx_drop(dev, skb);
3915 return NETDEV_TX_OK;
3919 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3921 struct bonding *bond = netdev_priv(dev);
3922 netdev_tx_t ret = NETDEV_TX_OK;
3924 /* If we risk deadlock from transmitting this in the
3925 * netpoll path, tell netpoll to queue the frame for later tx
3927 if (unlikely(is_netpoll_tx_blocked(dev)))
3928 return NETDEV_TX_BUSY;
3931 if (bond_has_slaves(bond))
3932 ret = __bond_start_xmit(skb, dev);
3934 bond_tx_drop(dev, skb);
3940 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3941 struct ethtool_cmd *ecmd)
3943 struct bonding *bond = netdev_priv(bond_dev);
3944 unsigned long speed = 0;
3945 struct list_head *iter;
3946 struct slave *slave;
3948 ecmd->duplex = DUPLEX_UNKNOWN;
3949 ecmd->port = PORT_OTHER;
3951 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3952 * do not need to check mode. Though link speed might not represent
3953 * the true receive or transmit bandwidth (not all modes are symmetric)
3954 * this is an accurate maximum.
3956 bond_for_each_slave(bond, slave, iter) {
3957 if (bond_slave_can_tx(slave)) {
3958 if (slave->speed != SPEED_UNKNOWN)
3959 speed += slave->speed;
3960 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3961 slave->duplex != DUPLEX_UNKNOWN)
3962 ecmd->duplex = slave->duplex;
3965 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3970 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3971 struct ethtool_drvinfo *drvinfo)
3973 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3974 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3975 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3979 static const struct ethtool_ops bond_ethtool_ops = {
3980 .get_drvinfo = bond_ethtool_get_drvinfo,
3981 .get_settings = bond_ethtool_get_settings,
3982 .get_link = ethtool_op_get_link,
3985 static const struct net_device_ops bond_netdev_ops = {
3986 .ndo_init = bond_init,
3987 .ndo_uninit = bond_uninit,
3988 .ndo_open = bond_open,
3989 .ndo_stop = bond_close,
3990 .ndo_start_xmit = bond_start_xmit,
3991 .ndo_select_queue = bond_select_queue,
3992 .ndo_get_stats64 = bond_get_stats,
3993 .ndo_do_ioctl = bond_do_ioctl,
3994 .ndo_change_rx_flags = bond_change_rx_flags,
3995 .ndo_set_rx_mode = bond_set_rx_mode,
3996 .ndo_change_mtu = bond_change_mtu,
3997 .ndo_set_mac_address = bond_set_mac_address,
3998 .ndo_neigh_setup = bond_neigh_setup,
3999 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4000 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4001 #ifdef CONFIG_NET_POLL_CONTROLLER
4002 .ndo_netpoll_setup = bond_netpoll_setup,
4003 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4004 .ndo_poll_controller = bond_poll_controller,
4006 .ndo_add_slave = bond_enslave,
4007 .ndo_del_slave = bond_release,
4008 .ndo_fix_features = bond_fix_features,
4009 .ndo_bridge_setlink = ndo_dflt_netdev_switch_port_bridge_setlink,
4010 .ndo_bridge_dellink = ndo_dflt_netdev_switch_port_bridge_dellink,
4013 static const struct device_type bond_type = {
4017 static void bond_destructor(struct net_device *bond_dev)
4019 struct bonding *bond = netdev_priv(bond_dev);
4021 destroy_workqueue(bond->wq);
4022 free_netdev(bond_dev);
4025 void bond_setup(struct net_device *bond_dev)
4027 struct bonding *bond = netdev_priv(bond_dev);
4029 spin_lock_init(&bond->mode_lock);
4030 bond->params = bonding_defaults;
4032 /* Initialize pointers */
4033 bond->dev = bond_dev;
4035 /* Initialize the device entry points */
4036 ether_setup(bond_dev);
4037 bond_dev->netdev_ops = &bond_netdev_ops;
4038 bond_dev->ethtool_ops = &bond_ethtool_ops;
4040 bond_dev->destructor = bond_destructor;
4042 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4044 /* Initialize the device options */
4045 bond_dev->tx_queue_len = 0;
4046 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4047 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
4048 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4050 /* don't acquire bond device's netif_tx_lock when transmitting */
4051 bond_dev->features |= NETIF_F_LLTX;
4053 /* By default, we declare the bond to be fully
4054 * VLAN hardware accelerated capable. Special
4055 * care is taken in the various xmit functions
4056 * when there are slaves that are not hw accel
4060 /* Don't allow bond devices to change network namespaces. */
4061 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4063 bond_dev->hw_features = BOND_VLAN_FEATURES |
4064 NETIF_F_HW_VLAN_CTAG_TX |
4065 NETIF_F_HW_VLAN_CTAG_RX |
4066 NETIF_F_HW_VLAN_CTAG_FILTER;
4068 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4069 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4070 bond_dev->features |= bond_dev->hw_features;
4073 /* Destroy a bonding device.
