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_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 #include "bonding_priv.h"
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 static int packets_per_slave = 1;
114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
116 module_param(max_bonds, int, 0);
117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118 module_param(tx_queues, int, 0);
119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126 module_param(miimon, int, 0);
127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128 module_param(updelay, int, 0);
129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130 module_param(downdelay, int, 0);
131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
133 module_param(use_carrier, int, 0);
134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136 module_param(mode, charp, 0);
137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141 module_param(primary, charp, 0);
142 MODULE_PARM_DESC(primary, "Primary network device to use");
143 module_param(primary_reselect, charp, 0);
144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
149 "2 for only on active slave "
151 module_param(lacp_rate, charp, 0);
152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154 module_param(ad_select, charp, 0);
155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
158 module_param(min_links, int, 0);
159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166 module_param(arp_interval, int, 0);
167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168 module_param_array(arp_ip_target, charp, NULL, 0);
169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170 module_param(arp_validate, charp, 0);
171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174 module_param(arp_all_targets, charp, 0);
175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac, charp, 0);
177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180 module_param(all_slaves_active, int, 0);
181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184 module_param(resend_igmp, int, 0);
185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
187 module_param(packets_per_slave, int, 0);
188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191 module_param(lp_interval, uint, 0);
192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
196 /*----------------------------- Global variables ----------------------------*/
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
202 unsigned int bond_net_id __read_mostly;
204 /*-------------------------- Forward declarations ---------------------------*/
206 static int bond_init(struct net_device *bond_dev);
207 static void bond_uninit(struct net_device *bond_dev);
208 static void bond_get_stats(struct net_device *bond_dev,
209 struct rtnl_link_stats64 *stats);
210 static void bond_slave_arr_handler(struct work_struct *work);
211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
213 static void bond_netdev_notify_work(struct work_struct *work);
215 /*---------------------------- General routines -----------------------------*/
217 const char *bond_mode_name(int mode)
219 static const char *names[] = {
220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
224 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
225 [BOND_MODE_TLB] = "transmit load balancing",
226 [BOND_MODE_ALB] = "adaptive load balancing",
229 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
235 /*---------------------------------- VLAN -----------------------------------*/
238 * bond_dev_queue_xmit - Prepare skb for xmit.
240 * @bond: bond device that got this skb for tx.
241 * @skb: hw accel VLAN tagged skb to transmit
242 * @slave_dev: slave that is supposed to xmit this skbuff
244 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
245 struct net_device *slave_dev)
247 skb->dev = slave_dev;
249 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
250 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
251 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
253 if (unlikely(netpoll_tx_running(bond->dev)))
254 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
259 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
260 * We don't protect the slave list iteration with a lock because:
261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
281 __be16 proto, u16 vid)
283 struct bonding *bond = netdev_priv(bond_dev);
284 struct slave *slave, *rollback_slave;
285 struct list_head *iter;
288 bond_for_each_slave(bond, slave, iter) {
289 res = vlan_vid_add(slave->dev, proto, vid);
297 /* unwind to the slave that failed */
298 bond_for_each_slave(bond, rollback_slave, iter) {
299 if (rollback_slave == slave)
302 vlan_vid_del(rollback_slave->dev, proto, vid);
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
314 __be16 proto, u16 vid)
316 struct bonding *bond = netdev_priv(bond_dev);
317 struct list_head *iter;
320 bond_for_each_slave(bond, slave, iter)
321 vlan_vid_del(slave->dev, proto, vid);
323 if (bond_is_lb(bond))
324 bond_alb_clear_vlan(bond, vid);
329 /*------------------------------- Link status -------------------------------*/
331 /* Set the carrier state for the master according to the state of its
332 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
333 * do special 802.3ad magic.
335 * Returns zero if carrier state does not change, nonzero if it does.
337 int bond_set_carrier(struct bonding *bond)
339 struct list_head *iter;
342 if (!bond_has_slaves(bond))
345 if (BOND_MODE(bond) == BOND_MODE_8023AD)
346 return bond_3ad_set_carrier(bond);
348 bond_for_each_slave(bond, slave, iter) {
349 if (slave->link == BOND_LINK_UP) {
350 if (!netif_carrier_ok(bond->dev)) {
351 netif_carrier_on(bond->dev);
359 if (netif_carrier_ok(bond->dev)) {
360 netif_carrier_off(bond->dev);
366 /* Get link speed and duplex from the slave's base driver
367 * using ethtool. If for some reason the call fails or the
368 * values are invalid, set speed and duplex to -1,
369 * and return. Return 1 if speed or duplex settings are
370 * UNKNOWN; 0 otherwise.
372 static int bond_update_speed_duplex(struct slave *slave)
374 struct net_device *slave_dev = slave->dev;
375 struct ethtool_link_ksettings ecmd;
378 slave->speed = SPEED_UNKNOWN;
379 slave->duplex = DUPLEX_UNKNOWN;
381 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
384 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
386 switch (ecmd.base.duplex) {
394 slave->speed = ecmd.base.speed;
395 slave->duplex = ecmd.base.duplex;
400 const char *bond_slave_link_status(s8 link)
416 /* if <dev> supports MII link status reporting, check its link status.
418 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
419 * depending upon the setting of the use_carrier parameter.
421 * Return either BMSR_LSTATUS, meaning that the link is up (or we
422 * can't tell and just pretend it is), or 0, meaning that the link is
425 * If reporting is non-zero, instead of faking link up, return -1 if
426 * both ETHTOOL and MII ioctls fail (meaning the device does not
427 * support them). If use_carrier is set, return whatever it says.
428 * It'd be nice if there was a good way to tell if a driver supports
429 * netif_carrier, but there really isn't.
431 static int bond_check_dev_link(struct bonding *bond,
432 struct net_device *slave_dev, int reporting)
434 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
435 int (*ioctl)(struct net_device *, struct ifreq *, int);
437 struct mii_ioctl_data *mii;
439 if (!reporting && !netif_running(slave_dev))
442 if (bond->params.use_carrier)
443 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
445 /* Try to get link status using Ethtool first. */
446 if (slave_dev->ethtool_ops->get_link)
447 return slave_dev->ethtool_ops->get_link(slave_dev) ?
450 /* Ethtool can't be used, fallback to MII ioctls. */
451 ioctl = slave_ops->ndo_do_ioctl;
453 /* TODO: set pointer to correct ioctl on a per team member
454 * bases to make this more efficient. that is, once
455 * we determine the correct ioctl, we will always
456 * call it and not the others for that team
460 /* We cannot assume that SIOCGMIIPHY will also read a
461 * register; not all network drivers (e.g., e100)
465 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
466 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
468 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
469 mii->reg_num = MII_BMSR;
470 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
471 return mii->val_out & BMSR_LSTATUS;
475 /* If reporting, report that either there's no dev->do_ioctl,
476 * or both SIOCGMIIREG and get_link failed (meaning that we
477 * cannot report link status). If not reporting, pretend
480 return reporting ? -1 : BMSR_LSTATUS;
483 /*----------------------------- Multicast list ------------------------------*/
485 /* Push the promiscuity flag down to appropriate slaves */
486 static int bond_set_promiscuity(struct bonding *bond, int inc)
488 struct list_head *iter;
491 if (bond_uses_primary(bond)) {
492 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
495 err = dev_set_promiscuity(curr_active->dev, inc);
499 bond_for_each_slave(bond, slave, iter) {
500 err = dev_set_promiscuity(slave->dev, inc);
508 /* Push the allmulti flag down to all slaves */
509 static int bond_set_allmulti(struct bonding *bond, int inc)
511 struct list_head *iter;
514 if (bond_uses_primary(bond)) {
515 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
518 err = dev_set_allmulti(curr_active->dev, inc);
522 bond_for_each_slave(bond, slave, iter) {
523 err = dev_set_allmulti(slave->dev, inc);
531 /* Retrieve the list of registered multicast addresses for the bonding
532 * device and retransmit an IGMP JOIN request to the current active
535 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
537 struct bonding *bond = container_of(work, struct bonding,
540 if (!rtnl_trylock()) {
541 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
544 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
546 if (bond->igmp_retrans > 1) {
547 bond->igmp_retrans--;
548 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
553 /* Flush bond's hardware addresses from slave */
554 static void bond_hw_addr_flush(struct net_device *bond_dev,
555 struct net_device *slave_dev)
557 struct bonding *bond = netdev_priv(bond_dev);
559 dev_uc_unsync(slave_dev, bond_dev);
560 dev_mc_unsync(slave_dev, bond_dev);
562 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
563 /* del lacpdu mc addr from mc list */
564 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
566 dev_mc_del(slave_dev, lacpdu_multicast);
570 /*--------------------------- Active slave change ---------------------------*/
572 /* Update the hardware address list and promisc/allmulti for the new and
573 * old active slaves (if any). Modes that are not using primary keep all
574 * slaves up date at all times; only the modes that use primary need to call
575 * this function to swap these settings during a failover.
577 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
578 struct slave *old_active)
581 if (bond->dev->flags & IFF_PROMISC)
582 dev_set_promiscuity(old_active->dev, -1);
584 if (bond->dev->flags & IFF_ALLMULTI)
585 dev_set_allmulti(old_active->dev, -1);
587 bond_hw_addr_flush(bond->dev, old_active->dev);
591 /* FIXME: Signal errors upstream. */
592 if (bond->dev->flags & IFF_PROMISC)
593 dev_set_promiscuity(new_active->dev, 1);
595 if (bond->dev->flags & IFF_ALLMULTI)
596 dev_set_allmulti(new_active->dev, 1);
598 netif_addr_lock_bh(bond->dev);
599 dev_uc_sync(new_active->dev, bond->dev);
600 dev_mc_sync(new_active->dev, bond->dev);
601 netif_addr_unlock_bh(bond->dev);
606 * bond_set_dev_addr - clone slave's address to bond
607 * @bond_dev: bond net device
608 * @slave_dev: slave net device
610 * Should be called with RTNL held.
612 static void bond_set_dev_addr(struct net_device *bond_dev,
613 struct net_device *slave_dev)
615 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
616 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
617 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
618 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
619 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
622 static struct slave *bond_get_old_active(struct bonding *bond,
623 struct slave *new_active)
626 struct list_head *iter;
628 bond_for_each_slave(bond, slave, iter) {
629 if (slave == new_active)
632 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
639 /* bond_do_fail_over_mac
641 * Perform special MAC address swapping for fail_over_mac settings
645 static void bond_do_fail_over_mac(struct bonding *bond,
646 struct slave *new_active,
647 struct slave *old_active)
649 u8 tmp_mac[MAX_ADDR_LEN];
650 struct sockaddr_storage ss;
653 switch (bond->params.fail_over_mac) {
654 case BOND_FOM_ACTIVE:
656 bond_set_dev_addr(bond->dev, new_active->dev);
658 case BOND_FOM_FOLLOW:
659 /* if new_active && old_active, swap them
660 * if just old_active, do nothing (going to no active slave)
661 * if just new_active, set new_active to bond's MAC
667 old_active = bond_get_old_active(bond, new_active);
670 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
671 new_active->dev->addr_len);
672 bond_hw_addr_copy(ss.__data,
673 old_active->dev->dev_addr,
674 old_active->dev->addr_len);
675 ss.ss_family = new_active->dev->type;
677 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
678 bond->dev->addr_len);
679 ss.ss_family = bond->dev->type;
682 rv = dev_set_mac_address(new_active->dev,
683 (struct sockaddr *)&ss);
685 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
686 -rv, new_active->dev->name);
693 bond_hw_addr_copy(ss.__data, tmp_mac,
694 new_active->dev->addr_len);
695 ss.ss_family = old_active->dev->type;
697 rv = dev_set_mac_address(old_active->dev,
698 (struct sockaddr *)&ss);
700 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
701 -rv, new_active->dev->name);
705 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
706 bond->params.fail_over_mac);
712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
714 struct slave *prim = rtnl_dereference(bond->primary_slave);
715 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
717 if (!prim || prim->link != BOND_LINK_UP) {
718 if (!curr || curr->link != BOND_LINK_UP)
723 if (bond->force_primary) {
724 bond->force_primary = false;
728 if (!curr || curr->link != BOND_LINK_UP)
731 /* At this point, prim and curr are both up */
732 switch (bond->params.primary_reselect) {
733 case BOND_PRI_RESELECT_ALWAYS:
735 case BOND_PRI_RESELECT_BETTER:
736 if (prim->speed < curr->speed)
738 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
741 case BOND_PRI_RESELECT_FAILURE:
744 netdev_err(bond->dev, "impossible primary_reselect %d\n",
745 bond->params.primary_reselect);
751 * bond_find_best_slave - select the best available slave to be the active one
752 * @bond: our bonding struct
754 static struct slave *bond_find_best_slave(struct bonding *bond)
756 struct slave *slave, *bestslave = NULL;
757 struct list_head *iter;
758 int mintime = bond->params.updelay;
760 slave = bond_choose_primary_or_current(bond);
764 bond_for_each_slave(bond, slave, iter) {
765 if (slave->link == BOND_LINK_UP)
767 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
768 slave->delay < mintime) {
769 mintime = slave->delay;
777 static bool bond_should_notify_peers(struct bonding *bond)
782 slave = rcu_dereference(bond->curr_active_slave);
785 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
786 slave ? slave->dev->name : "NULL");
788 if (!slave || !bond->send_peer_notif ||
789 !netif_carrier_ok(bond->dev) ||
790 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
797 * change_active_interface - change the active slave into the specified one
798 * @bond: our bonding struct
799 * @new: the new slave to make the active one
801 * Set the new slave to the bond's settings and unset them on the old
803 * Setting include flags, mc-list, promiscuity, allmulti, etc.
805 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
806 * because it is apparently the best available slave we have, even though its
807 * updelay hasn't timed out yet.
809 * Caller must hold RTNL.
