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/bonding.h>
81 #include <net/bond_3ad.h>
82 #include <net/bond_alb.h>
84 #include "bonding_priv.h"
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
96 static int use_carrier = 1;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *arp_all_targets;
108 static char *fail_over_mac;
109 static int all_slaves_active;
110 static struct bond_params bonding_defaults;
111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 static int packets_per_slave = 1;
113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125 module_param(miimon, int, 0);
126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127 module_param(updelay, int, 0);
128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129 module_param(downdelay, int, 0);
130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 module_param(use_carrier, int, 0);
133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135 module_param(mode, charp, 0);
136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140 module_param(primary, charp, 0);
141 MODULE_PARM_DESC(primary, "Primary network device to use");
142 module_param(primary_reselect, charp, 0);
143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
148 "2 for only on active slave "
150 module_param(lacp_rate, charp, 0);
151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153 module_param(ad_select, charp, 0);
154 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
157 module_param(min_links, int, 0);
158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160 module_param(xmit_hash_policy, charp, 0);
161 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165 module_param(arp_interval, int, 0);
166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167 module_param_array(arp_ip_target, charp, NULL, 0);
168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169 module_param(arp_validate, charp, 0);
170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173 module_param(arp_all_targets, charp, 0);
174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac, charp, 0);
176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179 module_param(all_slaves_active, int, 0);
180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183 module_param(resend_igmp, int, 0);
184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 module_param(packets_per_slave, int, 0);
187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190 module_param(lp_interval, uint, 0);
191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
195 /*----------------------------- Global variables ----------------------------*/
197 #ifdef CONFIG_NET_POLL_CONTROLLER
198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
201 unsigned int bond_net_id __read_mostly;
203 /*-------------------------- Forward declarations ---------------------------*/
205 static int bond_init(struct net_device *bond_dev);
206 static void bond_uninit(struct net_device *bond_dev);
207 static void bond_get_stats(struct net_device *bond_dev,
208 struct rtnl_link_stats64 *stats);
209 static void bond_slave_arr_handler(struct work_struct *work);
210 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
212 static void bond_netdev_notify_work(struct work_struct *work);
214 /*---------------------------- General routines -----------------------------*/
216 const char *bond_mode_name(int mode)
218 static const char *names[] = {
219 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
223 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
224 [BOND_MODE_TLB] = "transmit load balancing",
225 [BOND_MODE_ALB] = "adaptive load balancing",
228 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
234 /*---------------------------------- VLAN -----------------------------------*/
237 * bond_dev_queue_xmit - Prepare skb for xmit.
239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
244 struct net_device *slave_dev)
246 skb->dev = slave_dev;
248 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
249 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
250 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
252 if (unlikely(netpoll_tx_running(bond->dev)))
253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
258 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
259 * We don't protect the slave list iteration with a lock because:
260 * a. This operation is performed in IOCTL context,
261 * b. The operation is protected by the RTNL semaphore in the 8021q code,
262 * c. Holding a lock with BH disabled while directly calling a base driver
263 * entry point is generally a BAD idea.
265 * The design of synchronization/protection for this operation in the 8021q
266 * module is good for one or more VLAN devices over a single physical device
267 * and cannot be extended for a teaming solution like bonding, so there is a
268 * potential race condition here where a net device from the vlan group might
269 * be referenced (either by a base driver or the 8021q code) while it is being
270 * removed from the system. However, it turns out we're not making matters
271 * worse, and if it works for regular VLAN usage it will work here too.
275 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
276 * @bond_dev: bonding net device that got called
277 * @vid: vlan id being added
279 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
280 __be16 proto, u16 vid)
282 struct bonding *bond = netdev_priv(bond_dev);
283 struct slave *slave, *rollback_slave;
284 struct list_head *iter;
287 bond_for_each_slave(bond, slave, iter) {
288 res = vlan_vid_add(slave->dev, proto, vid);
296 /* unwind to the slave that failed */
297 bond_for_each_slave(bond, rollback_slave, iter) {
298 if (rollback_slave == slave)
301 vlan_vid_del(rollback_slave->dev, proto, vid);
308 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
309 * @bond_dev: bonding net device that got called
310 * @vid: vlan id being removed
312 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
313 __be16 proto, u16 vid)
315 struct bonding *bond = netdev_priv(bond_dev);
316 struct list_head *iter;
319 bond_for_each_slave(bond, slave, iter)
320 vlan_vid_del(slave->dev, proto, vid);
322 if (bond_is_lb(bond))
323 bond_alb_clear_vlan(bond, vid);
328 /*------------------------------- Link status -------------------------------*/
330 /* Set the carrier state for the master according to the state of its
331 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
332 * do special 802.3ad magic.
334 * Returns zero if carrier state does not change, nonzero if it does.
336 int bond_set_carrier(struct bonding *bond)
338 struct list_head *iter;
341 if (!bond_has_slaves(bond))
344 if (BOND_MODE(bond) == BOND_MODE_8023AD)
345 return bond_3ad_set_carrier(bond);
347 bond_for_each_slave(bond, slave, iter) {
348 if (slave->link == BOND_LINK_UP) {
349 if (!netif_carrier_ok(bond->dev)) {
350 netif_carrier_on(bond->dev);
358 if (netif_carrier_ok(bond->dev)) {
359 netif_carrier_off(bond->dev);
365 /* Get link speed and duplex from the slave's base driver
366 * using ethtool. If for some reason the call fails or the
367 * values are invalid, set speed and duplex to -1,
368 * and return. Return 1 if speed or duplex settings are
369 * UNKNOWN; 0 otherwise.
371 static int bond_update_speed_duplex(struct slave *slave)
373 struct net_device *slave_dev = slave->dev;
374 struct ethtool_link_ksettings ecmd;
377 slave->speed = SPEED_UNKNOWN;
378 slave->duplex = DUPLEX_UNKNOWN;
380 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
383 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
385 switch (ecmd.base.duplex) {
393 slave->speed = ecmd.base.speed;
394 slave->duplex = ecmd.base.duplex;
399 const char *bond_slave_link_status(s8 link)
415 /* if <dev> supports MII link status reporting, check its link status.
417 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
418 * depending upon the setting of the use_carrier parameter.
420 * Return either BMSR_LSTATUS, meaning that the link is up (or we
421 * can't tell and just pretend it is), or 0, meaning that the link is
424 * If reporting is non-zero, instead of faking link up, return -1 if
425 * both ETHTOOL and MII ioctls fail (meaning the device does not
426 * support them). If use_carrier is set, return whatever it says.
427 * It'd be nice if there was a good way to tell if a driver supports
428 * netif_carrier, but there really isn't.
430 static int bond_check_dev_link(struct bonding *bond,
431 struct net_device *slave_dev, int reporting)
433 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
434 int (*ioctl)(struct net_device *, struct ifreq *, int);
436 struct mii_ioctl_data *mii;
438 if (!reporting && !netif_running(slave_dev))
441 if (bond->params.use_carrier)
442 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
444 /* Try to get link status using Ethtool first. */
445 if (slave_dev->ethtool_ops->get_link)
446 return slave_dev->ethtool_ops->get_link(slave_dev) ?
449 /* Ethtool can't be used, fallback to MII ioctls. */
450 ioctl = slave_ops->ndo_do_ioctl;
452 /* TODO: set pointer to correct ioctl on a per team member
453 * bases to make this more efficient. that is, once
454 * we determine the correct ioctl, we will always
455 * call it and not the others for that team
459 /* We cannot assume that SIOCGMIIPHY will also read a
460 * register; not all network drivers (e.g., e100)
464 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
465 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
467 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
468 mii->reg_num = MII_BMSR;
469 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
470 return mii->val_out & BMSR_LSTATUS;
474 /* If reporting, report that either there's no dev->do_ioctl,
475 * or both SIOCGMIIREG and get_link failed (meaning that we
476 * cannot report link status). If not reporting, pretend
479 return reporting ? -1 : BMSR_LSTATUS;
482 /*----------------------------- Multicast list ------------------------------*/
484 /* Push the promiscuity flag down to appropriate slaves */
485 static int bond_set_promiscuity(struct bonding *bond, int inc)
487 struct list_head *iter;
490 if (bond_uses_primary(bond)) {
491 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
494 err = dev_set_promiscuity(curr_active->dev, inc);
498 bond_for_each_slave(bond, slave, iter) {
499 err = dev_set_promiscuity(slave->dev, inc);
507 /* Push the allmulti flag down to all slaves */
508 static int bond_set_allmulti(struct bonding *bond, int inc)
510 struct list_head *iter;
513 if (bond_uses_primary(bond)) {
514 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
517 err = dev_set_allmulti(curr_active->dev, inc);
521 bond_for_each_slave(bond, slave, iter) {
522 err = dev_set_allmulti(slave->dev, inc);
530 /* Retrieve the list of registered multicast addresses for the bonding
531 * device and retransmit an IGMP JOIN request to the current active
534 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
536 struct bonding *bond = container_of(work, struct bonding,
539 if (!rtnl_trylock()) {
540 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
543 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
545 if (bond->igmp_retrans > 1) {
546 bond->igmp_retrans--;
547 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
552 /* Flush bond's hardware addresses from slave */
553 static void bond_hw_addr_flush(struct net_device *bond_dev,
554 struct net_device *slave_dev)
556 struct bonding *bond = netdev_priv(bond_dev);
558 dev_uc_unsync(slave_dev, bond_dev);
559 dev_mc_unsync(slave_dev, bond_dev);
561 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
562 /* del lacpdu mc addr from mc list */
563 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
565 dev_mc_del(slave_dev, lacpdu_multicast);
569 /*--------------------------- Active slave change ---------------------------*/
571 /* Update the hardware address list and promisc/allmulti for the new and
572 * old active slaves (if any). Modes that are not using primary keep all
573 * slaves up date at all times; only the modes that use primary need to call
574 * this function to swap these settings during a failover.
576 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
577 struct slave *old_active)
580 if (bond->dev->flags & IFF_PROMISC)
581 dev_set_promiscuity(old_active->dev, -1);
583 if (bond->dev->flags & IFF_ALLMULTI)
584 dev_set_allmulti(old_active->dev, -1);
586 bond_hw_addr_flush(bond->dev, old_active->dev);
590 /* FIXME: Signal errors upstream. */
591 if (bond->dev->flags & IFF_PROMISC)
592 dev_set_promiscuity(new_active->dev, 1);
594 if (bond->dev->flags & IFF_ALLMULTI)
595 dev_set_allmulti(new_active->dev, 1);
597 netif_addr_lock_bh(bond->dev);
598 dev_uc_sync(new_active->dev, bond->dev);
599 dev_mc_sync(new_active->dev, bond->dev);
600 netif_addr_unlock_bh(bond->dev);
605 * bond_set_dev_addr - clone slave's address to bond
606 * @bond_dev: bond net device
607 * @slave_dev: slave net device
609 * Should be called with RTNL held.
611 static int bond_set_dev_addr(struct net_device *bond_dev,
612 struct net_device *slave_dev)
616 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
617 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
618 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
622 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
623 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
624 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
628 static struct slave *bond_get_old_active(struct bonding *bond,
629 struct slave *new_active)
632 struct list_head *iter;
634 bond_for_each_slave(bond, slave, iter) {
635 if (slave == new_active)
638 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
645 /* bond_do_fail_over_mac
647 * Perform special MAC address swapping for fail_over_mac settings
651 static void bond_do_fail_over_mac(struct bonding *bond,
652 struct slave *new_active,
653 struct slave *old_active)
655 u8 tmp_mac[MAX_ADDR_LEN];
656 struct sockaddr_storage ss;
659 switch (bond->params.fail_over_mac) {
660 case BOND_FOM_ACTIVE:
662 rv = bond_set_dev_addr(bond->dev, new_active->dev);
664 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
665 -rv, bond->dev->name);
668 case BOND_FOM_FOLLOW:
669 /* if new_active && old_active, swap them
670 * if just old_active, do nothing (going to no active slave)
671 * if just new_active, set new_active to bond's MAC
677 old_active = bond_get_old_active(bond, new_active);
680 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
681 new_active->dev->addr_len);
682 bond_hw_addr_copy(ss.__data,
683 old_active->dev->dev_addr,
684 old_active->dev->addr_len);
685 ss.ss_family = new_active->dev->type;
687 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
688 bond->dev->addr_len);
689 ss.ss_family = bond->dev->type;
692 rv = dev_set_mac_address(new_active->dev,
693 (struct sockaddr *)&ss, NULL);
695 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
696 -rv, new_active->dev->name);
703 bond_hw_addr_copy(ss.__data, tmp_mac,
704 new_active->dev->addr_len);
705 ss.ss_family = old_active->dev->type;
707 rv = dev_set_mac_address(old_active->dev,
708 (struct sockaddr *)&ss, NULL);
710 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
711 -rv, new_active->dev->name);
715 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
716 bond->params.fail_over_mac);
722 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
724 struct slave *prim = rtnl_dereference(bond->primary_slave);
725 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
727 if (!prim || prim->link != BOND_LINK_UP) {
728 if (!curr || curr->link != BOND_LINK_UP)
733 if (bond->force_primary) {
734 bond->force_primary = false;
738 if (!curr || curr->link != BOND_LINK_UP)
741 /* At this point, prim and curr are both up */
742 switch (bond->params.primary_reselect) {
743 case BOND_PRI_RESELECT_ALWAYS:
745 case BOND_PRI_RESELECT_BETTER:
746 if (prim->speed < curr->speed)
748 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
751 case BOND_PRI_RESELECT_FAILURE:
754 netdev_err(bond->dev, "impossible primary_reselect %d\n",
755 bond->params.primary_reselect);
761 * bond_find_best_slave - select the best available slave to be the active one
762 * @bond: our bonding struct
764 static struct slave *bond_find_best_slave(struct bonding *bond)
766 struct slave *slave, *bestslave = NULL;
767 struct list_head *iter;
768 int mintime = bond->params.updelay;
770 slave = bond_choose_primary_or_current(bond);
774 bond_for_each_slave(bond, slave, iter) {
775 if (slave->link == BOND_LINK_UP)
777 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
778 slave->delay < mintime) {
779 mintime = slave->delay;
787 static bool bond_should_notify_peers(struct bonding *bond)
792 slave = rcu_dereference(bond->curr_active_slave);
795 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
796 slave ? slave->dev->name : "NULL");
798 if (!slave || !bond->send_peer_notif ||
799 !netif_carrier_ok(bond->dev) ||
800 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
807 * change_active_interface - change the active slave into the specified one
808 * @bond: our bonding struct
809 * @new: the new slave to make the active one
811 * Set the new slave to the bond's settings and unset them on the old
813 * Setting include flags, mc-list, promiscuity, allmulti, etc.
