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-xor and 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->queue_mapping = 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;
1111 bond_dev->gso_max_segs = gso_max_segs;
1112 netif_set_gso_max_size(bond_dev, gso_max_size);
1114 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1115 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1116 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1117 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1119 netdev_change_features(bond_dev);
1122 static void bond_setup_by_slave(struct net_device *bond_dev,
1123 struct net_device *slave_dev)
1125 bond_dev->header_ops = slave_dev->header_ops;
1127 bond_dev->type = slave_dev->type;
1128 bond_dev->hard_header_len = slave_dev->hard_header_len;
1129 bond_dev->addr_len = slave_dev->addr_len;
1131 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1132 slave_dev->addr_len);
1135 /* On bonding slaves other than the currently active slave, suppress
1136 * duplicates except for alb non-mcast/bcast.
1138 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1139 struct slave *slave,
1140 struct bonding *bond)
1142 if (bond_is_slave_inactive(slave)) {
1143 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1144 skb->pkt_type != PACKET_BROADCAST &&
1145 skb->pkt_type != PACKET_MULTICAST)
1152 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1154 struct sk_buff *skb = *pskb;
1155 struct slave *slave;
1156 struct bonding *bond;
1157 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1159 int ret = RX_HANDLER_ANOTHER;
1161 skb = skb_share_check(skb, GFP_ATOMIC);
1163 return RX_HANDLER_CONSUMED;
1167 slave = bond_slave_get_rcu(skb->dev);
1170 recv_probe = ACCESS_ONCE(bond->recv_probe);
1172 ret = recv_probe(skb, bond, slave);
1173 if (ret == RX_HANDLER_CONSUMED) {
1179 /* don't change skb->dev for link-local packets */
1180 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1181 return RX_HANDLER_PASS;
1182 if (bond_should_deliver_exact_match(skb, slave, bond))
1183 return RX_HANDLER_EXACT;
1185 skb->dev = bond->dev;
1187 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1188 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1189 skb->pkt_type == PACKET_HOST) {
1191 if (unlikely(skb_cow_head(skb,
1192 skb->data - skb_mac_header(skb)))) {
1194 return RX_HANDLER_CONSUMED;
1196 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1197 bond->dev->addr_len);
1203 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1205 switch (BOND_MODE(bond)) {
1206 case BOND_MODE_ROUNDROBIN:
1207 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1208 case BOND_MODE_ACTIVEBACKUP:
1209 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1210 case BOND_MODE_BROADCAST:
1211 return NETDEV_LAG_TX_TYPE_BROADCAST;
1213 case BOND_MODE_8023AD:
1214 return NETDEV_LAG_TX_TYPE_HASH;
1216 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1220 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1222 struct netdev_lag_upper_info lag_upper_info;
1225 lag_upper_info.tx_type = bond_lag_tx_type(bond);
1226 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1230 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1234 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1236 netdev_upper_dev_unlink(slave->dev, bond->dev);
1237 slave->dev->flags &= ~IFF_SLAVE;
1238 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1241 static struct slave *bond_alloc_slave(struct bonding *bond)
1243 struct slave *slave = NULL;
1245 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1249 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1250 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1252 if (!SLAVE_AD_INFO(slave)) {
1260 static void bond_free_slave(struct slave *slave)
1262 struct bonding *bond = bond_get_bond_by_slave(slave);
1264 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1265 kfree(SLAVE_AD_INFO(slave));
1270 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1272 info->bond_mode = BOND_MODE(bond);
1273 info->miimon = bond->params.miimon;
1274 info->num_slaves = bond->slave_cnt;
1277 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1279 strcpy(info->slave_name, slave->dev->name);
1280 info->link = slave->link;
1281 info->state = bond_slave_state(slave);
1282 info->link_failure_count = slave->link_failure_count;
1285 static void bond_netdev_notify(struct net_device *dev,
1286 struct netdev_bonding_info *info)
1289 netdev_bonding_info_change(dev, info);
1293 static void bond_netdev_notify_work(struct work_struct *_work)
1295 struct netdev_notify_work *w =
1296 container_of(_work, struct netdev_notify_work, work.work);
1298 bond_netdev_notify(w->dev, &w->bonding_info);
1303 void bond_queue_slave_event(struct slave *slave)
1305 struct bonding *bond = slave->bond;
1306 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1311 dev_hold(slave->dev);
1312 nnw->dev = slave->dev;
1313 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1314 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1315 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1317 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1320 void bond_lower_state_changed(struct slave *slave)
1322 struct netdev_lag_lower_state_info info;
1324 info.link_up = slave->link == BOND_LINK_UP ||
1325 slave->link == BOND_LINK_FAIL;
1326 info.tx_enabled = bond_is_active_slave(slave);
1327 netdev_lower_state_changed(slave->dev, &info);
1330 /* enslave device <slave> to bond device <master> */
1331 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1333 struct bonding *bond = netdev_priv(bond_dev);