811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
813 struct slave *old_active;
817 old_active = rtnl_dereference(bond->curr_active_slave);
819 if (old_active == new_active)
823 new_active->last_link_up = jiffies;
825 if (new_active->link == BOND_LINK_BACK) {
826 if (bond_uses_primary(bond)) {
827 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
828 new_active->dev->name,
829 (bond->params.updelay - new_active->delay) * bond->params.miimon);
832 new_active->delay = 0;
833 bond_set_slave_link_state(new_active, BOND_LINK_UP,
834 BOND_SLAVE_NOTIFY_NOW);
836 if (BOND_MODE(bond) == BOND_MODE_8023AD)
837 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
839 if (bond_is_lb(bond))
840 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
842 if (bond_uses_primary(bond)) {
843 netdev_info(bond->dev, "making interface %s the new active one\n",
844 new_active->dev->name);
849 if (bond_uses_primary(bond))
850 bond_hw_addr_swap(bond, new_active, old_active);
852 if (bond_is_lb(bond)) {
853 bond_alb_handle_active_change(bond, new_active);
855 bond_set_slave_inactive_flags(old_active,
856 BOND_SLAVE_NOTIFY_NOW);
858 bond_set_slave_active_flags(new_active,
859 BOND_SLAVE_NOTIFY_NOW);
861 rcu_assign_pointer(bond->curr_active_slave, new_active);
864 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
866 bond_set_slave_inactive_flags(old_active,
867 BOND_SLAVE_NOTIFY_NOW);
870 bool should_notify_peers = false;
872 bond_set_slave_active_flags(new_active,
873 BOND_SLAVE_NOTIFY_NOW);
875 if (bond->params.fail_over_mac)
876 bond_do_fail_over_mac(bond, new_active,
879 if (netif_running(bond->dev)) {
880 bond->send_peer_notif =
881 bond->params.num_peer_notif;
882 should_notify_peers =
883 bond_should_notify_peers(bond);
886 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
887 if (should_notify_peers)
888 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
893 /* resend IGMP joins since active slave has changed or
894 * all were sent on curr_active_slave.
895 * resend only if bond is brought up with the affected
896 * bonding modes and the retransmission is enabled
898 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
899 ((bond_uses_primary(bond) && new_active) ||
900 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
901 bond->igmp_retrans = bond->params.resend_igmp;
902 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
907 * bond_select_active_slave - select a new active slave, if needed
908 * @bond: our bonding struct
910 * This functions should be called when one of the following occurs:
911 * - The old curr_active_slave has been released or lost its link.
912 * - The primary_slave has got its link back.
913 * - A slave has got its link back and there's no old curr_active_slave.
915 * Caller must hold RTNL.
917 void bond_select_active_slave(struct bonding *bond)
919 struct slave *best_slave;
924 best_slave = bond_find_best_slave(bond);
925 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
926 bond_change_active_slave(bond, best_slave);
927 rv = bond_set_carrier(bond);
931 if (netif_carrier_ok(bond->dev))
932 netdev_info(bond->dev, "first active interface up!\n");
934 netdev_info(bond->dev, "now running without any active interface!\n");
938 #ifdef CONFIG_NET_POLL_CONTROLLER
939 static inline int slave_enable_netpoll(struct slave *slave)
944 np = kzalloc(sizeof(*np), GFP_KERNEL);
949 err = __netpoll_setup(np, slave->dev);
958 static inline void slave_disable_netpoll(struct slave *slave)
960 struct netpoll *np = slave->np;
970 static void bond_poll_controller(struct net_device *bond_dev)
972 struct bonding *bond = netdev_priv(bond_dev);
973 struct slave *slave = NULL;
974 struct list_head *iter;
975 struct ad_info ad_info;
977 if (BOND_MODE(bond) == BOND_MODE_8023AD)
978 if (bond_3ad_get_active_agg_info(bond, &ad_info))
981 bond_for_each_slave_rcu(bond, slave, iter) {
982 if (!bond_slave_is_up(slave))
985 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
986 struct aggregator *agg =
987 SLAVE_AD_INFO(slave)->port.aggregator;
990 agg->aggregator_identifier != ad_info.aggregator_id)
994 netpoll_poll_dev(slave->dev);
998 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1000 struct bonding *bond = netdev_priv(bond_dev);
1001 struct list_head *iter;
1002 struct slave *slave;
1004 bond_for_each_slave(bond, slave, iter)
1005 if (bond_slave_is_up(slave))
1006 slave_disable_netpoll(slave);
1009 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1011 struct bonding *bond = netdev_priv(dev);
1012 struct list_head *iter;
1013 struct slave *slave;
1016 bond_for_each_slave(bond, slave, iter) {
1017 err = slave_enable_netpoll(slave);
1019 bond_netpoll_cleanup(dev);
1026 static inline int slave_enable_netpoll(struct slave *slave)
1030 static inline void slave_disable_netpoll(struct slave *slave)
1033 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1038 /*---------------------------------- IOCTL ----------------------------------*/
1040 static netdev_features_t bond_fix_features(struct net_device *dev,
1041 netdev_features_t features)
1043 struct bonding *bond = netdev_priv(dev);
1044 struct list_head *iter;
1045 netdev_features_t mask;
1046 struct slave *slave;
1050 features &= ~NETIF_F_ONE_FOR_ALL;
1051 features |= NETIF_F_ALL_FOR_ALL;
1053 bond_for_each_slave(bond, slave, iter) {
1054 features = netdev_increment_features(features,
1055 slave->dev->features,
1058 features = netdev_add_tso_features(features, mask);
1063 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1064 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1065 NETIF_F_HIGHDMA | NETIF_F_LRO)
1067 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1068 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1070 static void bond_compute_features(struct bonding *bond)
1072 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1073 IFF_XMIT_DST_RELEASE_PERM;
1074 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1075 netdev_features_t enc_features = BOND_ENC_FEATURES;
1076 struct net_device *bond_dev = bond->dev;
1077 struct list_head *iter;
1078 struct slave *slave;
1079 unsigned short max_hard_header_len = ETH_HLEN;
1080 unsigned int gso_max_size = GSO_MAX_SIZE;
1081 u16 gso_max_segs = GSO_MAX_SEGS;
1083 if (!bond_has_slaves(bond))
1085 vlan_features &= NETIF_F_ALL_FOR_ALL;
1087 bond_for_each_slave(bond, slave, iter) {
1088 vlan_features = netdev_increment_features(vlan_features,
1089 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1091 enc_features = netdev_increment_features(enc_features,
1092 slave->dev->hw_enc_features,
1094 dst_release_flag &= slave->dev->priv_flags;
1095 if (slave->dev->hard_header_len > max_hard_header_len)
1096 max_hard_header_len = slave->dev->hard_header_len;
1098 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1099 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1101 bond_dev->hard_header_len = max_hard_header_len;
1104 bond_dev->vlan_features = vlan_features;
1105 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1107 bond_dev->gso_max_segs = gso_max_segs;
1108 netif_set_gso_max_size(bond_dev, gso_max_size);
1110 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1111 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1112 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1113 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1115 netdev_change_features(bond_dev);
1118 static void bond_setup_by_slave(struct net_device *bond_dev,
1119 struct net_device *slave_dev)
1121 bond_dev->header_ops = slave_dev->header_ops;
1123 bond_dev->type = slave_dev->type;
1124 bond_dev->hard_header_len = slave_dev->hard_header_len;
1125 bond_dev->addr_len = slave_dev->addr_len;
1127 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1128 slave_dev->addr_len);
1131 /* On bonding slaves other than the currently active slave, suppress
1132 * duplicates except for alb non-mcast/bcast.
1134 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1135 struct slave *slave,
1136 struct bonding *bond)
1138 if (bond_is_slave_inactive(slave)) {
1139 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1140 skb->pkt_type != PACKET_BROADCAST &&
1141 skb->pkt_type != PACKET_MULTICAST)
1148 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1150 struct sk_buff *skb = *pskb;
1151 struct slave *slave;
1152 struct bonding *bond;
1153 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1155 int ret = RX_HANDLER_ANOTHER;
1157 skb = skb_share_check(skb, GFP_ATOMIC);
1159 return RX_HANDLER_CONSUMED;
1163 slave = bond_slave_get_rcu(skb->dev);
1166 recv_probe = READ_ONCE(bond->recv_probe);
1168 ret = recv_probe(skb, bond, slave);
1169 if (ret == RX_HANDLER_CONSUMED) {
1175 /* Link-local multicast packets should be passed to the
1176 * stack on the link they arrive as well as pass them to the
1177 * bond-master device. These packets are mostly usable when
1178 * stack receives it with the link on which they arrive
1179 * (e.g. LLDP) they also must be available on master. Some of
1180 * the use cases include (but are not limited to): LLDP agents
1181 * that must be able to operate both on enslaved interfaces as
1182 * well as on bonds themselves; linux bridges that must be able
1183 * to process/pass BPDUs from attached bonds when any kind of
1184 * STP version is enabled on the network.
1186 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) {
1187 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1190 nskb->dev = bond->dev;
1191 nskb->queue_mapping = 0;
1194 return RX_HANDLER_PASS;
1196 if (bond_should_deliver_exact_match(skb, slave, bond))
1197 return RX_HANDLER_EXACT;
1199 skb->dev = bond->dev;
1201 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1202 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1203 skb->pkt_type == PACKET_HOST) {
1205 if (unlikely(skb_cow_head(skb,
1206 skb->data - skb_mac_header(skb)))) {
1208 return RX_HANDLER_CONSUMED;
1210 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1211 bond->dev->addr_len);
1217 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1219 switch (BOND_MODE(bond)) {
1220 case BOND_MODE_ROUNDROBIN:
1221 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1222 case BOND_MODE_ACTIVEBACKUP:
1223 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1224 case BOND_MODE_BROADCAST:
1225 return NETDEV_LAG_TX_TYPE_BROADCAST;
1227 case BOND_MODE_8023AD:
1228 return NETDEV_LAG_TX_TYPE_HASH;
1230 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1234 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1235 enum netdev_lag_tx_type type)
1237 if (type != NETDEV_LAG_TX_TYPE_HASH)
1238 return NETDEV_LAG_HASH_NONE;
1240 switch (bond->params.xmit_policy) {
1241 case BOND_XMIT_POLICY_LAYER2:
1242 return NETDEV_LAG_HASH_L2;
1243 case BOND_XMIT_POLICY_LAYER34:
1244 return NETDEV_LAG_HASH_L34;
1245 case BOND_XMIT_POLICY_LAYER23:
1246 return NETDEV_LAG_HASH_L23;
1247 case BOND_XMIT_POLICY_ENCAP23:
1248 return NETDEV_LAG_HASH_E23;
1249 case BOND_XMIT_POLICY_ENCAP34:
1250 return NETDEV_LAG_HASH_E34;
1252 return NETDEV_LAG_HASH_UNKNOWN;
1256 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1257 struct netlink_ext_ack *extack)
1259 struct netdev_lag_upper_info lag_upper_info;
1260 enum netdev_lag_tx_type type;
1262 type = bond_lag_tx_type(bond);
1263 lag_upper_info.tx_type = type;
1264 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1266 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1267 &lag_upper_info, extack);
1270 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1272 netdev_upper_dev_unlink(slave->dev, bond->dev);
1273 slave->dev->flags &= ~IFF_SLAVE;
1276 static struct slave *bond_alloc_slave(struct bonding *bond)
1278 struct slave *slave = NULL;
1280 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1284 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1285 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1287 if (!SLAVE_AD_INFO(slave)) {
1292 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1297 static void bond_free_slave(struct slave *slave)
1299 struct bonding *bond = bond_get_bond_by_slave(slave);
1301 cancel_delayed_work_sync(&slave->notify_work);
1302 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1303 kfree(SLAVE_AD_INFO(slave));
1308 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1310 info->bond_mode = BOND_MODE(bond);
1311 info->miimon = bond->params.miimon;
1312 info->num_slaves = bond->slave_cnt;
1315 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1317 strcpy(info->slave_name, slave->dev->name);
1318 info->link = slave->link;
1319 info->state = bond_slave_state(slave);
1320 info->link_failure_count = slave->link_failure_count;
1323 static void bond_netdev_notify_work(struct work_struct *_work)
1325 struct slave *slave = container_of(_work, struct slave,
1328 if (rtnl_trylock()) {
1329 struct netdev_bonding_info binfo;
1331 bond_fill_ifslave(slave, &binfo.slave);
1332 bond_fill_ifbond(slave->bond, &binfo.master);
1333 netdev_bonding_info_change(slave->dev, &binfo);
1336 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1340 void bond_queue_slave_event(struct slave *slave)
1342 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1345 void bond_lower_state_changed(struct slave *slave)
1347 struct netdev_lag_lower_state_info info;
1349 info.link_up = slave->link == BOND_LINK_UP ||
1350 slave->link == BOND_LINK_FAIL;
1351 info.tx_enabled = bond_is_active_slave(slave);
1352 netdev_lower_state_changed(slave->dev, &info);
1355 /* enslave device <slave> to bond device <master> */
1356 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1357 struct netlink_ext_ack *extack)
1359 struct bonding *bond = netdev_priv(bond_dev);
1360 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1361 struct slave *new_slave = NULL, *prev_slave;
1362 struct sockaddr_storage ss;
1366 if (!bond->params.use_carrier &&
1367 slave_dev->ethtool_ops->get_link == NULL &&
1368 slave_ops->ndo_do_ioctl == NULL) {
1369 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1373 /* already in-use? */
1374 if (netdev_is_rx_handler_busy(slave_dev)) {
1375 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1376 netdev_err(bond_dev,
1377 "Error: Device is in use and cannot be enslaved\n");
1381 if (bond_dev == slave_dev) {
1382 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1383 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1387 /* vlan challenged mutual exclusion */
1388 /* no need to lock since we're protected by rtnl_lock */
1389 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1390 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1392 if (vlan_uses_dev(bond_dev)) {
1393 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1394 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1395 slave_dev->name, bond_dev->name);
1398 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1399 slave_dev->name, slave_dev->name,
1403 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1407 /* Old ifenslave binaries are no longer supported. These can
1408 * be identified with moderate accuracy by the state of the slave:
1409 * the current ifenslave will set the interface down prior to
1410 * enslaving it; the old ifenslave will not.