815 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
816 * because it is apparently the best available slave we have, even though its
817 * updelay hasn't timed out yet.
819 * Caller must hold RTNL.
821 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
823 struct slave *old_active;
827 old_active = rtnl_dereference(bond->curr_active_slave);
829 if (old_active == new_active)
833 new_active->last_link_up = jiffies;
835 if (new_active->link == BOND_LINK_BACK) {
836 if (bond_uses_primary(bond)) {
837 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
838 new_active->dev->name,
839 (bond->params.updelay - new_active->delay) * bond->params.miimon);
842 new_active->delay = 0;
843 bond_set_slave_link_state(new_active, BOND_LINK_UP,
844 BOND_SLAVE_NOTIFY_NOW);
846 if (BOND_MODE(bond) == BOND_MODE_8023AD)
847 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
849 if (bond_is_lb(bond))
850 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
852 if (bond_uses_primary(bond)) {
853 netdev_info(bond->dev, "making interface %s the new active one\n",
854 new_active->dev->name);
859 if (bond_uses_primary(bond))
860 bond_hw_addr_swap(bond, new_active, old_active);
862 if (bond_is_lb(bond)) {
863 bond_alb_handle_active_change(bond, new_active);
865 bond_set_slave_inactive_flags(old_active,
866 BOND_SLAVE_NOTIFY_NOW);
868 bond_set_slave_active_flags(new_active,
869 BOND_SLAVE_NOTIFY_NOW);
871 rcu_assign_pointer(bond->curr_active_slave, new_active);
874 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
876 bond_set_slave_inactive_flags(old_active,
877 BOND_SLAVE_NOTIFY_NOW);
880 bool should_notify_peers = false;
882 bond_set_slave_active_flags(new_active,
883 BOND_SLAVE_NOTIFY_NOW);
885 if (bond->params.fail_over_mac)
886 bond_do_fail_over_mac(bond, new_active,
889 if (netif_running(bond->dev)) {
890 bond->send_peer_notif =
891 bond->params.num_peer_notif;
892 should_notify_peers =
893 bond_should_notify_peers(bond);
896 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
897 if (should_notify_peers)
898 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
903 /* resend IGMP joins since active slave has changed or
904 * all were sent on curr_active_slave.
905 * resend only if bond is brought up with the affected
906 * bonding modes and the retransmission is enabled
908 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
909 ((bond_uses_primary(bond) && new_active) ||
910 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
911 bond->igmp_retrans = bond->params.resend_igmp;
912 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
917 * bond_select_active_slave - select a new active slave, if needed
918 * @bond: our bonding struct
920 * This functions should be called when one of the following occurs:
921 * - The old curr_active_slave has been released or lost its link.
922 * - The primary_slave has got its link back.
923 * - A slave has got its link back and there's no old curr_active_slave.
925 * Caller must hold RTNL.
927 void bond_select_active_slave(struct bonding *bond)
929 struct slave *best_slave;
934 best_slave = bond_find_best_slave(bond);
935 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
936 bond_change_active_slave(bond, best_slave);
937 rv = bond_set_carrier(bond);
941 if (netif_carrier_ok(bond->dev))
942 netdev_info(bond->dev, "first active interface up!\n");
944 netdev_info(bond->dev, "now running without any active interface!\n");
948 #ifdef CONFIG_NET_POLL_CONTROLLER
949 static inline int slave_enable_netpoll(struct slave *slave)
954 np = kzalloc(sizeof(*np), GFP_KERNEL);
959 err = __netpoll_setup(np, slave->dev);
968 static inline void slave_disable_netpoll(struct slave *slave)
970 struct netpoll *np = slave->np;
980 static void bond_poll_controller(struct net_device *bond_dev)
982 struct bonding *bond = netdev_priv(bond_dev);
983 struct slave *slave = NULL;
984 struct list_head *iter;
985 struct ad_info ad_info;
987 if (BOND_MODE(bond) == BOND_MODE_8023AD)
988 if (bond_3ad_get_active_agg_info(bond, &ad_info))
991 bond_for_each_slave_rcu(bond, slave, iter) {
992 if (!bond_slave_is_up(slave))
995 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
996 struct aggregator *agg =
997 SLAVE_AD_INFO(slave)->port.aggregator;
1000 agg->aggregator_identifier != ad_info.aggregator_id)
1004 netpoll_poll_dev(slave->dev);
1008 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1010 struct bonding *bond = netdev_priv(bond_dev);
1011 struct list_head *iter;
1012 struct slave *slave;
1014 bond_for_each_slave(bond, slave, iter)
1015 if (bond_slave_is_up(slave))
1016 slave_disable_netpoll(slave);
1019 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1021 struct bonding *bond = netdev_priv(dev);
1022 struct list_head *iter;
1023 struct slave *slave;
1026 bond_for_each_slave(bond, slave, iter) {
1027 err = slave_enable_netpoll(slave);
1029 bond_netpoll_cleanup(dev);
1036 static inline int slave_enable_netpoll(struct slave *slave)
1040 static inline void slave_disable_netpoll(struct slave *slave)
1043 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1048 /*---------------------------------- IOCTL ----------------------------------*/
1050 static netdev_features_t bond_fix_features(struct net_device *dev,
1051 netdev_features_t features)
1053 struct bonding *bond = netdev_priv(dev);
1054 struct list_head *iter;
1055 netdev_features_t mask;
1056 struct slave *slave;
1060 features &= ~NETIF_F_ONE_FOR_ALL;
1061 features |= NETIF_F_ALL_FOR_ALL;
1063 bond_for_each_slave(bond, slave, iter) {
1064 features = netdev_increment_features(features,
1065 slave->dev->features,
1068 features = netdev_add_tso_features(features, mask);
1073 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1074 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1075 NETIF_F_HIGHDMA | NETIF_F_LRO)
1077 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1078 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1080 static void bond_compute_features(struct bonding *bond)
1082 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1083 IFF_XMIT_DST_RELEASE_PERM;
1084 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1085 netdev_features_t enc_features = BOND_ENC_FEATURES;
1086 struct net_device *bond_dev = bond->dev;
1087 struct list_head *iter;
1088 struct slave *slave;
1089 unsigned short max_hard_header_len = ETH_HLEN;
1090 unsigned int gso_max_size = GSO_MAX_SIZE;
1091 u16 gso_max_segs = GSO_MAX_SEGS;
1093 if (!bond_has_slaves(bond))
1095 vlan_features &= NETIF_F_ALL_FOR_ALL;
1097 bond_for_each_slave(bond, slave, iter) {
1098 vlan_features = netdev_increment_features(vlan_features,
1099 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1101 enc_features = netdev_increment_features(enc_features,
1102 slave->dev->hw_enc_features,
1104 dst_release_flag &= slave->dev->priv_flags;
1105 if (slave->dev->hard_header_len > max_hard_header_len)
1106 max_hard_header_len = slave->dev->hard_header_len;
1108 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1109 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1111 bond_dev->hard_header_len = max_hard_header_len;
1114 bond_dev->vlan_features = vlan_features;
1115 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1117 bond_dev->gso_max_segs = gso_max_segs;
1118 netif_set_gso_max_size(bond_dev, gso_max_size);
1120 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1121 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1122 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1123 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1125 netdev_change_features(bond_dev);
1128 static void bond_setup_by_slave(struct net_device *bond_dev,
1129 struct net_device *slave_dev)
1131 bond_dev->header_ops = slave_dev->header_ops;
1133 bond_dev->type = slave_dev->type;
1134 bond_dev->hard_header_len = slave_dev->hard_header_len;
1135 bond_dev->addr_len = slave_dev->addr_len;
1137 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1138 slave_dev->addr_len);
1141 /* On bonding slaves other than the currently active slave, suppress
1142 * duplicates except for alb non-mcast/bcast.
1144 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1145 struct slave *slave,
1146 struct bonding *bond)
1148 if (bond_is_slave_inactive(slave)) {
1149 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1150 skb->pkt_type != PACKET_BROADCAST &&
1151 skb->pkt_type != PACKET_MULTICAST)
1158 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1160 struct sk_buff *skb = *pskb;
1161 struct slave *slave;
1162 struct bonding *bond;
1163 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1165 int ret = RX_HANDLER_ANOTHER;
1167 skb = skb_share_check(skb, GFP_ATOMIC);
1169 return RX_HANDLER_CONSUMED;
1173 slave = bond_slave_get_rcu(skb->dev);
1176 recv_probe = READ_ONCE(bond->recv_probe);
1178 ret = recv_probe(skb, bond, slave);
1179 if (ret == RX_HANDLER_CONSUMED) {
1186 * For packets determined by bond_should_deliver_exact_match() call to
1187 * be suppressed we want to make an exception for link-local packets.
1188 * This is necessary for e.g. LLDP daemons to be able to monitor
1189 * inactive slave links without being forced to bind to them
1192 * At the same time, packets that are passed to the bonding master
1193 * (including link-local ones) can have their originating interface
1194 * determined via PACKET_ORIGDEV socket option.
1196 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1197 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1198 return RX_HANDLER_PASS;
1199 return RX_HANDLER_EXACT;
1202 skb->dev = bond->dev;
1204 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1205 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1206 skb->pkt_type == PACKET_HOST) {
1208 if (unlikely(skb_cow_head(skb,
1209 skb->data - skb_mac_header(skb)))) {
1211 return RX_HANDLER_CONSUMED;
1213 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1214 bond->dev->addr_len);
1220 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1222 switch (BOND_MODE(bond)) {
1223 case BOND_MODE_ROUNDROBIN:
1224 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1225 case BOND_MODE_ACTIVEBACKUP:
1226 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1227 case BOND_MODE_BROADCAST:
1228 return NETDEV_LAG_TX_TYPE_BROADCAST;
1230 case BOND_MODE_8023AD:
1231 return NETDEV_LAG_TX_TYPE_HASH;
1233 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1237 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1238 enum netdev_lag_tx_type type)
1240 if (type != NETDEV_LAG_TX_TYPE_HASH)
1241 return NETDEV_LAG_HASH_NONE;
1243 switch (bond->params.xmit_policy) {
1244 case BOND_XMIT_POLICY_LAYER2:
1245 return NETDEV_LAG_HASH_L2;
1246 case BOND_XMIT_POLICY_LAYER34:
1247 return NETDEV_LAG_HASH_L34;
1248 case BOND_XMIT_POLICY_LAYER23:
1249 return NETDEV_LAG_HASH_L23;
1250 case BOND_XMIT_POLICY_ENCAP23:
1251 return NETDEV_LAG_HASH_E23;
1252 case BOND_XMIT_POLICY_ENCAP34:
1253 return NETDEV_LAG_HASH_E34;
1255 return NETDEV_LAG_HASH_UNKNOWN;
1259 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1260 struct netlink_ext_ack *extack)
1262 struct netdev_lag_upper_info lag_upper_info;
1263 enum netdev_lag_tx_type type;
1265 type = bond_lag_tx_type(bond);
1266 lag_upper_info.tx_type = type;
1267 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1269 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1270 &lag_upper_info, extack);
1273 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1275 netdev_upper_dev_unlink(slave->dev, bond->dev);
1276 slave->dev->flags &= ~IFF_SLAVE;
1279 static struct slave *bond_alloc_slave(struct bonding *bond)
1281 struct slave *slave = NULL;
1283 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1287 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1288 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1290 if (!SLAVE_AD_INFO(slave)) {
1295 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1300 static void bond_free_slave(struct slave *slave)
1302 struct bonding *bond = bond_get_bond_by_slave(slave);
1304 cancel_delayed_work_sync(&slave->notify_work);
1305 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1306 kfree(SLAVE_AD_INFO(slave));
1311 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1313 info->bond_mode = BOND_MODE(bond);
1314 info->miimon = bond->params.miimon;
1315 info->num_slaves = bond->slave_cnt;
1318 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1320 strcpy(info->slave_name, slave->dev->name);
1321 info->link = slave->link;
1322 info->state = bond_slave_state(slave);
1323 info->link_failure_count = slave->link_failure_count;
1326 static void bond_netdev_notify_work(struct work_struct *_work)
1328 struct slave *slave = container_of(_work, struct slave,
1331 if (rtnl_trylock()) {
1332 struct netdev_bonding_info binfo;
1334 bond_fill_ifslave(slave, &binfo.slave);
1335 bond_fill_ifbond(slave->bond, &binfo.master);
1336 netdev_bonding_info_change(slave->dev, &binfo);
1339 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1343 void bond_queue_slave_event(struct slave *slave)
1345 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1348 void bond_lower_state_changed(struct slave *slave)
1350 struct netdev_lag_lower_state_info info;
1352 info.link_up = slave->link == BOND_LINK_UP ||
1353 slave->link == BOND_LINK_FAIL;
1354 info.tx_enabled = bond_is_active_slave(slave);
1355 netdev_lower_state_changed(slave->dev, &info);
1358 /* enslave device <slave> to bond device <master> */
1359 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1360 struct netlink_ext_ack *extack)
1362 struct bonding *bond = netdev_priv(bond_dev);
1363 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1364 struct slave *new_slave = NULL, *prev_slave;
1365 struct sockaddr_storage ss;
1369 if (!bond->params.use_carrier &&
1370 slave_dev->ethtool_ops->get_link == NULL &&
1371 slave_ops->ndo_do_ioctl == NULL) {
1372 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1376 /* already in-use? */
1377 if (netdev_is_rx_handler_busy(slave_dev)) {
1378 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1379 netdev_err(bond_dev,
1380 "Error: Device is in use and cannot be enslaved\n");
1384 if (bond_dev == slave_dev) {
1385 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1386 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1390 /* vlan challenged mutual exclusion */
1391 /* no need to lock since we're protected by rtnl_lock */
1392 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1393 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1395 if (vlan_uses_dev(bond_dev)) {
1396 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1397 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1398 slave_dev->name, bond_dev->name);
1401 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1402 slave_dev->name, slave_dev->name,
1406 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1410 /* Old ifenslave binaries are no longer supported. These can
1411 * be identified with moderate accuracy by the state of the slave:
1412 * the current ifenslave will set the interface down prior to
1413 * enslaving it; the old ifenslave will not.
