2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 #include "bonding_priv.h"
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 static int packets_per_slave = 1;
114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
116 module_param(max_bonds, int, 0);
117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118 module_param(tx_queues, int, 0);
119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126 module_param(miimon, int, 0);
127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128 module_param(updelay, int, 0);
129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130 module_param(downdelay, int, 0);
131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
133 module_param(use_carrier, int, 0);
134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136 module_param(mode, charp, 0);
137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141 module_param(primary, charp, 0);
142 MODULE_PARM_DESC(primary, "Primary network device to use");
143 module_param(primary_reselect, charp, 0);
144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
149 "2 for only on active slave "
151 module_param(lacp_rate, charp, 0);
152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154 module_param(ad_select, charp, 0);
155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
158 module_param(min_links, int, 0);
159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166 module_param(arp_interval, int, 0);
167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168 module_param_array(arp_ip_target, charp, NULL, 0);
169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170 module_param(arp_validate, charp, 0);
171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174 module_param(arp_all_targets, charp, 0);
175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac, charp, 0);
177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180 module_param(all_slaves_active, int, 0);
181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184 module_param(resend_igmp, int, 0);
185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
187 module_param(packets_per_slave, int, 0);
188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191 module_param(lp_interval, uint, 0);
192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
196 /*----------------------------- Global variables ----------------------------*/
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
202 unsigned int bond_net_id __read_mostly;
204 /*-------------------------- Forward declarations ---------------------------*/
206 static int bond_init(struct net_device *bond_dev);
207 static void bond_uninit(struct net_device *bond_dev);
208 static void bond_get_stats(struct net_device *bond_dev,
209 struct rtnl_link_stats64 *stats);
210 static void bond_slave_arr_handler(struct work_struct *work);
211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
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 void bond_set_dev_addr(struct net_device *bond_dev,
612 struct net_device *slave_dev)
614 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
615 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
616 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
617 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
618 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
621 static struct slave *bond_get_old_active(struct bonding *bond,
622 struct slave *new_active)
625 struct list_head *iter;
627 bond_for_each_slave(bond, slave, iter) {
628 if (slave == new_active)
631 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
638 /* bond_do_fail_over_mac
640 * Perform special MAC address swapping for fail_over_mac settings
644 static void bond_do_fail_over_mac(struct bonding *bond,
645 struct slave *new_active,
646 struct slave *old_active)
648 u8 tmp_mac[MAX_ADDR_LEN];
649 struct sockaddr_storage ss;
652 switch (bond->params.fail_over_mac) {
653 case BOND_FOM_ACTIVE:
655 bond_set_dev_addr(bond->dev, new_active->dev);
657 case BOND_FOM_FOLLOW:
658 /* if new_active && old_active, swap them
659 * if just old_active, do nothing (going to no active slave)
660 * if just new_active, set new_active to bond's MAC
666 old_active = bond_get_old_active(bond, new_active);
669 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
670 new_active->dev->addr_len);
671 bond_hw_addr_copy(ss.__data,
672 old_active->dev->dev_addr,
673 old_active->dev->addr_len);
674 ss.ss_family = new_active->dev->type;
676 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
677 bond->dev->addr_len);
678 ss.ss_family = bond->dev->type;
681 rv = dev_set_mac_address(new_active->dev,
682 (struct sockaddr *)&ss);
684 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
685 -rv, new_active->dev->name);
692 bond_hw_addr_copy(ss.__data, tmp_mac,
693 new_active->dev->addr_len);
694 ss.ss_family = old_active->dev->type;
696 rv = dev_set_mac_address(old_active->dev,
697 (struct sockaddr *)&ss);
699 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
700 -rv, new_active->dev->name);
704 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
705 bond->params.fail_over_mac);
711 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
713 struct slave *prim = rtnl_dereference(bond->primary_slave);
714 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
716 if (!prim || prim->link != BOND_LINK_UP) {
717 if (!curr || curr->link != BOND_LINK_UP)
722 if (bond->force_primary) {
723 bond->force_primary = false;
727 if (!curr || curr->link != BOND_LINK_UP)
730 /* At this point, prim and curr are both up */
731 switch (bond->params.primary_reselect) {
732 case BOND_PRI_RESELECT_ALWAYS:
734 case BOND_PRI_RESELECT_BETTER:
735 if (prim->speed < curr->speed)
737 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
740 case BOND_PRI_RESELECT_FAILURE:
743 netdev_err(bond->dev, "impossible primary_reselect %d\n",
744 bond->params.primary_reselect);
750 * bond_find_best_slave - select the best available slave to be the active one
751 * @bond: our bonding struct
753 static struct slave *bond_find_best_slave(struct bonding *bond)
755 struct slave *slave, *bestslave = NULL;
756 struct list_head *iter;
757 int mintime = bond->params.updelay;
759 slave = bond_choose_primary_or_current(bond);
763 bond_for_each_slave(bond, slave, iter) {
764 if (slave->link == BOND_LINK_UP)
766 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
767 slave->delay < mintime) {
768 mintime = slave->delay;
776 static bool bond_should_notify_peers(struct bonding *bond)
781 slave = rcu_dereference(bond->curr_active_slave);
784 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
785 slave ? slave->dev->name : "NULL");
787 if (!slave || !bond->send_peer_notif ||
788 !netif_carrier_ok(bond->dev) ||
789 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
796 * change_active_interface - change the active slave into the specified one
797 * @bond: our bonding struct
798 * @new: the new slave to make the active one
800 * Set the new slave to the bond's settings and unset them on the old
802 * Setting include flags, mc-list, promiscuity, allmulti, etc.
804 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
805 * because it is apparently the best available slave we have, even though its
806 * updelay hasn't timed out yet.
808 * Caller must hold RTNL.
810 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
812 struct slave *old_active;
816 old_active = rtnl_dereference(bond->curr_active_slave);
818 if (old_active == new_active)
822 new_active->last_link_up = jiffies;
824 if (new_active->link == BOND_LINK_BACK) {
825 if (bond_uses_primary(bond)) {
826 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
827 new_active->dev->name,
828 (bond->params.updelay - new_active->delay) * bond->params.miimon);
831 new_active->delay = 0;
832 bond_set_slave_link_state(new_active, BOND_LINK_UP,
833 BOND_SLAVE_NOTIFY_NOW);
835 if (BOND_MODE(bond) == BOND_MODE_8023AD)
836 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
838 if (bond_is_lb(bond))
839 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
841 if (bond_uses_primary(bond)) {
842 netdev_info(bond->dev, "making interface %s the new active one\n",
843 new_active->dev->name);
848 if (bond_uses_primary(bond))
849 bond_hw_addr_swap(bond, new_active, old_active);
851 if (bond_is_lb(bond)) {
852 bond_alb_handle_active_change(bond, new_active);
854 bond_set_slave_inactive_flags(old_active,
855 BOND_SLAVE_NOTIFY_NOW);
857 bond_set_slave_active_flags(new_active,
858 BOND_SLAVE_NOTIFY_NOW);
860 rcu_assign_pointer(bond->curr_active_slave, new_active);
863 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
865 bond_set_slave_inactive_flags(old_active,
866 BOND_SLAVE_NOTIFY_NOW);
869 bool should_notify_peers = false;
871 bond_set_slave_active_flags(new_active,
872 BOND_SLAVE_NOTIFY_NOW);
874 if (bond->params.fail_over_mac)
875 bond_do_fail_over_mac(bond, new_active,
878 if (netif_running(bond->dev)) {
879 bond->send_peer_notif =
880 bond->params.num_peer_notif;
881 should_notify_peers =
882 bond_should_notify_peers(bond);
885 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
886 if (should_notify_peers)
887 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
892 /* resend IGMP joins since active slave has changed or
893 * all were sent on curr_active_slave.
894 * resend only if bond is brought up with the affected
895 * bonding modes and the retransmission is enabled
897 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
898 ((bond_uses_primary(bond) && new_active) ||
899 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
900 bond->igmp_retrans = bond->params.resend_igmp;
901 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
906 * bond_select_active_slave - select a new active slave, if needed
907 * @bond: our bonding struct
909 * This functions should be called when one of the following occurs:
910 * - The old curr_active_slave has been released or lost its link.
911 * - The primary_slave has got its link back.
912 * - A slave has got its link back and there's no old curr_active_slave.
914 * Caller must hold RTNL.
916 void bond_select_active_slave(struct bonding *bond)
918 struct slave *best_slave;
923 best_slave = bond_find_best_slave(bond);
924 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
925 bond_change_active_slave(bond, best_slave);
926 rv = bond_set_carrier(bond);
930 if (netif_carrier_ok(bond->dev))
931 netdev_info(bond->dev, "first active interface up!\n");
933 netdev_info(bond->dev, "now running without any active interface!\n");
937 #ifdef CONFIG_NET_POLL_CONTROLLER
938 static inline int slave_enable_netpoll(struct slave *slave)
943 np = kzalloc(sizeof(*np), GFP_KERNEL);
948 err = __netpoll_setup(np, slave->dev);
957 static inline void slave_disable_netpoll(struct slave *slave)
959 struct netpoll *np = slave->np;
965 __netpoll_free_async(np);
968 static void bond_poll_controller(struct net_device *bond_dev)
970 struct bonding *bond = netdev_priv(bond_dev);
971 struct slave *slave = NULL;
972 struct list_head *iter;
973 struct ad_info ad_info;
974 struct netpoll_info *ni;
975 const struct net_device_ops *ops;
977 if (BOND_MODE(bond) == BOND_MODE_8023AD)
978 if (bond_3ad_get_active_agg_info(bond, &ad_info))
981 bond_for_each_slave_rcu(bond, slave, iter) {
982 ops = slave->dev->netdev_ops;
983 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
986 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
987 struct aggregator *agg =
988 SLAVE_AD_INFO(slave)->port.aggregator;
991 agg->aggregator_identifier != ad_info.aggregator_id)
995 ni = rcu_dereference_bh(slave->dev->npinfo);
996 if (down_trylock(&ni->dev_lock))
998 ops->ndo_poll_controller(slave->dev);
1003 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1005 struct bonding *bond = netdev_priv(bond_dev);
1006 struct list_head *iter;
1007 struct slave *slave;
1009 bond_for_each_slave(bond, slave, iter)
1010 if (bond_slave_is_up(slave))
1011 slave_disable_netpoll(slave);
1014 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1016 struct bonding *bond = netdev_priv(dev);
1017 struct list_head *iter;
1018 struct slave *slave;
1021 bond_for_each_slave(bond, slave, iter) {
1022 err = slave_enable_netpoll(slave);
1024 bond_netpoll_cleanup(dev);
1031 static inline int slave_enable_netpoll(struct slave *slave)
1035 static inline void slave_disable_netpoll(struct slave *slave)
1038 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1043 /*---------------------------------- IOCTL ----------------------------------*/
1045 static netdev_features_t bond_fix_features(struct net_device *dev,
1046 netdev_features_t features)
1048 struct bonding *bond = netdev_priv(dev);
1049 struct list_head *iter;
1050 netdev_features_t mask;
1051 struct slave *slave;
1055 features &= ~NETIF_F_ONE_FOR_ALL;
1056 features |= NETIF_F_ALL_FOR_ALL;
1058 bond_for_each_slave(bond, slave, iter) {
1059 features = netdev_increment_features(features,
1060 slave->dev->features,
1063 features = netdev_add_tso_features(features, mask);
1068 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1069 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1070 NETIF_F_HIGHDMA | NETIF_F_LRO)
1072 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1073 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1075 static void bond_compute_features(struct bonding *bond)
1077 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1078 IFF_XMIT_DST_RELEASE_PERM;
1079 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1080 netdev_features_t enc_features = BOND_ENC_FEATURES;
1081 struct net_device *bond_dev = bond->dev;
1082 struct list_head *iter;
1083 struct slave *slave;
1084 unsigned short max_hard_header_len = ETH_HLEN;
1085 unsigned int gso_max_size = GSO_MAX_SIZE;
1086 u16 gso_max_segs = GSO_MAX_SEGS;
1088 if (!bond_has_slaves(bond))
1090 vlan_features &= NETIF_F_ALL_FOR_ALL;
1092 bond_for_each_slave(bond, slave, iter) {
1093 vlan_features = netdev_increment_features(vlan_features,
1094 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1096 enc_features = netdev_increment_features(enc_features,
1097 slave->dev->hw_enc_features,
1099 dst_release_flag &= slave->dev->priv_flags;
1100 if (slave->dev->hard_header_len > max_hard_header_len)
1101 max_hard_header_len = slave->dev->hard_header_len;
1103 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1104 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1106 bond_dev->hard_header_len = max_hard_header_len;
1109 bond_dev->vlan_features = vlan_features;
1110 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1112 bond_dev->gso_max_segs = gso_max_segs;
1113 netif_set_gso_max_size(bond_dev, gso_max_size);
1115 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1116 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1117 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1118 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1120 netdev_change_features(bond_dev);
1123 static void bond_setup_by_slave(struct net_device *bond_dev,
1124 struct net_device *slave_dev)
1126 bond_dev->header_ops = slave_dev->header_ops;
1128 bond_dev->type = slave_dev->type;
1129 bond_dev->hard_header_len = slave_dev->hard_header_len;
1130 bond_dev->addr_len = slave_dev->addr_len;
1132 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1133 slave_dev->addr_len);
1136 /* On bonding slaves other than the currently active slave, suppress
1137 * duplicates except for alb non-mcast/bcast.