1412 if (slave_dev->flags & IFF_UP) {
1413 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1414 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1419 /* set bonding device ether type by slave - bonding netdevices are
1420 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1421 * there is a need to override some of the type dependent attribs/funcs.
1423 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1424 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1426 if (!bond_has_slaves(bond)) {
1427 if (bond_dev->type != slave_dev->type) {
1428 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1429 bond_dev->type, slave_dev->type);
1431 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1433 res = notifier_to_errno(res);
1435 netdev_err(bond_dev, "refused to change device type\n");
1439 /* Flush unicast and multicast addresses */
1440 dev_uc_flush(bond_dev);
1441 dev_mc_flush(bond_dev);
1443 if (slave_dev->type != ARPHRD_ETHER)
1444 bond_setup_by_slave(bond_dev, slave_dev);
1446 ether_setup(bond_dev);
1447 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1450 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1453 } else if (bond_dev->type != slave_dev->type) {
1454 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1455 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1456 slave_dev->name, slave_dev->type, bond_dev->type);
1460 if (slave_dev->type == ARPHRD_INFINIBAND &&
1461 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1462 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1463 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1466 goto err_undo_flags;
1469 if (!slave_ops->ndo_set_mac_address ||
1470 slave_dev->type == ARPHRD_INFINIBAND) {
1471 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1472 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1473 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1474 if (!bond_has_slaves(bond)) {
1475 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1476 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1478 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1479 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");
1481 goto err_undo_flags;
1486 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1488 /* If this is the first slave, then we need to set the master's hardware
1489 * address to be the same as the slave's.
1491 if (!bond_has_slaves(bond) &&
1492 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1493 bond_set_dev_addr(bond->dev, slave_dev);
1495 new_slave = bond_alloc_slave(bond);
1498 goto err_undo_flags;
1501 new_slave->bond = bond;
1502 new_slave->dev = slave_dev;
1503 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1504 * is set via sysfs or module option if desired.
1506 new_slave->queue_id = 0;
1508 /* Save slave's original mtu and then set it to match the bond */
1509 new_slave->original_mtu = slave_dev->mtu;
1510 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1512 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1516 /* Save slave's original ("permanent") mac address for modes
1517 * that need it, and for restoring it upon release, and then
1518 * set it to the master's address
1520 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1521 slave_dev->addr_len);
1523 if (!bond->params.fail_over_mac ||
1524 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1525 /* Set slave to master's mac address. The application already
1526 * set the master's mac address to that of the first slave
1528 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1529 ss.ss_family = slave_dev->type;
1530 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1532 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1533 goto err_restore_mtu;
1537 /* set slave flag before open to prevent IPv6 addrconf */
1538 slave_dev->flags |= IFF_SLAVE;
1540 /* open the slave since the application closed it */
1541 res = dev_open(slave_dev);
1543 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1544 goto err_restore_mac;
1547 slave_dev->priv_flags |= IFF_BONDING;
1548 /* initialize slave stats */
1549 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1551 if (bond_is_lb(bond)) {
1552 /* bond_alb_init_slave() must be called before all other stages since
1553 * it might fail and we do not want to have to undo everything
1555 res = bond_alb_init_slave(bond, new_slave);
1560 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1562 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1567 prev_slave = bond_last_slave(bond);
1569 new_slave->delay = 0;
1570 new_slave->link_failure_count = 0;
1572 if (bond_update_speed_duplex(new_slave) &&
1573 bond_needs_speed_duplex(bond))
1574 new_slave->link = BOND_LINK_DOWN;
1576 new_slave->last_rx = jiffies -
1577 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1578 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1579 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1581 if (bond->params.miimon && !bond->params.use_carrier) {
1582 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1584 if ((link_reporting == -1) && !bond->params.arp_interval) {
1585 /* miimon is set but a bonded network driver
1586 * does not support ETHTOOL/MII and
1587 * arp_interval is not set. Note: if
1588 * use_carrier is enabled, we will never go
1589 * here (because netif_carrier is always
1590 * supported); thus, we don't need to change
1591 * the messages for netif_carrier.
1593 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",
1595 } else if (link_reporting == -1) {
1596 /* unable get link status using mii/ethtool */
1597 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",
1602 /* check for initial state */
1603 new_slave->link = BOND_LINK_NOCHANGE;
1604 if (bond->params.miimon) {
1605 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1606 if (bond->params.updelay) {
1607 bond_set_slave_link_state(new_slave,
1609 BOND_SLAVE_NOTIFY_NOW);
1610 new_slave->delay = bond->params.updelay;
1612 bond_set_slave_link_state(new_slave,
1614 BOND_SLAVE_NOTIFY_NOW);
1617 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1618 BOND_SLAVE_NOTIFY_NOW);
1620 } else if (bond->params.arp_interval) {
1621 bond_set_slave_link_state(new_slave,
1622 (netif_carrier_ok(slave_dev) ?
1623 BOND_LINK_UP : BOND_LINK_DOWN),
1624 BOND_SLAVE_NOTIFY_NOW);
1626 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1627 BOND_SLAVE_NOTIFY_NOW);
1630 if (new_slave->link != BOND_LINK_DOWN)
1631 new_slave->last_link_up = jiffies;
1632 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1633 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1634 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1636 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1637 /* if there is a primary slave, remember it */
1638 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1639 rcu_assign_pointer(bond->primary_slave, new_slave);
1640 bond->force_primary = true;
1644 switch (BOND_MODE(bond)) {
1645 case BOND_MODE_ACTIVEBACKUP:
1646 bond_set_slave_inactive_flags(new_slave,
1647 BOND_SLAVE_NOTIFY_NOW);
1649 case BOND_MODE_8023AD:
1650 /* in 802.3ad mode, the internal mechanism
1651 * will activate the slaves in the selected
1654 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1655 /* if this is the first slave */
1657 SLAVE_AD_INFO(new_slave)->id = 1;
1658 /* Initialize AD with the number of times that the AD timer is called in 1 second
1659 * can be called only after the mac address of the bond is set
1661 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1663 SLAVE_AD_INFO(new_slave)->id =
1664 SLAVE_AD_INFO(prev_slave)->id + 1;
1667 bond_3ad_bind_slave(new_slave);
1671 bond_set_active_slave(new_slave);
1672 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1675 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1677 /* always active in trunk mode */
1678 bond_set_active_slave(new_slave);
1680 /* In trunking mode there is little meaning to curr_active_slave
1681 * anyway (it holds no special properties of the bond device),
1682 * so we can change it without calling change_active_interface()
1684 if (!rcu_access_pointer(bond->curr_active_slave) &&
1685 new_slave->link == BOND_LINK_UP)
1686 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1689 } /* switch(bond_mode) */
1691 #ifdef CONFIG_NET_POLL_CONTROLLER
1692 if (bond->dev->npinfo) {
1693 if (slave_enable_netpoll(new_slave)) {
1694 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1701 if (!(bond_dev->features & NETIF_F_LRO))
1702 dev_disable_lro(slave_dev);
1704 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1707 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1711 res = bond_master_upper_dev_link(bond, new_slave, extack);
1713 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1714 goto err_unregister;
1717 res = bond_sysfs_slave_add(new_slave);
1719 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1720 goto err_upper_unlink;
1723 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1725 /* If the mode uses primary, then the following is handled by
1726 * bond_change_active_slave().
1728 if (!bond_uses_primary(bond)) {
1729 /* set promiscuity level to new slave */
1730 if (bond_dev->flags & IFF_PROMISC) {
1731 res = dev_set_promiscuity(slave_dev, 1);
1736 /* set allmulti level to new slave */
1737 if (bond_dev->flags & IFF_ALLMULTI) {
1738 res = dev_set_allmulti(slave_dev, 1);
1740 if (bond_dev->flags & IFF_PROMISC)
1741 dev_set_promiscuity(slave_dev, -1);
1746 netif_addr_lock_bh(bond_dev);
1747 dev_mc_sync_multiple(slave_dev, bond_dev);
1748 dev_uc_sync_multiple(slave_dev, bond_dev);
1749 netif_addr_unlock_bh(bond_dev);
1751 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1752 /* add lacpdu mc addr to mc list */
1753 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1755 dev_mc_add(slave_dev, lacpdu_multicast);
1760 bond_compute_features(bond);
1761 bond_set_carrier(bond);
1763 if (bond_uses_primary(bond)) {
1765 bond_select_active_slave(bond);
1766 unblock_netpoll_tx();
1769 if (bond_mode_can_use_xmit_hash(bond))
1770 bond_update_slave_arr(bond, NULL);
1773 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1775 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1776 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1778 /* enslave is successful */
1779 bond_queue_slave_event(new_slave);
1782 /* Undo stages on error */
1784 bond_sysfs_slave_del(new_slave);
1787 bond_upper_dev_unlink(bond, new_slave);
1790 netdev_rx_handler_unregister(slave_dev);
1793 vlan_vids_del_by_dev(slave_dev, bond_dev);
1794 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1795 RCU_INIT_POINTER(bond->primary_slave, NULL);
1796 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1798 bond_change_active_slave(bond, NULL);
1799 bond_select_active_slave(bond);
1800 unblock_netpoll_tx();
1802 /* either primary_slave or curr_active_slave might've changed */
1804 slave_disable_netpoll(new_slave);
1807 slave_dev->priv_flags &= ~IFF_BONDING;
1808 dev_close(slave_dev);
1811 slave_dev->flags &= ~IFF_SLAVE;
1812 if (!bond->params.fail_over_mac ||
1813 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1814 /* XXX TODO - fom follow mode needs to change master's
1815 * MAC if this slave's MAC is in use by the bond, or at
1816 * least print a warning.
1818 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1819 new_slave->dev->addr_len);
1820 ss.ss_family = slave_dev->type;
1821 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1825 dev_set_mtu(slave_dev, new_slave->original_mtu);
1828 bond_free_slave(new_slave);
1831 /* Enslave of first slave has failed and we need to fix master's mac */
1832 if (!bond_has_slaves(bond)) {
1833 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1834 slave_dev->dev_addr))
1835 eth_hw_addr_random(bond_dev);
1836 if (bond_dev->type != ARPHRD_ETHER) {
1837 dev_close(bond_dev);
1838 ether_setup(bond_dev);
1839 bond_dev->flags |= IFF_MASTER;
1840 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1847 /* Try to release the slave device <slave> from the bond device <master>
1848 * It is legal to access curr_active_slave without a lock because all the function
1849 * is RTNL-locked. If "all" is true it means that the function is being called
1850 * while destroying a bond interface and all slaves are being released.
1852 * The rules for slave state should be:
1853 * for Active/Backup:
1854 * Active stays on all backups go down
1855 * for Bonded connections:
1856 * The first up interface should be left on and all others downed.
1858 static int __bond_release_one(struct net_device *bond_dev,
1859 struct net_device *slave_dev,
1860 bool all, bool unregister)
1862 struct bonding *bond = netdev_priv(bond_dev);
1863 struct slave *slave, *oldcurrent;
1864 struct sockaddr_storage ss;
1865 int old_flags = bond_dev->flags;
1866 netdev_features_t old_features = bond_dev->features;
1868 /* slave is not a slave or master is not master of this slave */
1869 if (!(slave_dev->flags & IFF_SLAVE) ||
1870 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1871 netdev_dbg(bond_dev, "cannot release %s\n",
1878 slave = bond_get_slave_by_dev(bond, slave_dev);
1880 /* not a slave of this bond */
1881 netdev_info(bond_dev, "%s not enslaved\n",
1883 unblock_netpoll_tx();
1887 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1889 bond_sysfs_slave_del(slave);
1891 /* recompute stats just before removing the slave */
1892 bond_get_stats(bond->dev, &bond->bond_stats);
1894 bond_upper_dev_unlink(bond, slave);
1895 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1896 * for this slave anymore.
1898 netdev_rx_handler_unregister(slave_dev);
1900 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1901 bond_3ad_unbind_slave(slave);
1903 if (bond_mode_can_use_xmit_hash(bond))
1904 bond_update_slave_arr(bond, slave);
1906 netdev_info(bond_dev, "Releasing %s interface %s\n",
1907 bond_is_active_slave(slave) ? "active" : "backup",
1910 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1912 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1914 if (!all && (!bond->params.fail_over_mac ||
1915 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1916 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1917 bond_has_slaves(bond))
1918 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",
1919 slave_dev->name, slave->perm_hwaddr,
1920 bond_dev->name, slave_dev->name);
1923 if (rtnl_dereference(bond->primary_slave) == slave)
1924 RCU_INIT_POINTER(bond->primary_slave, NULL);
1926 if (oldcurrent == slave)
1927 bond_change_active_slave(bond, NULL);
1929 if (bond_is_lb(bond)) {
1930 /* Must be called only after the slave has been
1931 * detached from the list and the curr_active_slave
1932 * has been cleared (if our_slave == old_current),
1933 * but before a new active slave is selected.