1415 if (slave_dev->flags & IFF_UP) {
1416 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1417 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1422 /* set bonding device ether type by slave - bonding netdevices are
1423 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1424 * there is a need to override some of the type dependent attribs/funcs.
1426 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1427 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1429 if (!bond_has_slaves(bond)) {
1430 if (bond_dev->type != slave_dev->type) {
1431 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1432 bond_dev->type, slave_dev->type);
1434 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1436 res = notifier_to_errno(res);
1438 netdev_err(bond_dev, "refused to change device type\n");
1442 /* Flush unicast and multicast addresses */
1443 dev_uc_flush(bond_dev);
1444 dev_mc_flush(bond_dev);
1446 if (slave_dev->type != ARPHRD_ETHER)
1447 bond_setup_by_slave(bond_dev, slave_dev);
1449 ether_setup(bond_dev);
1450 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1453 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1456 } else if (bond_dev->type != slave_dev->type) {
1457 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1458 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1459 slave_dev->name, slave_dev->type, bond_dev->type);
1463 if (slave_dev->type == ARPHRD_INFINIBAND &&
1464 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1465 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1466 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1469 goto err_undo_flags;
1472 if (!slave_ops->ndo_set_mac_address ||
1473 slave_dev->type == ARPHRD_INFINIBAND) {
1474 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1475 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1476 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1477 if (!bond_has_slaves(bond)) {
1478 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1479 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1481 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1482 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");
1484 goto err_undo_flags;
1489 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1491 /* If this is the first slave, then we need to set the master's hardware
1492 * address to be the same as the slave's.
1494 if (!bond_has_slaves(bond) &&
1495 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1496 res = bond_set_dev_addr(bond->dev, slave_dev);
1498 goto err_undo_flags;
1501 new_slave = bond_alloc_slave(bond);
1504 goto err_undo_flags;
1507 new_slave->bond = bond;
1508 new_slave->dev = slave_dev;
1509 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1510 * is set via sysfs or module option if desired.
1512 new_slave->queue_id = 0;
1514 /* Save slave's original mtu and then set it to match the bond */
1515 new_slave->original_mtu = slave_dev->mtu;
1516 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1518 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1522 /* Save slave's original ("permanent") mac address for modes
1523 * that need it, and for restoring it upon release, and then
1524 * set it to the master's address
1526 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1527 slave_dev->addr_len);
1529 if (!bond->params.fail_over_mac ||
1530 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1531 /* Set slave to master's mac address. The application already
1532 * set the master's mac address to that of the first slave
1534 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1535 ss.ss_family = slave_dev->type;
1536 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1539 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1540 goto err_restore_mtu;
1544 /* set slave flag before open to prevent IPv6 addrconf */
1545 slave_dev->flags |= IFF_SLAVE;
1547 /* open the slave since the application closed it */
1548 res = dev_open(slave_dev, extack);
1550 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1551 goto err_restore_mac;
1554 slave_dev->priv_flags |= IFF_BONDING;
1555 /* initialize slave stats */
1556 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1558 if (bond_is_lb(bond)) {
1559 /* bond_alb_init_slave() must be called before all other stages since
1560 * it might fail and we do not want to have to undo everything
1562 res = bond_alb_init_slave(bond, new_slave);
1567 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1569 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1574 prev_slave = bond_last_slave(bond);
1576 new_slave->delay = 0;
1577 new_slave->link_failure_count = 0;
1579 if (bond_update_speed_duplex(new_slave) &&
1580 bond_needs_speed_duplex(bond))
1581 new_slave->link = BOND_LINK_DOWN;
1583 new_slave->last_rx = jiffies -
1584 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1585 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1586 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1588 if (bond->params.miimon && !bond->params.use_carrier) {
1589 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1591 if ((link_reporting == -1) && !bond->params.arp_interval) {
1592 /* miimon is set but a bonded network driver
1593 * does not support ETHTOOL/MII and
1594 * arp_interval is not set. Note: if
1595 * use_carrier is enabled, we will never go
1596 * here (because netif_carrier is always
1597 * supported); thus, we don't need to change
1598 * the messages for netif_carrier.
1600 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",
1602 } else if (link_reporting == -1) {
1603 /* unable get link status using mii/ethtool */
1604 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",
1609 /* check for initial state */
1610 new_slave->link = BOND_LINK_NOCHANGE;
1611 if (bond->params.miimon) {
1612 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1613 if (bond->params.updelay) {
1614 bond_set_slave_link_state(new_slave,
1616 BOND_SLAVE_NOTIFY_NOW);
1617 new_slave->delay = bond->params.updelay;
1619 bond_set_slave_link_state(new_slave,
1621 BOND_SLAVE_NOTIFY_NOW);
1624 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1625 BOND_SLAVE_NOTIFY_NOW);
1627 } else if (bond->params.arp_interval) {
1628 bond_set_slave_link_state(new_slave,
1629 (netif_carrier_ok(slave_dev) ?
1630 BOND_LINK_UP : BOND_LINK_DOWN),
1631 BOND_SLAVE_NOTIFY_NOW);
1633 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1634 BOND_SLAVE_NOTIFY_NOW);
1637 if (new_slave->link != BOND_LINK_DOWN)
1638 new_slave->last_link_up = jiffies;
1639 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1640 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1641 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1643 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1644 /* if there is a primary slave, remember it */
1645 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1646 rcu_assign_pointer(bond->primary_slave, new_slave);
1647 bond->force_primary = true;
1651 switch (BOND_MODE(bond)) {
1652 case BOND_MODE_ACTIVEBACKUP:
1653 bond_set_slave_inactive_flags(new_slave,
1654 BOND_SLAVE_NOTIFY_NOW);
1656 case BOND_MODE_8023AD:
1657 /* in 802.3ad mode, the internal mechanism
1658 * will activate the slaves in the selected
1661 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1662 /* if this is the first slave */
1664 SLAVE_AD_INFO(new_slave)->id = 1;
1665 /* Initialize AD with the number of times that the AD timer is called in 1 second
1666 * can be called only after the mac address of the bond is set
1668 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1670 SLAVE_AD_INFO(new_slave)->id =
1671 SLAVE_AD_INFO(prev_slave)->id + 1;
1674 bond_3ad_bind_slave(new_slave);
1678 bond_set_active_slave(new_slave);
1679 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1682 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1684 /* always active in trunk mode */
1685 bond_set_active_slave(new_slave);
1687 /* In trunking mode there is little meaning to curr_active_slave
1688 * anyway (it holds no special properties of the bond device),
1689 * so we can change it without calling change_active_interface()
1691 if (!rcu_access_pointer(bond->curr_active_slave) &&
1692 new_slave->link == BOND_LINK_UP)
1693 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1696 } /* switch(bond_mode) */
1698 #ifdef CONFIG_NET_POLL_CONTROLLER
1699 if (bond->dev->npinfo) {
1700 if (slave_enable_netpoll(new_slave)) {
1701 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1708 if (!(bond_dev->features & NETIF_F_LRO))
1709 dev_disable_lro(slave_dev);
1711 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1714 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1718 res = bond_master_upper_dev_link(bond, new_slave, extack);
1720 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1721 goto err_unregister;
1724 res = bond_sysfs_slave_add(new_slave);
1726 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1727 goto err_upper_unlink;
1730 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1732 /* If the mode uses primary, then the following is handled by
1733 * bond_change_active_slave().
1735 if (!bond_uses_primary(bond)) {
1736 /* set promiscuity level to new slave */
1737 if (bond_dev->flags & IFF_PROMISC) {
1738 res = dev_set_promiscuity(slave_dev, 1);
1743 /* set allmulti level to new slave */
1744 if (bond_dev->flags & IFF_ALLMULTI) {
1745 res = dev_set_allmulti(slave_dev, 1);
1747 if (bond_dev->flags & IFF_PROMISC)
1748 dev_set_promiscuity(slave_dev, -1);
1753 netif_addr_lock_bh(bond_dev);
1754 dev_mc_sync_multiple(slave_dev, bond_dev);
1755 dev_uc_sync_multiple(slave_dev, bond_dev);
1756 netif_addr_unlock_bh(bond_dev);
1758 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1759 /* add lacpdu mc addr to mc list */
1760 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1762 dev_mc_add(slave_dev, lacpdu_multicast);
1767 bond_compute_features(bond);
1768 bond_set_carrier(bond);
1770 if (bond_uses_primary(bond)) {
1772 bond_select_active_slave(bond);
1773 unblock_netpoll_tx();
1776 if (bond_mode_can_use_xmit_hash(bond))
1777 bond_update_slave_arr(bond, NULL);
1780 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1782 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1783 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1785 /* enslave is successful */
1786 bond_queue_slave_event(new_slave);
1789 /* Undo stages on error */
1791 bond_sysfs_slave_del(new_slave);
1794 bond_upper_dev_unlink(bond, new_slave);
1797 netdev_rx_handler_unregister(slave_dev);
1800 vlan_vids_del_by_dev(slave_dev, bond_dev);
1801 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1802 RCU_INIT_POINTER(bond->primary_slave, NULL);
1803 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1805 bond_change_active_slave(bond, NULL);
1806 bond_select_active_slave(bond);
1807 unblock_netpoll_tx();
1809 /* either primary_slave or curr_active_slave might've changed */
1811 slave_disable_netpoll(new_slave);
1814 slave_dev->priv_flags &= ~IFF_BONDING;
1815 dev_close(slave_dev);
1818 slave_dev->flags &= ~IFF_SLAVE;
1819 if (!bond->params.fail_over_mac ||
1820 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1821 /* XXX TODO - fom follow mode needs to change master's
1822 * MAC if this slave's MAC is in use by the bond, or at
1823 * least print a warning.
1825 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1826 new_slave->dev->addr_len);
1827 ss.ss_family = slave_dev->type;
1828 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
1832 dev_set_mtu(slave_dev, new_slave->original_mtu);
1835 bond_free_slave(new_slave);
1838 /* Enslave of first slave has failed and we need to fix master's mac */
1839 if (!bond_has_slaves(bond)) {
1840 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1841 slave_dev->dev_addr))
1842 eth_hw_addr_random(bond_dev);
1843 if (bond_dev->type != ARPHRD_ETHER) {
1844 dev_close(bond_dev);
1845 ether_setup(bond_dev);
1846 bond_dev->flags |= IFF_MASTER;
1847 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1854 /* Try to release the slave device <slave> from the bond device <master>
1855 * It is legal to access curr_active_slave without a lock because all the function
1856 * is RTNL-locked. If "all" is true it means that the function is being called
1857 * while destroying a bond interface and all slaves are being released.
1859 * The rules for slave state should be:
1860 * for Active/Backup:
1861 * Active stays on all backups go down
1862 * for Bonded connections:
1863 * The first up interface should be left on and all others downed.
1865 static int __bond_release_one(struct net_device *bond_dev,
1866 struct net_device *slave_dev,
1867 bool all, bool unregister)
1869 struct bonding *bond = netdev_priv(bond_dev);
1870 struct slave *slave, *oldcurrent;
1871 struct sockaddr_storage ss;
1872 int old_flags = bond_dev->flags;
1873 netdev_features_t old_features = bond_dev->features;
1875 /* slave is not a slave or master is not master of this slave */
1876 if (!(slave_dev->flags & IFF_SLAVE) ||
1877 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1878 netdev_dbg(bond_dev, "cannot release %s\n",
1885 slave = bond_get_slave_by_dev(bond, slave_dev);
1887 /* not a slave of this bond */
1888 netdev_info(bond_dev, "%s not enslaved\n",
1890 unblock_netpoll_tx();
1894 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1896 bond_sysfs_slave_del(slave);
1898 /* recompute stats just before removing the slave */
1899 bond_get_stats(bond->dev, &bond->bond_stats);
1901 bond_upper_dev_unlink(bond, slave);
1902 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1903 * for this slave anymore.
1905 netdev_rx_handler_unregister(slave_dev);
1907 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1908 bond_3ad_unbind_slave(slave);
1910 if (bond_mode_can_use_xmit_hash(bond))
1911 bond_update_slave_arr(bond, slave);
1913 netdev_info(bond_dev, "Releasing %s interface %s\n",
1914 bond_is_active_slave(slave) ? "active" : "backup",
1917 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1919 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1921 if (!all && (!bond->params.fail_over_mac ||
1922 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1923 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1924 bond_has_slaves(bond))
1925 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",
1926 slave_dev->name, slave->perm_hwaddr,
1927 bond_dev->name, slave_dev->name);
1930 if (rtnl_dereference(bond->primary_slave) == slave)
1931 RCU_INIT_POINTER(bond->primary_slave, NULL);
1933 if (oldcurrent == slave)
1934 bond_change_active_slave(bond, NULL);
1936 if (bond_is_lb(bond)) {
1937 /* Must be called only after the slave has been
1938 * detached from the list and the curr_active_slave
1939 * has been cleared (if our_slave == old_current),
1940 * but before a new active slave is selected.