1334 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1335 struct slave *new_slave = NULL, *prev_slave;
1336 struct sockaddr_storage ss;
1340 if (!bond->params.use_carrier &&
1341 slave_dev->ethtool_ops->get_link == NULL &&
1342 slave_ops->ndo_do_ioctl == NULL) {
1343 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1347 /* already in-use? */
1348 if (netdev_is_rx_handler_busy(slave_dev)) {
1349 netdev_err(bond_dev,
1350 "Error: Device is in use and cannot be enslaved\n");
1354 if (bond_dev == slave_dev) {
1355 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1359 /* vlan challenged mutual exclusion */
1360 /* no need to lock since we're protected by rtnl_lock */
1361 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1362 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1364 if (vlan_uses_dev(bond_dev)) {
1365 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1366 slave_dev->name, bond_dev->name);
1369 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1370 slave_dev->name, slave_dev->name,
1374 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1378 /* Old ifenslave binaries are no longer supported. These can
1379 * be identified with moderate accuracy by the state of the slave:
1380 * the current ifenslave will set the interface down prior to
1381 * enslaving it; the old ifenslave will not.
1383 if (slave_dev->flags & IFF_UP) {
1384 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1389 /* set bonding device ether type by slave - bonding netdevices are
1390 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1391 * there is a need to override some of the type dependent attribs/funcs.
1393 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1394 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1396 if (!bond_has_slaves(bond)) {
1397 if (bond_dev->type != slave_dev->type) {
1398 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1399 bond_dev->type, slave_dev->type);
1401 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1403 res = notifier_to_errno(res);
1405 netdev_err(bond_dev, "refused to change device type\n");
1409 /* Flush unicast and multicast addresses */
1410 dev_uc_flush(bond_dev);
1411 dev_mc_flush(bond_dev);
1413 if (slave_dev->type != ARPHRD_ETHER)
1414 bond_setup_by_slave(bond_dev, slave_dev);
1416 ether_setup(bond_dev);
1417 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1420 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1423 } else if (bond_dev->type != slave_dev->type) {
1424 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1425 slave_dev->name, slave_dev->type, bond_dev->type);
1429 if (slave_dev->type == ARPHRD_INFINIBAND &&
1430 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1431 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1434 goto err_undo_flags;
1437 if (!slave_ops->ndo_set_mac_address ||
1438 slave_dev->type == ARPHRD_INFINIBAND) {
1439 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1440 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1441 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1442 if (!bond_has_slaves(bond)) {
1443 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1444 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1446 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");
1448 goto err_undo_flags;
1453 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1455 /* If this is the first slave, then we need to set the master's hardware
1456 * address to be the same as the slave's.
1458 if (!bond_has_slaves(bond) &&
1459 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1460 bond_set_dev_addr(bond->dev, slave_dev);
1462 new_slave = bond_alloc_slave(bond);
1465 goto err_undo_flags;
1468 new_slave->bond = bond;
1469 new_slave->dev = slave_dev;
1470 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1471 * is set via sysfs or module option if desired.
1473 new_slave->queue_id = 0;
1475 /* Save slave's original mtu and then set it to match the bond */
1476 new_slave->original_mtu = slave_dev->mtu;
1477 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1479 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1483 /* Save slave's original ("permanent") mac address for modes
1484 * that need it, and for restoring it upon release, and then
1485 * set it to the master's address
1487 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1488 slave_dev->addr_len);
1490 if (!bond->params.fail_over_mac ||
1491 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1492 /* Set slave to master's mac address. The application already
1493 * set the master's mac address to that of the first slave
1495 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1496 ss.ss_family = slave_dev->type;
1497 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1499 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1500 goto err_restore_mtu;
1504 /* set slave flag before open to prevent IPv6 addrconf */
1505 slave_dev->flags |= IFF_SLAVE;
1507 /* open the slave since the application closed it */
1508 res = dev_open(slave_dev);
1510 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1511 goto err_restore_mac;
1514 slave_dev->priv_flags |= IFF_BONDING;
1515 /* initialize slave stats */
1516 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1518 if (bond_is_lb(bond)) {
1519 /* bond_alb_init_slave() must be called before all other stages since
1520 * it might fail and we do not want to have to undo everything
1522 res = bond_alb_init_slave(bond, new_slave);
1527 /* If the mode uses primary, then the following is handled by
1528 * bond_change_active_slave().