1139 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1140 struct slave *slave,
1141 struct bonding *bond)
1143 if (bond_is_slave_inactive(slave)) {
1144 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1145 skb->pkt_type != PACKET_BROADCAST &&
1146 skb->pkt_type != PACKET_MULTICAST)
1153 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1155 struct sk_buff *skb = *pskb;
1156 struct slave *slave;
1157 struct bonding *bond;
1158 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1160 int ret = RX_HANDLER_ANOTHER;
1162 skb = skb_share_check(skb, GFP_ATOMIC);
1164 return RX_HANDLER_CONSUMED;
1168 slave = bond_slave_get_rcu(skb->dev);
1171 recv_probe = READ_ONCE(bond->recv_probe);
1173 ret = recv_probe(skb, bond, slave);
1174 if (ret == RX_HANDLER_CONSUMED) {
1180 /* don't change skb->dev for link-local packets */
1181 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1182 return RX_HANDLER_PASS;
1183 if (bond_should_deliver_exact_match(skb, slave, bond))
1184 return RX_HANDLER_EXACT;
1186 skb->dev = bond->dev;
1188 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1189 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1190 skb->pkt_type == PACKET_HOST) {
1192 if (unlikely(skb_cow_head(skb,
1193 skb->data - skb_mac_header(skb)))) {
1195 return RX_HANDLER_CONSUMED;
1197 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1198 bond->dev->addr_len);
1204 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1206 switch (BOND_MODE(bond)) {
1207 case BOND_MODE_ROUNDROBIN:
1208 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1209 case BOND_MODE_ACTIVEBACKUP:
1210 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1211 case BOND_MODE_BROADCAST:
1212 return NETDEV_LAG_TX_TYPE_BROADCAST;
1214 case BOND_MODE_8023AD:
1215 return NETDEV_LAG_TX_TYPE_HASH;
1217 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1221 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1222 enum netdev_lag_tx_type type)
1224 if (type != NETDEV_LAG_TX_TYPE_HASH)
1225 return NETDEV_LAG_HASH_NONE;
1227 switch (bond->params.xmit_policy) {
1228 case BOND_XMIT_POLICY_LAYER2:
1229 return NETDEV_LAG_HASH_L2;
1230 case BOND_XMIT_POLICY_LAYER34:
1231 return NETDEV_LAG_HASH_L34;
1232 case BOND_XMIT_POLICY_LAYER23:
1233 return NETDEV_LAG_HASH_L23;
1234 case BOND_XMIT_POLICY_ENCAP23:
1235 return NETDEV_LAG_HASH_E23;
1236 case BOND_XMIT_POLICY_ENCAP34:
1237 return NETDEV_LAG_HASH_E34;
1239 return NETDEV_LAG_HASH_UNKNOWN;
1243 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1244 struct netlink_ext_ack *extack)
1246 struct netdev_lag_upper_info lag_upper_info;
1247 enum netdev_lag_tx_type type;
1249 type = bond_lag_tx_type(bond);
1250 lag_upper_info.tx_type = type;
1251 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1253 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1254 &lag_upper_info, extack);
1257 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1259 netdev_upper_dev_unlink(slave->dev, bond->dev);
1260 slave->dev->flags &= ~IFF_SLAVE;
1263 static struct slave *bond_alloc_slave(struct bonding *bond)
1265 struct slave *slave = NULL;
1267 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1271 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1272 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1274 if (!SLAVE_AD_INFO(slave)) {
1282 static void bond_free_slave(struct slave *slave)
1284 struct bonding *bond = bond_get_bond_by_slave(slave);
1286 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1287 kfree(SLAVE_AD_INFO(slave));
1292 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1294 info->bond_mode = BOND_MODE(bond);
1295 info->miimon = bond->params.miimon;
1296 info->num_slaves = bond->slave_cnt;
1299 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1301 strcpy(info->slave_name, slave->dev->name);
1302 info->link = slave->link;
1303 info->state = bond_slave_state(slave);
1304 info->link_failure_count = slave->link_failure_count;
1307 static void bond_netdev_notify(struct net_device *dev,
1308 struct netdev_bonding_info *info)
1311 netdev_bonding_info_change(dev, info);
1315 static void bond_netdev_notify_work(struct work_struct *_work)
1317 struct netdev_notify_work *w =
1318 container_of(_work, struct netdev_notify_work, work.work);
1320 bond_netdev_notify(w->dev, &w->bonding_info);
1325 void bond_queue_slave_event(struct slave *slave)
1327 struct bonding *bond = slave->bond;
1328 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1333 dev_hold(slave->dev);
1334 nnw->dev = slave->dev;
1335 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1336 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1337 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1339 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1342 void bond_lower_state_changed(struct slave *slave)
1344 struct netdev_lag_lower_state_info info;
1346 info.link_up = slave->link == BOND_LINK_UP ||
1347 slave->link == BOND_LINK_FAIL;
1348 info.tx_enabled = bond_is_active_slave(slave);
1349 netdev_lower_state_changed(slave->dev, &info);
1352 /* enslave device <slave> to bond device <master> */
1353 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1354 struct netlink_ext_ack *extack)
1356 struct bonding *bond = netdev_priv(bond_dev);
1357 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1358 struct slave *new_slave = NULL, *prev_slave;
1359 struct sockaddr_storage ss;
1363 if (!bond->params.use_carrier &&
1364 slave_dev->ethtool_ops->get_link == NULL &&
1365 slave_ops->ndo_do_ioctl == NULL) {
1366 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1370 /* already in-use? */
1371 if (netdev_is_rx_handler_busy(slave_dev)) {
1372 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1373 netdev_err(bond_dev,
1374 "Error: Device is in use and cannot be enslaved\n");
1378 if (bond_dev == slave_dev) {
1379 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1380 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1384 /* vlan challenged mutual exclusion */
1385 /* no need to lock since we're protected by rtnl_lock */
1386 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1387 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1389 if (vlan_uses_dev(bond_dev)) {
1390 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1391 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1392 slave_dev->name, bond_dev->name);
1395 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1396 slave_dev->name, slave_dev->name,
1400 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1404 /* Old ifenslave binaries are no longer supported. These can
1405 * be identified with moderate accuracy by the state of the slave:
1406 * the current ifenslave will set the interface down prior to
1407 * enslaving it; the old ifenslave will not.
1409 if (slave_dev->flags & IFF_UP) {
1410 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1411 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1416 /* set bonding device ether type by slave - bonding netdevices are
1417 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1418 * there is a need to override some of the type dependent attribs/funcs.
1420 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1421 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1423 if (!bond_has_slaves(bond)) {
1424 if (bond_dev->type != slave_dev->type) {
1425 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1426 bond_dev->type, slave_dev->type);
1428 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1430 res = notifier_to_errno(res);
1432 netdev_err(bond_dev, "refused to change device type\n");
1436 /* Flush unicast and multicast addresses */
1437 dev_uc_flush(bond_dev);
1438 dev_mc_flush(bond_dev);
1440 if (slave_dev->type != ARPHRD_ETHER)
1441 bond_setup_by_slave(bond_dev, slave_dev);
1443 ether_setup(bond_dev);
1444 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1447 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1450 } else if (bond_dev->type != slave_dev->type) {
1451 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1452 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1453 slave_dev->name, slave_dev->type, bond_dev->type);
1457 if (slave_dev->type == ARPHRD_INFINIBAND &&
1458 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1459 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1460 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1463 goto err_undo_flags;
1466 if (!slave_ops->ndo_set_mac_address ||
1467 slave_dev->type == ARPHRD_INFINIBAND) {
1468 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1469 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1470 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1471 if (!bond_has_slaves(bond)) {
1472 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1473 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1475 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1476 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");
1478 goto err_undo_flags;
1483 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1485 /* If this is the first slave, then we need to set the master's hardware
1486 * address to be the same as the slave's.
1488 if (!bond_has_slaves(bond) &&
1489 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1490 bond_set_dev_addr(bond->dev, slave_dev);
1492 new_slave = bond_alloc_slave(bond);
1495 goto err_undo_flags;
1498 new_slave->bond = bond;
1499 new_slave->dev = slave_dev;
1500 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1501 * is set via sysfs or module option if desired.
1503 new_slave->queue_id = 0;
1505 /* Save slave's original mtu and then set it to match the bond */
1506 new_slave->original_mtu = slave_dev->mtu;
1507 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1509 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1513 /* Save slave's original ("permanent") mac address for modes
1514 * that need it, and for restoring it upon release, and then
1515 * set it to the master's address
1517 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1518 slave_dev->addr_len);
1520 if (!bond->params.fail_over_mac ||
1521 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1522 /* Set slave to master's mac address. The application already
1523 * set the master's mac address to that of the first slave
1525 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1526 ss.ss_family = slave_dev->type;
1527 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1529 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1530 goto err_restore_mtu;
1534 /* set slave flag before open to prevent IPv6 addrconf */
1535 slave_dev->flags |= IFF_SLAVE;
1537 /* open the slave since the application closed it */
1538 res = dev_open(slave_dev);
1540 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1541 goto err_restore_mac;
1544 slave_dev->priv_flags |= IFF_BONDING;
1545 /* initialize slave stats */
1546 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1548 if (bond_is_lb(bond)) {
1549 /* bond_alb_init_slave() must be called before all other stages since
1550 * it might fail and we do not want to have to undo everything
1552 res = bond_alb_init_slave(bond, new_slave);
1557 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1559 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1564 prev_slave = bond_last_slave(bond);
1566 new_slave->delay = 0;
1567 new_slave->link_failure_count = 0;
1569 if (bond_update_speed_duplex(new_slave) &&
1570 bond_needs_speed_duplex(bond))
1571 new_slave->link = BOND_LINK_DOWN;
1573 new_slave->last_rx = jiffies -
1574 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1575 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1576 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1578 if (bond->params.miimon && !bond->params.use_carrier) {
1579 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1581 if ((link_reporting == -1) && !bond->params.arp_interval) {
1582 /* miimon is set but a bonded network driver
1583 * does not support ETHTOOL/MII and
1584 * arp_interval is not set. Note: if
1585 * use_carrier is enabled, we will never go
1586 * here (because netif_carrier is always
1587 * supported); thus, we don't need to change
1588 * the messages for netif_carrier.
1590 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",
1592 } else if (link_reporting == -1) {
1593 /* unable get link status using mii/ethtool */
1594 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",
1599 /* check for initial state */
1600 new_slave->link = BOND_LINK_NOCHANGE;
1601 if (bond->params.miimon) {
1602 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1603 if (bond->params.updelay) {
1604 bond_set_slave_link_state(new_slave,
1606 BOND_SLAVE_NOTIFY_NOW);
1607 new_slave->delay = bond->params.updelay;
1609 bond_set_slave_link_state(new_slave,
1611 BOND_SLAVE_NOTIFY_NOW);
1614 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1615 BOND_SLAVE_NOTIFY_NOW);
1617 } else if (bond->params.arp_interval) {
1618 bond_set_slave_link_state(new_slave,
1619 (netif_carrier_ok(slave_dev) ?