1942 bond_alb_deinit_slave(bond, slave);
1946 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1947 } else if (oldcurrent == slave) {
1948 /* Note that we hold RTNL over this sequence, so there
1949 * is no concern that another slave add/remove event
1952 bond_select_active_slave(bond);
1955 if (!bond_has_slaves(bond)) {
1956 bond_set_carrier(bond);
1957 eth_hw_addr_random(bond_dev);
1958 bond->nest_level = SINGLE_DEPTH_NESTING;
1960 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1963 unblock_netpoll_tx();
1967 if (!bond_has_slaves(bond)) {
1968 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1969 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1972 bond_compute_features(bond);
1973 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1974 (old_features & NETIF_F_VLAN_CHALLENGED))
1975 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1976 slave_dev->name, bond_dev->name);
1978 vlan_vids_del_by_dev(slave_dev, bond_dev);
1980 /* If the mode uses primary, then this case was handled above by
1981 * bond_change_active_slave(..., NULL)
1983 if (!bond_uses_primary(bond)) {
1984 /* unset promiscuity level from slave
1985 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1986 * of the IFF_PROMISC flag in the bond_dev, but we need the
1987 * value of that flag before that change, as that was the value
1988 * when this slave was attached, so we cache at the start of the
1989 * function and use it here. Same goes for ALLMULTI below
1991 if (old_flags & IFF_PROMISC)
1992 dev_set_promiscuity(slave_dev, -1);
1994 /* unset allmulti level from slave */
1995 if (old_flags & IFF_ALLMULTI)
1996 dev_set_allmulti(slave_dev, -1);
1998 bond_hw_addr_flush(bond_dev, slave_dev);
2001 slave_disable_netpoll(slave);
2003 /* close slave before restoring its mac address */
2004 dev_close(slave_dev);
2006 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2007 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2008 /* restore original ("permanent") mac address */
2009 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2010 slave->dev->addr_len);
2011 ss.ss_family = slave_dev->type;
2012 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2016 __dev_set_mtu(slave_dev, slave->original_mtu);
2018 dev_set_mtu(slave_dev, slave->original_mtu);
2020 slave_dev->priv_flags &= ~IFF_BONDING;
2022 bond_free_slave(slave);
2027 /* A wrapper used because of ndo_del_link */
2028 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2030 return __bond_release_one(bond_dev, slave_dev, false, false);
2033 /* First release a slave and then destroy the bond if no more slaves are left.
2034 * Must be under rtnl_lock when this function is called.
2036 static int bond_release_and_destroy(struct net_device *bond_dev,
2037 struct net_device *slave_dev)
2039 struct bonding *bond = netdev_priv(bond_dev);
2042 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2043 if (ret == 0 && !bond_has_slaves(bond)) {
2044 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2045 netdev_info(bond_dev, "Destroying bond %s\n",
2047 bond_remove_proc_entry(bond);
2048 unregister_netdevice(bond_dev);
2053 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2055 struct bonding *bond = netdev_priv(bond_dev);
2056 bond_fill_ifbond(bond, info);
2059 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2061 struct bonding *bond = netdev_priv(bond_dev);
2062 struct list_head *iter;
2063 int i = 0, res = -ENODEV;
2064 struct slave *slave;
2066 bond_for_each_slave(bond, slave, iter) {
2067 if (i++ == (int)info->slave_id) {
2069 bond_fill_ifslave(slave, info);
2077 /*-------------------------------- Monitoring -------------------------------*/
2079 /* called with rcu_read_lock() */
2080 static int bond_miimon_inspect(struct bonding *bond)
2082 int link_state, commit = 0;
2083 struct list_head *iter;
2084 struct slave *slave;
2085 bool ignore_updelay;
2087 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2089 bond_for_each_slave_rcu(bond, slave, iter) {
2090 slave->new_link = BOND_LINK_NOCHANGE;
2091 slave->link_new_state = slave->link;
2093 link_state = bond_check_dev_link(bond, slave->dev, 0);
2095 switch (slave->link) {
2100 bond_propose_link_state(slave, BOND_LINK_FAIL);
2102 slave->delay = bond->params.downdelay;
2104 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2106 BOND_MODE_ACTIVEBACKUP) ?
2107 (bond_is_active_slave(slave) ?
2108 "active " : "backup ") : "",
2110 bond->params.downdelay * bond->params.miimon);
2113 case BOND_LINK_FAIL:
2115 /* recovered before downdelay expired */
2116 bond_propose_link_state(slave, BOND_LINK_UP);
2117 slave->last_link_up = jiffies;
2118 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2119 (bond->params.downdelay - slave->delay) *
2120 bond->params.miimon,
2126 if (slave->delay <= 0) {
2127 slave->new_link = BOND_LINK_DOWN;
2135 case BOND_LINK_DOWN:
2139 bond_propose_link_state(slave, BOND_LINK_BACK);
2141 slave->delay = bond->params.updelay;
2144 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2146 ignore_updelay ? 0 :
2147 bond->params.updelay *
2148 bond->params.miimon);
2151 case BOND_LINK_BACK:
2153 bond_propose_link_state(slave, BOND_LINK_DOWN);
2154 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2155 (bond->params.updelay - slave->delay) *
2156 bond->params.miimon,
2165 if (slave->delay <= 0) {
2166 slave->new_link = BOND_LINK_UP;
2168 ignore_updelay = false;
2180 static void bond_miimon_link_change(struct bonding *bond,
2181 struct slave *slave,
2184 switch (BOND_MODE(bond)) {
2185 case BOND_MODE_8023AD:
2186 bond_3ad_handle_link_change(slave, link);
2190 bond_alb_handle_link_change(bond, slave, link);
2193 bond_update_slave_arr(bond, NULL);
2198 static void bond_miimon_commit(struct bonding *bond)
2200 struct list_head *iter;
2201 struct slave *slave, *primary;
2203 bond_for_each_slave(bond, slave, iter) {
2204 switch (slave->new_link) {
2205 case BOND_LINK_NOCHANGE:
2209 if (bond_update_speed_duplex(slave) &&
2210 bond_needs_speed_duplex(bond)) {
2211 slave->link = BOND_LINK_DOWN;
2212 if (net_ratelimit())
2213 netdev_warn(bond->dev,
2214 "failed to get link speed/duplex for %s\n",
2218 bond_set_slave_link_state(slave, BOND_LINK_UP,
2219 BOND_SLAVE_NOTIFY_NOW);
2220 slave->last_link_up = jiffies;
2222 primary = rtnl_dereference(bond->primary_slave);
2223 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2224 /* prevent it from being the active one */
2225 bond_set_backup_slave(slave);
2226 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2227 /* make it immediately active */
2228 bond_set_active_slave(slave);
2229 } else if (slave != primary) {
2230 /* prevent it from being the active one */
2231 bond_set_backup_slave(slave);
2234 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2236 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2237 slave->duplex ? "full" : "half");
2239 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2241 if (!bond->curr_active_slave || slave == primary)
2246 case BOND_LINK_DOWN:
2247 if (slave->link_failure_count < UINT_MAX)
2248 slave->link_failure_count++;
2250 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2251 BOND_SLAVE_NOTIFY_NOW);
2253 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2254 BOND_MODE(bond) == BOND_MODE_8023AD)
2255 bond_set_slave_inactive_flags(slave,
2256 BOND_SLAVE_NOTIFY_NOW);
2258 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2261 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2263 if (slave == rcu_access_pointer(bond->curr_active_slave))
2269 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2270 slave->new_link, slave->dev->name);
2271 slave->new_link = BOND_LINK_NOCHANGE;
2278 bond_select_active_slave(bond);
2279 unblock_netpoll_tx();
2282 bond_set_carrier(bond);
2287 * Really a wrapper that splits the mii monitor into two phases: an
2288 * inspection, then (if inspection indicates something needs to be done)
2289 * an acquisition of appropriate locks followed by a commit phase to
2290 * implement whatever link state changes are indicated.
2292 static void bond_mii_monitor(struct work_struct *work)
2294 struct bonding *bond = container_of(work, struct bonding,
2296 bool should_notify_peers = false;
2297 unsigned long delay;
2298 struct slave *slave;
2299 struct list_head *iter;
2301 delay = msecs_to_jiffies(bond->params.miimon);
2303 if (!bond_has_slaves(bond))
2308 should_notify_peers = bond_should_notify_peers(bond);
2310 if (bond_miimon_inspect(bond)) {
2313 /* Race avoidance with bond_close cancel of workqueue */
2314 if (!rtnl_trylock()) {
2316 should_notify_peers = false;
2320 bond_for_each_slave(bond, slave, iter) {
2321 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2323 bond_miimon_commit(bond);
2325 rtnl_unlock(); /* might sleep, hold no other locks */
2330 if (bond->params.miimon)
2331 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2333 if (should_notify_peers) {
2334 if (!rtnl_trylock())
2336 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2341 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2343 __be32 ip = *((__be32 *)data);
2345 return ip == bond_confirm_addr(upper, 0, ip);
2348 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2352 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2356 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2363 /* We go to the (large) trouble of VLAN tagging ARP frames because
2364 * switches in VLAN mode (especially if ports are configured as
2365 * "native" to a VLAN) might not pass non-tagged frames.
2367 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2368 __be32 dest_ip, __be32 src_ip,
2369 struct bond_vlan_tag *tags)
2371 struct sk_buff *skb;
2372 struct bond_vlan_tag *outer_tag = tags;
2374 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2375 arp_op, slave_dev->name, &dest_ip, &src_ip);
2377 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2378 NULL, slave_dev->dev_addr, NULL);
2381 net_err_ratelimited("ARP packet allocation failed\n");
2385 if (!tags || tags->vlan_proto == VLAN_N_VID)
2390 /* Go through all the tags backwards and add them to the packet */
2391 while (tags->vlan_proto != VLAN_N_VID) {
2392 if (!tags->vlan_id) {
2397 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2398 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2399 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2402 net_err_ratelimited("failed to insert inner VLAN tag\n");
2408 /* Set the outer tag */
2409 if (outer_tag->vlan_id) {
2410 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2411 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2412 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2413 outer_tag->vlan_id);
2420 /* Validate the device path between the @start_dev and the @end_dev.
2421 * The path is valid if the @end_dev is reachable through device
2423 * When the path is validated, collect any vlan information in the
2426 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2427 struct net_device *end_dev,
2430 struct bond_vlan_tag *tags;
2431 struct net_device *upper;
2432 struct list_head *iter;
2434 if (start_dev == end_dev) {
2435 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2437 return ERR_PTR(-ENOMEM);
2438 tags[level].vlan_proto = VLAN_N_VID;
2442 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2443 tags = bond_verify_device_path(upper, end_dev, level + 1);
2444 if (IS_ERR_OR_NULL(tags)) {
2449 if (is_vlan_dev(upper)) {
2450 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2451 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2460 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2463 struct bond_vlan_tag *tags;
2464 __be32 *targets = bond->params.arp_targets, addr;
2467 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2468 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2471 /* Find out through which dev should the packet go */
2472 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2475 /* there's no route to target - try to send arp
2476 * probe to generate any traffic (arp_validate=0)
2478 if (bond->params.arp_validate)
2479 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2482 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2487 /* bond device itself */
2488 if (rt->dst.dev == bond->dev)
2492 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2495 if (!IS_ERR_OR_NULL(tags))
2498 /* Not our device - skip */
2499 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2500 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2506 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2508 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2514 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2518 if (!sip || !bond_has_this_ip(bond, tip)) {
2519 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2524 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2526 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2530 slave->last_rx = jiffies;
2531 slave->target_last_arp_rx[i] = jiffies;
2534 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2535 struct slave *slave)
2537 struct arphdr *arp = (struct arphdr *)skb->data;
2538 struct slave *curr_active_slave, *curr_arp_slave;
2539 unsigned char *arp_ptr;
2541 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2544 if (!slave_do_arp_validate(bond, slave)) {
2545 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2546 !slave_do_arp_validate_only(bond))
2547 slave->last_rx = jiffies;
2548 return RX_HANDLER_ANOTHER;
2549 } else if (!is_arp) {
2550 return RX_HANDLER_ANOTHER;
2553 alen = arp_hdr_len(bond->dev);
2555 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2558 if (alen > skb_headlen(skb)) {
2559 arp = kmalloc(alen, GFP_ATOMIC);
2562 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2566 if (arp->ar_hln != bond->dev->addr_len ||
2567 skb->pkt_type == PACKET_OTHERHOST ||
2568 skb->pkt_type == PACKET_LOOPBACK ||
2569 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2570 arp->ar_pro != htons(ETH_P_IP) ||
2574 arp_ptr = (unsigned char *)(arp + 1);
2575 arp_ptr += bond->dev->addr_len;
2576 memcpy(&sip, arp_ptr, 4);
2577 arp_ptr += 4 + bond->dev->addr_len;
2578 memcpy(&tip, arp_ptr, 4);
2580 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2581 slave->dev->name, bond_slave_state(slave),
2582 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2585 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2586 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2588 /* We 'trust' the received ARP enough to validate it if:
2590 * (a) the slave receiving the ARP is active (which includes the
2591 * current ARP slave, if any), or
2593 * (b) the receiving slave isn't active, but there is a currently
2594 * active slave and it received valid arp reply(s) after it became
2595 * the currently active slave, or
2597 * (c) there is an ARP slave that sent an ARP during the prior ARP
2598 * interval, and we receive an ARP reply on any slave. We accept
2599 * these because switch FDB update delays may deliver the ARP
2600 * reply to a slave other than the sender of the ARP request.
2602 * Note: for (b), backup slaves are receiving the broadcast ARP
2603 * request, not a reply. This request passes from the sending
2604 * slave through the L2 switch(es) to the receiving slave. Since
2605 * this is checking the request, sip/tip are swapped for
2608 * This is done to avoid endless looping when we can't reach the
2609 * arp_ip_target and fool ourselves with our own arp requests.