1935 bond_alb_deinit_slave(bond, slave);
1939 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1940 } else if (oldcurrent == slave) {
1941 /* Note that we hold RTNL over this sequence, so there
1942 * is no concern that another slave add/remove event
1945 bond_select_active_slave(bond);
1948 if (!bond_has_slaves(bond)) {
1949 bond_set_carrier(bond);
1950 eth_hw_addr_random(bond_dev);
1953 unblock_netpoll_tx();
1957 if (!bond_has_slaves(bond)) {
1958 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1959 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1962 bond_compute_features(bond);
1963 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1964 (old_features & NETIF_F_VLAN_CHALLENGED))
1965 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1966 slave_dev->name, bond_dev->name);
1968 vlan_vids_del_by_dev(slave_dev, bond_dev);
1970 /* If the mode uses primary, then this case was handled above by
1971 * bond_change_active_slave(..., NULL)
1973 if (!bond_uses_primary(bond)) {
1974 /* unset promiscuity level from slave
1975 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1976 * of the IFF_PROMISC flag in the bond_dev, but we need the
1977 * value of that flag before that change, as that was the value
1978 * when this slave was attached, so we cache at the start of the
1979 * function and use it here. Same goes for ALLMULTI below
1981 if (old_flags & IFF_PROMISC)
1982 dev_set_promiscuity(slave_dev, -1);
1984 /* unset allmulti level from slave */
1985 if (old_flags & IFF_ALLMULTI)
1986 dev_set_allmulti(slave_dev, -1);
1988 bond_hw_addr_flush(bond_dev, slave_dev);
1991 slave_disable_netpoll(slave);
1993 /* close slave before restoring its mac address */
1994 dev_close(slave_dev);
1996 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1997 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1998 /* restore original ("permanent") mac address */
1999 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2000 slave->dev->addr_len);
2001 ss.ss_family = slave_dev->type;
2002 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
2006 __dev_set_mtu(slave_dev, slave->original_mtu);
2008 dev_set_mtu(slave_dev, slave->original_mtu);
2010 slave_dev->priv_flags &= ~IFF_BONDING;
2012 bond_free_slave(slave);
2017 /* A wrapper used because of ndo_del_link */
2018 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2020 return __bond_release_one(bond_dev, slave_dev, false, false);
2023 /* First release a slave and then destroy the bond if no more slaves are left.
2024 * Must be under rtnl_lock when this function is called.
2026 static int bond_release_and_destroy(struct net_device *bond_dev,
2027 struct net_device *slave_dev)
2029 struct bonding *bond = netdev_priv(bond_dev);
2032 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2033 if (ret == 0 && !bond_has_slaves(bond)) {
2034 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2035 netdev_info(bond_dev, "Destroying bond %s\n",
2037 bond_remove_proc_entry(bond);
2038 unregister_netdevice(bond_dev);
2043 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2045 struct bonding *bond = netdev_priv(bond_dev);
2046 bond_fill_ifbond(bond, info);
2049 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2051 struct bonding *bond = netdev_priv(bond_dev);
2052 struct list_head *iter;
2053 int i = 0, res = -ENODEV;
2054 struct slave *slave;
2056 bond_for_each_slave(bond, slave, iter) {
2057 if (i++ == (int)info->slave_id) {
2059 bond_fill_ifslave(slave, info);
2067 /*-------------------------------- Monitoring -------------------------------*/
2069 /* called with rcu_read_lock() */
2070 static int bond_miimon_inspect(struct bonding *bond)
2072 int link_state, commit = 0;
2073 struct list_head *iter;
2074 struct slave *slave;
2075 bool ignore_updelay;
2077 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2079 bond_for_each_slave_rcu(bond, slave, iter) {
2080 slave->new_link = BOND_LINK_NOCHANGE;
2081 slave->link_new_state = slave->link;
2083 link_state = bond_check_dev_link(bond, slave->dev, 0);
2085 switch (slave->link) {
2090 bond_propose_link_state(slave, BOND_LINK_FAIL);
2092 slave->delay = bond->params.downdelay;
2094 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2096 BOND_MODE_ACTIVEBACKUP) ?
2097 (bond_is_active_slave(slave) ?
2098 "active " : "backup ") : "",
2100 bond->params.downdelay * bond->params.miimon);
2103 case BOND_LINK_FAIL:
2105 /* recovered before downdelay expired */
2106 bond_propose_link_state(slave, BOND_LINK_UP);
2107 slave->last_link_up = jiffies;
2108 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2109 (bond->params.downdelay - slave->delay) *
2110 bond->params.miimon,
2116 if (slave->delay <= 0) {
2117 slave->new_link = BOND_LINK_DOWN;
2125 case BOND_LINK_DOWN:
2129 bond_propose_link_state(slave, BOND_LINK_BACK);
2131 slave->delay = bond->params.updelay;
2134 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2136 ignore_updelay ? 0 :
2137 bond->params.updelay *
2138 bond->params.miimon);
2141 case BOND_LINK_BACK:
2143 bond_propose_link_state(slave, BOND_LINK_DOWN);
2144 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2145 (bond->params.updelay - slave->delay) *
2146 bond->params.miimon,
2155 if (slave->delay <= 0) {
2156 slave->new_link = BOND_LINK_UP;
2158 ignore_updelay = false;
2170 static void bond_miimon_link_change(struct bonding *bond,
2171 struct slave *slave,
2174 switch (BOND_MODE(bond)) {
2175 case BOND_MODE_8023AD:
2176 bond_3ad_handle_link_change(slave, link);
2180 bond_alb_handle_link_change(bond, slave, link);
2183 bond_update_slave_arr(bond, NULL);
2188 static void bond_miimon_commit(struct bonding *bond)
2190 struct list_head *iter;
2191 struct slave *slave, *primary;
2193 bond_for_each_slave(bond, slave, iter) {
2194 switch (slave->new_link) {
2195 case BOND_LINK_NOCHANGE:
2199 if (bond_update_speed_duplex(slave) &&
2200 bond_needs_speed_duplex(bond)) {
2201 slave->link = BOND_LINK_DOWN;
2202 if (net_ratelimit())
2203 netdev_warn(bond->dev,
2204 "failed to get link speed/duplex for %s\n",
2208 bond_set_slave_link_state(slave, BOND_LINK_UP,
2209 BOND_SLAVE_NOTIFY_NOW);
2210 slave->last_link_up = jiffies;
2212 primary = rtnl_dereference(bond->primary_slave);
2213 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2214 /* prevent it from being the active one */
2215 bond_set_backup_slave(slave);
2216 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2217 /* make it immediately active */
2218 bond_set_active_slave(slave);
2219 } else if (slave != primary) {
2220 /* prevent it from being the active one */
2221 bond_set_backup_slave(slave);
2224 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2226 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2227 slave->duplex ? "full" : "half");
2229 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2231 if (!bond->curr_active_slave || slave == primary)
2236 case BOND_LINK_DOWN:
2237 if (slave->link_failure_count < UINT_MAX)
2238 slave->link_failure_count++;
2240 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2241 BOND_SLAVE_NOTIFY_NOW);
2243 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2244 BOND_MODE(bond) == BOND_MODE_8023AD)
2245 bond_set_slave_inactive_flags(slave,
2246 BOND_SLAVE_NOTIFY_NOW);
2248 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2251 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2253 if (slave == rcu_access_pointer(bond->curr_active_slave))
2259 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2260 slave->new_link, slave->dev->name);
2261 slave->new_link = BOND_LINK_NOCHANGE;
2268 bond_select_active_slave(bond);
2269 unblock_netpoll_tx();
2272 bond_set_carrier(bond);
2277 * Really a wrapper that splits the mii monitor into two phases: an
2278 * inspection, then (if inspection indicates something needs to be done)
2279 * an acquisition of appropriate locks followed by a commit phase to
2280 * implement whatever link state changes are indicated.
2282 static void bond_mii_monitor(struct work_struct *work)
2284 struct bonding *bond = container_of(work, struct bonding,
2286 bool should_notify_peers = false;
2287 unsigned long delay;
2288 struct slave *slave;
2289 struct list_head *iter;
2291 delay = msecs_to_jiffies(bond->params.miimon);
2293 if (!bond_has_slaves(bond))
2298 should_notify_peers = bond_should_notify_peers(bond);
2300 if (bond_miimon_inspect(bond)) {
2303 /* Race avoidance with bond_close cancel of workqueue */
2304 if (!rtnl_trylock()) {
2306 should_notify_peers = false;
2310 bond_for_each_slave(bond, slave, iter) {
2311 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2313 bond_miimon_commit(bond);
2315 rtnl_unlock(); /* might sleep, hold no other locks */
2320 if (bond->params.miimon)
2321 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2323 if (should_notify_peers) {
2324 if (!rtnl_trylock())
2326 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2331 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2333 __be32 ip = *((__be32 *)data);
2335 return ip == bond_confirm_addr(upper, 0, ip);
2338 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2342 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2346 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2353 /* We go to the (large) trouble of VLAN tagging ARP frames because
2354 * switches in VLAN mode (especially if ports are configured as
2355 * "native" to a VLAN) might not pass non-tagged frames.
2357 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2358 __be32 dest_ip, __be32 src_ip,
2359 struct bond_vlan_tag *tags)
2361 struct sk_buff *skb;
2362 struct bond_vlan_tag *outer_tag = tags;
2364 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2365 arp_op, slave_dev->name, &dest_ip, &src_ip);
2367 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2368 NULL, slave_dev->dev_addr, NULL);
2371 net_err_ratelimited("ARP packet allocation failed\n");
2375 if (!tags || tags->vlan_proto == VLAN_N_VID)
2380 /* Go through all the tags backwards and add them to the packet */
2381 while (tags->vlan_proto != VLAN_N_VID) {
2382 if (!tags->vlan_id) {
2387 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2388 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2389 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2392 net_err_ratelimited("failed to insert inner VLAN tag\n");
2398 /* Set the outer tag */
2399 if (outer_tag->vlan_id) {
2400 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2401 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2402 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2403 outer_tag->vlan_id);
2410 /* Validate the device path between the @start_dev and the @end_dev.
2411 * The path is valid if the @end_dev is reachable through device
2413 * When the path is validated, collect any vlan information in the
2416 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2417 struct net_device *end_dev,
2420 struct bond_vlan_tag *tags;
2421 struct net_device *upper;
2422 struct list_head *iter;
2424 if (start_dev == end_dev) {
2425 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2427 return ERR_PTR(-ENOMEM);
2428 tags[level].vlan_proto = VLAN_N_VID;
2432 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2433 tags = bond_verify_device_path(upper, end_dev, level + 1);
2434 if (IS_ERR_OR_NULL(tags)) {
2439 if (is_vlan_dev(upper)) {
2440 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2441 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2450 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2453 struct bond_vlan_tag *tags;
2454 __be32 *targets = bond->params.arp_targets, addr;
2457 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2458 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2461 /* Find out through which dev should the packet go */
2462 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2465 /* there's no route to target - try to send arp
2466 * probe to generate any traffic (arp_validate=0)
2468 if (bond->params.arp_validate)
2469 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2472 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2477 /* bond device itself */
2478 if (rt->dst.dev == bond->dev)
2482 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2485 if (!IS_ERR_OR_NULL(tags))
2488 /* Not our device - skip */
2489 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2490 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2496 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2498 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2504 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2508 if (!sip || !bond_has_this_ip(bond, tip)) {
2509 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2514 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2516 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2520 slave->last_rx = jiffies;
2521 slave->target_last_arp_rx[i] = jiffies;
2524 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2525 struct slave *slave)
2527 struct arphdr *arp = (struct arphdr *)skb->data;
2528 struct slave *curr_active_slave, *curr_arp_slave;
2529 unsigned char *arp_ptr;
2531 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2534 if (!slave_do_arp_validate(bond, slave)) {
2535 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2536 !slave_do_arp_validate_only(bond))
2537 slave->last_rx = jiffies;
2538 return RX_HANDLER_ANOTHER;
2539 } else if (!is_arp) {
2540 return RX_HANDLER_ANOTHER;
2543 alen = arp_hdr_len(bond->dev);
2545 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2548 if (alen > skb_headlen(skb)) {
2549 arp = kmalloc(alen, GFP_ATOMIC);
2552 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2556 if (arp->ar_hln != bond->dev->addr_len ||
2557 skb->pkt_type == PACKET_OTHERHOST ||
2558 skb->pkt_type == PACKET_LOOPBACK ||
2559 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2560 arp->ar_pro != htons(ETH_P_IP) ||
2564 arp_ptr = (unsigned char *)(arp + 1);
2565 arp_ptr += bond->dev->addr_len;
2566 memcpy(&sip, arp_ptr, 4);
2567 arp_ptr += 4 + bond->dev->addr_len;
2568 memcpy(&tip, arp_ptr, 4);
2570 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2571 slave->dev->name, bond_slave_state(slave),
2572 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2575 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2576 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2578 /* We 'trust' the received ARP enough to validate it if:
2580 * (a) the slave receiving the ARP is active (which includes the
2581 * current ARP slave, if any), or
2583 * (b) the receiving slave isn't active, but there is a currently
2584 * active slave and it received valid arp reply(s) after it became
2585 * the currently active slave, or
2587 * (c) there is an ARP slave that sent an ARP during the prior ARP
2588 * interval, and we receive an ARP reply on any slave. We accept
2589 * these because switch FDB update delays may deliver the ARP
2590 * reply to a slave other than the sender of the ARP request.
2592 * Note: for (b), backup slaves are receiving the broadcast ARP
2593 * request, not a reply. This request passes from the sending
2594 * slave through the L2 switch(es) to the receiving slave. Since
2595 * this is checking the request, sip/tip are swapped for
2598 * This is done to avoid endless looping when we can't reach the
2599 * arp_ip_target and fool ourselves with our own arp requests.
2601 if (bond_is_active_slave(slave))
2602 bond_validate_arp(bond, slave, sip, tip);
2603 else if (curr_active_slave &&
2604 time_after(slave_last_rx(bond, curr_active_slave),
2605 curr_active_slave->last_link_up))
2606 bond_validate_arp(bond, slave, tip, sip);
2607 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2608 bond_time_in_interval(bond,
2609 dev_trans_start(curr_arp_slave->dev), 1))
2610 bond_validate_arp(bond, slave, sip, tip);
2613 if (arp != (struct arphdr *)skb->data)
2615 return RX_HANDLER_ANOTHER;
2618 /* function to verify if we're in the arp_interval timeslice, returns true if
2619 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2620 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2622 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2625 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2627 return time_in_range(jiffies,
2628 last_act - delta_in_ticks,
2629 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2632 /* This function is called regularly to monitor each slave's link
2633 * ensuring that traffic is being sent and received when arp monitoring
2634 * is used in load-balancing mode. if the adapter has been dormant, then an
2635 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2636 * arp monitoring in active backup mode.