1620 BOND_LINK_UP : BOND_LINK_DOWN),
1621 BOND_SLAVE_NOTIFY_NOW);
1623 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1624 BOND_SLAVE_NOTIFY_NOW);
1627 if (new_slave->link != BOND_LINK_DOWN)
1628 new_slave->last_link_up = jiffies;
1629 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1630 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1631 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1633 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1634 /* if there is a primary slave, remember it */
1635 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1636 rcu_assign_pointer(bond->primary_slave, new_slave);
1637 bond->force_primary = true;
1641 switch (BOND_MODE(bond)) {
1642 case BOND_MODE_ACTIVEBACKUP:
1643 bond_set_slave_inactive_flags(new_slave,
1644 BOND_SLAVE_NOTIFY_NOW);
1646 case BOND_MODE_8023AD:
1647 /* in 802.3ad mode, the internal mechanism
1648 * will activate the slaves in the selected
1651 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1652 /* if this is the first slave */
1654 SLAVE_AD_INFO(new_slave)->id = 1;
1655 /* Initialize AD with the number of times that the AD timer is called in 1 second
1656 * can be called only after the mac address of the bond is set
1658 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1660 SLAVE_AD_INFO(new_slave)->id =
1661 SLAVE_AD_INFO(prev_slave)->id + 1;
1664 bond_3ad_bind_slave(new_slave);
1668 bond_set_active_slave(new_slave);
1669 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1672 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1674 /* always active in trunk mode */
1675 bond_set_active_slave(new_slave);
1677 /* In trunking mode there is little meaning to curr_active_slave
1678 * anyway (it holds no special properties of the bond device),
1679 * so we can change it without calling change_active_interface()
1681 if (!rcu_access_pointer(bond->curr_active_slave) &&
1682 new_slave->link == BOND_LINK_UP)
1683 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1686 } /* switch(bond_mode) */
1688 #ifdef CONFIG_NET_POLL_CONTROLLER
1689 if (bond->dev->npinfo) {
1690 if (slave_enable_netpoll(new_slave)) {
1691 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1698 if (!(bond_dev->features & NETIF_F_LRO))
1699 dev_disable_lro(slave_dev);
1701 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1704 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1708 res = bond_master_upper_dev_link(bond, new_slave, extack);
1710 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1711 goto err_unregister;
1714 res = bond_sysfs_slave_add(new_slave);
1716 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1717 goto err_upper_unlink;
1720 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1722 /* If the mode uses primary, then the following is handled by
1723 * bond_change_active_slave().
1725 if (!bond_uses_primary(bond)) {
1726 /* set promiscuity level to new slave */
1727 if (bond_dev->flags & IFF_PROMISC) {
1728 res = dev_set_promiscuity(slave_dev, 1);
1733 /* set allmulti level to new slave */
1734 if (bond_dev->flags & IFF_ALLMULTI) {
1735 res = dev_set_allmulti(slave_dev, 1);
1737 if (bond_dev->flags & IFF_PROMISC)
1738 dev_set_promiscuity(slave_dev, -1);
1743 netif_addr_lock_bh(bond_dev);
1744 dev_mc_sync_multiple(slave_dev, bond_dev);
1745 dev_uc_sync_multiple(slave_dev, bond_dev);
1746 netif_addr_unlock_bh(bond_dev);
1748 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1749 /* add lacpdu mc addr to mc list */
1750 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1752 dev_mc_add(slave_dev, lacpdu_multicast);
1757 bond_compute_features(bond);
1758 bond_set_carrier(bond);
1760 if (bond_uses_primary(bond)) {
1762 bond_select_active_slave(bond);
1763 unblock_netpoll_tx();
1766 if (bond_mode_can_use_xmit_hash(bond))
1767 bond_update_slave_arr(bond, NULL);
1770 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1772 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1773 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1775 /* enslave is successful */
1776 bond_queue_slave_event(new_slave);
1779 /* Undo stages on error */
1781 bond_sysfs_slave_del(new_slave);
1784 bond_upper_dev_unlink(bond, new_slave);
1787 netdev_rx_handler_unregister(slave_dev);
1790 vlan_vids_del_by_dev(slave_dev, bond_dev);
1791 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1792 RCU_INIT_POINTER(bond->primary_slave, NULL);
1793 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1795 bond_change_active_slave(bond, NULL);
1796 bond_select_active_slave(bond);
1797 unblock_netpoll_tx();
1799 /* either primary_slave or curr_active_slave might've changed */
1801 slave_disable_netpoll(new_slave);
1804 slave_dev->priv_flags &= ~IFF_BONDING;
1805 dev_close(slave_dev);
1808 slave_dev->flags &= ~IFF_SLAVE;
1809 if (!bond->params.fail_over_mac ||
1810 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1811 /* XXX TODO - fom follow mode needs to change master's
1812 * MAC if this slave's MAC is in use by the bond, or at
1813 * least print a warning.
1815 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1816 new_slave->dev->addr_len);
1817 ss.ss_family = slave_dev->type;
1818 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1822 dev_set_mtu(slave_dev, new_slave->original_mtu);
1825 bond_free_slave(new_slave);
1828 /* Enslave of first slave has failed and we need to fix master's mac */
1829 if (!bond_has_slaves(bond)) {
1830 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1831 slave_dev->dev_addr))
1832 eth_hw_addr_random(bond_dev);
1833 if (bond_dev->type != ARPHRD_ETHER) {
1834 dev_close(bond_dev);
1835 ether_setup(bond_dev);
1836 bond_dev->flags |= IFF_MASTER;
1837 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1844 /* Try to release the slave device <slave> from the bond device <master>
1845 * It is legal to access curr_active_slave without a lock because all the function
1846 * is RTNL-locked. If "all" is true it means that the function is being called
1847 * while destroying a bond interface and all slaves are being released.
1849 * The rules for slave state should be:
1850 * for Active/Backup:
1851 * Active stays on all backups go down
1852 * for Bonded connections:
1853 * The first up interface should be left on and all others downed.
1855 static int __bond_release_one(struct net_device *bond_dev,
1856 struct net_device *slave_dev,
1857 bool all, bool unregister)
1859 struct bonding *bond = netdev_priv(bond_dev);
1860 struct slave *slave, *oldcurrent;
1861 struct sockaddr_storage ss;
1862 int old_flags = bond_dev->flags;
1863 netdev_features_t old_features = bond_dev->features;
1865 /* slave is not a slave or master is not master of this slave */
1866 if (!(slave_dev->flags & IFF_SLAVE) ||
1867 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1868 netdev_dbg(bond_dev, "cannot release %s\n",
1875 slave = bond_get_slave_by_dev(bond, slave_dev);
1877 /* not a slave of this bond */
1878 netdev_info(bond_dev, "%s not enslaved\n",
1880 unblock_netpoll_tx();
1884 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1886 bond_sysfs_slave_del(slave);
1888 /* recompute stats just before removing the slave */
1889 bond_get_stats(bond->dev, &bond->bond_stats);
1891 bond_upper_dev_unlink(bond, slave);
1892 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1893 * for this slave anymore.
1895 netdev_rx_handler_unregister(slave_dev);
1897 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1898 bond_3ad_unbind_slave(slave);
1900 if (bond_mode_can_use_xmit_hash(bond))
1901 bond_update_slave_arr(bond, slave);
1903 netdev_info(bond_dev, "Releasing %s interface %s\n",
1904 bond_is_active_slave(slave) ? "active" : "backup",
1907 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1909 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1911 if (!all && (!bond->params.fail_over_mac ||
1912 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1913 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1914 bond_has_slaves(bond))
1915 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",
1916 slave_dev->name, slave->perm_hwaddr,
1917 bond_dev->name, slave_dev->name);
1920 if (rtnl_dereference(bond->primary_slave) == slave)
1921 RCU_INIT_POINTER(bond->primary_slave, NULL);
1923 if (oldcurrent == slave)
1924 bond_change_active_slave(bond, NULL);
1926 if (bond_is_lb(bond)) {
1927 /* Must be called only after the slave has been
1928 * detached from the list and the curr_active_slave
1929 * has been cleared (if our_slave == old_current),
1930 * but before a new active slave is selected.
1932 bond_alb_deinit_slave(bond, slave);
1936 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1937 } else if (oldcurrent == slave) {
1938 /* Note that we hold RTNL over this sequence, so there
1939 * is no concern that another slave add/remove event
1942 bond_select_active_slave(bond);
1945 if (!bond_has_slaves(bond)) {
1946 bond_set_carrier(bond);
1947 eth_hw_addr_random(bond_dev);
1950 unblock_netpoll_tx();
1954 if (!bond_has_slaves(bond)) {
1955 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1956 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1959 bond_compute_features(bond);
1960 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1961 (old_features & NETIF_F_VLAN_CHALLENGED))
1962 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1963 slave_dev->name, bond_dev->name);
1965 vlan_vids_del_by_dev(slave_dev, bond_dev);
1967 /* If the mode uses primary, then this case was handled above by
1968 * bond_change_active_slave(..., NULL)
1970 if (!bond_uses_primary(bond)) {
1971 /* unset promiscuity level from slave
1972 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1973 * of the IFF_PROMISC flag in the bond_dev, but we need the
1974 * value of that flag before that change, as that was the value
1975 * when this slave was attached, so we cache at the start of the
1976 * function and use it here. Same goes for ALLMULTI below
1978 if (old_flags & IFF_PROMISC)
1979 dev_set_promiscuity(slave_dev, -1);
1981 /* unset allmulti level from slave */
1982 if (old_flags & IFF_ALLMULTI)
1983 dev_set_allmulti(slave_dev, -1);
1985 bond_hw_addr_flush(bond_dev, slave_dev);
1988 slave_disable_netpoll(slave);
1990 /* close slave before restoring its mac address */
1991 dev_close(slave_dev);
1993 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1994 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1995 /* restore original ("permanent") mac address */
1996 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
1997 slave->dev->addr_len);
1998 ss.ss_family = slave_dev->type;
1999 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
2003 __dev_set_mtu(slave_dev, slave->original_mtu);
2005 dev_set_mtu(slave_dev, slave->original_mtu);
2007 slave_dev->priv_flags &= ~IFF_BONDING;
2009 bond_free_slave(slave);
2014 /* A wrapper used because of ndo_del_link */
2015 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2017 return __bond_release_one(bond_dev, slave_dev, false, false);
2020 /* First release a slave and then destroy the bond if no more slaves are left.
2021 * Must be under rtnl_lock when this function is called.
2023 static int bond_release_and_destroy(struct net_device *bond_dev,
2024 struct net_device *slave_dev)
2026 struct bonding *bond = netdev_priv(bond_dev);
2029 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2030 if (ret == 0 && !bond_has_slaves(bond)) {
2031 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2032 netdev_info(bond_dev, "Destroying bond %s\n",
2034 bond_remove_proc_entry(bond);
2035 unregister_netdevice(bond_dev);
2040 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2042 struct bonding *bond = netdev_priv(bond_dev);
2043 bond_fill_ifbond(bond, info);
2046 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2048 struct bonding *bond = netdev_priv(bond_dev);
2049 struct list_head *iter;
2050 int i = 0, res = -ENODEV;
2051 struct slave *slave;
2053 bond_for_each_slave(bond, slave, iter) {
2054 if (i++ == (int)info->slave_id) {
2056 bond_fill_ifslave(slave, info);
2064 /*-------------------------------- Monitoring -------------------------------*/
2066 /* called with rcu_read_lock() */
2067 static int bond_miimon_inspect(struct bonding *bond)
2069 int link_state, commit = 0;
2070 struct list_head *iter;
2071 struct slave *slave;
2072 bool ignore_updelay;
2074 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2076 bond_for_each_slave_rcu(bond, slave, iter) {
2077 slave->new_link = BOND_LINK_NOCHANGE;
2078 slave->link_new_state = slave->link;
2080 link_state = bond_check_dev_link(bond, slave->dev, 0);
2082 switch (slave->link) {
2087 bond_propose_link_state(slave, BOND_LINK_FAIL);
2089 slave->delay = bond->params.downdelay;
2091 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2093 BOND_MODE_ACTIVEBACKUP) ?
2094 (bond_is_active_slave(slave) ?
2095 "active " : "backup ") : "",
2097 bond->params.downdelay * bond->params.miimon);
2100 case BOND_LINK_FAIL:
2102 /* recovered before downdelay expired */
2103 bond_propose_link_state(slave, BOND_LINK_UP);
2104 slave->last_link_up = jiffies;
2105 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2106 (bond->params.downdelay - slave->delay) *
2107 bond->params.miimon,
2113 if (slave->delay <= 0) {
2114 slave->new_link = BOND_LINK_DOWN;
2122 case BOND_LINK_DOWN:
2126 bond_propose_link_state(slave, BOND_LINK_BACK);
2128 slave->delay = bond->params.updelay;
2131 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2133 ignore_updelay ? 0 :
2134 bond->params.updelay *
2135 bond->params.miimon);
2138 case BOND_LINK_BACK:
2140 bond_propose_link_state(slave, BOND_LINK_DOWN);
2141 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2142 (bond->params.updelay - slave->delay) *
2143 bond->params.miimon,
2152 if (slave->delay <= 0) {
2153 slave->new_link = BOND_LINK_UP;
2155 ignore_updelay = false;
2167 static void bond_miimon_link_change(struct bonding *bond,
2168 struct slave *slave,
2171 switch (BOND_MODE(bond)) {
2172 case BOND_MODE_8023AD:
2173 bond_3ad_handle_link_change(slave, link);
2177 bond_alb_handle_link_change(bond, slave, link);
2180 bond_update_slave_arr(bond, NULL);
2185 static void bond_miimon_commit(struct bonding *bond)
2187 struct list_head *iter;
2188 struct slave *slave, *primary;
2190 bond_for_each_slave(bond, slave, iter) {
2191 switch (slave->new_link) {
2192 case BOND_LINK_NOCHANGE:
2196 if (bond_update_speed_duplex(slave) &&
2197 bond_needs_speed_duplex(bond)) {
2198 slave->link = BOND_LINK_DOWN;
2199 if (net_ratelimit())
2200 netdev_warn(bond->dev,
2201 "failed to get link speed/duplex for %s\n",
2205 bond_set_slave_link_state(slave, BOND_LINK_UP,
2206 BOND_SLAVE_NOTIFY_NOW);
2207 slave->last_link_up = jiffies;
2209 primary = rtnl_dereference(bond->primary_slave);
2210 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2211 /* prevent it from being the active one */
2212 bond_set_backup_slave(slave);
2213 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2214 /* make it immediately active */
2215 bond_set_active_slave(slave);
2216 } else if (slave != primary) {
2217 /* prevent it from being the active one */
2218 bond_set_backup_slave(slave);
2221 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2223 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2224 slave->duplex ? "full" : "half");
2226 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2228 if (!bond->curr_active_slave || slave == primary)
2233 case BOND_LINK_DOWN:
2234 if (slave->link_failure_count < UINT_MAX)
2235 slave->link_failure_count++;
2237 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2238 BOND_SLAVE_NOTIFY_NOW);
2240 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2241 BOND_MODE(bond) == BOND_MODE_8023AD)
2242 bond_set_slave_inactive_flags(slave,
2243 BOND_SLAVE_NOTIFY_NOW);
2245 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2248 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2250 if (slave == rcu_access_pointer(bond->curr_active_slave))
2256 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2257 slave->new_link, slave->dev->name);
2258 slave->new_link = BOND_LINK_NOCHANGE;
2265 bond_select_active_slave(bond);
2266 unblock_netpoll_tx();
2269 bond_set_carrier(bond);
2274 * Really a wrapper that splits the mii monitor into two phases: an
2275 * inspection, then (if inspection indicates something needs to be done)
2276 * an acquisition of appropriate locks followed by a commit phase to
2277 * implement whatever link state changes are indicated.