2611 if (bond_is_active_slave(slave))
2612 bond_validate_arp(bond, slave, sip, tip);
2613 else if (curr_active_slave &&
2614 time_after(slave_last_rx(bond, curr_active_slave),
2615 curr_active_slave->last_link_up))
2616 bond_validate_arp(bond, slave, tip, sip);
2617 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2618 bond_time_in_interval(bond,
2619 dev_trans_start(curr_arp_slave->dev), 1))
2620 bond_validate_arp(bond, slave, sip, tip);
2623 if (arp != (struct arphdr *)skb->data)
2625 return RX_HANDLER_ANOTHER;
2628 /* function to verify if we're in the arp_interval timeslice, returns true if
2629 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2630 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2632 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2635 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2637 return time_in_range(jiffies,
2638 last_act - delta_in_ticks,
2639 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2642 /* This function is called regularly to monitor each slave's link
2643 * ensuring that traffic is being sent and received when arp monitoring
2644 * is used in load-balancing mode. if the adapter has been dormant, then an
2645 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2646 * arp monitoring in active backup mode.
2648 static void bond_loadbalance_arp_mon(struct bonding *bond)
2650 struct slave *slave, *oldcurrent;
2651 struct list_head *iter;
2652 int do_failover = 0, slave_state_changed = 0;
2654 if (!bond_has_slaves(bond))
2659 oldcurrent = rcu_dereference(bond->curr_active_slave);
2660 /* see if any of the previous devices are up now (i.e. they have
2661 * xmt and rcv traffic). the curr_active_slave does not come into
2662 * the picture unless it is null. also, slave->last_link_up is not
2663 * needed here because we send an arp on each slave and give a slave
2664 * as long as it needs to get the tx/rx within the delta.
2665 * TODO: what about up/down delay in arp mode? it wasn't here before
2668 bond_for_each_slave_rcu(bond, slave, iter) {
2669 unsigned long trans_start = dev_trans_start(slave->dev);
2671 slave->new_link = BOND_LINK_NOCHANGE;
2673 if (slave->link != BOND_LINK_UP) {
2674 if (bond_time_in_interval(bond, trans_start, 1) &&
2675 bond_time_in_interval(bond, slave->last_rx, 1)) {
2677 slave->new_link = BOND_LINK_UP;
2678 slave_state_changed = 1;
2680 /* primary_slave has no meaning in round-robin
2681 * mode. the window of a slave being up and
2682 * curr_active_slave being null after enslaving
2686 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2690 netdev_info(bond->dev, "interface %s is now up\n",
2695 /* slave->link == BOND_LINK_UP */
2697 /* not all switches will respond to an arp request
2698 * when the source ip is 0, so don't take the link down
2699 * if we don't know our ip yet
2701 if (!bond_time_in_interval(bond, trans_start, 2) ||
2702 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2704 slave->new_link = BOND_LINK_DOWN;
2705 slave_state_changed = 1;
2707 if (slave->link_failure_count < UINT_MAX)
2708 slave->link_failure_count++;
2710 netdev_info(bond->dev, "interface %s is now down\n",
2713 if (slave == oldcurrent)
2718 /* note: if switch is in round-robin mode, all links
2719 * must tx arp to ensure all links rx an arp - otherwise
2720 * links may oscillate or not come up at all; if switch is
2721 * in something like xor mode, there is nothing we can
2722 * do - all replies will be rx'ed on same link causing slaves
2723 * to be unstable during low/no traffic periods
2725 if (bond_slave_is_up(slave))
2726 bond_arp_send_all(bond, slave);
2731 if (do_failover || slave_state_changed) {
2732 if (!rtnl_trylock())
2735 bond_for_each_slave(bond, slave, iter) {
2736 if (slave->new_link != BOND_LINK_NOCHANGE)
2737 slave->link = slave->new_link;
2740 if (slave_state_changed) {
2741 bond_slave_state_change(bond);
2742 if (BOND_MODE(bond) == BOND_MODE_XOR)
2743 bond_update_slave_arr(bond, NULL);
2747 bond_select_active_slave(bond);
2748 unblock_netpoll_tx();
2754 if (bond->params.arp_interval)
2755 queue_delayed_work(bond->wq, &bond->arp_work,
2756 msecs_to_jiffies(bond->params.arp_interval));
2759 /* Called to inspect slaves for active-backup mode ARP monitor link state
2760 * changes. Sets new_link in slaves to specify what action should take
2761 * place for the slave. Returns 0 if no changes are found, >0 if changes
2762 * to link states must be committed.
2764 * Called with rcu_read_lock held.
2766 static int bond_ab_arp_inspect(struct bonding *bond)
2768 unsigned long trans_start, last_rx;
2769 struct list_head *iter;
2770 struct slave *slave;
2773 bond_for_each_slave_rcu(bond, slave, iter) {
2774 slave->new_link = BOND_LINK_NOCHANGE;
2775 last_rx = slave_last_rx(bond, slave);
2777 if (slave->link != BOND_LINK_UP) {
2778 if (bond_time_in_interval(bond, last_rx, 1)) {
2779 slave->new_link = BOND_LINK_UP;
2785 /* Give slaves 2*delta after being enslaved or made
2786 * active. This avoids bouncing, as the last receive
2787 * times need a full ARP monitor cycle to be updated.
2789 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2792 /* Backup slave is down if:
2793 * - No current_arp_slave AND
2794 * - more than 3*delta since last receive AND
2795 * - the bond has an IP address
2797 * Note: a non-null current_arp_slave indicates
2798 * the curr_active_slave went down and we are
2799 * searching for a new one; under this condition
2800 * we only take the curr_active_slave down - this
2801 * gives each slave a chance to tx/rx traffic
2802 * before being taken out
2804 if (!bond_is_active_slave(slave) &&
2805 !rcu_access_pointer(bond->current_arp_slave) &&
2806 !bond_time_in_interval(bond, last_rx, 3)) {
2807 slave->new_link = BOND_LINK_DOWN;
2811 /* Active slave is down if:
2812 * - more than 2*delta since transmitting OR
2813 * - (more than 2*delta since receive AND
2814 * the bond has an IP address)
2816 trans_start = dev_trans_start(slave->dev);
2817 if (bond_is_active_slave(slave) &&
2818 (!bond_time_in_interval(bond, trans_start, 2) ||
2819 !bond_time_in_interval(bond, last_rx, 2))) {
2820 slave->new_link = BOND_LINK_DOWN;
2828 /* Called to commit link state changes noted by inspection step of
2829 * active-backup mode ARP monitor.
2831 * Called with RTNL hold.
2833 static void bond_ab_arp_commit(struct bonding *bond)
2835 unsigned long trans_start;
2836 struct list_head *iter;
2837 struct slave *slave;
2839 bond_for_each_slave(bond, slave, iter) {
2840 switch (slave->new_link) {
2841 case BOND_LINK_NOCHANGE:
2845 trans_start = dev_trans_start(slave->dev);
2846 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2847 (!rtnl_dereference(bond->curr_active_slave) &&
2848 bond_time_in_interval(bond, trans_start, 1))) {
2849 struct slave *current_arp_slave;
2851 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2852 bond_set_slave_link_state(slave, BOND_LINK_UP,
2853 BOND_SLAVE_NOTIFY_NOW);
2854 if (current_arp_slave) {
2855 bond_set_slave_inactive_flags(
2857 BOND_SLAVE_NOTIFY_NOW);
2858 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2861 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2864 if (!rtnl_dereference(bond->curr_active_slave) ||
2865 slave == rtnl_dereference(bond->primary_slave))
2872 case BOND_LINK_DOWN:
2873 if (slave->link_failure_count < UINT_MAX)
2874 slave->link_failure_count++;
2876 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2877 BOND_SLAVE_NOTIFY_NOW);
2878 bond_set_slave_inactive_flags(slave,
2879 BOND_SLAVE_NOTIFY_NOW);
2881 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2884 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2885 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2892 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2893 slave->new_link, slave->dev->name);
2899 bond_select_active_slave(bond);
2900 unblock_netpoll_tx();
2903 bond_set_carrier(bond);
2906 /* Send ARP probes for active-backup mode ARP monitor.
2908 * Called with rcu_read_lock held.
2910 static bool bond_ab_arp_probe(struct bonding *bond)
2912 struct slave *slave, *before = NULL, *new_slave = NULL,
2913 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2914 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2915 struct list_head *iter;
2917 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2919 if (curr_arp_slave && curr_active_slave)
2920 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2921 curr_arp_slave->dev->name,
2922 curr_active_slave->dev->name);
2924 if (curr_active_slave) {
2925 bond_arp_send_all(bond, curr_active_slave);
2926 return should_notify_rtnl;
2929 /* if we don't have a curr_active_slave, search for the next available
2930 * backup slave from the current_arp_slave and make it the candidate
2931 * for becoming the curr_active_slave
2934 if (!curr_arp_slave) {
2935 curr_arp_slave = bond_first_slave_rcu(bond);
2936 if (!curr_arp_slave)
2937 return should_notify_rtnl;
2940 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2942 bond_for_each_slave_rcu(bond, slave, iter) {
2943 if (!found && !before && bond_slave_is_up(slave))
2946 if (found && !new_slave && bond_slave_is_up(slave))
2948 /* if the link state is up at this point, we
2949 * mark it down - this can happen if we have
2950 * simultaneous link failures and
2951 * reselect_active_interface doesn't make this
2952 * one the current slave so it is still marked
2953 * up when it is actually down
2955 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2956 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2957 BOND_SLAVE_NOTIFY_LATER);
2958 if (slave->link_failure_count < UINT_MAX)
2959 slave->link_failure_count++;
2961 bond_set_slave_inactive_flags(slave,
2962 BOND_SLAVE_NOTIFY_LATER);
2964 netdev_info(bond->dev, "backup interface %s is now down\n",
2967 if (slave == curr_arp_slave)
2971 if (!new_slave && before)
2977 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2978 BOND_SLAVE_NOTIFY_LATER);
2979 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2980 bond_arp_send_all(bond, new_slave);
2981 new_slave->last_link_up = jiffies;
2982 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2985 bond_for_each_slave_rcu(bond, slave, iter) {
2986 if (slave->should_notify || slave->should_notify_link) {
2987 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2991 return should_notify_rtnl;
2994 static void bond_activebackup_arp_mon(struct bonding *bond)
2996 bool should_notify_peers = false;
2997 bool should_notify_rtnl = false;
3000 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3002 if (!bond_has_slaves(bond))
3007 should_notify_peers = bond_should_notify_peers(bond);
3009 if (bond_ab_arp_inspect(bond)) {
3012 /* Race avoidance with bond_close flush of workqueue */
3013 if (!rtnl_trylock()) {
3015 should_notify_peers = false;
3019 bond_ab_arp_commit(bond);
3025 should_notify_rtnl = bond_ab_arp_probe(bond);
3029 if (bond->params.arp_interval)
3030 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3032 if (should_notify_peers || should_notify_rtnl) {
3033 if (!rtnl_trylock())
3036 if (should_notify_peers)
3037 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3039 if (should_notify_rtnl) {
3040 bond_slave_state_notify(bond);
3041 bond_slave_link_notify(bond);
3048 static void bond_arp_monitor(struct work_struct *work)
3050 struct bonding *bond = container_of(work, struct bonding,
3053 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3054 bond_activebackup_arp_mon(bond);
3056 bond_loadbalance_arp_mon(bond);
3059 /*-------------------------- netdev event handling --------------------------*/
3061 /* Change device name */
3062 static int bond_event_changename(struct bonding *bond)
3064 bond_remove_proc_entry(bond);
3065 bond_create_proc_entry(bond);
3067 bond_debug_reregister(bond);
3072 static int bond_master_netdev_event(unsigned long event,
3073 struct net_device *bond_dev)
3075 struct bonding *event_bond = netdev_priv(bond_dev);
3078 case NETDEV_CHANGENAME:
3079 return bond_event_changename(event_bond);
3080 case NETDEV_UNREGISTER:
3081 bond_remove_proc_entry(event_bond);
3083 case NETDEV_REGISTER:
3084 bond_create_proc_entry(event_bond);
3086 case NETDEV_NOTIFY_PEERS:
3087 if (event_bond->send_peer_notif)
3088 event_bond->send_peer_notif--;
3097 static int bond_slave_netdev_event(unsigned long event,
3098 struct net_device *slave_dev)
3100 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3101 struct bonding *bond;
3102 struct net_device *bond_dev;
3104 /* A netdev event can be generated while enslaving a device
3105 * before netdev_rx_handler_register is called in which case
3106 * slave will be NULL
3110 bond_dev = slave->bond->dev;
3112 primary = rtnl_dereference(bond->primary_slave);
3115 case NETDEV_UNREGISTER:
3116 if (bond_dev->type != ARPHRD_ETHER)
3117 bond_release_and_destroy(bond_dev, slave_dev);
3119 __bond_release_one(bond_dev, slave_dev, false, true);
3123 /* For 802.3ad mode only:
3124 * Getting invalid Speed/Duplex values here will put slave
3125 * in weird state. So mark it as link-fail for the time
3126 * being and let link-monitoring (miimon) set it right when
3127 * correct speeds/duplex are available.
3129 if (bond_update_speed_duplex(slave) &&
3130 BOND_MODE(bond) == BOND_MODE_8023AD)
3131 slave->link = BOND_LINK_FAIL;
3133 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3134 bond_3ad_adapter_speed_duplex_changed(slave);
3137 /* Refresh slave-array if applicable!
3138 * If the setup does not use miimon or arpmon (mode-specific!),
3139 * then these events will not cause the slave-array to be
3140 * refreshed. This will cause xmit to use a slave that is not
3141 * usable. Avoid such situation by refeshing the array at these
3142 * events. If these (miimon/arpmon) parameters are configured
3143 * then array gets refreshed twice and that should be fine!