4074 * Must be under rtnl_lock when this function is called.
4076 static void bond_uninit(struct net_device *bond_dev)
4078 struct bonding *bond = netdev_priv(bond_dev);
4079 struct list_head *iter;
4080 struct slave *slave;
4081 struct bond_up_slave *arr;
4083 bond_netpoll_cleanup(bond_dev);
4085 /* Release the bonded slaves */
4086 bond_for_each_slave(bond, slave, iter)
4087 __bond_release_one(bond_dev, slave->dev, true);
4088 netdev_info(bond_dev, "Released all slaves\n");
4090 arr = rtnl_dereference(bond->slave_arr);
4092 RCU_INIT_POINTER(bond->slave_arr, NULL);
4093 kfree_rcu(arr, rcu);
4096 list_del(&bond->bond_list);
4098 bond_debug_unregister(bond);
4101 /*------------------------- Module initialization ---------------------------*/
4103 static int bond_check_params(struct bond_params *params)
4105 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4106 struct bond_opt_value newval;
4107 const struct bond_opt_value *valptr;
4108 int arp_all_targets_value;
4110 /* Convert string parameters. */
4112 bond_opt_initstr(&newval, mode);
4113 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4115 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4118 bond_mode = valptr->value;
4121 if (xmit_hash_policy) {
4122 if ((bond_mode != BOND_MODE_XOR) &&
4123 (bond_mode != BOND_MODE_8023AD) &&
4124 (bond_mode != BOND_MODE_TLB)) {
4125 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4126 bond_mode_name(bond_mode));
4128 bond_opt_initstr(&newval, xmit_hash_policy);
4129 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4132 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4136 xmit_hashtype = valptr->value;
4141 if (bond_mode != BOND_MODE_8023AD) {
4142 pr_info("lacp_rate param is irrelevant in mode %s\n",
4143 bond_mode_name(bond_mode));
4145 bond_opt_initstr(&newval, lacp_rate);
4146 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4149 pr_err("Error: Invalid lacp rate \"%s\"\n",
4153 lacp_fast = valptr->value;
4158 bond_opt_initstr(&newval, ad_select);
4159 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4162 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4165 params->ad_select = valptr->value;
4166 if (bond_mode != BOND_MODE_8023AD)
4167 pr_warn("ad_select param only affects 802.3ad mode\n");
4169 params->ad_select = BOND_AD_STABLE;
4172 if (max_bonds < 0) {
4173 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4174 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4175 max_bonds = BOND_DEFAULT_MAX_BONDS;
4179 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4185 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4190 if (downdelay < 0) {
4191 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4192 downdelay, INT_MAX);
4196 if ((use_carrier != 0) && (use_carrier != 1)) {
4197 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4202 if (num_peer_notif < 0 || num_peer_notif > 255) {
4203 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4208 /* reset values for 802.3ad/TLB/ALB */
4209 if (!bond_mode_uses_arp(bond_mode)) {
4211 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4212 pr_warn("Forcing miimon to 100msec\n");
4213 miimon = BOND_DEFAULT_MIIMON;
4217 if (tx_queues < 1 || tx_queues > 255) {
4218 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4219 tx_queues, BOND_DEFAULT_TX_QUEUES);
4220 tx_queues = BOND_DEFAULT_TX_QUEUES;
4223 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4224 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4226 all_slaves_active = 0;
4229 if (resend_igmp < 0 || resend_igmp > 255) {
4230 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4231 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4232 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4235 bond_opt_initval(&newval, packets_per_slave);
4236 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4237 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4238 packets_per_slave, USHRT_MAX);
4239 packets_per_slave = 1;
4242 if (bond_mode == BOND_MODE_ALB) {
4243 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",
4248 if (updelay || downdelay) {
4249 /* just warn the user the up/down delay will have
4250 * no effect since miimon is zero...