2279 static void bond_mii_monitor(struct work_struct *work)
2281 struct bonding *bond = container_of(work, struct bonding,
2283 bool should_notify_peers = false;
2284 unsigned long delay;
2285 struct slave *slave;
2286 struct list_head *iter;
2288 delay = msecs_to_jiffies(bond->params.miimon);
2290 if (!bond_has_slaves(bond))
2295 should_notify_peers = bond_should_notify_peers(bond);
2297 if (bond_miimon_inspect(bond)) {
2300 /* Race avoidance with bond_close cancel of workqueue */
2301 if (!rtnl_trylock()) {
2303 should_notify_peers = false;
2307 bond_for_each_slave(bond, slave, iter) {
2308 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2310 bond_miimon_commit(bond);
2312 rtnl_unlock(); /* might sleep, hold no other locks */
2317 if (bond->params.miimon)
2318 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2320 if (should_notify_peers) {
2321 if (!rtnl_trylock())
2323 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2328 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2330 __be32 ip = *((__be32 *)data);
2332 return ip == bond_confirm_addr(upper, 0, ip);
2335 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2339 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2343 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2350 /* We go to the (large) trouble of VLAN tagging ARP frames because
2351 * switches in VLAN mode (especially if ports are configured as
2352 * "native" to a VLAN) might not pass non-tagged frames.
2354 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2355 __be32 dest_ip, __be32 src_ip,
2356 struct bond_vlan_tag *tags)
2358 struct sk_buff *skb;
2359 struct bond_vlan_tag *outer_tag = tags;
2361 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2362 arp_op, slave_dev->name, &dest_ip, &src_ip);
2364 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2365 NULL, slave_dev->dev_addr, NULL);
2368 net_err_ratelimited("ARP packet allocation failed\n");
2372 if (!tags || tags->vlan_proto == VLAN_N_VID)
2377 /* Go through all the tags backwards and add them to the packet */
2378 while (tags->vlan_proto != VLAN_N_VID) {
2379 if (!tags->vlan_id) {
2384 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2385 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2386 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2389 net_err_ratelimited("failed to insert inner VLAN tag\n");
2395 /* Set the outer tag */
2396 if (outer_tag->vlan_id) {
2397 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2398 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2399 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2400 outer_tag->vlan_id);
2407 /* Validate the device path between the @start_dev and the @end_dev.
2408 * The path is valid if the @end_dev is reachable through device
2410 * When the path is validated, collect any vlan information in the
2413 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2414 struct net_device *end_dev,
2417 struct bond_vlan_tag *tags;
2418 struct net_device *upper;
2419 struct list_head *iter;
2421 if (start_dev == end_dev) {
2422 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2424 return ERR_PTR(-ENOMEM);
2425 tags[level].vlan_proto = VLAN_N_VID;
2429 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2430 tags = bond_verify_device_path(upper, end_dev, level + 1);
2431 if (IS_ERR_OR_NULL(tags)) {
2436 if (is_vlan_dev(upper)) {
2437 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2438 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2447 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2450 struct bond_vlan_tag *tags;
2451 __be32 *targets = bond->params.arp_targets, addr;
2454 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2455 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2458 /* Find out through which dev should the packet go */
2459 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2462 /* there's no route to target - try to send arp
2463 * probe to generate any traffic (arp_validate=0)
2465 if (bond->params.arp_validate)
2466 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2469 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2474 /* bond device itself */
2475 if (rt->dst.dev == bond->dev)
2479 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2482 if (!IS_ERR_OR_NULL(tags))
2485 /* Not our device - skip */
2486 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2487 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2493 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2495 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2501 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2505 if (!sip || !bond_has_this_ip(bond, tip)) {
2506 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2511 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2513 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2517 slave->last_rx = jiffies;
2518 slave->target_last_arp_rx[i] = jiffies;
2521 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2522 struct slave *slave)
2524 struct arphdr *arp = (struct arphdr *)skb->data;
2525 struct slave *curr_active_slave, *curr_arp_slave;
2526 unsigned char *arp_ptr;
2528 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2531 if (!slave_do_arp_validate(bond, slave)) {
2532 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2533 !slave_do_arp_validate_only(bond))
2534 slave->last_rx = jiffies;
2535 return RX_HANDLER_ANOTHER;
2536 } else if (!is_arp) {
2537 return RX_HANDLER_ANOTHER;
2540 alen = arp_hdr_len(bond->dev);
2542 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2545 if (alen > skb_headlen(skb)) {
2546 arp = kmalloc(alen, GFP_ATOMIC);
2549 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2553 if (arp->ar_hln != bond->dev->addr_len ||
2554 skb->pkt_type == PACKET_OTHERHOST ||
2555 skb->pkt_type == PACKET_LOOPBACK ||
2556 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2557 arp->ar_pro != htons(ETH_P_IP) ||
2561 arp_ptr = (unsigned char *)(arp + 1);
2562 arp_ptr += bond->dev->addr_len;
2563 memcpy(&sip, arp_ptr, 4);
2564 arp_ptr += 4 + bond->dev->addr_len;
2565 memcpy(&tip, arp_ptr, 4);
2567 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2568 slave->dev->name, bond_slave_state(slave),
2569 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2572 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2573 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2575 /* We 'trust' the received ARP enough to validate it if:
2577 * (a) the slave receiving the ARP is active (which includes the
2578 * current ARP slave, if any), or
2580 * (b) the receiving slave isn't active, but there is a currently
2581 * active slave and it received valid arp reply(s) after it became
2582 * the currently active slave, or
2584 * (c) there is an ARP slave that sent an ARP during the prior ARP
2585 * interval, and we receive an ARP reply on any slave. We accept
2586 * these because switch FDB update delays may deliver the ARP
2587 * reply to a slave other than the sender of the ARP request.
2589 * Note: for (b), backup slaves are receiving the broadcast ARP
2590 * request, not a reply. This request passes from the sending
2591 * slave through the L2 switch(es) to the receiving slave. Since
2592 * this is checking the request, sip/tip are swapped for
2595 * This is done to avoid endless looping when we can't reach the
2596 * arp_ip_target and fool ourselves with our own arp requests.
2598 if (bond_is_active_slave(slave))
2599 bond_validate_arp(bond, slave, sip, tip);
2600 else if (curr_active_slave &&
2601 time_after(slave_last_rx(bond, curr_active_slave),
2602 curr_active_slave->last_link_up))
2603 bond_validate_arp(bond, slave, tip, sip);
2604 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2605 bond_time_in_interval(bond,
2606 dev_trans_start(curr_arp_slave->dev), 1))
2607 bond_validate_arp(bond, slave, sip, tip);
2610 if (arp != (struct arphdr *)skb->data)
2612 return RX_HANDLER_ANOTHER;
2615 /* function to verify if we're in the arp_interval timeslice, returns true if
2616 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2617 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2619 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2622 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2624 return time_in_range(jiffies,
2625 last_act - delta_in_ticks,
2626 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2629 /* This function is called regularly to monitor each slave's link
2630 * ensuring that traffic is being sent and received when arp monitoring
2631 * is used in load-balancing mode. if the adapter has been dormant, then an
2632 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2633 * arp monitoring in active backup mode.
2635 static void bond_loadbalance_arp_mon(struct bonding *bond)
2637 struct slave *slave, *oldcurrent;
2638 struct list_head *iter;
2639 int do_failover = 0, slave_state_changed = 0;
2641 if (!bond_has_slaves(bond))
2646 oldcurrent = rcu_dereference(bond->curr_active_slave);
2647 /* see if any of the previous devices are up now (i.e. they have
2648 * xmt and rcv traffic). the curr_active_slave does not come into
2649 * the picture unless it is null. also, slave->last_link_up is not
2650 * needed here because we send an arp on each slave and give a slave
2651 * as long as it needs to get the tx/rx within the delta.
2652 * TODO: what about up/down delay in arp mode? it wasn't here before
2655 bond_for_each_slave_rcu(bond, slave, iter) {
2656 unsigned long trans_start = dev_trans_start(slave->dev);
2658 slave->new_link = BOND_LINK_NOCHANGE;
2660 if (slave->link != BOND_LINK_UP) {
2661 if (bond_time_in_interval(bond, trans_start, 1) &&
2662 bond_time_in_interval(bond, slave->last_rx, 1)) {
2664 slave->new_link = BOND_LINK_UP;
2665 slave_state_changed = 1;
2667 /* primary_slave has no meaning in round-robin
2668 * mode. the window of a slave being up and
2669 * curr_active_slave being null after enslaving
2673 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2677 netdev_info(bond->dev, "interface %s is now up\n",
2682 /* slave->link == BOND_LINK_UP */
2684 /* not all switches will respond to an arp request
2685 * when the source ip is 0, so don't take the link down
2686 * if we don't know our ip yet
2688 if (!bond_time_in_interval(bond, trans_start, 2) ||
2689 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2691 slave->new_link = BOND_LINK_DOWN;
2692 slave_state_changed = 1;
2694 if (slave->link_failure_count < UINT_MAX)
2695 slave->link_failure_count++;
2697 netdev_info(bond->dev, "interface %s is now down\n",
2700 if (slave == oldcurrent)
2705 /* note: if switch is in round-robin mode, all links
2706 * must tx arp to ensure all links rx an arp - otherwise
2707 * links may oscillate or not come up at all; if switch is
2708 * in something like xor mode, there is nothing we can
2709 * do - all replies will be rx'ed on same link causing slaves
2710 * to be unstable during low/no traffic periods
2712 if (bond_slave_is_up(slave))
2713 bond_arp_send_all(bond, slave);
2718 if (do_failover || slave_state_changed) {
2719 if (!rtnl_trylock())
2722 bond_for_each_slave(bond, slave, iter) {
2723 if (slave->new_link != BOND_LINK_NOCHANGE)
2724 slave->link = slave->new_link;
2727 if (slave_state_changed) {
2728 bond_slave_state_change(bond);
2729 if (BOND_MODE(bond) == BOND_MODE_XOR)
2730 bond_update_slave_arr(bond, NULL);
2734 bond_select_active_slave(bond);
2735 unblock_netpoll_tx();
2741 if (bond->params.arp_interval)
2742 queue_delayed_work(bond->wq, &bond->arp_work,
2743 msecs_to_jiffies(bond->params.arp_interval));
2746 /* Called to inspect slaves for active-backup mode ARP monitor link state
2747 * changes. Sets new_link in slaves to specify what action should take
2748 * place for the slave. Returns 0 if no changes are found, >0 if changes
2749 * to link states must be committed.
2751 * Called with rcu_read_lock held.
2753 static int bond_ab_arp_inspect(struct bonding *bond)
2755 unsigned long trans_start, last_rx;
2756 struct list_head *iter;
2757 struct slave *slave;
2760 bond_for_each_slave_rcu(bond, slave, iter) {
2761 slave->new_link = BOND_LINK_NOCHANGE;
2762 last_rx = slave_last_rx(bond, slave);
2764 if (slave->link != BOND_LINK_UP) {
2765 if (bond_time_in_interval(bond, last_rx, 1)) {
2766 slave->new_link = BOND_LINK_UP;
2772 /* Give slaves 2*delta after being enslaved or made
2773 * active. This avoids bouncing, as the last receive
2774 * times need a full ARP monitor cycle to be updated.