3145 if (bond_mode_can_use_xmit_hash(bond))
3146 bond_update_slave_arr(bond, NULL);
3148 case NETDEV_CHANGEMTU:
3149 /* TODO: Should slaves be allowed to
3150 * independently alter their MTU? For
3151 * an active-backup bond, slaves need
3152 * not be the same type of device, so
3153 * MTUs may vary. For other modes,
3154 * slaves arguably should have the
3155 * same MTUs. To do this, we'd need to
3156 * take over the slave's change_mtu
3157 * function for the duration of their
3161 case NETDEV_CHANGENAME:
3162 /* we don't care if we don't have primary set */
3163 if (!bond_uses_primary(bond) ||
3164 !bond->params.primary[0])
3167 if (slave == primary) {
3168 /* slave's name changed - he's no longer primary */
3169 RCU_INIT_POINTER(bond->primary_slave, NULL);
3170 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3171 /* we have a new primary slave */
3172 rcu_assign_pointer(bond->primary_slave, slave);
3173 } else { /* we didn't change primary - exit */
3177 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3178 primary ? slave_dev->name : "none");
3181 bond_select_active_slave(bond);
3182 unblock_netpoll_tx();
3184 case NETDEV_FEAT_CHANGE:
3185 bond_compute_features(bond);
3187 case NETDEV_RESEND_IGMP:
3188 /* Propagate to master device */
3189 call_netdevice_notifiers(event, slave->bond->dev);
3198 /* bond_netdev_event: handle netdev notifier chain events.
3200 * This function receives events for the netdev chain. The caller (an
3201 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3202 * locks for us to safely manipulate the slave devices (RTNL lock,
3205 static int bond_netdev_event(struct notifier_block *this,
3206 unsigned long event, void *ptr)
3208 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3210 netdev_dbg(event_dev, "event: %lx\n", event);
3212 if (!(event_dev->priv_flags & IFF_BONDING))
3215 if (event_dev->flags & IFF_MASTER) {
3216 netdev_dbg(event_dev, "IFF_MASTER\n");
3217 return bond_master_netdev_event(event, event_dev);
3220 if (event_dev->flags & IFF_SLAVE) {
3221 netdev_dbg(event_dev, "IFF_SLAVE\n");
3222 return bond_slave_netdev_event(event, event_dev);
3228 static struct notifier_block bond_netdev_notifier = {
3229 .notifier_call = bond_netdev_event,
3232 /*---------------------------- Hashing Policies -----------------------------*/
3234 /* L2 hash helper */
3235 static inline u32 bond_eth_hash(struct sk_buff *skb)
3237 struct ethhdr *ep, hdr_tmp;
3239 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3241 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3245 /* Extract the appropriate headers based on bond's xmit policy */
3246 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3247 struct flow_keys *fk)
3249 const struct ipv6hdr *iph6;
3250 const struct iphdr *iph;
3251 int noff, proto = -1;
3253 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3254 return skb_flow_dissect_flow_keys(skb, fk, 0);
3256 fk->ports.ports = 0;
3257 noff = skb_network_offset(skb);
3258 if (skb->protocol == htons(ETH_P_IP)) {
3259 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3262 iph_to_flow_copy_v4addrs(fk, iph);
3263 noff += iph->ihl << 2;
3264 if (!ip_is_fragment(iph))
3265 proto = iph->protocol;
3266 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3267 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3269 iph6 = ipv6_hdr(skb);
3270 iph_to_flow_copy_v6addrs(fk, iph6);
3271 noff += sizeof(*iph6);
3272 proto = iph6->nexthdr;
3276 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3277 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3283 * bond_xmit_hash - generate a hash value based on the xmit policy
3284 * @bond: bonding device
3285 * @skb: buffer to use for headers
3287 * This function will extract the necessary headers from the skb buffer and use
3288 * them to generate a hash based on the xmit_policy set in the bonding device
3290 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3292 struct flow_keys flow;
3295 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3299 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3300 !bond_flow_dissect(bond, skb, &flow))
3301 return bond_eth_hash(skb);
3303 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3304 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3305 hash = bond_eth_hash(skb);
3307 hash = (__force u32)flow.ports.ports;
3308 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3309 (__force u32)flow_get_u32_src(&flow);
3310 hash ^= (hash >> 16);
3311 hash ^= (hash >> 8);
3316 /*-------------------------- Device entry points ----------------------------*/
3318 void bond_work_init_all(struct bonding *bond)
3320 INIT_DELAYED_WORK(&bond->mcast_work,
3321 bond_resend_igmp_join_requests_delayed);
3322 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3323 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3324 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3325 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3326 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3329 static void bond_work_cancel_all(struct bonding *bond)
3331 cancel_delayed_work_sync(&bond->mii_work);
3332 cancel_delayed_work_sync(&bond->arp_work);
3333 cancel_delayed_work_sync(&bond->alb_work);
3334 cancel_delayed_work_sync(&bond->ad_work);
3335 cancel_delayed_work_sync(&bond->mcast_work);
3336 cancel_delayed_work_sync(&bond->slave_arr_work);
3339 static int bond_open(struct net_device *bond_dev)
3341 struct bonding *bond = netdev_priv(bond_dev);
3342 struct list_head *iter;
3343 struct slave *slave;
3345 /* reset slave->backup and slave->inactive */
3346 if (bond_has_slaves(bond)) {
3347 bond_for_each_slave(bond, slave, iter) {
3348 if (bond_uses_primary(bond) &&
3349 slave != rcu_access_pointer(bond->curr_active_slave)) {
3350 bond_set_slave_inactive_flags(slave,
3351 BOND_SLAVE_NOTIFY_NOW);
3352 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3353 bond_set_slave_active_flags(slave,
3354 BOND_SLAVE_NOTIFY_NOW);
3359 if (bond_is_lb(bond)) {
3360 /* bond_alb_initialize must be called before the timer
3363 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3365 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3366 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3369 if (bond->params.miimon) /* link check interval, in milliseconds. */
3370 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3372 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3373 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3374 bond->recv_probe = bond_arp_rcv;
3377 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3378 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3379 /* register to receive LACPDUs */
3380 bond->recv_probe = bond_3ad_lacpdu_recv;
3381 bond_3ad_initiate_agg_selection(bond, 1);
3384 if (bond_mode_can_use_xmit_hash(bond))
3385 bond_update_slave_arr(bond, NULL);
3390 static int bond_close(struct net_device *bond_dev)
3392 struct bonding *bond = netdev_priv(bond_dev);
3394 bond_work_cancel_all(bond);
3395 bond->send_peer_notif = 0;
3396 if (bond_is_lb(bond))
3397 bond_alb_deinitialize(bond);
3398 bond->recv_probe = NULL;
3403 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3404 * that some drivers can provide 32bit values only.
3406 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3407 const struct rtnl_link_stats64 *_new,
3408 const struct rtnl_link_stats64 *_old)
3410 const u64 *new = (const u64 *)_new;
3411 const u64 *old = (const u64 *)_old;
3412 u64 *res = (u64 *)_res;
3415 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3418 s64 delta = nv - ov;
3420 /* detects if this particular field is 32bit only */
3421 if (((nv | ov) >> 32) == 0)
3422 delta = (s64)(s32)((u32)nv - (u32)ov);
3424 /* filter anomalies, some drivers reset their stats
3425 * at down/up events.
3432 static int bond_get_nest_level(struct net_device *bond_dev)
3434 struct bonding *bond = netdev_priv(bond_dev);
3436 return bond->nest_level;
3439 static void bond_get_stats(struct net_device *bond_dev,
3440 struct rtnl_link_stats64 *stats)
3442 struct bonding *bond = netdev_priv(bond_dev);
3443 struct rtnl_link_stats64 temp;
3444 struct list_head *iter;
3445 struct slave *slave;
3447 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3448 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3451 bond_for_each_slave_rcu(bond, slave, iter) {
3452 const struct rtnl_link_stats64 *new =
3453 dev_get_stats(slave->dev, &temp);
3455 bond_fold_stats(stats, new, &slave->slave_stats);
3457 /* save off the slave stats for the next run */
3458 memcpy(&slave->slave_stats, new, sizeof(*new));
3462 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3463 spin_unlock(&bond->stats_lock);
3466 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3468 struct bonding *bond = netdev_priv(bond_dev);
3469 struct net_device *slave_dev = NULL;
3470 struct ifbond k_binfo;
3471 struct ifbond __user *u_binfo = NULL;
3472 struct ifslave k_sinfo;
3473 struct ifslave __user *u_sinfo = NULL;
3474 struct mii_ioctl_data *mii = NULL;
3475 struct bond_opt_value newval;
3479 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3490 /* We do this again just in case we were called by SIOCGMIIREG
3491 * instead of SIOCGMIIPHY.
3497 if (mii->reg_num == 1) {
3499 if (netif_carrier_ok(bond->dev))
3500 mii->val_out = BMSR_LSTATUS;
3504 case BOND_INFO_QUERY_OLD:
3505 case SIOCBONDINFOQUERY:
3506 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3508 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3511 bond_info_query(bond_dev, &k_binfo);
3512 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3516 case BOND_SLAVE_INFO_QUERY_OLD:
3517 case SIOCBONDSLAVEINFOQUERY:
3518 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3520 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3523 res = bond_slave_info_query(bond_dev, &k_sinfo);
3525 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3533 net = dev_net(bond_dev);
3535 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3538 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3540 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3545 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3547 case BOND_ENSLAVE_OLD:
3548 case SIOCBONDENSLAVE:
3549 res = bond_enslave(bond_dev, slave_dev, NULL);
3551 case BOND_RELEASE_OLD:
3552 case SIOCBONDRELEASE:
3553 res = bond_release(bond_dev, slave_dev);
3555 case BOND_SETHWADDR_OLD:
3556 case SIOCBONDSETHWADDR:
3557 res = bond_set_dev_addr(bond_dev, slave_dev);
3559 case BOND_CHANGE_ACTIVE_OLD:
3560 case SIOCBONDCHANGEACTIVE:
3561 bond_opt_initstr(&newval, slave_dev->name);
3562 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3572 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3574 struct bonding *bond = netdev_priv(bond_dev);
3576 if (change & IFF_PROMISC)
3577 bond_set_promiscuity(bond,
3578 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3580 if (change & IFF_ALLMULTI)
3581 bond_set_allmulti(bond,
3582 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3585 static void bond_set_rx_mode(struct net_device *bond_dev)
3587 struct bonding *bond = netdev_priv(bond_dev);
3588 struct list_head *iter;
3589 struct slave *slave;
3592 if (bond_uses_primary(bond)) {
3593 slave = rcu_dereference(bond->curr_active_slave);
3595 dev_uc_sync(slave->dev, bond_dev);
3596 dev_mc_sync(slave->dev, bond_dev);
3599 bond_for_each_slave_rcu(bond, slave, iter) {
3600 dev_uc_sync_multiple(slave->dev, bond_dev);
3601 dev_mc_sync_multiple(slave->dev, bond_dev);
3607 static int bond_neigh_init(struct neighbour *n)
3609 struct bonding *bond = netdev_priv(n->dev);
3610 const struct net_device_ops *slave_ops;
3611 struct neigh_parms parms;
3612 struct slave *slave;
3615 slave = bond_first_slave(bond);
3618 slave_ops = slave->dev->netdev_ops;
3619 if (!slave_ops->ndo_neigh_setup)
3622 parms.neigh_setup = NULL;
3623 parms.neigh_cleanup = NULL;
3624 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3628 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3629 * after the last slave has been detached. Assumes that all slaves
3630 * utilize the same neigh_cleanup (true at this writing as only user
3633 n->parms->neigh_cleanup = parms.neigh_cleanup;
3635 if (!parms.neigh_setup)
3638 return parms.neigh_setup(n);
3641 /* The bonding ndo_neigh_setup is called at init time beofre any
3642 * slave exists. So we must declare proxy setup function which will
3643 * be used at run time to resolve the actual slave neigh param setup.
3645 * It's also called by master devices (such as vlans) to setup their
3646 * underlying devices. In that case - do nothing, we're already set up from
3649 static int bond_neigh_setup(struct net_device *dev,
3650 struct neigh_parms *parms)
3652 /* modify only our neigh_parms */
3653 if (parms->dev == dev)
3654 parms->neigh_setup = bond_neigh_init;
3659 /* Change the MTU of all of a master's slaves to match the master */
3660 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3662 struct bonding *bond = netdev_priv(bond_dev);
3663 struct slave *slave, *rollback_slave;
3664 struct list_head *iter;
3667 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3669 bond_for_each_slave(bond, slave, iter) {
3670 netdev_dbg(bond_dev, "s %p c_m %p\n",
3671 slave, slave->dev->netdev_ops->ndo_change_mtu);
3673 res = dev_set_mtu(slave->dev, new_mtu);
3676 /* If we failed to set the slave's mtu to the new value
3677 * we must abort the operation even in ACTIVE_BACKUP
3678 * mode, because if we allow the backup slaves to have
3679 * different mtu values than the active slave we'll
3680 * need to change their mtu when doing a failover. That
3681 * means changing their mtu from timer context, which
3682 * is probably not a good idea.
3684 netdev_dbg(bond_dev, "err %d %s\n", res,
3690 bond_dev->mtu = new_mtu;
3695 /* unwind from head to the slave that failed */
3696 bond_for_each_slave(bond, rollback_slave, iter) {
3699 if (rollback_slave == slave)
3702 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3704 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3705 tmp_res, rollback_slave->dev->name);
3712 /* Change HW address
3714 * Note that many devices must be down to change the HW address, and
3715 * downing the master releases all slaves. We can make bonds full of
3716 * bonding devices to test this, however.
3718 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3720 struct bonding *bond = netdev_priv(bond_dev);
3721 struct slave *slave, *rollback_slave;
3722 struct sockaddr_storage *ss = addr, tmp_ss;
3723 struct list_head *iter;
3726 if (BOND_MODE(bond) == BOND_MODE_ALB)
3727 return bond_alb_set_mac_address(bond_dev, addr);
3730 netdev_dbg(bond_dev, "bond=%p\n", bond);
3732 /* If fail_over_mac is enabled, do nothing and return success.
3733 * Returning an error causes ifenslave to fail.