4252 pr_warn("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",
4253 updelay, downdelay);
4256 /* don't allow arp monitoring */
4258 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4259 miimon, arp_interval);
4263 if ((updelay % miimon) != 0) {
4264 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4265 updelay, miimon, (updelay / miimon) * miimon);
4270 if ((downdelay % miimon) != 0) {
4271 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4273 (downdelay / miimon) * miimon);
4276 downdelay /= miimon;
4279 if (arp_interval < 0) {
4280 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4281 arp_interval, INT_MAX);
4285 for (arp_ip_count = 0, i = 0;
4286 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4289 /* not a complete check, but good enough to catch mistakes */
4290 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4291 !bond_is_ip_target_ok(ip)) {
4292 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4296 if (bond_get_targets_ip(arp_target, ip) == -1)
4297 arp_target[arp_ip_count++] = ip;
4299 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4304 if (arp_interval && !arp_ip_count) {
4305 /* don't allow arping if no arp_ip_target given... */
4306 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4312 if (!arp_interval) {
4313 pr_err("arp_validate requires arp_interval\n");
4317 bond_opt_initstr(&newval, arp_validate);
4318 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4321 pr_err("Error: invalid arp_validate \"%s\"\n",
4325 arp_validate_value = valptr->value;
4327 arp_validate_value = 0;
4330 arp_all_targets_value = 0;
4331 if (arp_all_targets) {
4332 bond_opt_initstr(&newval, arp_all_targets);
4333 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4336 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4338 arp_all_targets_value = 0;
4340 arp_all_targets_value = valptr->value;
4345 pr_info("MII link monitoring set to %d ms\n", miimon);
4346 } else if (arp_interval) {
4347 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4348 arp_validate_value);
4349 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4350 arp_interval, valptr->string, arp_ip_count);
4352 for (i = 0; i < arp_ip_count; i++)
4353 pr_cont(" %s", arp_ip_target[i]);
4357 } else if (max_bonds) {
4358 /* miimon and arp_interval not set, we need one so things
4359 * work as expected, see bonding.txt for details
4361 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");
4364 if (primary && !bond_mode_uses_primary(bond_mode)) {
4365 /* currently, using a primary only makes sense
4366 * in active backup, TLB or ALB modes
4368 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4369 primary, bond_mode_name(bond_mode));
4373 if (primary && primary_reselect) {
4374 bond_opt_initstr(&newval, primary_reselect);
4375 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4378 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4382 primary_reselect_value = valptr->value;
4384 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4387 if (fail_over_mac) {
4388 bond_opt_initstr(&newval, fail_over_mac);
4389 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4392 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4396 fail_over_mac_value = valptr->value;
4397 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4398 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4400 fail_over_mac_value = BOND_FOM_NONE;
4403 if (lp_interval == 0) {
4404 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4405 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4406 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4409 /* fill params struct with the proper values */
4410 params->mode = bond_mode;
4411 params->xmit_policy = xmit_hashtype;
4412 params->miimon = miimon;
4413 params->num_peer_notif = num_peer_notif;
4414 params->arp_interval = arp_interval;
4415 params->arp_validate = arp_validate_value;
4416 params->arp_all_targets = arp_all_targets_value;
4417 params->updelay = updelay;
4418 params->downdelay = downdelay;
4419 params->use_carrier = use_carrier;
4420 params->lacp_fast = lacp_fast;
4421 params->primary[0] = 0;
4422 params->primary_reselect = primary_reselect_value;
4423 params->fail_over_mac = fail_over_mac_value;
4424 params->tx_queues = tx_queues;
4425 params->all_slaves_active = all_slaves_active;
4426 params->resend_igmp = resend_igmp;
4427 params->min_links = min_links;
4428 params->lp_interval = lp_interval;
4429 params->packets_per_slave = packets_per_slave;