2776 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2779 /* Backup slave is down if:
2780 * - No current_arp_slave AND
2781 * - more than 3*delta since last receive AND
2782 * - the bond has an IP address
2784 * Note: a non-null current_arp_slave indicates
2785 * the curr_active_slave went down and we are
2786 * searching for a new one; under this condition
2787 * we only take the curr_active_slave down - this
2788 * gives each slave a chance to tx/rx traffic
2789 * before being taken out
2791 if (!bond_is_active_slave(slave) &&
2792 !rcu_access_pointer(bond->current_arp_slave) &&
2793 !bond_time_in_interval(bond, last_rx, 3)) {
2794 slave->new_link = BOND_LINK_DOWN;
2798 /* Active slave is down if:
2799 * - more than 2*delta since transmitting OR
2800 * - (more than 2*delta since receive AND
2801 * the bond has an IP address)
2803 trans_start = dev_trans_start(slave->dev);
2804 if (bond_is_active_slave(slave) &&
2805 (!bond_time_in_interval(bond, trans_start, 2) ||
2806 !bond_time_in_interval(bond, last_rx, 2))) {
2807 slave->new_link = BOND_LINK_DOWN;
2815 /* Called to commit link state changes noted by inspection step of
2816 * active-backup mode ARP monitor.
2818 * Called with RTNL hold.
2820 static void bond_ab_arp_commit(struct bonding *bond)
2822 unsigned long trans_start;
2823 struct list_head *iter;
2824 struct slave *slave;
2826 bond_for_each_slave(bond, slave, iter) {
2827 switch (slave->new_link) {
2828 case BOND_LINK_NOCHANGE:
2832 trans_start = dev_trans_start(slave->dev);
2833 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2834 (!rtnl_dereference(bond->curr_active_slave) &&
2835 bond_time_in_interval(bond, trans_start, 1))) {
2836 struct slave *current_arp_slave;
2838 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2839 bond_set_slave_link_state(slave, BOND_LINK_UP,
2840 BOND_SLAVE_NOTIFY_NOW);
2841 if (current_arp_slave) {
2842 bond_set_slave_inactive_flags(
2844 BOND_SLAVE_NOTIFY_NOW);
2845 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2848 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2851 if (!rtnl_dereference(bond->curr_active_slave) ||
2852 slave == rtnl_dereference(bond->primary_slave))
2859 case BOND_LINK_DOWN:
2860 if (slave->link_failure_count < UINT_MAX)
2861 slave->link_failure_count++;
2863 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2864 BOND_SLAVE_NOTIFY_NOW);
2865 bond_set_slave_inactive_flags(slave,
2866 BOND_SLAVE_NOTIFY_NOW);
2868 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2871 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2872 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2879 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2880 slave->new_link, slave->dev->name);
2886 bond_select_active_slave(bond);
2887 unblock_netpoll_tx();
2890 bond_set_carrier(bond);
2893 /* Send ARP probes for active-backup mode ARP monitor.
2895 * Called with rcu_read_lock held.
2897 static bool bond_ab_arp_probe(struct bonding *bond)
2899 struct slave *slave, *before = NULL, *new_slave = NULL,
2900 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2901 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2902 struct list_head *iter;
2904 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2906 if (curr_arp_slave && curr_active_slave)
2907 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2908 curr_arp_slave->dev->name,
2909 curr_active_slave->dev->name);
2911 if (curr_active_slave) {
2912 bond_arp_send_all(bond, curr_active_slave);
2913 return should_notify_rtnl;
2916 /* if we don't have a curr_active_slave, search for the next available
2917 * backup slave from the current_arp_slave and make it the candidate
2918 * for becoming the curr_active_slave
2921 if (!curr_arp_slave) {
2922 curr_arp_slave = bond_first_slave_rcu(bond);
2923 if (!curr_arp_slave)
2924 return should_notify_rtnl;
2927 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2929 bond_for_each_slave_rcu(bond, slave, iter) {
2930 if (!found && !before && bond_slave_is_up(slave))
2933 if (found && !new_slave && bond_slave_is_up(slave))
2935 /* if the link state is up at this point, we
2936 * mark it down - this can happen if we have
2937 * simultaneous link failures and
2938 * reselect_active_interface doesn't make this
2939 * one the current slave so it is still marked
2940 * up when it is actually down
2942 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2943 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2944 BOND_SLAVE_NOTIFY_LATER);
2945 if (slave->link_failure_count < UINT_MAX)
2946 slave->link_failure_count++;
2948 bond_set_slave_inactive_flags(slave,
2949 BOND_SLAVE_NOTIFY_LATER);
2951 netdev_info(bond->dev, "backup interface %s is now down\n",
2954 if (slave == curr_arp_slave)
2958 if (!new_slave && before)
2964 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2965 BOND_SLAVE_NOTIFY_LATER);
2966 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2967 bond_arp_send_all(bond, new_slave);
2968 new_slave->last_link_up = jiffies;
2969 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2972 bond_for_each_slave_rcu(bond, slave, iter) {
2973 if (slave->should_notify || slave->should_notify_link) {
2974 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2978 return should_notify_rtnl;
2981 static void bond_activebackup_arp_mon(struct bonding *bond)
2983 bool should_notify_peers = false;
2984 bool should_notify_rtnl = false;
2987 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2989 if (!bond_has_slaves(bond))
2994 should_notify_peers = bond_should_notify_peers(bond);
2996 if (bond_ab_arp_inspect(bond)) {
2999 /* Race avoidance with bond_close flush of workqueue */
3000 if (!rtnl_trylock()) {
3002 should_notify_peers = false;
3006 bond_ab_arp_commit(bond);
3012 should_notify_rtnl = bond_ab_arp_probe(bond);
3016 if (bond->params.arp_interval)
3017 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3019 if (should_notify_peers || should_notify_rtnl) {
3020 if (!rtnl_trylock())
3023 if (should_notify_peers)
3024 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3026 if (should_notify_rtnl) {
3027 bond_slave_state_notify(bond);
3028 bond_slave_link_notify(bond);
3035 static void bond_arp_monitor(struct work_struct *work)
3037 struct bonding *bond = container_of(work, struct bonding,
3040 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3041 bond_activebackup_arp_mon(bond);
3043 bond_loadbalance_arp_mon(bond);
3046 /*-------------------------- netdev event handling --------------------------*/
3048 /* Change device name */
3049 static int bond_event_changename(struct bonding *bond)
3051 bond_remove_proc_entry(bond);
3052 bond_create_proc_entry(bond);
3054 bond_debug_reregister(bond);
3059 static int bond_master_netdev_event(unsigned long event,
3060 struct net_device *bond_dev)
3062 struct bonding *event_bond = netdev_priv(bond_dev);
3065 case NETDEV_CHANGENAME:
3066 return bond_event_changename(event_bond);
3067 case NETDEV_UNREGISTER:
3068 bond_remove_proc_entry(event_bond);
3070 case NETDEV_REGISTER:
3071 bond_create_proc_entry(event_bond);
3073 case NETDEV_NOTIFY_PEERS:
3074 if (event_bond->send_peer_notif)
3075 event_bond->send_peer_notif--;
3084 static int bond_slave_netdev_event(unsigned long event,
3085 struct net_device *slave_dev)
3087 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3088 struct bonding *bond;
3089 struct net_device *bond_dev;
3091 /* A netdev event can be generated while enslaving a device
3092 * before netdev_rx_handler_register is called in which case
3093 * slave will be NULL
3097 bond_dev = slave->bond->dev;
3099 primary = rtnl_dereference(bond->primary_slave);
3102 case NETDEV_UNREGISTER:
3103 if (bond_dev->type != ARPHRD_ETHER)
3104 bond_release_and_destroy(bond_dev, slave_dev);
3106 __bond_release_one(bond_dev, slave_dev, false, true);
3110 /* For 802.3ad mode only:
3111 * Getting invalid Speed/Duplex values here will put slave
3112 * in weird state. So mark it as link-down for the time
3113 * being and let link-monitoring (miimon) set it right when
3114 * correct speeds/duplex are available.
3116 if (bond_update_speed_duplex(slave) &&
3117 BOND_MODE(bond) == BOND_MODE_8023AD)
3118 slave->link = BOND_LINK_DOWN;
3120 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3121 bond_3ad_adapter_speed_duplex_changed(slave);
3124 /* Refresh slave-array if applicable!
3125 * If the setup does not use miimon or arpmon (mode-specific!),
3126 * then these events will not cause the slave-array to be
3127 * refreshed. This will cause xmit to use a slave that is not
3128 * usable. Avoid such situation by refeshing the array at these
3129 * events. If these (miimon/arpmon) parameters are configured
3130 * then array gets refreshed twice and that should be fine!
3132 if (bond_mode_can_use_xmit_hash(bond))
3133 bond_update_slave_arr(bond, NULL);
3135 case NETDEV_CHANGEMTU:
3136 /* TODO: Should slaves be allowed to
3137 * independently alter their MTU? For
3138 * an active-backup bond, slaves need
3139 * not be the same type of device, so
3140 * MTUs may vary. For other modes,
3141 * slaves arguably should have the
3142 * same MTUs. To do this, we'd need to
3143 * take over the slave's change_mtu
3144 * function for the duration of their
3148 case NETDEV_CHANGENAME:
3149 /* we don't care if we don't have primary set */
3150 if (!bond_uses_primary(bond) ||
3151 !bond->params.primary[0])
3154 if (slave == primary) {
3155 /* slave's name changed - he's no longer primary */
3156 RCU_INIT_POINTER(bond->primary_slave, NULL);
3157 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3158 /* we have a new primary slave */
3159 rcu_assign_pointer(bond->primary_slave, slave);
3160 } else { /* we didn't change primary - exit */
3164 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3165 primary ? slave_dev->name : "none");
3168 bond_select_active_slave(bond);
3169 unblock_netpoll_tx();
3171 case NETDEV_FEAT_CHANGE:
3172 bond_compute_features(bond);
3174 case NETDEV_RESEND_IGMP:
3175 /* Propagate to master device */
3176 call_netdevice_notifiers(event, slave->bond->dev);
3185 /* bond_netdev_event: handle netdev notifier chain events.