1530 if (!bond_uses_primary(bond)) {
1531 /* set promiscuity level to new slave */
1532 if (bond_dev->flags & IFF_PROMISC) {
1533 res = dev_set_promiscuity(slave_dev, 1);
1538 /* set allmulti level to new slave */
1539 if (bond_dev->flags & IFF_ALLMULTI) {
1540 res = dev_set_allmulti(slave_dev, 1);
1545 netif_addr_lock_bh(bond_dev);
1547 dev_mc_sync_multiple(slave_dev, bond_dev);
1548 dev_uc_sync_multiple(slave_dev, bond_dev);
1550 netif_addr_unlock_bh(bond_dev);
1553 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1554 /* add lacpdu mc addr to mc list */
1555 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1557 dev_mc_add(slave_dev, lacpdu_multicast);
1560 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1562 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1567 prev_slave = bond_last_slave(bond);
1569 new_slave->delay = 0;
1570 new_slave->link_failure_count = 0;
1572 if (bond_update_speed_duplex(new_slave))
1573 new_slave->link = BOND_LINK_DOWN;
1575 new_slave->last_rx = jiffies -
1576 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1577 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1578 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1580 if (bond->params.miimon && !bond->params.use_carrier) {
1581 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1583 if ((link_reporting == -1) && !bond->params.arp_interval) {
1584 /* miimon is set but a bonded network driver
1585 * does not support ETHTOOL/MII and
1586 * arp_interval is not set. Note: if
1587 * use_carrier is enabled, we will never go
1588 * here (because netif_carrier is always
1589 * supported); thus, we don't need to change
1590 * the messages for netif_carrier.
1592 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",
1594 } else if (link_reporting == -1) {
1595 /* unable get link status using mii/ethtool */
1596 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",
1601 /* check for initial state */
1602 new_slave->link = BOND_LINK_NOCHANGE;
1603 if (bond->params.miimon) {
1604 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1605 if (bond->params.updelay) {
1606 bond_set_slave_link_state(new_slave,
1608 BOND_SLAVE_NOTIFY_NOW);
1609 new_slave->delay = bond->params.updelay;
1611 bond_set_slave_link_state(new_slave,
1613 BOND_SLAVE_NOTIFY_NOW);
1616 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1617 BOND_SLAVE_NOTIFY_NOW);
1619 } else if (bond->params.arp_interval) {
1620 bond_set_slave_link_state(new_slave,
1621 (netif_carrier_ok(slave_dev) ?
1622 BOND_LINK_UP : BOND_LINK_DOWN),
1623 BOND_SLAVE_NOTIFY_NOW);
1625 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1626 BOND_SLAVE_NOTIFY_NOW);
1629 if (new_slave->link != BOND_LINK_DOWN)
1630 new_slave->last_link_up = jiffies;
1631 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1632 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1633 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1635 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1636 /* if there is a primary slave, remember it */
1637 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1638 rcu_assign_pointer(bond->primary_slave, new_slave);
1639 bond->force_primary = true;
1643 switch (BOND_MODE(bond)) {
1644 case BOND_MODE_ACTIVEBACKUP:
1645 bond_set_slave_inactive_flags(new_slave,
1646 BOND_SLAVE_NOTIFY_NOW);
1648 case BOND_MODE_8023AD:
1649 /* in 802.3ad mode, the internal mechanism
1650 * will activate the slaves in the selected
1653 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1654 /* if this is the first slave */
1656 SLAVE_AD_INFO(new_slave)->id = 1;
1657 /* Initialize AD with the number of times that the AD timer is called in 1 second
1658 * can be called only after the mac address of the bond is set
1660 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1662 SLAVE_AD_INFO(new_slave)->id =
1663 SLAVE_AD_INFO(prev_slave)->id + 1;
1666 bond_3ad_bind_slave(new_slave);
1670 bond_set_active_slave(new_slave);
1671 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1674 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1676 /* always active in trunk mode */
1677 bond_set_active_slave(new_slave);
1679 /* In trunking mode there is little meaning to curr_active_slave