3735 if (bond->params.fail_over_mac &&
3736 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3739 if (!is_valid_ether_addr(ss->__data))
3740 return -EADDRNOTAVAIL;
3742 bond_for_each_slave(bond, slave, iter) {
3743 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3744 res = dev_set_mac_address(slave->dev, addr, NULL);
3746 /* TODO: consider downing the slave
3748 * User should expect communications
3749 * breakage anyway until ARP finish
3752 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3758 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3762 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3763 tmp_ss.ss_family = bond_dev->type;
3765 /* unwind from head to the slave that failed */
3766 bond_for_each_slave(bond, rollback_slave, iter) {
3769 if (rollback_slave == slave)
3772 tmp_res = dev_set_mac_address(rollback_slave->dev,
3773 (struct sockaddr *)&tmp_ss, NULL);
3775 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3776 tmp_res, rollback_slave->dev->name);
3784 * bond_xmit_slave_id - transmit skb through slave with slave_id
3785 * @bond: bonding device that is transmitting
3786 * @skb: buffer to transmit
3787 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3789 * This function tries to transmit through slave with slave_id but in case
3790 * it fails, it tries to find the first available slave for transmission.
3791 * The skb is consumed in all cases, thus the function is void.
3793 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3795 struct list_head *iter;
3796 struct slave *slave;
3799 /* Here we start from the slave with slave_id */
3800 bond_for_each_slave_rcu(bond, slave, iter) {
3802 if (bond_slave_can_tx(slave)) {
3803 bond_dev_queue_xmit(bond, skb, slave->dev);
3809 /* Here we start from the first slave up to slave_id */
3811 bond_for_each_slave_rcu(bond, slave, iter) {
3814 if (bond_slave_can_tx(slave)) {
3815 bond_dev_queue_xmit(bond, skb, slave->dev);
3819 /* no slave that can tx has been found */
3820 bond_tx_drop(bond->dev, skb);
3824 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3825 * @bond: bonding device to use
3827 * Based on the value of the bonding device's packets_per_slave parameter
3828 * this function generates a slave id, which is usually used as the next
3829 * slave to transmit through.
3831 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3834 struct reciprocal_value reciprocal_packets_per_slave;
3835 int packets_per_slave = bond->params.packets_per_slave;
3837 switch (packets_per_slave) {
3839 slave_id = prandom_u32();
3842 slave_id = bond->rr_tx_counter;
3845 reciprocal_packets_per_slave =
3846 bond->params.reciprocal_packets_per_slave;
3847 slave_id = reciprocal_divide(bond->rr_tx_counter,
3848 reciprocal_packets_per_slave);
3851 bond->rr_tx_counter++;
3856 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3857 struct net_device *bond_dev)
3859 struct bonding *bond = netdev_priv(bond_dev);
3860 struct iphdr *iph = ip_hdr(skb);
3861 struct slave *slave;
3864 /* Start with the curr_active_slave that joined the bond as the
3865 * default for sending IGMP traffic. For failover purposes one
3866 * needs to maintain some consistency for the interface that will
3867 * send the join/membership reports. The curr_active_slave found
3868 * will send all of this type of traffic.
3870 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3871 slave = rcu_dereference(bond->curr_active_slave);
3873 bond_dev_queue_xmit(bond, skb, slave->dev);
3875 bond_xmit_slave_id(bond, skb, 0);
3877 int slave_cnt = READ_ONCE(bond->slave_cnt);
3879 if (likely(slave_cnt)) {
3880 slave_id = bond_rr_gen_slave_id(bond);
3881 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3883 bond_tx_drop(bond_dev, skb);
3887 return NETDEV_TX_OK;
3890 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3891 * the bond has a usable interface.
3893 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3894 struct net_device *bond_dev)
3896 struct bonding *bond = netdev_priv(bond_dev);
3897 struct slave *slave;
3899 slave = rcu_dereference(bond->curr_active_slave);
3901 bond_dev_queue_xmit(bond, skb, slave->dev);
3903 bond_tx_drop(bond_dev, skb);
3905 return NETDEV_TX_OK;
3908 /* Use this to update slave_array when (a) it's not appropriate to update
3909 * slave_array right away (note that update_slave_array() may sleep)
3910 * and / or (b) RTNL is not held.
3912 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3914 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3917 /* Slave array work handler. Holds only RTNL */
3918 static void bond_slave_arr_handler(struct work_struct *work)
3920 struct bonding *bond = container_of(work, struct bonding,
3921 slave_arr_work.work);
3924 if (!rtnl_trylock())
3927 ret = bond_update_slave_arr(bond, NULL);
3930 pr_warn_ratelimited("Failed to update slave array from WT\n");
3936 bond_slave_arr_work_rearm(bond, 1);
3939 /* Build the usable slaves array in control path for modes that use xmit-hash
3940 * to determine the slave interface -
3941 * (a) BOND_MODE_8023AD
3943 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3945 * The caller is expected to hold RTNL only and NO other lock!
3947 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3949 struct slave *slave;
3950 struct list_head *iter;
3951 struct bond_up_slave *new_arr, *old_arr;
3955 #ifdef CONFIG_LOCKDEP
3956 WARN_ON(lockdep_is_held(&bond->mode_lock));
3959 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3963 pr_err("Failed to build slave-array.\n");
3966 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3967 struct ad_info ad_info;
3969 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3970 pr_debug("bond_3ad_get_active_agg_info failed\n");
3971 kfree_rcu(new_arr, rcu);
3972 /* No active aggragator means it's not safe to use
3973 * the previous array.
3975 old_arr = rtnl_dereference(bond->slave_arr);
3977 RCU_INIT_POINTER(bond->slave_arr, NULL);
3978 kfree_rcu(old_arr, rcu);
3982 agg_id = ad_info.aggregator_id;
3984 bond_for_each_slave(bond, slave, iter) {
3985 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3986 struct aggregator *agg;
3988 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3989 if (!agg || agg->aggregator_identifier != agg_id)
3992 if (!bond_slave_can_tx(slave))
3994 if (skipslave == slave)
3997 netdev_dbg(bond->dev,
3998 "Adding slave dev %s to tx hash array[%d]\n",
3999 slave->dev->name, new_arr->count);
4001 new_arr->arr[new_arr->count++] = slave;
4004 old_arr = rtnl_dereference(bond->slave_arr);
4005 rcu_assign_pointer(bond->slave_arr, new_arr);
4007 kfree_rcu(old_arr, rcu);
4009 if (ret != 0 && skipslave) {
4012 /* Rare situation where caller has asked to skip a specific
4013 * slave but allocation failed (most likely!). BTW this is
4014 * only possible when the call is initiated from
4015 * __bond_release_one(). In this situation; overwrite the
4016 * skipslave entry in the array with the last entry from the
4017 * array to avoid a situation where the xmit path may choose
4018 * this to-be-skipped slave to send a packet out.
4020 old_arr = rtnl_dereference(bond->slave_arr);
4021 for (idx = 0; idx < old_arr->count; idx++) {
4022 if (skipslave == old_arr->arr[idx]) {
4024 old_arr->arr[old_arr->count-1];
4033 /* Use this Xmit function for 3AD as well as XOR modes. The current
4034 * usable slave array is formed in the control path. The xmit function
4035 * just calculates hash and sends the packet out.
4037 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4038 struct net_device *dev)
4040 struct bonding *bond = netdev_priv(dev);
4041 struct slave *slave;
4042 struct bond_up_slave *slaves;
4045 slaves = rcu_dereference(bond->slave_arr);
4046 count = slaves ? READ_ONCE(slaves->count) : 0;
4047 if (likely(count)) {
4048 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4049 bond_dev_queue_xmit(bond, skb, slave->dev);
4051 bond_tx_drop(dev, skb);
4054 return NETDEV_TX_OK;
4057 /* in broadcast mode, we send everything to all usable interfaces. */
4058 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4059 struct net_device *bond_dev)
4061 struct bonding *bond = netdev_priv(bond_dev);
4062 struct slave *slave = NULL;
4063 struct list_head *iter;
4065 bond_for_each_slave_rcu(bond, slave, iter) {
4066 if (bond_is_last_slave(bond, slave))
4068 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4069 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4072 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4073 bond_dev->name, __func__);
4076 bond_dev_queue_xmit(bond, skb2, slave->dev);
4079 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4080 bond_dev_queue_xmit(bond, skb, slave->dev);
4082 bond_tx_drop(bond_dev, skb);
4084 return NETDEV_TX_OK;
4087 /*------------------------- Device initialization ---------------------------*/
4089 /* Lookup the slave that corresponds to a qid */
4090 static inline int bond_slave_override(struct bonding *bond,
4091 struct sk_buff *skb)
4093 struct slave *slave = NULL;
4094 struct list_head *iter;
4096 if (!skb_rx_queue_recorded(skb))
4099 /* Find out if any slaves have the same mapping as this skb. */
4100 bond_for_each_slave_rcu(bond, slave, iter) {
4101 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4102 if (bond_slave_is_up(slave) &&
4103 slave->link == BOND_LINK_UP) {
4104 bond_dev_queue_xmit(bond, skb, slave->dev);
4107 /* If the slave isn't UP, use default transmit policy. */
4116 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4117 struct net_device *sb_dev)
4119 /* This helper function exists to help dev_pick_tx get the correct
4120 * destination queue. Using a helper function skips a call to
4121 * skb_tx_hash and will put the skbs in the queue we expect on their
4122 * way down to the bonding driver.
4124 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4126 /* Save the original txq to restore before passing to the driver */
4127 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4129 if (unlikely(txq >= dev->real_num_tx_queues)) {
4131 txq -= dev->real_num_tx_queues;
4132 } while (txq >= dev->real_num_tx_queues);
4137 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4139 struct bonding *bond = netdev_priv(dev);
4141 if (bond_should_override_tx_queue(bond) &&
4142 !bond_slave_override(bond, skb))
4143 return NETDEV_TX_OK;
4145 switch (BOND_MODE(bond)) {
4146 case BOND_MODE_ROUNDROBIN:
4147 return bond_xmit_roundrobin(skb, dev);
4148 case BOND_MODE_ACTIVEBACKUP:
4149 return bond_xmit_activebackup(skb, dev);
4150 case BOND_MODE_8023AD:
4152 return bond_3ad_xor_xmit(skb, dev);
4153 case BOND_MODE_BROADCAST:
4154 return bond_xmit_broadcast(skb, dev);
4156 return bond_alb_xmit(skb, dev);
4158 return bond_tlb_xmit(skb, dev);
4160 /* Should never happen, mode already checked */
4161 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4163 bond_tx_drop(dev, skb);
4164 return NETDEV_TX_OK;
4168 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4170 struct bonding *bond = netdev_priv(dev);
4171 netdev_tx_t ret = NETDEV_TX_OK;
4173 /* If we risk deadlock from transmitting this in the
4174 * netpoll path, tell netpoll to queue the frame for later tx
4176 if (unlikely(is_netpoll_tx_blocked(dev)))
4177 return NETDEV_TX_BUSY;
4180 if (bond_has_slaves(bond))
4181 ret = __bond_start_xmit(skb, dev);
4183 bond_tx_drop(dev, skb);
4189 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4190 struct ethtool_link_ksettings *cmd)
4192 struct bonding *bond = netdev_priv(bond_dev);
4193 unsigned long speed = 0;
4194 struct list_head *iter;
4195 struct slave *slave;
4197 cmd->base.duplex = DUPLEX_UNKNOWN;
4198 cmd->base.port = PORT_OTHER;
4200 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4201 * do not need to check mode. Though link speed might not represent
4202 * the true receive or transmit bandwidth (not all modes are symmetric)
4203 * this is an accurate maximum.
4205 bond_for_each_slave(bond, slave, iter) {
4206 if (bond_slave_can_tx(slave)) {
4207 if (slave->speed != SPEED_UNKNOWN)
4208 speed += slave->speed;
4209 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4210 slave->duplex != DUPLEX_UNKNOWN)
4211 cmd->base.duplex = slave->duplex;
4214 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4219 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4220 struct ethtool_drvinfo *drvinfo)
4222 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4223 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4224 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4228 static const struct ethtool_ops bond_ethtool_ops = {
4229 .get_drvinfo = bond_ethtool_get_drvinfo,
4230 .get_link = ethtool_op_get_link,
4231 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4234 static const struct net_device_ops bond_netdev_ops = {
4235 .ndo_init = bond_init,
4236 .ndo_uninit = bond_uninit,
4237 .ndo_open = bond_open,
4238 .ndo_stop = bond_close,
4239 .ndo_start_xmit = bond_start_xmit,
4240 .ndo_select_queue = bond_select_queue,
4241 .ndo_get_stats64 = bond_get_stats,
4242 .ndo_do_ioctl = bond_do_ioctl,
4243 .ndo_change_rx_flags = bond_change_rx_flags,
4244 .ndo_set_rx_mode = bond_set_rx_mode,
4245 .ndo_change_mtu = bond_change_mtu,
4246 .ndo_set_mac_address = bond_set_mac_address,
4247 .ndo_neigh_setup = bond_neigh_setup,
4248 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4249 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4250 .ndo_get_lock_subclass = bond_get_nest_level,
4251 #ifdef CONFIG_NET_POLL_CONTROLLER
4252 .ndo_netpoll_setup = bond_netpoll_setup,
4253 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4254 .ndo_poll_controller = bond_poll_controller,
4256 .ndo_add_slave = bond_enslave,
4257 .ndo_del_slave = bond_release,
4258 .ndo_fix_features = bond_fix_features,
4259 .ndo_features_check = passthru_features_check,
4262 static const struct device_type bond_type = {
4266 static void bond_destructor(struct net_device *bond_dev)
4268 struct bonding *bond = netdev_priv(bond_dev);
4270 destroy_workqueue(bond->wq);
4273 void bond_setup(struct net_device *bond_dev)
4275 struct bonding *bond = netdev_priv(bond_dev);
4277 spin_lock_init(&bond->mode_lock);
4278 spin_lock_init(&bond->stats_lock);
4279 bond->params = bonding_defaults;
4281 /* Initialize pointers */
4282 bond->dev = bond_dev;
4284 /* Initialize the device entry points */
4285 ether_setup(bond_dev);
4286 bond_dev->max_mtu = ETH_MAX_MTU;
4287 bond_dev->netdev_ops = &bond_netdev_ops;
4288 bond_dev->ethtool_ops = &bond_ethtool_ops;
4290 bond_dev->needs_free_netdev = true;
4291 bond_dev->priv_destructor = bond_destructor;
4293 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4295 /* Initialize the device options */
4296 bond_dev->flags |= IFF_MASTER;
4297 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4298 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4300 /* don't acquire bond device's netif_tx_lock when transmitting */
4301 bond_dev->features |= NETIF_F_LLTX;
4303 /* By default, we declare the bond to be fully
4304 * VLAN hardware accelerated capable. Special
4305 * care is taken in the various xmit functions
4306 * when there are slaves that are not hw accel
4310 /* Don't allow bond devices to change network namespaces. */
4311 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4313 bond_dev->hw_features = BOND_VLAN_FEATURES |
4314 NETIF_F_HW_VLAN_CTAG_TX |
4315 NETIF_F_HW_VLAN_CTAG_RX |
4316 NETIF_F_HW_VLAN_CTAG_FILTER;
4318 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4319 bond_dev->features |= bond_dev->hw_features;
4322 /* Destroy a bonding device.