3187 * This function receives events for the netdev chain. The caller (an
3188 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3189 * locks for us to safely manipulate the slave devices (RTNL lock,
3192 static int bond_netdev_event(struct notifier_block *this,
3193 unsigned long event, void *ptr)
3195 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3197 netdev_dbg(event_dev, "event: %lx\n", event);
3199 if (!(event_dev->priv_flags & IFF_BONDING))
3202 if (event_dev->flags & IFF_MASTER) {
3203 netdev_dbg(event_dev, "IFF_MASTER\n");
3204 return bond_master_netdev_event(event, event_dev);
3207 if (event_dev->flags & IFF_SLAVE) {
3208 netdev_dbg(event_dev, "IFF_SLAVE\n");
3209 return bond_slave_netdev_event(event, event_dev);
3215 static struct notifier_block bond_netdev_notifier = {
3216 .notifier_call = bond_netdev_event,
3219 /*---------------------------- Hashing Policies -----------------------------*/
3221 /* L2 hash helper */
3222 static inline u32 bond_eth_hash(struct sk_buff *skb)
3224 struct ethhdr *ep, hdr_tmp;
3226 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3228 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3232 /* Extract the appropriate headers based on bond's xmit policy */
3233 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3234 struct flow_keys *fk)
3236 const struct ipv6hdr *iph6;
3237 const struct iphdr *iph;
3238 int noff, proto = -1;
3240 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3241 return skb_flow_dissect_flow_keys(skb, fk, 0);
3243 fk->ports.ports = 0;
3244 noff = skb_network_offset(skb);
3245 if (skb->protocol == htons(ETH_P_IP)) {
3246 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3249 iph_to_flow_copy_v4addrs(fk, iph);
3250 noff += iph->ihl << 2;
3251 if (!ip_is_fragment(iph))
3252 proto = iph->protocol;
3253 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3254 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3256 iph6 = ipv6_hdr(skb);
3257 iph_to_flow_copy_v6addrs(fk, iph6);
3258 noff += sizeof(*iph6);
3259 proto = iph6->nexthdr;
3263 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3264 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3270 * bond_xmit_hash - generate a hash value based on the xmit policy
3271 * @bond: bonding device
3272 * @skb: buffer to use for headers
3274 * This function will extract the necessary headers from the skb buffer and use
3275 * them to generate a hash based on the xmit_policy set in the bonding device
3277 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3279 struct flow_keys flow;
3282 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3286 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3287 !bond_flow_dissect(bond, skb, &flow))
3288 return bond_eth_hash(skb);
3290 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3291 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3292 hash = bond_eth_hash(skb);
3294 hash = (__force u32)flow.ports.ports;
3295 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3296 (__force u32)flow_get_u32_src(&flow);
3297 hash ^= (hash >> 16);
3298 hash ^= (hash >> 8);
3303 /*-------------------------- Device entry points ----------------------------*/
3305 void bond_work_init_all(struct bonding *bond)
3307 INIT_DELAYED_WORK(&bond->mcast_work,
3308 bond_resend_igmp_join_requests_delayed);
3309 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3310 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3311 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3312 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3313 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3316 static void bond_work_cancel_all(struct bonding *bond)
3318 cancel_delayed_work_sync(&bond->mii_work);
3319 cancel_delayed_work_sync(&bond->arp_work);
3320 cancel_delayed_work_sync(&bond->alb_work);
3321 cancel_delayed_work_sync(&bond->ad_work);
3322 cancel_delayed_work_sync(&bond->mcast_work);
3323 cancel_delayed_work_sync(&bond->slave_arr_work);
3326 static int bond_open(struct net_device *bond_dev)
3328 struct bonding *bond = netdev_priv(bond_dev);
3329 struct list_head *iter;
3330 struct slave *slave;
3332 /* reset slave->backup and slave->inactive */
3333 if (bond_has_slaves(bond)) {
3334 bond_for_each_slave(bond, slave, iter) {
3335 if (bond_uses_primary(bond) &&
3336 slave != rcu_access_pointer(bond->curr_active_slave)) {
3337 bond_set_slave_inactive_flags(slave,
3338 BOND_SLAVE_NOTIFY_NOW);
3339 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3340 bond_set_slave_active_flags(slave,
3341 BOND_SLAVE_NOTIFY_NOW);
3346 if (bond_is_lb(bond)) {
3347 /* bond_alb_initialize must be called before the timer
3350 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3352 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3353 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3356 if (bond->params.miimon) /* link check interval, in milliseconds. */
3357 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3359 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3360 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3361 bond->recv_probe = bond_arp_rcv;
3364 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3365 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3366 /* register to receive LACPDUs */
3367 bond->recv_probe = bond_3ad_lacpdu_recv;
3368 bond_3ad_initiate_agg_selection(bond, 1);
3371 if (bond_mode_can_use_xmit_hash(bond))
3372 bond_update_slave_arr(bond, NULL);
3377 static int bond_close(struct net_device *bond_dev)
3379 struct bonding *bond = netdev_priv(bond_dev);
3381 bond_work_cancel_all(bond);
3382 bond->send_peer_notif = 0;
3383 if (bond_is_lb(bond))
3384 bond_alb_deinitialize(bond);
3385 bond->recv_probe = NULL;
3390 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3391 * that some drivers can provide 32bit values only.
3393 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3394 const struct rtnl_link_stats64 *_new,
3395 const struct rtnl_link_stats64 *_old)
3397 const u64 *new = (const u64 *)_new;
3398 const u64 *old = (const u64 *)_old;
3399 u64 *res = (u64 *)_res;
3402 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3405 s64 delta = nv - ov;
3407 /* detects if this particular field is 32bit only */
3408 if (((nv | ov) >> 32) == 0)
3409 delta = (s64)(s32)((u32)nv - (u32)ov);
3411 /* filter anomalies, some drivers reset their stats
3412 * at down/up events.
3419 static int bond_get_nest_level(struct net_device *bond_dev)
3421 struct bonding *bond = netdev_priv(bond_dev);
3423 return bond->nest_level;
3426 static void bond_get_stats(struct net_device *bond_dev,
3427 struct rtnl_link_stats64 *stats)
3429 struct bonding *bond = netdev_priv(bond_dev);
3430 struct rtnl_link_stats64 temp;
3431 struct list_head *iter;
3432 struct slave *slave;
3434 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3435 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3438 bond_for_each_slave_rcu(bond, slave, iter) {
3439 const struct rtnl_link_stats64 *new =
3440 dev_get_stats(slave->dev, &temp);
3442 bond_fold_stats(stats, new, &slave->slave_stats);
3444 /* save off the slave stats for the next run */
3445 memcpy(&slave->slave_stats, new, sizeof(*new));
3449 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3450 spin_unlock(&bond->stats_lock);
3453 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3455 struct bonding *bond = netdev_priv(bond_dev);
3456 struct net_device *slave_dev = NULL;
3457 struct ifbond k_binfo;
3458 struct ifbond __user *u_binfo = NULL;
3459 struct ifslave k_sinfo;
3460 struct ifslave __user *u_sinfo = NULL;
3461 struct mii_ioctl_data *mii = NULL;
3462 struct bond_opt_value newval;
3466 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3477 /* We do this again just in case we were called by SIOCGMIIREG
3478 * instead of SIOCGMIIPHY.
3484 if (mii->reg_num == 1) {
3486 if (netif_carrier_ok(bond->dev))
3487 mii->val_out = BMSR_LSTATUS;
3491 case BOND_INFO_QUERY_OLD:
3492 case SIOCBONDINFOQUERY:
3493 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3495 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3498 bond_info_query(bond_dev, &k_binfo);
3499 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3503 case BOND_SLAVE_INFO_QUERY_OLD:
3504 case SIOCBONDSLAVEINFOQUERY:
3505 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3507 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3510 res = bond_slave_info_query(bond_dev, &k_sinfo);
3512 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3520 net = dev_net(bond_dev);
3522 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3525 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3527 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3532 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3534 case BOND_ENSLAVE_OLD:
3535 case SIOCBONDENSLAVE:
3536 res = bond_enslave(bond_dev, slave_dev, NULL);
3538 case BOND_RELEASE_OLD:
3539 case SIOCBONDRELEASE:
3540 res = bond_release(bond_dev, slave_dev);
3542 case BOND_SETHWADDR_OLD:
3543 case SIOCBONDSETHWADDR:
3544 bond_set_dev_addr(bond_dev, slave_dev);
3547 case BOND_CHANGE_ACTIVE_OLD:
3548 case SIOCBONDCHANGEACTIVE:
3549 bond_opt_initstr(&newval, slave_dev->name);
3550 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3560 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3562 struct bonding *bond = netdev_priv(bond_dev);
3564 if (change & IFF_PROMISC)
3565 bond_set_promiscuity(bond,
3566 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3568 if (change & IFF_ALLMULTI)
3569 bond_set_allmulti(bond,
3570 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3573 static void bond_set_rx_mode(struct net_device *bond_dev)
3575 struct bonding *bond = netdev_priv(bond_dev);
3576 struct list_head *iter;
3577 struct slave *slave;
3580 if (bond_uses_primary(bond)) {
3581 slave = rcu_dereference(bond->curr_active_slave);
3583 dev_uc_sync(slave->dev, bond_dev);
3584 dev_mc_sync(slave->dev, bond_dev);
3587 bond_for_each_slave_rcu(bond, slave, iter) {
3588 dev_uc_sync_multiple(slave->dev, bond_dev);
3589 dev_mc_sync_multiple(slave->dev, bond_dev);
3595 static int bond_neigh_init(struct neighbour *n)
3597 struct bonding *bond = netdev_priv(n->dev);
3598 const struct net_device_ops *slave_ops;
3599 struct neigh_parms parms;
3600 struct slave *slave;
3603 slave = bond_first_slave(bond);
3606 slave_ops = slave->dev->netdev_ops;
3607 if (!slave_ops->ndo_neigh_setup)
3610 parms.neigh_setup = NULL;
3611 parms.neigh_cleanup = NULL;
3612 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3616 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3617 * after the last slave has been detached. Assumes that all slaves
3618 * utilize the same neigh_cleanup (true at this writing as only user
3621 n->parms->neigh_cleanup = parms.neigh_cleanup;
3623 if (!parms.neigh_setup)
3626 return parms.neigh_setup(n);
3629 /* The bonding ndo_neigh_setup is called at init time beofre any
3630 * slave exists. So we must declare proxy setup function which will
3631 * be used at run time to resolve the actual slave neigh param setup.
3633 * It's also called by master devices (such as vlans) to setup their
3634 * underlying devices. In that case - do nothing, we're already set up from
3637 static int bond_neigh_setup(struct net_device *dev,
3638 struct neigh_parms *parms)
3640 /* modify only our neigh_parms */
3641 if (parms->dev == dev)
3642 parms->neigh_setup = bond_neigh_init;
3647 /* Change the MTU of all of a master's slaves to match the master */
3648 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3650 struct bonding *bond = netdev_priv(bond_dev);
3651 struct slave *slave, *rollback_slave;
3652 struct list_head *iter;
3655 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3657 bond_for_each_slave(bond, slave, iter) {
3658 netdev_dbg(bond_dev, "s %p c_m %p\n",
3659 slave, slave->dev->netdev_ops->ndo_change_mtu);
3661 res = dev_set_mtu(slave->dev, new_mtu);
3664 /* If we failed to set the slave's mtu to the new value
3665 * we must abort the operation even in ACTIVE_BACKUP
3666 * mode, because if we allow the backup slaves to have
3667 * different mtu values than the active slave we'll
3668 * need to change their mtu when doing a failover. That
3669 * means changing their mtu from timer context, which
3670 * is probably not a good idea.
3672 netdev_dbg(bond_dev, "err %d %s\n", res,
3678 bond_dev->mtu = new_mtu;
3683 /* unwind from head to the slave that failed */
3684 bond_for_each_slave(bond, rollback_slave, iter) {
3687 if (rollback_slave == slave)
3690 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3692 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3693 tmp_res, rollback_slave->dev->name);
3700 /* Change HW address
3702 * Note that many devices must be down to change the HW address, and
3703 * downing the master releases all slaves. We can make bonds full of
3704 * bonding devices to test this, however.
3706 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3708 struct bonding *bond = netdev_priv(bond_dev);
3709 struct slave *slave, *rollback_slave;
3710 struct sockaddr_storage *ss = addr, tmp_ss;
3711 struct list_head *iter;
3714 if (BOND_MODE(bond) == BOND_MODE_ALB)
3715 return bond_alb_set_mac_address(bond_dev, addr);
3718 netdev_dbg(bond_dev, "bond=%p\n", bond);
3720 /* If fail_over_mac is enabled, do nothing and return success.
3721 * Returning an error causes ifenslave to fail.
3723 if (bond->params.fail_over_mac &&
3724 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3727 if (!is_valid_ether_addr(ss->__data))
3728 return -EADDRNOTAVAIL;
3730 bond_for_each_slave(bond, slave, iter) {
3731 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3732 res = dev_set_mac_address(slave->dev, addr);
3734 /* TODO: consider downing the slave
3736 * User should expect communications
3737 * breakage anyway until ARP finish
3740 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3746 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3750 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3751 tmp_ss.ss_family = bond_dev->type;
3753 /* unwind from head to the slave that failed */
3754 bond_for_each_slave(bond, rollback_slave, iter) {
3757 if (rollback_slave == slave)
3760 tmp_res = dev_set_mac_address(rollback_slave->dev,
3761 (struct sockaddr *)&tmp_ss);
3763 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3764 tmp_res, rollback_slave->dev->name);
3772 * bond_xmit_slave_id - transmit skb through slave with slave_id
3773 * @bond: bonding device that is transmitting
3774 * @skb: buffer to transmit
3775 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3777 * This function tries to transmit through slave with slave_id but in case
3778 * it fails, it tries to find the first available slave for transmission.
3779 * The skb is consumed in all cases, thus the function is void.
3781 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3783 struct list_head *iter;
3784 struct slave *slave;
3787 /* Here we start from the slave with slave_id */
3788 bond_for_each_slave_rcu(bond, slave, iter) {
3790 if (bond_slave_can_tx(slave)) {
3791 bond_dev_queue_xmit(bond, skb, slave->dev);
3797 /* Here we start from the first slave up to slave_id */
3799 bond_for_each_slave_rcu(bond, slave, iter) {
3802 if (bond_slave_can_tx(slave)) {
3803 bond_dev_queue_xmit(bond, skb, slave->dev);
3807 /* no slave that can tx has been found */
3808 bond_tx_drop(bond->dev, skb);
3812 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3813 * @bond: bonding device to use
3815 * Based on the value of the bonding device's packets_per_slave parameter
3816 * this function generates a slave id, which is usually used as the next
3817 * slave to transmit through.
3819 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3822 struct reciprocal_value reciprocal_packets_per_slave;
3823 int packets_per_slave = bond->params.packets_per_slave;
3825 switch (packets_per_slave) {
3827 slave_id = prandom_u32();
3830 slave_id = bond->rr_tx_counter;
3833 reciprocal_packets_per_slave =
3834 bond->params.reciprocal_packets_per_slave;
3835 slave_id = reciprocal_divide(bond->rr_tx_counter,
3836 reciprocal_packets_per_slave);
3839 bond->rr_tx_counter++;
3844 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3845 struct net_device *bond_dev)
3847 struct bonding *bond = netdev_priv(bond_dev);
3848 struct iphdr *iph = ip_hdr(skb);
3849 struct slave *slave;
3852 /* Start with the curr_active_slave that joined the bond as the
3853 * default for sending IGMP traffic. For failover purposes one
3854 * needs to maintain some consistency for the interface that will
3855 * send the join/membership reports. The curr_active_slave found
3856 * will send all of this type of traffic.