2638 static void bond_loadbalance_arp_mon(struct bonding *bond)
2640 struct slave *slave, *oldcurrent;
2641 struct list_head *iter;
2642 int do_failover = 0, slave_state_changed = 0;
2644 if (!bond_has_slaves(bond))
2649 oldcurrent = rcu_dereference(bond->curr_active_slave);
2650 /* see if any of the previous devices are up now (i.e. they have
2651 * xmt and rcv traffic). the curr_active_slave does not come into
2652 * the picture unless it is null. also, slave->last_link_up is not
2653 * needed here because we send an arp on each slave and give a slave
2654 * as long as it needs to get the tx/rx within the delta.
2655 * TODO: what about up/down delay in arp mode? it wasn't here before
2658 bond_for_each_slave_rcu(bond, slave, iter) {
2659 unsigned long trans_start = dev_trans_start(slave->dev);
2661 slave->new_link = BOND_LINK_NOCHANGE;
2663 if (slave->link != BOND_LINK_UP) {
2664 if (bond_time_in_interval(bond, trans_start, 1) &&
2665 bond_time_in_interval(bond, slave->last_rx, 1)) {
2667 slave->new_link = BOND_LINK_UP;
2668 slave_state_changed = 1;
2670 /* primary_slave has no meaning in round-robin
2671 * mode. the window of a slave being up and
2672 * curr_active_slave being null after enslaving
2676 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2680 netdev_info(bond->dev, "interface %s is now up\n",
2685 /* slave->link == BOND_LINK_UP */
2687 /* not all switches will respond to an arp request
2688 * when the source ip is 0, so don't take the link down
2689 * if we don't know our ip yet
2691 if (!bond_time_in_interval(bond, trans_start, 2) ||
2692 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2694 slave->new_link = BOND_LINK_DOWN;
2695 slave_state_changed = 1;
2697 if (slave->link_failure_count < UINT_MAX)
2698 slave->link_failure_count++;
2700 netdev_info(bond->dev, "interface %s is now down\n",
2703 if (slave == oldcurrent)
2708 /* note: if switch is in round-robin mode, all links
2709 * must tx arp to ensure all links rx an arp - otherwise
2710 * links may oscillate or not come up at all; if switch is
2711 * in something like xor mode, there is nothing we can
2712 * do - all replies will be rx'ed on same link causing slaves
2713 * to be unstable during low/no traffic periods
2715 if (bond_slave_is_up(slave))
2716 bond_arp_send_all(bond, slave);
2721 if (do_failover || slave_state_changed) {
2722 if (!rtnl_trylock())
2725 bond_for_each_slave(bond, slave, iter) {
2726 if (slave->new_link != BOND_LINK_NOCHANGE)
2727 slave->link = slave->new_link;
2730 if (slave_state_changed) {
2731 bond_slave_state_change(bond);
2732 if (BOND_MODE(bond) == BOND_MODE_XOR)
2733 bond_update_slave_arr(bond, NULL);
2737 bond_select_active_slave(bond);
2738 unblock_netpoll_tx();
2744 if (bond->params.arp_interval)
2745 queue_delayed_work(bond->wq, &bond->arp_work,
2746 msecs_to_jiffies(bond->params.arp_interval));
2749 /* Called to inspect slaves for active-backup mode ARP monitor link state
2750 * changes. Sets new_link in slaves to specify what action should take
2751 * place for the slave. Returns 0 if no changes are found, >0 if changes
2752 * to link states must be committed.
2754 * Called with rcu_read_lock held.
2756 static int bond_ab_arp_inspect(struct bonding *bond)
2758 unsigned long trans_start, last_rx;
2759 struct list_head *iter;
2760 struct slave *slave;
2763 bond_for_each_slave_rcu(bond, slave, iter) {
2764 slave->new_link = BOND_LINK_NOCHANGE;
2765 last_rx = slave_last_rx(bond, slave);
2767 if (slave->link != BOND_LINK_UP) {
2768 if (bond_time_in_interval(bond, last_rx, 1)) {
2769 slave->new_link = BOND_LINK_UP;
2775 /* Give slaves 2*delta after being enslaved or made
2776 * active. This avoids bouncing, as the last receive
2777 * times need a full ARP monitor cycle to be updated.
2779 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2782 /* Backup slave is down if:
2783 * - No current_arp_slave AND
2784 * - more than 3*delta since last receive AND
2785 * - the bond has an IP address
2787 * Note: a non-null current_arp_slave indicates
2788 * the curr_active_slave went down and we are
2789 * searching for a new one; under this condition
2790 * we only take the curr_active_slave down - this
2791 * gives each slave a chance to tx/rx traffic
2792 * before being taken out
2794 if (!bond_is_active_slave(slave) &&
2795 !rcu_access_pointer(bond->current_arp_slave) &&
2796 !bond_time_in_interval(bond, last_rx, 3)) {
2797 slave->new_link = BOND_LINK_DOWN;
2801 /* Active slave is down if:
2802 * - more than 2*delta since transmitting OR
2803 * - (more than 2*delta since receive AND
2804 * the bond has an IP address)
2806 trans_start = dev_trans_start(slave->dev);
2807 if (bond_is_active_slave(slave) &&
2808 (!bond_time_in_interval(bond, trans_start, 2) ||
2809 !bond_time_in_interval(bond, last_rx, 2))) {
2810 slave->new_link = BOND_LINK_DOWN;
2818 /* Called to commit link state changes noted by inspection step of
2819 * active-backup mode ARP monitor.
2821 * Called with RTNL hold.
2823 static void bond_ab_arp_commit(struct bonding *bond)
2825 unsigned long trans_start;
2826 struct list_head *iter;
2827 struct slave *slave;
2829 bond_for_each_slave(bond, slave, iter) {
2830 switch (slave->new_link) {
2831 case BOND_LINK_NOCHANGE:
2835 trans_start = dev_trans_start(slave->dev);
2836 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2837 (!rtnl_dereference(bond->curr_active_slave) &&
2838 bond_time_in_interval(bond, trans_start, 1))) {
2839 struct slave *current_arp_slave;
2841 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2842 bond_set_slave_link_state(slave, BOND_LINK_UP,
2843 BOND_SLAVE_NOTIFY_NOW);
2844 if (current_arp_slave) {
2845 bond_set_slave_inactive_flags(
2847 BOND_SLAVE_NOTIFY_NOW);
2848 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2851 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2854 if (!rtnl_dereference(bond->curr_active_slave) ||
2855 slave == rtnl_dereference(bond->primary_slave))
2862 case BOND_LINK_DOWN:
2863 if (slave->link_failure_count < UINT_MAX)
2864 slave->link_failure_count++;
2866 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2867 BOND_SLAVE_NOTIFY_NOW);
2868 bond_set_slave_inactive_flags(slave,
2869 BOND_SLAVE_NOTIFY_NOW);
2871 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2874 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2875 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2882 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2883 slave->new_link, slave->dev->name);
2889 bond_select_active_slave(bond);
2890 unblock_netpoll_tx();
2893 bond_set_carrier(bond);
2896 /* Send ARP probes for active-backup mode ARP monitor.
2898 * Called with rcu_read_lock held.
2900 static bool bond_ab_arp_probe(struct bonding *bond)
2902 struct slave *slave, *before = NULL, *new_slave = NULL,
2903 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2904 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2905 struct list_head *iter;
2907 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2909 if (curr_arp_slave && curr_active_slave)
2910 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2911 curr_arp_slave->dev->name,
2912 curr_active_slave->dev->name);
2914 if (curr_active_slave) {
2915 bond_arp_send_all(bond, curr_active_slave);
2916 return should_notify_rtnl;
2919 /* if we don't have a curr_active_slave, search for the next available
2920 * backup slave from the current_arp_slave and make it the candidate
2921 * for becoming the curr_active_slave
2924 if (!curr_arp_slave) {
2925 curr_arp_slave = bond_first_slave_rcu(bond);
2926 if (!curr_arp_slave)
2927 return should_notify_rtnl;
2930 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2932 bond_for_each_slave_rcu(bond, slave, iter) {
2933 if (!found && !before && bond_slave_is_up(slave))
2936 if (found && !new_slave && bond_slave_is_up(slave))
2938 /* if the link state is up at this point, we
2939 * mark it down - this can happen if we have
2940 * simultaneous link failures and
2941 * reselect_active_interface doesn't make this
2942 * one the current slave so it is still marked
2943 * up when it is actually down
2945 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2946 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2947 BOND_SLAVE_NOTIFY_LATER);
2948 if (slave->link_failure_count < UINT_MAX)
2949 slave->link_failure_count++;
2951 bond_set_slave_inactive_flags(slave,
2952 BOND_SLAVE_NOTIFY_LATER);
2954 netdev_info(bond->dev, "backup interface %s is now down\n",
2957 if (slave == curr_arp_slave)
2961 if (!new_slave && before)
2967 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2968 BOND_SLAVE_NOTIFY_LATER);
2969 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2970 bond_arp_send_all(bond, new_slave);
2971 new_slave->last_link_up = jiffies;
2972 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2975 bond_for_each_slave_rcu(bond, slave, iter) {
2976 if (slave->should_notify || slave->should_notify_link) {
2977 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2981 return should_notify_rtnl;
2984 static void bond_activebackup_arp_mon(struct bonding *bond)
2986 bool should_notify_peers = false;
2987 bool should_notify_rtnl = false;
2990 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2992 if (!bond_has_slaves(bond))
2997 should_notify_peers = bond_should_notify_peers(bond);
2999 if (bond_ab_arp_inspect(bond)) {
3002 /* Race avoidance with bond_close flush of workqueue */
3003 if (!rtnl_trylock()) {
3005 should_notify_peers = false;
3009 bond_ab_arp_commit(bond);
3015 should_notify_rtnl = bond_ab_arp_probe(bond);
3019 if (bond->params.arp_interval)
3020 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3022 if (should_notify_peers || should_notify_rtnl) {
3023 if (!rtnl_trylock())
3026 if (should_notify_peers)
3027 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3029 if (should_notify_rtnl) {
3030 bond_slave_state_notify(bond);
3031 bond_slave_link_notify(bond);
3038 static void bond_arp_monitor(struct work_struct *work)
3040 struct bonding *bond = container_of(work, struct bonding,
3043 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3044 bond_activebackup_arp_mon(bond);
3046 bond_loadbalance_arp_mon(bond);
3049 /*-------------------------- netdev event handling --------------------------*/
3051 /* Change device name */
3052 static int bond_event_changename(struct bonding *bond)
3054 bond_remove_proc_entry(bond);
3055 bond_create_proc_entry(bond);
3057 bond_debug_reregister(bond);
3062 static int bond_master_netdev_event(unsigned long event,
3063 struct net_device *bond_dev)
3065 struct bonding *event_bond = netdev_priv(bond_dev);
3068 case NETDEV_CHANGENAME:
3069 return bond_event_changename(event_bond);
3070 case NETDEV_UNREGISTER:
3071 bond_remove_proc_entry(event_bond);
3073 case NETDEV_REGISTER:
3074 bond_create_proc_entry(event_bond);
3076 case NETDEV_NOTIFY_PEERS:
3077 if (event_bond->send_peer_notif)
3078 event_bond->send_peer_notif--;
3087 static int bond_slave_netdev_event(unsigned long event,
3088 struct net_device *slave_dev)
3090 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3091 struct bonding *bond;
3092 struct net_device *bond_dev;
3094 /* A netdev event can be generated while enslaving a device
3095 * before netdev_rx_handler_register is called in which case
3096 * slave will be NULL
3100 bond_dev = slave->bond->dev;
3102 primary = rtnl_dereference(bond->primary_slave);
3105 case NETDEV_UNREGISTER:
3106 if (bond_dev->type != ARPHRD_ETHER)
3107 bond_release_and_destroy(bond_dev, slave_dev);
3109 __bond_release_one(bond_dev, slave_dev, false, true);
3113 /* For 802.3ad mode only:
3114 * Getting invalid Speed/Duplex values here will put slave
3115 * in weird state. So mark it as link-fail for the time
3116 * being and let link-monitoring (miimon) set it right when
3117 * correct speeds/duplex are available.
3119 if (bond_update_speed_duplex(slave) &&
3120 BOND_MODE(bond) == BOND_MODE_8023AD)
3121 slave->link = BOND_LINK_FAIL;
3123 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3124 bond_3ad_adapter_speed_duplex_changed(slave);
3127 /* Refresh slave-array if applicable!
3128 * If the setup does not use miimon or arpmon (mode-specific!),
3129 * then these events will not cause the slave-array to be
3130 * refreshed. This will cause xmit to use a slave that is not
3131 * usable. Avoid such situation by refeshing the array at these
3132 * events. If these (miimon/arpmon) parameters are configured
3133 * then array gets refreshed twice and that should be fine!
3135 if (bond_mode_can_use_xmit_hash(bond))
3136 bond_update_slave_arr(bond, NULL);
3138 case NETDEV_CHANGEMTU:
3139 /* TODO: Should slaves be allowed to
3140 * independently alter their MTU? For
3141 * an active-backup bond, slaves need
3142 * not be the same type of device, so
3143 * MTUs may vary. For other modes,
3144 * slaves arguably should have the
3145 * same MTUs. To do this, we'd need to
3146 * take over the slave's change_mtu
3147 * function for the duration of their
3151 case NETDEV_CHANGENAME:
3152 /* we don't care if we don't have primary set */
3153 if (!bond_uses_primary(bond) ||
3154 !bond->params.primary[0])
3157 if (slave == primary) {
3158 /* slave's name changed - he's no longer primary */
3159 RCU_INIT_POINTER(bond->primary_slave, NULL);
3160 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3161 /* we have a new primary slave */
3162 rcu_assign_pointer(bond->primary_slave, slave);
3163 } else { /* we didn't change primary - exit */
3167 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3168 primary ? slave_dev->name : "none");
3171 bond_select_active_slave(bond);
3172 unblock_netpoll_tx();
3174 case NETDEV_FEAT_CHANGE:
3175 bond_compute_features(bond);
3177 case NETDEV_RESEND_IGMP:
3178 /* Propagate to master device */
3179 call_netdevice_notifiers(event, slave->bond->dev);
3188 /* bond_netdev_event: handle netdev notifier chain events.