4323 * Must be under rtnl_lock when this function is called.
4325 static void bond_uninit(struct net_device *bond_dev)
4327 struct bonding *bond = netdev_priv(bond_dev);
4328 struct list_head *iter;
4329 struct slave *slave;
4330 struct bond_up_slave *arr;
4332 bond_netpoll_cleanup(bond_dev);
4334 /* Release the bonded slaves */
4335 bond_for_each_slave(bond, slave, iter)
4336 __bond_release_one(bond_dev, slave->dev, true, true);
4337 netdev_info(bond_dev, "Released all slaves\n");
4339 arr = rtnl_dereference(bond->slave_arr);
4341 RCU_INIT_POINTER(bond->slave_arr, NULL);
4342 kfree_rcu(arr, rcu);
4345 list_del(&bond->bond_list);
4347 bond_debug_unregister(bond);
4350 /*------------------------- Module initialization ---------------------------*/
4352 static int bond_check_params(struct bond_params *params)
4354 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4355 struct bond_opt_value newval;
4356 const struct bond_opt_value *valptr;
4357 int arp_all_targets_value = 0;
4358 u16 ad_actor_sys_prio = 0;
4359 u16 ad_user_port_key = 0;
4360 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4362 int bond_mode = BOND_MODE_ROUNDROBIN;
4363 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4367 /* Convert string parameters. */
4369 bond_opt_initstr(&newval, mode);
4370 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4372 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4375 bond_mode = valptr->value;
4378 if (xmit_hash_policy) {
4379 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4380 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4381 bond_mode == BOND_MODE_BROADCAST) {
4382 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4383 bond_mode_name(bond_mode));
4385 bond_opt_initstr(&newval, xmit_hash_policy);
4386 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4389 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4393 xmit_hashtype = valptr->value;
4398 if (bond_mode != BOND_MODE_8023AD) {
4399 pr_info("lacp_rate param is irrelevant in mode %s\n",
4400 bond_mode_name(bond_mode));
4402 bond_opt_initstr(&newval, lacp_rate);
4403 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4406 pr_err("Error: Invalid lacp rate \"%s\"\n",
4410 lacp_fast = valptr->value;
4415 bond_opt_initstr(&newval, ad_select);
4416 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4419 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4422 params->ad_select = valptr->value;
4423 if (bond_mode != BOND_MODE_8023AD)
4424 pr_warn("ad_select param only affects 802.3ad mode\n");
4426 params->ad_select = BOND_AD_STABLE;
4429 if (max_bonds < 0) {
4430 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4431 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4432 max_bonds = BOND_DEFAULT_MAX_BONDS;
4436 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4442 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4447 if (downdelay < 0) {
4448 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4449 downdelay, INT_MAX);
4453 if ((use_carrier != 0) && (use_carrier != 1)) {
4454 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4459 if (num_peer_notif < 0 || num_peer_notif > 255) {
4460 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4465 /* reset values for 802.3ad/TLB/ALB */
4466 if (!bond_mode_uses_arp(bond_mode)) {
4468 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");
4469 pr_warn("Forcing miimon to 100msec\n");
4470 miimon = BOND_DEFAULT_MIIMON;
4474 if (tx_queues < 1 || tx_queues > 255) {
4475 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4476 tx_queues, BOND_DEFAULT_TX_QUEUES);
4477 tx_queues = BOND_DEFAULT_TX_QUEUES;
4480 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4481 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4483 all_slaves_active = 0;
4486 if (resend_igmp < 0 || resend_igmp > 255) {
4487 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4488 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4489 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4492 bond_opt_initval(&newval, packets_per_slave);
4493 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4494 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4495 packets_per_slave, USHRT_MAX);
4496 packets_per_slave = 1;
4499 if (bond_mode == BOND_MODE_ALB) {
4500 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",
4505 if (updelay || downdelay) {
4506 /* just warn the user the up/down delay will have
4507 * no effect since miimon is zero...
4509 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",
4510 updelay, downdelay);
4513 /* don't allow arp monitoring */
4515 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4516 miimon, arp_interval);
4520 if ((updelay % miimon) != 0) {
4521 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4522 updelay, miimon, (updelay / miimon) * miimon);
4527 if ((downdelay % miimon) != 0) {
4528 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4530 (downdelay / miimon) * miimon);
4533 downdelay /= miimon;
4536 if (arp_interval < 0) {
4537 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4538 arp_interval, INT_MAX);
4542 for (arp_ip_count = 0, i = 0;
4543 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4546 /* not a complete check, but good enough to catch mistakes */
4547 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4548 !bond_is_ip_target_ok(ip)) {
4549 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4553 if (bond_get_targets_ip(arp_target, ip) == -1)
4554 arp_target[arp_ip_count++] = ip;
4556 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4561 if (arp_interval && !arp_ip_count) {
4562 /* don't allow arping if no arp_ip_target given... */
4563 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4569 if (!arp_interval) {
4570 pr_err("arp_validate requires arp_interval\n");
4574 bond_opt_initstr(&newval, arp_validate);
4575 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4578 pr_err("Error: invalid arp_validate \"%s\"\n",
4582 arp_validate_value = valptr->value;
4584 arp_validate_value = 0;
4587 if (arp_all_targets) {
4588 bond_opt_initstr(&newval, arp_all_targets);
4589 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4592 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4594 arp_all_targets_value = 0;
4596 arp_all_targets_value = valptr->value;
4601 pr_info("MII link monitoring set to %d ms\n", miimon);
4602 } else if (arp_interval) {
4603 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4604 arp_validate_value);
4605 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4606 arp_interval, valptr->string, arp_ip_count);
4608 for (i = 0; i < arp_ip_count; i++)
4609 pr_cont(" %s", arp_ip_target[i]);
4613 } else if (max_bonds) {
4614 /* miimon and arp_interval not set, we need one so things
4615 * work as expected, see bonding.txt for details
4617 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");
4620 if (primary && !bond_mode_uses_primary(bond_mode)) {
4621 /* currently, using a primary only makes sense
4622 * in active backup, TLB or ALB modes
4624 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4625 primary, bond_mode_name(bond_mode));
4629 if (primary && primary_reselect) {
4630 bond_opt_initstr(&newval, primary_reselect);
4631 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4634 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4638 primary_reselect_value = valptr->value;
4640 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4643 if (fail_over_mac) {
4644 bond_opt_initstr(&newval, fail_over_mac);
4645 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4648 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4652 fail_over_mac_value = valptr->value;
4653 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4654 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4656 fail_over_mac_value = BOND_FOM_NONE;
4659 bond_opt_initstr(&newval, "default");
4660 valptr = bond_opt_parse(
4661 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4664 pr_err("Error: No ad_actor_sys_prio default value");
4667 ad_actor_sys_prio = valptr->value;
4669 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4672 pr_err("Error: No ad_user_port_key default value");
4675 ad_user_port_key = valptr->value;
4677 bond_opt_initstr(&newval, "default");
4678 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4680 pr_err("Error: No tlb_dynamic_lb default value");
4683 tlb_dynamic_lb = valptr->value;
4685 if (lp_interval == 0) {
4686 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4687 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4688 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4691 /* fill params struct with the proper values */
4692 params->mode = bond_mode;
4693 params->xmit_policy = xmit_hashtype;
4694 params->miimon = miimon;
4695 params->num_peer_notif = num_peer_notif;
4696 params->arp_interval = arp_interval;
4697 params->arp_validate = arp_validate_value;
4698 params->arp_all_targets = arp_all_targets_value;
4699 params->updelay = updelay;
4700 params->downdelay = downdelay;
4701 params->use_carrier = use_carrier;
4702 params->lacp_fast = lacp_fast;
4703 params->primary[0] = 0;
4704 params->primary_reselect = primary_reselect_value;
4705 params->fail_over_mac = fail_over_mac_value;
4706 params->tx_queues = tx_queues;
4707 params->all_slaves_active = all_slaves_active;
4708 params->resend_igmp = resend_igmp;
4709 params->min_links = min_links;
4710 params->lp_interval = lp_interval;
4711 params->packets_per_slave = packets_per_slave;
4712 params->tlb_dynamic_lb = tlb_dynamic_lb;
4713 params->ad_actor_sys_prio = ad_actor_sys_prio;
4714 eth_zero_addr(params->ad_actor_system);
4715 params->ad_user_port_key = ad_user_port_key;
4716 if (packets_per_slave > 0) {
4717 params->reciprocal_packets_per_slave =
4718 reciprocal_value(packets_per_slave);
4720 /* reciprocal_packets_per_slave is unused if
4721 * packets_per_slave is 0 or 1, just initialize it
4723 params->reciprocal_packets_per_slave =
4724 (struct reciprocal_value) { 0 };
4728 strncpy(params->primary, primary, IFNAMSIZ);
4729 params->primary[IFNAMSIZ - 1] = 0;
4732 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4737 /* Called from registration process */
4738 static int bond_init(struct net_device *bond_dev)
4740 struct bonding *bond = netdev_priv(bond_dev);
4741 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4743 netdev_dbg(bond_dev, "Begin bond_init\n");
4745 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4749 bond->nest_level = SINGLE_DEPTH_NESTING;
4750 netdev_lockdep_set_classes(bond_dev);
4752 list_add_tail(&bond->bond_list, &bn->dev_list);
4754 bond_prepare_sysfs_group(bond);
4756 bond_debug_register(bond);
4758 /* Ensure valid dev_addr */
4759 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4760 bond_dev->addr_assign_type == NET_ADDR_PERM)
4761 eth_hw_addr_random(bond_dev);
4766 unsigned int bond_get_num_tx_queues(void)
4771 /* Create a new bond based on the specified name and bonding parameters.
4772 * If name is NULL, obtain a suitable "bond%d" name for us.
4773 * Caller must NOT hold rtnl_lock; we need to release it here before we
4774 * set up our sysfs entries.
4776 int bond_create(struct net *net, const char *name)
4778 struct net_device *bond_dev;
4779 struct bonding *bond;
4780 struct alb_bond_info *bond_info;
4785 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4786 name ? name : "bond%d", NET_NAME_UNKNOWN,
4787 bond_setup, tx_queues);
4789 pr_err("%s: eek! can't alloc netdev!\n", name);
4795 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4796 * It is set to 0 by default which is wrong.
4798 bond = netdev_priv(bond_dev);
4799 bond_info = &(BOND_ALB_INFO(bond));
4800 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4802 dev_net_set(bond_dev, net);
4803 bond_dev->rtnl_link_ops = &bond_link_ops;
4805 res = register_netdevice(bond_dev);
4807 netif_carrier_off(bond_dev);
4809 bond_work_init_all(bond);
4813 free_netdev(bond_dev);
4817 static int __net_init bond_net_init(struct net *net)
4819 struct bond_net *bn = net_generic(net, bond_net_id);
4822 INIT_LIST_HEAD(&bn->dev_list);
4824 bond_create_proc_dir(bn);
4825 bond_create_sysfs(bn);
4830 static void __net_exit bond_net_exit(struct net *net)
4832 struct bond_net *bn = net_generic(net, bond_net_id);
4833 struct bonding *bond, *tmp_bond;
4836 bond_destroy_sysfs(bn);
4838 /* Kill off any bonds created after unregistering bond rtnl ops */
4840 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4841 unregister_netdevice_queue(bond->dev, &list);
4842 unregister_netdevice_many(&list);
4845 bond_destroy_proc_dir(bn);
4848 static struct pernet_operations bond_net_ops = {
4849 .init = bond_net_init,
4850 .exit = bond_net_exit,
4852 .size = sizeof(struct bond_net),
4855 static int __init bonding_init(void)
4860 pr_info("%s", bond_version);
4862 res = bond_check_params(&bonding_defaults);
4866 res = register_pernet_subsys(&bond_net_ops);
4870 res = bond_netlink_init();
4874 bond_create_debugfs();
4876 for (i = 0; i < max_bonds; i++) {
4877 res = bond_create(&init_net, NULL);
4882 register_netdevice_notifier(&bond_netdev_notifier);
4886 bond_destroy_debugfs();
4887 bond_netlink_fini();
4889 unregister_pernet_subsys(&bond_net_ops);
4894 static void __exit bonding_exit(void)
4896 unregister_netdevice_notifier(&bond_netdev_notifier);
4898 bond_destroy_debugfs();
4900 bond_netlink_fini();
4901 unregister_pernet_subsys(&bond_net_ops);
4903 #ifdef CONFIG_NET_POLL_CONTROLLER
4904 /* Make sure we don't have an imbalance on our netpoll blocking */
4905 WARN_ON(atomic_read(&netpoll_block_tx));
4909 module_init(bonding_init);
4910 module_exit(bonding_exit);
4911 MODULE_LICENSE("GPL");
4912 MODULE_VERSION(DRV_VERSION);
4913 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4914 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");