3858 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3859 slave = rcu_dereference(bond->curr_active_slave);
3861 bond_dev_queue_xmit(bond, skb, slave->dev);
3863 bond_xmit_slave_id(bond, skb, 0);
3865 int slave_cnt = READ_ONCE(bond->slave_cnt);
3867 if (likely(slave_cnt)) {
3868 slave_id = bond_rr_gen_slave_id(bond);
3869 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3871 bond_tx_drop(bond_dev, skb);
3875 return NETDEV_TX_OK;
3878 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3879 * the bond has a usable interface.
3881 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3882 struct net_device *bond_dev)
3884 struct bonding *bond = netdev_priv(bond_dev);
3885 struct slave *slave;
3887 slave = rcu_dereference(bond->curr_active_slave);
3889 bond_dev_queue_xmit(bond, skb, slave->dev);
3891 bond_tx_drop(bond_dev, skb);
3893 return NETDEV_TX_OK;
3896 /* Use this to update slave_array when (a) it's not appropriate to update
3897 * slave_array right away (note that update_slave_array() may sleep)
3898 * and / or (b) RTNL is not held.
3900 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3902 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3905 /* Slave array work handler. Holds only RTNL */
3906 static void bond_slave_arr_handler(struct work_struct *work)
3908 struct bonding *bond = container_of(work, struct bonding,
3909 slave_arr_work.work);
3912 if (!rtnl_trylock())
3915 ret = bond_update_slave_arr(bond, NULL);
3918 pr_warn_ratelimited("Failed to update slave array from WT\n");
3924 bond_slave_arr_work_rearm(bond, 1);
3927 /* Build the usable slaves array in control path for modes that use xmit-hash
3928 * to determine the slave interface -
3929 * (a) BOND_MODE_8023AD
3931 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3933 * The caller is expected to hold RTNL only and NO other lock!
3935 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3937 struct slave *slave;
3938 struct list_head *iter;
3939 struct bond_up_slave *new_arr, *old_arr;
3943 #ifdef CONFIG_LOCKDEP
3944 WARN_ON(lockdep_is_held(&bond->mode_lock));
3947 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3951 pr_err("Failed to build slave-array.\n");
3954 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3955 struct ad_info ad_info;
3957 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3958 pr_debug("bond_3ad_get_active_agg_info failed\n");
3959 kfree_rcu(new_arr, rcu);
3960 /* No active aggragator means it's not safe to use
3961 * the previous array.
3963 old_arr = rtnl_dereference(bond->slave_arr);
3965 RCU_INIT_POINTER(bond->slave_arr, NULL);
3966 kfree_rcu(old_arr, rcu);
3970 agg_id = ad_info.aggregator_id;
3972 bond_for_each_slave(bond, slave, iter) {
3973 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3974 struct aggregator *agg;
3976 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3977 if (!agg || agg->aggregator_identifier != agg_id)
3980 if (!bond_slave_can_tx(slave))
3982 if (skipslave == slave)
3985 netdev_dbg(bond->dev,
3986 "Adding slave dev %s to tx hash array[%d]\n",
3987 slave->dev->name, new_arr->count);
3989 new_arr->arr[new_arr->count++] = slave;
3992 old_arr = rtnl_dereference(bond->slave_arr);
3993 rcu_assign_pointer(bond->slave_arr, new_arr);
3995 kfree_rcu(old_arr, rcu);
3997 if (ret != 0 && skipslave) {
4000 /* Rare situation where caller has asked to skip a specific
4001 * slave but allocation failed (most likely!). BTW this is
4002 * only possible when the call is initiated from
4003 * __bond_release_one(). In this situation; overwrite the
4004 * skipslave entry in the array with the last entry from the
4005 * array to avoid a situation where the xmit path may choose
4006 * this to-be-skipped slave to send a packet out.
4008 old_arr = rtnl_dereference(bond->slave_arr);
4009 for (idx = 0; idx < old_arr->count; idx++) {
4010 if (skipslave == old_arr->arr[idx]) {
4012 old_arr->arr[old_arr->count-1];
4021 /* Use this Xmit function for 3AD as well as XOR modes. The current
4022 * usable slave array is formed in the control path. The xmit function
4023 * just calculates hash and sends the packet out.
4025 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4026 struct net_device *dev)
4028 struct bonding *bond = netdev_priv(dev);
4029 struct slave *slave;
4030 struct bond_up_slave *slaves;
4033 slaves = rcu_dereference(bond->slave_arr);
4034 count = slaves ? READ_ONCE(slaves->count) : 0;
4035 if (likely(count)) {
4036 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4037 bond_dev_queue_xmit(bond, skb, slave->dev);
4039 bond_tx_drop(dev, skb);
4042 return NETDEV_TX_OK;
4045 /* in broadcast mode, we send everything to all usable interfaces. */
4046 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4047 struct net_device *bond_dev)
4049 struct bonding *bond = netdev_priv(bond_dev);
4050 struct slave *slave = NULL;
4051 struct list_head *iter;
4053 bond_for_each_slave_rcu(bond, slave, iter) {
4054 if (bond_is_last_slave(bond, slave))
4056 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4057 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4060 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4061 bond_dev->name, __func__);
4064 bond_dev_queue_xmit(bond, skb2, slave->dev);
4067 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4068 bond_dev_queue_xmit(bond, skb, slave->dev);
4070 bond_tx_drop(bond_dev, skb);
4072 return NETDEV_TX_OK;
4075 /*------------------------- Device initialization ---------------------------*/
4077 /* Lookup the slave that corresponds to a qid */
4078 static inline int bond_slave_override(struct bonding *bond,
4079 struct sk_buff *skb)
4081 struct slave *slave = NULL;
4082 struct list_head *iter;
4084 if (!skb_rx_queue_recorded(skb))
4087 /* Find out if any slaves have the same mapping as this skb. */
4088 bond_for_each_slave_rcu(bond, slave, iter) {
4089 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4090 if (bond_slave_is_up(slave) &&
4091 slave->link == BOND_LINK_UP) {
4092 bond_dev_queue_xmit(bond, skb, slave->dev);
4095 /* If the slave isn't UP, use default transmit policy. */
4104 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4105 struct net_device *sb_dev,
4106 select_queue_fallback_t fallback)
4108 /* This helper function exists to help dev_pick_tx get the correct
4109 * destination queue. Using a helper function skips a call to
4110 * skb_tx_hash and will put the skbs in the queue we expect on their
4111 * way down to the bonding driver.
4113 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4115 /* Save the original txq to restore before passing to the driver */
4116 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4118 if (unlikely(txq >= dev->real_num_tx_queues)) {
4120 txq -= dev->real_num_tx_queues;
4121 } while (txq >= dev->real_num_tx_queues);
4126 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4128 struct bonding *bond = netdev_priv(dev);
4130 if (bond_should_override_tx_queue(bond) &&
4131 !bond_slave_override(bond, skb))
4132 return NETDEV_TX_OK;
4134 switch (BOND_MODE(bond)) {
4135 case BOND_MODE_ROUNDROBIN:
4136 return bond_xmit_roundrobin(skb, dev);
4137 case BOND_MODE_ACTIVEBACKUP:
4138 return bond_xmit_activebackup(skb, dev);
4139 case BOND_MODE_8023AD:
4141 return bond_3ad_xor_xmit(skb, dev);
4142 case BOND_MODE_BROADCAST:
4143 return bond_xmit_broadcast(skb, dev);
4145 return bond_alb_xmit(skb, dev);
4147 return bond_tlb_xmit(skb, dev);
4149 /* Should never happen, mode already checked */
4150 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4152 bond_tx_drop(dev, skb);
4153 return NETDEV_TX_OK;
4157 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4159 struct bonding *bond = netdev_priv(dev);
4160 netdev_tx_t ret = NETDEV_TX_OK;
4162 /* If we risk deadlock from transmitting this in the
4163 * netpoll path, tell netpoll to queue the frame for later tx
4165 if (unlikely(is_netpoll_tx_blocked(dev)))
4166 return NETDEV_TX_BUSY;
4169 if (bond_has_slaves(bond))
4170 ret = __bond_start_xmit(skb, dev);
4172 bond_tx_drop(dev, skb);
4178 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4179 struct ethtool_link_ksettings *cmd)
4181 struct bonding *bond = netdev_priv(bond_dev);
4182 unsigned long speed = 0;
4183 struct list_head *iter;
4184 struct slave *slave;
4186 cmd->base.duplex = DUPLEX_UNKNOWN;
4187 cmd->base.port = PORT_OTHER;
4189 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4190 * do not need to check mode. Though link speed might not represent
4191 * the true receive or transmit bandwidth (not all modes are symmetric)
4192 * this is an accurate maximum.
4194 bond_for_each_slave(bond, slave, iter) {
4195 if (bond_slave_can_tx(slave)) {
4196 if (slave->speed != SPEED_UNKNOWN)
4197 speed += slave->speed;
4198 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4199 slave->duplex != DUPLEX_UNKNOWN)
4200 cmd->base.duplex = slave->duplex;
4203 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4208 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4209 struct ethtool_drvinfo *drvinfo)
4211 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4212 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4213 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4217 static const struct ethtool_ops bond_ethtool_ops = {
4218 .get_drvinfo = bond_ethtool_get_drvinfo,
4219 .get_link = ethtool_op_get_link,
4220 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4223 static const struct net_device_ops bond_netdev_ops = {
4224 .ndo_init = bond_init,
4225 .ndo_uninit = bond_uninit,
4226 .ndo_open = bond_open,
4227 .ndo_stop = bond_close,
4228 .ndo_start_xmit = bond_start_xmit,
4229 .ndo_select_queue = bond_select_queue,
4230 .ndo_get_stats64 = bond_get_stats,
4231 .ndo_do_ioctl = bond_do_ioctl,
4232 .ndo_change_rx_flags = bond_change_rx_flags,
4233 .ndo_set_rx_mode = bond_set_rx_mode,
4234 .ndo_change_mtu = bond_change_mtu,
4235 .ndo_set_mac_address = bond_set_mac_address,
4236 .ndo_neigh_setup = bond_neigh_setup,
4237 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4238 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4239 .ndo_get_lock_subclass = bond_get_nest_level,
4240 #ifdef CONFIG_NET_POLL_CONTROLLER
4241 .ndo_netpoll_setup = bond_netpoll_setup,
4242 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4243 .ndo_poll_controller = bond_poll_controller,
4245 .ndo_add_slave = bond_enslave,
4246 .ndo_del_slave = bond_release,
4247 .ndo_fix_features = bond_fix_features,
4248 .ndo_features_check = passthru_features_check,
4251 static const struct device_type bond_type = {
4255 static void bond_destructor(struct net_device *bond_dev)
4257 struct bonding *bond = netdev_priv(bond_dev);
4259 destroy_workqueue(bond->wq);
4262 void bond_setup(struct net_device *bond_dev)
4264 struct bonding *bond = netdev_priv(bond_dev);
4266 spin_lock_init(&bond->mode_lock);
4267 spin_lock_init(&bond->stats_lock);
4268 bond->params = bonding_defaults;
4270 /* Initialize pointers */
4271 bond->dev = bond_dev;
4273 /* Initialize the device entry points */
4274 ether_setup(bond_dev);
4275 bond_dev->max_mtu = ETH_MAX_MTU;
4276 bond_dev->netdev_ops = &bond_netdev_ops;
4277 bond_dev->ethtool_ops = &bond_ethtool_ops;
4279 bond_dev->needs_free_netdev = true;
4280 bond_dev->priv_destructor = bond_destructor;
4282 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4284 /* Initialize the device options */
4285 bond_dev->flags |= IFF_MASTER;
4286 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4287 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4289 /* don't acquire bond device's netif_tx_lock when transmitting */
4290 bond_dev->features |= NETIF_F_LLTX;
4292 /* By default, we declare the bond to be fully
4293 * VLAN hardware accelerated capable. Special
4294 * care is taken in the various xmit functions
4295 * when there are slaves that are not hw accel
4299 /* Don't allow bond devices to change network namespaces. */
4300 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4302 bond_dev->hw_features = BOND_VLAN_FEATURES |
4303 NETIF_F_HW_VLAN_CTAG_TX |
4304 NETIF_F_HW_VLAN_CTAG_RX |
4305 NETIF_F_HW_VLAN_CTAG_FILTER;
4307 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4308 bond_dev->features |= bond_dev->hw_features;
4311 /* Destroy a bonding device.