3190 * This function receives events for the netdev chain. The caller (an
3191 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3192 * locks for us to safely manipulate the slave devices (RTNL lock,
3195 static int bond_netdev_event(struct notifier_block *this,
3196 unsigned long event, void *ptr)
3198 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3200 netdev_dbg(event_dev, "event: %lx\n", event);
3202 if (!(event_dev->priv_flags & IFF_BONDING))
3205 if (event_dev->flags & IFF_MASTER) {
3206 netdev_dbg(event_dev, "IFF_MASTER\n");
3207 return bond_master_netdev_event(event, event_dev);
3210 if (event_dev->flags & IFF_SLAVE) {
3211 netdev_dbg(event_dev, "IFF_SLAVE\n");
3212 return bond_slave_netdev_event(event, event_dev);
3218 static struct notifier_block bond_netdev_notifier = {
3219 .notifier_call = bond_netdev_event,
3222 /*---------------------------- Hashing Policies -----------------------------*/
3224 /* L2 hash helper */
3225 static inline u32 bond_eth_hash(struct sk_buff *skb)
3227 struct ethhdr *ep, hdr_tmp;
3229 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3231 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3235 /* Extract the appropriate headers based on bond's xmit policy */
3236 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3237 struct flow_keys *fk)
3239 const struct ipv6hdr *iph6;
3240 const struct iphdr *iph;
3241 int noff, proto = -1;
3243 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3244 return skb_flow_dissect_flow_keys(skb, fk, 0);
3246 fk->ports.ports = 0;
3247 noff = skb_network_offset(skb);
3248 if (skb->protocol == htons(ETH_P_IP)) {
3249 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3252 iph_to_flow_copy_v4addrs(fk, iph);
3253 noff += iph->ihl << 2;
3254 if (!ip_is_fragment(iph))
3255 proto = iph->protocol;
3256 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3257 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3259 iph6 = ipv6_hdr(skb);
3260 iph_to_flow_copy_v6addrs(fk, iph6);
3261 noff += sizeof(*iph6);
3262 proto = iph6->nexthdr;
3266 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3267 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3273 * bond_xmit_hash - generate a hash value based on the xmit policy
3274 * @bond: bonding device
3275 * @skb: buffer to use for headers
3277 * This function will extract the necessary headers from the skb buffer and use
3278 * them to generate a hash based on the xmit_policy set in the bonding device
3280 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3282 struct flow_keys flow;
3285 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3289 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3290 !bond_flow_dissect(bond, skb, &flow))
3291 return bond_eth_hash(skb);
3293 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3294 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3295 hash = bond_eth_hash(skb);
3297 hash = (__force u32)flow.ports.ports;
3298 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3299 (__force u32)flow_get_u32_src(&flow);
3300 hash ^= (hash >> 16);
3301 hash ^= (hash >> 8);
3306 /*-------------------------- Device entry points ----------------------------*/
3308 void bond_work_init_all(struct bonding *bond)
3310 INIT_DELAYED_WORK(&bond->mcast_work,
3311 bond_resend_igmp_join_requests_delayed);
3312 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3313 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3314 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3315 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3316 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3319 static void bond_work_cancel_all(struct bonding *bond)
3321 cancel_delayed_work_sync(&bond->mii_work);
3322 cancel_delayed_work_sync(&bond->arp_work);
3323 cancel_delayed_work_sync(&bond->alb_work);
3324 cancel_delayed_work_sync(&bond->ad_work);
3325 cancel_delayed_work_sync(&bond->mcast_work);
3326 cancel_delayed_work_sync(&bond->slave_arr_work);
3329 static int bond_open(struct net_device *bond_dev)
3331 struct bonding *bond = netdev_priv(bond_dev);
3332 struct list_head *iter;
3333 struct slave *slave;
3335 /* reset slave->backup and slave->inactive */
3336 if (bond_has_slaves(bond)) {
3337 bond_for_each_slave(bond, slave, iter) {
3338 if (bond_uses_primary(bond) &&
3339 slave != rcu_access_pointer(bond->curr_active_slave)) {
3340 bond_set_slave_inactive_flags(slave,
3341 BOND_SLAVE_NOTIFY_NOW);
3342 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3343 bond_set_slave_active_flags(slave,
3344 BOND_SLAVE_NOTIFY_NOW);
3349 if (bond_is_lb(bond)) {
3350 /* bond_alb_initialize must be called before the timer
3353 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3355 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3356 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3359 if (bond->params.miimon) /* link check interval, in milliseconds. */
3360 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3362 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3363 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3364 bond->recv_probe = bond_arp_rcv;
3367 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3368 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3369 /* register to receive LACPDUs */
3370 bond->recv_probe = bond_3ad_lacpdu_recv;
3371 bond_3ad_initiate_agg_selection(bond, 1);
3374 if (bond_mode_can_use_xmit_hash(bond))
3375 bond_update_slave_arr(bond, NULL);
3380 static int bond_close(struct net_device *bond_dev)
3382 struct bonding *bond = netdev_priv(bond_dev);
3384 bond_work_cancel_all(bond);
3385 bond->send_peer_notif = 0;
3386 if (bond_is_lb(bond))
3387 bond_alb_deinitialize(bond);
3388 bond->recv_probe = NULL;
3393 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3394 * that some drivers can provide 32bit values only.
3396 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3397 const struct rtnl_link_stats64 *_new,
3398 const struct rtnl_link_stats64 *_old)
3400 const u64 *new = (const u64 *)_new;
3401 const u64 *old = (const u64 *)_old;
3402 u64 *res = (u64 *)_res;
3405 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3408 s64 delta = nv - ov;
3410 /* detects if this particular field is 32bit only */
3411 if (((nv | ov) >> 32) == 0)
3412 delta = (s64)(s32)((u32)nv - (u32)ov);
3414 /* filter anomalies, some drivers reset their stats
3415 * at down/up events.
3422 static int bond_get_nest_level(struct net_device *bond_dev)
3424 struct bonding *bond = netdev_priv(bond_dev);
3426 return bond->nest_level;
3429 static void bond_get_stats(struct net_device *bond_dev,
3430 struct rtnl_link_stats64 *stats)
3432 struct bonding *bond = netdev_priv(bond_dev);
3433 struct rtnl_link_stats64 temp;
3434 struct list_head *iter;
3435 struct slave *slave;
3437 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3438 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3441 bond_for_each_slave_rcu(bond, slave, iter) {
3442 const struct rtnl_link_stats64 *new =
3443 dev_get_stats(slave->dev, &temp);
3445 bond_fold_stats(stats, new, &slave->slave_stats);
3447 /* save off the slave stats for the next run */
3448 memcpy(&slave->slave_stats, new, sizeof(*new));
3452 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3453 spin_unlock(&bond->stats_lock);
3456 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3458 struct bonding *bond = netdev_priv(bond_dev);
3459 struct net_device *slave_dev = NULL;
3460 struct ifbond k_binfo;
3461 struct ifbond __user *u_binfo = NULL;
3462 struct ifslave k_sinfo;
3463 struct ifslave __user *u_sinfo = NULL;
3464 struct mii_ioctl_data *mii = NULL;
3465 struct bond_opt_value newval;
3469 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3480 /* We do this again just in case we were called by SIOCGMIIREG
3481 * instead of SIOCGMIIPHY.
3487 if (mii->reg_num == 1) {
3489 if (netif_carrier_ok(bond->dev))
3490 mii->val_out = BMSR_LSTATUS;
3494 case BOND_INFO_QUERY_OLD:
3495 case SIOCBONDINFOQUERY:
3496 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3498 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3501 bond_info_query(bond_dev, &k_binfo);
3502 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3506 case BOND_SLAVE_INFO_QUERY_OLD:
3507 case SIOCBONDSLAVEINFOQUERY:
3508 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3510 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3513 res = bond_slave_info_query(bond_dev, &k_sinfo);
3515 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3523 net = dev_net(bond_dev);
3525 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3528 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3530 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3535 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3537 case BOND_ENSLAVE_OLD:
3538 case SIOCBONDENSLAVE:
3539 res = bond_enslave(bond_dev, slave_dev, NULL);
3541 case BOND_RELEASE_OLD:
3542 case SIOCBONDRELEASE:
3543 res = bond_release(bond_dev, slave_dev);
3545 case BOND_SETHWADDR_OLD:
3546 case SIOCBONDSETHWADDR:
3547 bond_set_dev_addr(bond_dev, slave_dev);
3550 case BOND_CHANGE_ACTIVE_OLD:
3551 case SIOCBONDCHANGEACTIVE:
3552 bond_opt_initstr(&newval, slave_dev->name);
3553 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3563 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3565 struct bonding *bond = netdev_priv(bond_dev);
3567 if (change & IFF_PROMISC)
3568 bond_set_promiscuity(bond,
3569 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3571 if (change & IFF_ALLMULTI)
3572 bond_set_allmulti(bond,
3573 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3576 static void bond_set_rx_mode(struct net_device *bond_dev)
3578 struct bonding *bond = netdev_priv(bond_dev);
3579 struct list_head *iter;
3580 struct slave *slave;
3583 if (bond_uses_primary(bond)) {
3584 slave = rcu_dereference(bond->curr_active_slave);
3586 dev_uc_sync(slave->dev, bond_dev);
3587 dev_mc_sync(slave->dev, bond_dev);
3590 bond_for_each_slave_rcu(bond, slave, iter) {
3591 dev_uc_sync_multiple(slave->dev, bond_dev);
3592 dev_mc_sync_multiple(slave->dev, bond_dev);
3598 static int bond_neigh_init(struct neighbour *n)
3600 struct bonding *bond = netdev_priv(n->dev);
3601 const struct net_device_ops *slave_ops;
3602 struct neigh_parms parms;
3603 struct slave *slave;
3606 slave = bond_first_slave(bond);
3609 slave_ops = slave->dev->netdev_ops;
3610 if (!slave_ops->ndo_neigh_setup)
3613 parms.neigh_setup = NULL;
3614 parms.neigh_cleanup = NULL;
3615 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3619 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3620 * after the last slave has been detached. Assumes that all slaves
3621 * utilize the same neigh_cleanup (true at this writing as only user
3624 n->parms->neigh_cleanup = parms.neigh_cleanup;
3626 if (!parms.neigh_setup)
3629 return parms.neigh_setup(n);
3632 /* The bonding ndo_neigh_setup is called at init time beofre any
3633 * slave exists. So we must declare proxy setup function which will
3634 * be used at run time to resolve the actual slave neigh param setup.
3636 * It's also called by master devices (such as vlans) to setup their
3637 * underlying devices. In that case - do nothing, we're already set up from
3640 static int bond_neigh_setup(struct net_device *dev,
3641 struct neigh_parms *parms)
3643 /* modify only our neigh_parms */
3644 if (parms->dev == dev)
3645 parms->neigh_setup = bond_neigh_init;
3650 /* Change the MTU of all of a master's slaves to match the master */
3651 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3653 struct bonding *bond = netdev_priv(bond_dev);
3654 struct slave *slave, *rollback_slave;
3655 struct list_head *iter;
3658 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3660 bond_for_each_slave(bond, slave, iter) {
3661 netdev_dbg(bond_dev, "s %p c_m %p\n",
3662 slave, slave->dev->netdev_ops->ndo_change_mtu);
3664 res = dev_set_mtu(slave->dev, new_mtu);
3667 /* If we failed to set the slave's mtu to the new value
3668 * we must abort the operation even in ACTIVE_BACKUP
3669 * mode, because if we allow the backup slaves to have
3670 * different mtu values than the active slave we'll
3671 * need to change their mtu when doing a failover. That
3672 * means changing their mtu from timer context, which
3673 * is probably not a good idea.
3675 netdev_dbg(bond_dev, "err %d %s\n", res,
3681 bond_dev->mtu = new_mtu;
3686 /* unwind from head to the slave that failed */
3687 bond_for_each_slave(bond, rollback_slave, iter) {
3690 if (rollback_slave == slave)
3693 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3695 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3696 tmp_res, rollback_slave->dev->name);
3703 /* Change HW address
3705 * Note that many devices must be down to change the HW address, and
3706 * downing the master releases all slaves. We can make bonds full of
3707 * bonding devices to test this, however.
3709 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3711 struct bonding *bond = netdev_priv(bond_dev);
3712 struct slave *slave, *rollback_slave;
3713 struct sockaddr_storage *ss = addr, tmp_ss;
3714 struct list_head *iter;
3717 if (BOND_MODE(bond) == BOND_MODE_ALB)
3718 return bond_alb_set_mac_address(bond_dev, addr);
3721 netdev_dbg(bond_dev, "bond=%p\n", bond);
3723 /* If fail_over_mac is enabled, do nothing and return success.
3724 * Returning an error causes ifenslave to fail.