4430 params->tlb_dynamic_lb = 1; /* Default value */
4431 if (packets_per_slave > 0) {
4432 params->reciprocal_packets_per_slave =
4433 reciprocal_value(packets_per_slave);
4435 /* reciprocal_packets_per_slave is unused if
4436 * packets_per_slave is 0 or 1, just initialize it
4438 params->reciprocal_packets_per_slave =
4439 (struct reciprocal_value) { 0 };
4443 strncpy(params->primary, primary, IFNAMSIZ);
4444 params->primary[IFNAMSIZ - 1] = 0;
4447 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4452 static struct lock_class_key bonding_netdev_xmit_lock_key;
4453 static struct lock_class_key bonding_netdev_addr_lock_key;
4454 static struct lock_class_key bonding_tx_busylock_key;
4456 static void bond_set_lockdep_class_one(struct net_device *dev,
4457 struct netdev_queue *txq,
4460 lockdep_set_class(&txq->_xmit_lock,
4461 &bonding_netdev_xmit_lock_key);
4464 static void bond_set_lockdep_class(struct net_device *dev)
4466 lockdep_set_class(&dev->addr_list_lock,
4467 &bonding_netdev_addr_lock_key);
4468 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4469 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4472 /* Called from registration process */
4473 static int bond_init(struct net_device *bond_dev)
4475 struct bonding *bond = netdev_priv(bond_dev);
4476 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4478 netdev_dbg(bond_dev, "Begin bond_init\n");
4480 bond->wq = create_singlethread_workqueue(bond_dev->name);
4484 bond_set_lockdep_class(bond_dev);
4486 list_add_tail(&bond->bond_list, &bn->dev_list);
4488 bond_prepare_sysfs_group(bond);
4490 bond_debug_register(bond);
4492 /* Ensure valid dev_addr */
4493 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4494 bond_dev->addr_assign_type == NET_ADDR_PERM)
4495 eth_hw_addr_random(bond_dev);
4500 unsigned int bond_get_num_tx_queues(void)
4505 /* Create a new bond based on the specified name and bonding parameters.
4506 * If name is NULL, obtain a suitable "bond%d" name for us.
4507 * Caller must NOT hold rtnl_lock; we need to release it here before we
4508 * set up our sysfs entries.
4510 int bond_create(struct net *net, const char *name)
4512 struct net_device *bond_dev;
4517 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4518 name ? name : "bond%d", NET_NAME_UNKNOWN,
4519 bond_setup, tx_queues);
4521 pr_err("%s: eek! can't alloc netdev!\n", name);
4526 dev_net_set(bond_dev, net);
4527 bond_dev->rtnl_link_ops = &bond_link_ops;
4529 res = register_netdevice(bond_dev);
4531 netif_carrier_off(bond_dev);
4535 bond_destructor(bond_dev);
4539 static int __net_init bond_net_init(struct net *net)
4541 struct bond_net *bn = net_generic(net, bond_net_id);
4544 INIT_LIST_HEAD(&bn->dev_list);
4546 bond_create_proc_dir(bn);
4547 bond_create_sysfs(bn);
4552 static void __net_exit bond_net_exit(struct net *net)
4554 struct bond_net *bn = net_generic(net, bond_net_id);
4555 struct bonding *bond, *tmp_bond;
4558 bond_destroy_sysfs(bn);
4560 /* Kill off any bonds created after unregistering bond rtnl ops */
4562 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4563 unregister_netdevice_queue(bond->dev, &list);
4564 unregister_netdevice_many(&list);
4567 bond_destroy_proc_dir(bn);
4570 static struct pernet_operations bond_net_ops = {
4571 .init = bond_net_init,
4572 .exit = bond_net_exit,
4574 .size = sizeof(struct bond_net),
4577 static int __init bonding_init(void)
4582 pr_info("%s", bond_version);
4584 res = bond_check_params(&bonding_defaults);
4588 res = register_pernet_subsys(&bond_net_ops);
4592 res = bond_netlink_init();
4596 bond_create_debugfs();
4598 for (i = 0; i < max_bonds; i++) {
4599 res = bond_create(&init_net, NULL);
4604 register_netdevice_notifier(&bond_netdev_notifier);
4608 bond_destroy_debugfs();
4609 bond_netlink_fini();
4611 unregister_pernet_subsys(&bond_net_ops);
4616 static void __exit bonding_exit(void)
4618 unregister_netdevice_notifier(&bond_netdev_notifier);
4620 bond_destroy_debugfs();
4622 bond_netlink_fini();
4623 unregister_pernet_subsys(&bond_net_ops);
4625 #ifdef CONFIG_NET_POLL_CONTROLLER
4626 /* Make sure we don't have an imbalance on our netpoll blocking */
4627 WARN_ON(atomic_read(&netpoll_block_tx));
4631 module_init(bonding_init);
4632 module_exit(bonding_exit);
4633 MODULE_LICENSE("GPL");
4634 MODULE_VERSION(DRV_VERSION);
4635 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4636 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");