4312 * Must be under rtnl_lock when this function is called.
4314 static void bond_uninit(struct net_device *bond_dev)
4316 struct bonding *bond = netdev_priv(bond_dev);
4317 struct list_head *iter;
4318 struct slave *slave;
4319 struct bond_up_slave *arr;
4321 bond_netpoll_cleanup(bond_dev);
4323 /* Release the bonded slaves */
4324 bond_for_each_slave(bond, slave, iter)
4325 __bond_release_one(bond_dev, slave->dev, true, true);
4326 netdev_info(bond_dev, "Released all slaves\n");
4328 arr = rtnl_dereference(bond->slave_arr);
4330 RCU_INIT_POINTER(bond->slave_arr, NULL);
4331 kfree_rcu(arr, rcu);
4334 list_del(&bond->bond_list);
4336 bond_debug_unregister(bond);
4339 /*------------------------- Module initialization ---------------------------*/
4341 static int bond_check_params(struct bond_params *params)
4343 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4344 struct bond_opt_value newval;
4345 const struct bond_opt_value *valptr;
4346 int arp_all_targets_value = 0;
4347 u16 ad_actor_sys_prio = 0;
4348 u16 ad_user_port_key = 0;
4349 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4351 int bond_mode = BOND_MODE_ROUNDROBIN;
4352 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4356 /* Convert string parameters. */
4358 bond_opt_initstr(&newval, mode);
4359 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4361 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4364 bond_mode = valptr->value;
4367 if (xmit_hash_policy) {
4368 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4369 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4370 bond_mode == BOND_MODE_BROADCAST) {
4371 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4372 bond_mode_name(bond_mode));
4374 bond_opt_initstr(&newval, xmit_hash_policy);
4375 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4378 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4382 xmit_hashtype = valptr->value;
4387 if (bond_mode != BOND_MODE_8023AD) {
4388 pr_info("lacp_rate param is irrelevant in mode %s\n",
4389 bond_mode_name(bond_mode));
4391 bond_opt_initstr(&newval, lacp_rate);
4392 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4395 pr_err("Error: Invalid lacp rate \"%s\"\n",
4399 lacp_fast = valptr->value;
4404 bond_opt_initstr(&newval, ad_select);
4405 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4408 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4411 params->ad_select = valptr->value;
4412 if (bond_mode != BOND_MODE_8023AD)
4413 pr_warn("ad_select param only affects 802.3ad mode\n");
4415 params->ad_select = BOND_AD_STABLE;
4418 if (max_bonds < 0) {
4419 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4420 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4421 max_bonds = BOND_DEFAULT_MAX_BONDS;
4425 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4431 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4436 if (downdelay < 0) {
4437 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4438 downdelay, INT_MAX);
4442 if ((use_carrier != 0) && (use_carrier != 1)) {
4443 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4448 if (num_peer_notif < 0 || num_peer_notif > 255) {
4449 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4454 /* reset values for 802.3ad/TLB/ALB */
4455 if (!bond_mode_uses_arp(bond_mode)) {
4457 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");
4458 pr_warn("Forcing miimon to 100msec\n");
4459 miimon = BOND_DEFAULT_MIIMON;
4463 if (tx_queues < 1 || tx_queues > 255) {
4464 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4465 tx_queues, BOND_DEFAULT_TX_QUEUES);
4466 tx_queues = BOND_DEFAULT_TX_QUEUES;
4469 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4470 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4472 all_slaves_active = 0;
4475 if (resend_igmp < 0 || resend_igmp > 255) {
4476 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4477 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4478 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4481 bond_opt_initval(&newval, packets_per_slave);
4482 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4483 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4484 packets_per_slave, USHRT_MAX);
4485 packets_per_slave = 1;
4488 if (bond_mode == BOND_MODE_ALB) {
4489 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",
4494 if (updelay || downdelay) {
4495 /* just warn the user the up/down delay will have
4496 * no effect since miimon is zero...
4498 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",
4499 updelay, downdelay);
4502 /* don't allow arp monitoring */
4504 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4505 miimon, arp_interval);
4509 if ((updelay % miimon) != 0) {
4510 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4511 updelay, miimon, (updelay / miimon) * miimon);
4516 if ((downdelay % miimon) != 0) {
4517 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4519 (downdelay / miimon) * miimon);
4522 downdelay /= miimon;
4525 if (arp_interval < 0) {
4526 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4527 arp_interval, INT_MAX);
4531 for (arp_ip_count = 0, i = 0;
4532 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4535 /* not a complete check, but good enough to catch mistakes */
4536 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4537 !bond_is_ip_target_ok(ip)) {
4538 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4542 if (bond_get_targets_ip(arp_target, ip) == -1)
4543 arp_target[arp_ip_count++] = ip;
4545 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4550 if (arp_interval && !arp_ip_count) {
4551 /* don't allow arping if no arp_ip_target given... */
4552 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4558 if (!arp_interval) {
4559 pr_err("arp_validate requires arp_interval\n");
4563 bond_opt_initstr(&newval, arp_validate);
4564 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4567 pr_err("Error: invalid arp_validate \"%s\"\n",
4571 arp_validate_value = valptr->value;
4573 arp_validate_value = 0;
4576 if (arp_all_targets) {
4577 bond_opt_initstr(&newval, arp_all_targets);
4578 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4581 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4583 arp_all_targets_value = 0;
4585 arp_all_targets_value = valptr->value;
4590 pr_info("MII link monitoring set to %d ms\n", miimon);
4591 } else if (arp_interval) {
4592 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4593 arp_validate_value);
4594 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4595 arp_interval, valptr->string, arp_ip_count);
4597 for (i = 0; i < arp_ip_count; i++)
4598 pr_cont(" %s", arp_ip_target[i]);
4602 } else if (max_bonds) {
4603 /* miimon and arp_interval not set, we need one so things
4604 * work as expected, see bonding.txt for details
4606 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");
4609 if (primary && !bond_mode_uses_primary(bond_mode)) {
4610 /* currently, using a primary only makes sense
4611 * in active backup, TLB or ALB modes
4613 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4614 primary, bond_mode_name(bond_mode));
4618 if (primary && primary_reselect) {
4619 bond_opt_initstr(&newval, primary_reselect);
4620 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4623 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4627 primary_reselect_value = valptr->value;
4629 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4632 if (fail_over_mac) {
4633 bond_opt_initstr(&newval, fail_over_mac);
4634 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4637 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4641 fail_over_mac_value = valptr->value;
4642 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4643 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4645 fail_over_mac_value = BOND_FOM_NONE;
4648 bond_opt_initstr(&newval, "default");
4649 valptr = bond_opt_parse(
4650 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4653 pr_err("Error: No ad_actor_sys_prio default value");
4656 ad_actor_sys_prio = valptr->value;
4658 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4661 pr_err("Error: No ad_user_port_key default value");
4664 ad_user_port_key = valptr->value;
4666 bond_opt_initstr(&newval, "default");
4667 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4669 pr_err("Error: No tlb_dynamic_lb default value");
4672 tlb_dynamic_lb = valptr->value;
4674 if (lp_interval == 0) {
4675 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4676 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4677 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4680 /* fill params struct with the proper values */
4681 params->mode = bond_mode;
4682 params->xmit_policy = xmit_hashtype;
4683 params->miimon = miimon;
4684 params->num_peer_notif = num_peer_notif;
4685 params->arp_interval = arp_interval;
4686 params->arp_validate = arp_validate_value;
4687 params->arp_all_targets = arp_all_targets_value;
4688 params->updelay = updelay;
4689 params->downdelay = downdelay;
4690 params->use_carrier = use_carrier;
4691 params->lacp_fast = lacp_fast;
4692 params->primary[0] = 0;
4693 params->primary_reselect = primary_reselect_value;
4694 params->fail_over_mac = fail_over_mac_value;
4695 params->tx_queues = tx_queues;
4696 params->all_slaves_active = all_slaves_active;
4697 params->resend_igmp = resend_igmp;
4698 params->min_links = min_links;
4699 params->lp_interval = lp_interval;
4700 params->packets_per_slave = packets_per_slave;
4701 params->tlb_dynamic_lb = tlb_dynamic_lb;
4702 params->ad_actor_sys_prio = ad_actor_sys_prio;
4703 eth_zero_addr(params->ad_actor_system);
4704 params->ad_user_port_key = ad_user_port_key;
4705 if (packets_per_slave > 0) {
4706 params->reciprocal_packets_per_slave =
4707 reciprocal_value(packets_per_slave);
4709 /* reciprocal_packets_per_slave is unused if
4710 * packets_per_slave is 0 or 1, just initialize it
4712 params->reciprocal_packets_per_slave =
4713 (struct reciprocal_value) { 0 };
4717 strncpy(params->primary, primary, IFNAMSIZ);
4718 params->primary[IFNAMSIZ - 1] = 0;
4721 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4726 /* Called from registration process */
4727 static int bond_init(struct net_device *bond_dev)
4729 struct bonding *bond = netdev_priv(bond_dev);
4730 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4732 netdev_dbg(bond_dev, "Begin bond_init\n");
4734 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4738 bond->nest_level = SINGLE_DEPTH_NESTING;
4739 netdev_lockdep_set_classes(bond_dev);
4741 list_add_tail(&bond->bond_list, &bn->dev_list);
4743 bond_prepare_sysfs_group(bond);
4745 bond_debug_register(bond);
4747 /* Ensure valid dev_addr */
4748 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4749 bond_dev->addr_assign_type == NET_ADDR_PERM)
4750 eth_hw_addr_random(bond_dev);
4755 unsigned int bond_get_num_tx_queues(void)
4760 /* Create a new bond based on the specified name and bonding parameters.
4761 * If name is NULL, obtain a suitable "bond%d" name for us.
4762 * Caller must NOT hold rtnl_lock; we need to release it here before we
4763 * set up our sysfs entries.
4765 int bond_create(struct net *net, const char *name)
4767 struct net_device *bond_dev;
4768 struct bonding *bond;
4769 struct alb_bond_info *bond_info;
4774 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4775 name ? name : "bond%d", NET_NAME_UNKNOWN,
4776 bond_setup, tx_queues);
4778 pr_err("%s: eek! can't alloc netdev!\n", name);
4784 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4785 * It is set to 0 by default which is wrong.
4787 bond = netdev_priv(bond_dev);
4788 bond_info = &(BOND_ALB_INFO(bond));
4789 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4791 dev_net_set(bond_dev, net);
4792 bond_dev->rtnl_link_ops = &bond_link_ops;
4794 res = register_netdevice(bond_dev);
4796 netif_carrier_off(bond_dev);
4798 bond_work_init_all(bond);
4802 free_netdev(bond_dev);
4806 static int __net_init bond_net_init(struct net *net)
4808 struct bond_net *bn = net_generic(net, bond_net_id);
4811 INIT_LIST_HEAD(&bn->dev_list);
4813 bond_create_proc_dir(bn);
4814 bond_create_sysfs(bn);
4819 static void __net_exit bond_net_exit(struct net *net)
4821 struct bond_net *bn = net_generic(net, bond_net_id);
4822 struct bonding *bond, *tmp_bond;
4825 bond_destroy_sysfs(bn);
4827 /* Kill off any bonds created after unregistering bond rtnl ops */
4829 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4830 unregister_netdevice_queue(bond->dev, &list);
4831 unregister_netdevice_many(&list);
4834 bond_destroy_proc_dir(bn);
4837 static struct pernet_operations bond_net_ops = {
4838 .init = bond_net_init,
4839 .exit = bond_net_exit,
4841 .size = sizeof(struct bond_net),
4844 static int __init bonding_init(void)
4849 pr_info("%s", bond_version);
4851 res = bond_check_params(&bonding_defaults);
4855 res = register_pernet_subsys(&bond_net_ops);
4859 res = bond_netlink_init();
4863 bond_create_debugfs();
4865 for (i = 0; i < max_bonds; i++) {
4866 res = bond_create(&init_net, NULL);
4871 register_netdevice_notifier(&bond_netdev_notifier);
4875 bond_destroy_debugfs();
4876 bond_netlink_fini();
4878 unregister_pernet_subsys(&bond_net_ops);
4883 static void __exit bonding_exit(void)
4885 unregister_netdevice_notifier(&bond_netdev_notifier);
4887 bond_destroy_debugfs();
4889 bond_netlink_fini();
4890 unregister_pernet_subsys(&bond_net_ops);
4892 #ifdef CONFIG_NET_POLL_CONTROLLER
4893 /* Make sure we don't have an imbalance on our netpoll blocking */
4894 WARN_ON(atomic_read(&netpoll_block_tx));
4898 module_init(bonding_init);
4899 module_exit(bonding_exit);
4900 MODULE_LICENSE("GPL");
4901 MODULE_VERSION(DRV_VERSION);
4902 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4903 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");