3726 if (bond->params.fail_over_mac &&
3727 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3730 if (!is_valid_ether_addr(ss->__data))
3731 return -EADDRNOTAVAIL;
3733 bond_for_each_slave(bond, slave, iter) {
3734 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3735 res = dev_set_mac_address(slave->dev, addr);
3737 /* TODO: consider downing the slave
3739 * User should expect communications
3740 * breakage anyway until ARP finish
3743 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3749 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3753 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3754 tmp_ss.ss_family = bond_dev->type;
3756 /* unwind from head to the slave that failed */
3757 bond_for_each_slave(bond, rollback_slave, iter) {
3760 if (rollback_slave == slave)
3763 tmp_res = dev_set_mac_address(rollback_slave->dev,
3764 (struct sockaddr *)&tmp_ss);
3766 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3767 tmp_res, rollback_slave->dev->name);
3775 * bond_xmit_slave_id - transmit skb through slave with slave_id
3776 * @bond: bonding device that is transmitting
3777 * @skb: buffer to transmit
3778 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3780 * This function tries to transmit through slave with slave_id but in case
3781 * it fails, it tries to find the first available slave for transmission.
3782 * The skb is consumed in all cases, thus the function is void.
3784 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3786 struct list_head *iter;
3787 struct slave *slave;
3790 /* Here we start from the slave with slave_id */
3791 bond_for_each_slave_rcu(bond, slave, iter) {
3793 if (bond_slave_can_tx(slave)) {
3794 bond_dev_queue_xmit(bond, skb, slave->dev);
3800 /* Here we start from the first slave up to slave_id */
3802 bond_for_each_slave_rcu(bond, slave, iter) {
3805 if (bond_slave_can_tx(slave)) {
3806 bond_dev_queue_xmit(bond, skb, slave->dev);
3810 /* no slave that can tx has been found */
3811 bond_tx_drop(bond->dev, skb);
3815 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3816 * @bond: bonding device to use
3818 * Based on the value of the bonding device's packets_per_slave parameter
3819 * this function generates a slave id, which is usually used as the next
3820 * slave to transmit through.
3822 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3825 struct reciprocal_value reciprocal_packets_per_slave;
3826 int packets_per_slave = bond->params.packets_per_slave;
3828 switch (packets_per_slave) {
3830 slave_id = prandom_u32();
3833 slave_id = bond->rr_tx_counter;
3836 reciprocal_packets_per_slave =
3837 bond->params.reciprocal_packets_per_slave;
3838 slave_id = reciprocal_divide(bond->rr_tx_counter,
3839 reciprocal_packets_per_slave);
3842 bond->rr_tx_counter++;
3847 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3848 struct net_device *bond_dev)
3850 struct bonding *bond = netdev_priv(bond_dev);
3851 struct iphdr *iph = ip_hdr(skb);
3852 struct slave *slave;
3855 /* Start with the curr_active_slave that joined the bond as the
3856 * default for sending IGMP traffic. For failover purposes one
3857 * needs to maintain some consistency for the interface that will
3858 * send the join/membership reports. The curr_active_slave found
3859 * will send all of this type of traffic.
3861 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3862 slave = rcu_dereference(bond->curr_active_slave);
3864 bond_dev_queue_xmit(bond, skb, slave->dev);
3866 bond_xmit_slave_id(bond, skb, 0);
3868 int slave_cnt = READ_ONCE(bond->slave_cnt);
3870 if (likely(slave_cnt)) {
3871 slave_id = bond_rr_gen_slave_id(bond);
3872 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3874 bond_tx_drop(bond_dev, skb);
3878 return NETDEV_TX_OK;
3881 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3882 * the bond has a usable interface.
3884 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3885 struct net_device *bond_dev)
3887 struct bonding *bond = netdev_priv(bond_dev);
3888 struct slave *slave;
3890 slave = rcu_dereference(bond->curr_active_slave);
3892 bond_dev_queue_xmit(bond, skb, slave->dev);
3894 bond_tx_drop(bond_dev, skb);
3896 return NETDEV_TX_OK;
3899 /* Use this to update slave_array when (a) it's not appropriate to update
3900 * slave_array right away (note that update_slave_array() may sleep)
3901 * and / or (b) RTNL is not held.
3903 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3905 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3908 /* Slave array work handler. Holds only RTNL */
3909 static void bond_slave_arr_handler(struct work_struct *work)
3911 struct bonding *bond = container_of(work, struct bonding,
3912 slave_arr_work.work);
3915 if (!rtnl_trylock())
3918 ret = bond_update_slave_arr(bond, NULL);
3921 pr_warn_ratelimited("Failed to update slave array from WT\n");
3927 bond_slave_arr_work_rearm(bond, 1);
3930 /* Build the usable slaves array in control path for modes that use xmit-hash
3931 * to determine the slave interface -
3932 * (a) BOND_MODE_8023AD
3934 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3936 * The caller is expected to hold RTNL only and NO other lock!
3938 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3940 struct slave *slave;
3941 struct list_head *iter;
3942 struct bond_up_slave *new_arr, *old_arr;
3946 #ifdef CONFIG_LOCKDEP
3947 WARN_ON(lockdep_is_held(&bond->mode_lock));
3950 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3954 pr_err("Failed to build slave-array.\n");
3957 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3958 struct ad_info ad_info;
3960 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3961 pr_debug("bond_3ad_get_active_agg_info failed\n");
3962 kfree_rcu(new_arr, rcu);
3963 /* No active aggragator means it's not safe to use
3964 * the previous array.
3966 old_arr = rtnl_dereference(bond->slave_arr);
3968 RCU_INIT_POINTER(bond->slave_arr, NULL);
3969 kfree_rcu(old_arr, rcu);
3973 agg_id = ad_info.aggregator_id;
3975 bond_for_each_slave(bond, slave, iter) {
3976 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3977 struct aggregator *agg;
3979 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3980 if (!agg || agg->aggregator_identifier != agg_id)
3983 if (!bond_slave_can_tx(slave))
3985 if (skipslave == slave)
3988 netdev_dbg(bond->dev,
3989 "Adding slave dev %s to tx hash array[%d]\n",
3990 slave->dev->name, new_arr->count);
3992 new_arr->arr[new_arr->count++] = slave;
3995 old_arr = rtnl_dereference(bond->slave_arr);
3996 rcu_assign_pointer(bond->slave_arr, new_arr);
3998 kfree_rcu(old_arr, rcu);
4000 if (ret != 0 && skipslave) {
4003 /* Rare situation where caller has asked to skip a specific
4004 * slave but allocation failed (most likely!). BTW this is
4005 * only possible when the call is initiated from
4006 * __bond_release_one(). In this situation; overwrite the
4007 * skipslave entry in the array with the last entry from the
4008 * array to avoid a situation where the xmit path may choose
4009 * this to-be-skipped slave to send a packet out.
4011 old_arr = rtnl_dereference(bond->slave_arr);
4012 for (idx = 0; idx < old_arr->count; idx++) {
4013 if (skipslave == old_arr->arr[idx]) {
4015 old_arr->arr[old_arr->count-1];
4024 /* Use this Xmit function for 3AD as well as XOR modes. The current
4025 * usable slave array is formed in the control path. The xmit function
4026 * just calculates hash and sends the packet out.
4028 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4029 struct net_device *dev)
4031 struct bonding *bond = netdev_priv(dev);
4032 struct slave *slave;
4033 struct bond_up_slave *slaves;
4036 slaves = rcu_dereference(bond->slave_arr);
4037 count = slaves ? READ_ONCE(slaves->count) : 0;
4038 if (likely(count)) {
4039 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4040 bond_dev_queue_xmit(bond, skb, slave->dev);
4042 bond_tx_drop(dev, skb);
4045 return NETDEV_TX_OK;
4048 /* in broadcast mode, we send everything to all usable interfaces. */
4049 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4050 struct net_device *bond_dev)
4052 struct bonding *bond = netdev_priv(bond_dev);
4053 struct slave *slave = NULL;
4054 struct list_head *iter;
4056 bond_for_each_slave_rcu(bond, slave, iter) {
4057 if (bond_is_last_slave(bond, slave))
4059 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4060 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4063 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4064 bond_dev->name, __func__);
4067 bond_dev_queue_xmit(bond, skb2, slave->dev);
4070 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4071 bond_dev_queue_xmit(bond, skb, slave->dev);
4073 bond_tx_drop(bond_dev, skb);
4075 return NETDEV_TX_OK;
4078 /*------------------------- Device initialization ---------------------------*/
4080 /* Lookup the slave that corresponds to a qid */
4081 static inline int bond_slave_override(struct bonding *bond,
4082 struct sk_buff *skb)
4084 struct slave *slave = NULL;
4085 struct list_head *iter;
4087 if (!skb_rx_queue_recorded(skb))
4090 /* Find out if any slaves have the same mapping as this skb. */
4091 bond_for_each_slave_rcu(bond, slave, iter) {
4092 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4093 if (bond_slave_is_up(slave) &&
4094 slave->link == BOND_LINK_UP) {
4095 bond_dev_queue_xmit(bond, skb, slave->dev);
4098 /* If the slave isn't UP, use default transmit policy. */
4107 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4108 struct net_device *sb_dev,
4109 select_queue_fallback_t fallback)
4111 /* This helper function exists to help dev_pick_tx get the correct
4112 * destination queue. Using a helper function skips a call to
4113 * skb_tx_hash and will put the skbs in the queue we expect on their
4114 * way down to the bonding driver.
4116 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4118 /* Save the original txq to restore before passing to the driver */
4119 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4121 if (unlikely(txq >= dev->real_num_tx_queues)) {
4123 txq -= dev->real_num_tx_queues;
4124 } while (txq >= dev->real_num_tx_queues);
4129 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4131 struct bonding *bond = netdev_priv(dev);
4133 if (bond_should_override_tx_queue(bond) &&
4134 !bond_slave_override(bond, skb))
4135 return NETDEV_TX_OK;
4137 switch (BOND_MODE(bond)) {
4138 case BOND_MODE_ROUNDROBIN:
4139 return bond_xmit_roundrobin(skb, dev);
4140 case BOND_MODE_ACTIVEBACKUP:
4141 return bond_xmit_activebackup(skb, dev);
4142 case BOND_MODE_8023AD:
4144 return bond_3ad_xor_xmit(skb, dev);
4145 case BOND_MODE_BROADCAST:
4146 return bond_xmit_broadcast(skb, dev);
4148 return bond_alb_xmit(skb, dev);
4150 return bond_tlb_xmit(skb, dev);
4152 /* Should never happen, mode already checked */
4153 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4155 bond_tx_drop(dev, skb);
4156 return NETDEV_TX_OK;
4160 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4162 struct bonding *bond = netdev_priv(dev);
4163 netdev_tx_t ret = NETDEV_TX_OK;
4165 /* If we risk deadlock from transmitting this in the
4166 * netpoll path, tell netpoll to queue the frame for later tx
4168 if (unlikely(is_netpoll_tx_blocked(dev)))
4169 return NETDEV_TX_BUSY;
4172 if (bond_has_slaves(bond))
4173 ret = __bond_start_xmit(skb, dev);
4175 bond_tx_drop(dev, skb);
4181 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4182 struct ethtool_link_ksettings *cmd)
4184 struct bonding *bond = netdev_priv(bond_dev);
4185 unsigned long speed = 0;
4186 struct list_head *iter;
4187 struct slave *slave;
4189 cmd->base.duplex = DUPLEX_UNKNOWN;
4190 cmd->base.port = PORT_OTHER;
4192 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4193 * do not need to check mode. Though link speed might not represent
4194 * the true receive or transmit bandwidth (not all modes are symmetric)
4195 * this is an accurate maximum.
4197 bond_for_each_slave(bond, slave, iter) {
4198 if (bond_slave_can_tx(slave)) {
4199 if (slave->speed != SPEED_UNKNOWN)
4200 speed += slave->speed;
4201 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4202 slave->duplex != DUPLEX_UNKNOWN)
4203 cmd->base.duplex = slave->duplex;
4206 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4211 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4212 struct ethtool_drvinfo *drvinfo)
4214 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4215 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4216 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4220 static const struct ethtool_ops bond_ethtool_ops = {
4221 .get_drvinfo = bond_ethtool_get_drvinfo,
4222 .get_link = ethtool_op_get_link,
4223 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4226 static const struct net_device_ops bond_netdev_ops = {
4227 .ndo_init = bond_init,
4228 .ndo_uninit = bond_uninit,
4229 .ndo_open = bond_open,
4230 .ndo_stop = bond_close,
4231 .ndo_start_xmit = bond_start_xmit,
4232 .ndo_select_queue = bond_select_queue,
4233 .ndo_get_stats64 = bond_get_stats,
4234 .ndo_do_ioctl = bond_do_ioctl,
4235 .ndo_change_rx_flags = bond_change_rx_flags,
4236 .ndo_set_rx_mode = bond_set_rx_mode,
4237 .ndo_change_mtu = bond_change_mtu,
4238 .ndo_set_mac_address = bond_set_mac_address,
4239 .ndo_neigh_setup = bond_neigh_setup,
4240 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4241 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4242 .ndo_get_lock_subclass = bond_get_nest_level,
4243 #ifdef CONFIG_NET_POLL_CONTROLLER
4244 .ndo_netpoll_setup = bond_netpoll_setup,
4245 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4246 .ndo_poll_controller = bond_poll_controller,
4248 .ndo_add_slave = bond_enslave,
4249 .ndo_del_slave = bond_release,
4250 .ndo_fix_features = bond_fix_features,
4251 .ndo_features_check = passthru_features_check,
4254 static const struct device_type bond_type = {
4258 static void bond_destructor(struct net_device *bond_dev)
4260 struct bonding *bond = netdev_priv(bond_dev);
4262 destroy_workqueue(bond->wq);
4265 void bond_setup(struct net_device *bond_dev)
4267 struct bonding *bond = netdev_priv(bond_dev);
4269 spin_lock_init(&bond->mode_lock);
4270 spin_lock_init(&bond->stats_lock);
4271 bond->params = bonding_defaults;
4273 /* Initialize pointers */
4274 bond->dev = bond_dev;
4276 /* Initialize the device entry points */
4277 ether_setup(bond_dev);
4278 bond_dev->max_mtu = ETH_MAX_MTU;
4279 bond_dev->netdev_ops = &bond_netdev_ops;
4280 bond_dev->ethtool_ops = &bond_ethtool_ops;
4282 bond_dev->needs_free_netdev = true;
4283 bond_dev->priv_destructor = bond_destructor;
4285 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4287 /* Initialize the device options */
4288 bond_dev->flags |= IFF_MASTER;
4289 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4290 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4292 /* don't acquire bond device's netif_tx_lock when transmitting */
4293 bond_dev->features |= NETIF_F_LLTX;
4295 /* By default, we declare the bond to be fully
4296 * VLAN hardware accelerated capable. Special
4297 * care is taken in the various xmit functions
4298 * when there are slaves that are not hw accel
4302 /* Don't allow bond devices to change network namespaces. */
4303 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4305 bond_dev->hw_features = BOND_VLAN_FEATURES |
4306 NETIF_F_HW_VLAN_CTAG_TX |
4307 NETIF_F_HW_VLAN_CTAG_RX |
4308 NETIF_F_HW_VLAN_CTAG_FILTER;
4310 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4311 bond_dev->features |= bond_dev->hw_features;
4314 /* Destroy a bonding device.
4315 * Must be under rtnl_lock when this function is called.
4317 static void bond_uninit(struct net_device *bond_dev)
4319 struct bonding *bond = netdev_priv(bond_dev);
4320 struct list_head *iter;
4321 struct slave *slave;
4322 struct bond_up_slave *arr;
4324 bond_netpoll_cleanup(bond_dev);
4326 /* Release the bonded slaves */
4327 bond_for_each_slave(bond, slave, iter)
4328 __bond_release_one(bond_dev, slave->dev, true, true);
4329 netdev_info(bond_dev, "Released all slaves\n");
4331 arr = rtnl_dereference(bond->slave_arr);
4333 RCU_INIT_POINTER(bond->slave_arr, NULL);
4334 kfree_rcu(arr, rcu);
4337 list_del(&bond->bond_list);
4339 bond_debug_unregister(bond);
4342 /*------------------------- Module initialization ---------------------------*/
4344 static int bond_check_params(struct bond_params *params)
4346 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4347 struct bond_opt_value newval;
4348 const struct bond_opt_value *valptr;
4349 int arp_all_targets_value = 0;
4350 u16 ad_actor_sys_prio = 0;
4351 u16 ad_user_port_key = 0;
4352 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4354 int bond_mode = BOND_MODE_ROUNDROBIN;
4355 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4359 /* Convert string parameters. */
4361 bond_opt_initstr(&newval, mode);
4362 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4364 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4367 bond_mode = valptr->value;
4370 if (xmit_hash_policy) {
4371 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4372 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4373 bond_mode == BOND_MODE_BROADCAST) {
4374 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4375 bond_mode_name(bond_mode));
4377 bond_opt_initstr(&newval, xmit_hash_policy);
4378 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4381 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4385 xmit_hashtype = valptr->value;
4390 if (bond_mode != BOND_MODE_8023AD) {
4391 pr_info("lacp_rate param is irrelevant in mode %s\n",
4392 bond_mode_name(bond_mode));
4394 bond_opt_initstr(&newval, lacp_rate);
4395 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4398 pr_err("Error: Invalid lacp rate \"%s\"\n",
4402 lacp_fast = valptr->value;
4407 bond_opt_initstr(&newval, ad_select);
4408 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4411 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4414 params->ad_select = valptr->value;
4415 if (bond_mode != BOND_MODE_8023AD)
4416 pr_warn("ad_select param only affects 802.3ad mode\n");
4418 params->ad_select = BOND_AD_STABLE;
4421 if (max_bonds < 0) {
4422 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4423 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4424 max_bonds = BOND_DEFAULT_MAX_BONDS;
4428 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4434 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4439 if (downdelay < 0) {
4440 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4441 downdelay, INT_MAX);
4445 if ((use_carrier != 0) && (use_carrier != 1)) {
4446 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4451 if (num_peer_notif < 0 || num_peer_notif > 255) {
4452 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4457 /* reset values for 802.3ad/TLB/ALB */
4458 if (!bond_mode_uses_arp(bond_mode)) {
4460 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");
4461 pr_warn("Forcing miimon to 100msec\n");
4462 miimon = BOND_DEFAULT_MIIMON;
4466 if (tx_queues < 1 || tx_queues > 255) {
4467 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4468 tx_queues, BOND_DEFAULT_TX_QUEUES);
4469 tx_queues = BOND_DEFAULT_TX_QUEUES;
4472 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4473 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4475 all_slaves_active = 0;
4478 if (resend_igmp < 0 || resend_igmp > 255) {
4479 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4480 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4481 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4484 bond_opt_initval(&newval, packets_per_slave);
4485 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4486 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4487 packets_per_slave, USHRT_MAX);
4488 packets_per_slave = 1;
4491 if (bond_mode == BOND_MODE_ALB) {
4492 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",
4497 if (updelay || downdelay) {
4498 /* just warn the user the up/down delay will have
4499 * no effect since miimon is zero...
4501 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",
4502 updelay, downdelay);
4505 /* don't allow arp monitoring */
4507 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4508 miimon, arp_interval);
4512 if ((updelay % miimon) != 0) {
4513 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4514 updelay, miimon, (updelay / miimon) * miimon);
4519 if ((downdelay % miimon) != 0) {
4520 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4522 (downdelay / miimon) * miimon);
4525 downdelay /= miimon;
4528 if (arp_interval < 0) {
4529 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4530 arp_interval, INT_MAX);
4534 for (arp_ip_count = 0, i = 0;
4535 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4538 /* not a complete check, but good enough to catch mistakes */
4539 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4540 !bond_is_ip_target_ok(ip)) {
4541 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4545 if (bond_get_targets_ip(arp_target, ip) == -1)
4546 arp_target[arp_ip_count++] = ip;
4548 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4553 if (arp_interval && !arp_ip_count) {
4554 /* don't allow arping if no arp_ip_target given... */
4555 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4561 if (!arp_interval) {
4562 pr_err("arp_validate requires arp_interval\n");
4566 bond_opt_initstr(&newval, arp_validate);
4567 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4570 pr_err("Error: invalid arp_validate \"%s\"\n",
4574 arp_validate_value = valptr->value;
4576 arp_validate_value = 0;
4579 if (arp_all_targets) {
4580 bond_opt_initstr(&newval, arp_all_targets);
4581 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4584 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4586 arp_all_targets_value = 0;
4588 arp_all_targets_value = valptr->value;
4593 pr_info("MII link monitoring set to %d ms\n", miimon);
4594 } else if (arp_interval) {
4595 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4596 arp_validate_value);
4597 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4598 arp_interval, valptr->string, arp_ip_count);
4600 for (i = 0; i < arp_ip_count; i++)
4601 pr_cont(" %s", arp_ip_target[i]);
4605 } else if (max_bonds) {
4606 /* miimon and arp_interval not set, we need one so things
4607 * work as expected, see bonding.txt for details
4609 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");
4612 if (primary && !bond_mode_uses_primary(bond_mode)) {
4613 /* currently, using a primary only makes sense
4614 * in active backup, TLB or ALB modes
4616 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4617 primary, bond_mode_name(bond_mode));
4621 if (primary && primary_reselect) {
4622 bond_opt_initstr(&newval, primary_reselect);
4623 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4626 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4630 primary_reselect_value = valptr->value;
4632 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4635 if (fail_over_mac) {
4636 bond_opt_initstr(&newval, fail_over_mac);
4637 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4640 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4644 fail_over_mac_value = valptr->value;
4645 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4646 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4648 fail_over_mac_value = BOND_FOM_NONE;
4651 bond_opt_initstr(&newval, "default");
4652 valptr = bond_opt_parse(
4653 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4656 pr_err("Error: No ad_actor_sys_prio default value");
4659 ad_actor_sys_prio = valptr->value;
4661 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4664 pr_err("Error: No ad_user_port_key default value");
4667 ad_user_port_key = valptr->value;
4669 bond_opt_initstr(&newval, "default");
4670 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4672 pr_err("Error: No tlb_dynamic_lb default value");
4675 tlb_dynamic_lb = valptr->value;
4677 if (lp_interval == 0) {
4678 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4679 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4680 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4683 /* fill params struct with the proper values */
4684 params->mode = bond_mode;
4685 params->xmit_policy = xmit_hashtype;
4686 params->miimon = miimon;
4687 params->num_peer_notif = num_peer_notif;
4688 params->arp_interval = arp_interval;
4689 params->arp_validate = arp_validate_value;
4690 params->arp_all_targets = arp_all_targets_value;
4691 params->updelay = updelay;
4692 params->downdelay = downdelay;
4693 params->use_carrier = use_carrier;
4694 params->lacp_fast = lacp_fast;
4695 params->primary[0] = 0;
4696 params->primary_reselect = primary_reselect_value;
4697 params->fail_over_mac = fail_over_mac_value;
4698 params->tx_queues = tx_queues;
4699 params->all_slaves_active = all_slaves_active;
4700 params->resend_igmp = resend_igmp;
4701 params->min_links = min_links;
4702 params->lp_interval = lp_interval;
4703 params->packets_per_slave = packets_per_slave;
4704 params->tlb_dynamic_lb = tlb_dynamic_lb;
4705 params->ad_actor_sys_prio = ad_actor_sys_prio;
4706 eth_zero_addr(params->ad_actor_system);
4707 params->ad_user_port_key = ad_user_port_key;
4708 if (packets_per_slave > 0) {
4709 params->reciprocal_packets_per_slave =
4710 reciprocal_value(packets_per_slave);
4712 /* reciprocal_packets_per_slave is unused if
4713 * packets_per_slave is 0 or 1, just initialize it
4715 params->reciprocal_packets_per_slave =
4716 (struct reciprocal_value) { 0 };
4720 strncpy(params->primary, primary, IFNAMSIZ);
4721 params->primary[IFNAMSIZ - 1] = 0;
4724 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4729 /* Called from registration process */
4730 static int bond_init(struct net_device *bond_dev)
4732 struct bonding *bond = netdev_priv(bond_dev);
4733 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4735 netdev_dbg(bond_dev, "Begin bond_init\n");
4737 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4741 bond->nest_level = SINGLE_DEPTH_NESTING;
4742 netdev_lockdep_set_classes(bond_dev);
4744 list_add_tail(&bond->bond_list, &bn->dev_list);
4746 bond_prepare_sysfs_group(bond);
4748 bond_debug_register(bond);
4750 /* Ensure valid dev_addr */
4751 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4752 bond_dev->addr_assign_type == NET_ADDR_PERM)
4753 eth_hw_addr_random(bond_dev);
4758 unsigned int bond_get_num_tx_queues(void)
4763 /* Create a new bond based on the specified name and bonding parameters.
4764 * If name is NULL, obtain a suitable "bond%d" name for us.
4765 * Caller must NOT hold rtnl_lock; we need to release it here before we
4766 * set up our sysfs entries.
4768 int bond_create(struct net *net, const char *name)
4770 struct net_device *bond_dev;
4771 struct bonding *bond;
4772 struct alb_bond_info *bond_info;
4777 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4778 name ? name : "bond%d", NET_NAME_UNKNOWN,
4779 bond_setup, tx_queues);
4781 pr_err("%s: eek! can't alloc netdev!\n", name);
4787 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4788 * It is set to 0 by default which is wrong.
4790 bond = netdev_priv(bond_dev);
4791 bond_info = &(BOND_ALB_INFO(bond));
4792 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4794 dev_net_set(bond_dev, net);
4795 bond_dev->rtnl_link_ops = &bond_link_ops;
4797 res = register_netdevice(bond_dev);
4799 netif_carrier_off(bond_dev);
4801 bond_work_init_all(bond);
4805 free_netdev(bond_dev);
4809 static int __net_init bond_net_init(struct net *net)
4811 struct bond_net *bn = net_generic(net, bond_net_id);
4814 INIT_LIST_HEAD(&bn->dev_list);
4816 bond_create_proc_dir(bn);
4817 bond_create_sysfs(bn);
4822 static void __net_exit bond_net_exit(struct net *net)
4824 struct bond_net *bn = net_generic(net, bond_net_id);
4825 struct bonding *bond, *tmp_bond;
4828 bond_destroy_sysfs(bn);
4830 /* Kill off any bonds created after unregistering bond rtnl ops */
4832 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4833 unregister_netdevice_queue(bond->dev, &list);
4834 unregister_netdevice_many(&list);
4837 bond_destroy_proc_dir(bn);
4840 static struct pernet_operations bond_net_ops = {
4841 .init = bond_net_init,
4842 .exit = bond_net_exit,
4844 .size = sizeof(struct bond_net),
4847 static int __init bonding_init(void)
4852 pr_info("%s", bond_version);
4854 res = bond_check_params(&bonding_defaults);
4858 res = register_pernet_subsys(&bond_net_ops);
4862 res = bond_netlink_init();
4866 bond_create_debugfs();
4868 for (i = 0; i < max_bonds; i++) {
4869 res = bond_create(&init_net, NULL);
4874 register_netdevice_notifier(&bond_netdev_notifier);
4878 bond_destroy_debugfs();
4879 bond_netlink_fini();
4881 unregister_pernet_subsys(&bond_net_ops);
4886 static void __exit bonding_exit(void)
4888 unregister_netdevice_notifier(&bond_netdev_notifier);
4890 bond_destroy_debugfs();
4892 bond_netlink_fini();
4893 unregister_pernet_subsys(&bond_net_ops);
4895 #ifdef CONFIG_NET_POLL_CONTROLLER
4896 /* Make sure we don't have an imbalance on our netpoll blocking */
4897 WARN_ON(atomic_read(&netpoll_block_tx));
4901 module_init(bonding_init);
4902 module_exit(bonding_exit);
4903 MODULE_LICENSE("GPL");
4904 MODULE_VERSION(DRV_VERSION);
4905 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4906 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");