1680 * anyway (it holds no special properties of the bond device),
1681 * so we can change it without calling change_active_interface()
1683 if (!rcu_access_pointer(bond->curr_active_slave) &&
1684 new_slave->link == BOND_LINK_UP)
1685 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1688 } /* switch(bond_mode) */
1690 #ifdef CONFIG_NET_POLL_CONTROLLER
1691 slave_dev->npinfo = bond->dev->npinfo;
1692 if (slave_dev->npinfo) {
1693 if (slave_enable_netpoll(new_slave)) {
1694 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1701 if (!(bond_dev->features & NETIF_F_LRO))
1702 dev_disable_lro(slave_dev);
1704 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1707 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1711 res = bond_master_upper_dev_link(bond, new_slave);
1713 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1714 goto err_unregister;
1717 res = bond_sysfs_slave_add(new_slave);
1719 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1720 goto err_upper_unlink;
1724 bond_compute_features(bond);
1725 bond_set_carrier(bond);
1727 if (bond_uses_primary(bond)) {
1729 bond_select_active_slave(bond);
1730 unblock_netpoll_tx();
1733 if (bond_mode_uses_xmit_hash(bond))
1734 bond_update_slave_arr(bond, NULL);
1736 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1738 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1739 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1741 /* enslave is successful */
1742 bond_queue_slave_event(new_slave);
1745 /* Undo stages on error */
1747 bond_upper_dev_unlink(bond, new_slave);
1750 netdev_rx_handler_unregister(slave_dev);
1753 if (!bond_uses_primary(bond))
1754 bond_hw_addr_flush(bond_dev, slave_dev);
1756 vlan_vids_del_by_dev(slave_dev, bond_dev);
1757 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1758 RCU_INIT_POINTER(bond->primary_slave, NULL);
1759 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1761 bond_change_active_slave(bond, NULL);
1762 bond_select_active_slave(bond);
1763 unblock_netpoll_tx();
1765 /* either primary_slave or curr_active_slave might've changed */
1767 slave_disable_netpoll(new_slave);
1770 slave_dev->priv_flags &= ~IFF_BONDING;
1771 dev_close(slave_dev);
1774 slave_dev->flags &= ~IFF_SLAVE;
1775 if (!bond->params.fail_over_mac ||
1776 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1777 /* XXX TODO - fom follow mode needs to change master's
1778 * MAC if this slave's MAC is in use by the bond, or at
1779 * least print a warning.
1781 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1782 new_slave->dev->addr_len);
1783 ss.ss_family = slave_dev->type;
1784 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1788 dev_set_mtu(slave_dev, new_slave->original_mtu);
1791 bond_free_slave(new_slave);
1794 /* Enslave of first slave has failed and we need to fix master's mac */
1795 if (!bond_has_slaves(bond)) {
1796 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1797 slave_dev->dev_addr))
1798 eth_hw_addr_random(bond_dev);
1799 if (bond_dev->type != ARPHRD_ETHER) {
1800 dev_close(bond_dev);
1801 ether_setup(bond_dev);
1802 bond_dev->flags |= IFF_MASTER;
1803 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1810 /* Try to release the slave device <slave> from the bond device <master>
1811 * It is legal to access curr_active_slave without a lock because all the function
1812 * is RTNL-locked. If "all" is true it means that the function is being called
1813 * while destroying a bond interface and all slaves are being released.
1815 * The rules for slave state should be:
1816 * for Active/Backup:
1817 * Active stays on all backups go down
1818 * for Bonded connections:
1819 * The first up interface should be left on and all others downed.
1821 static int __bond_release_one(struct net_device *bond_dev,
1822 struct net_device *slave_dev,
1825 struct bonding *bond = netdev_priv(bond_dev);
1826 struct slave *slave, *oldcurrent;
1827 struct sockaddr_storage ss;
1828 int old_flags = bond_dev->flags;
1829 netdev_features_t old_features = bond_dev->features;
1831 /* slave is not a slave or master is not master of this slave */
1832 if (!(slave_dev->flags & IFF_SLAVE) ||
1833 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1834 netdev_dbg(bond_dev, "cannot release %s\n",
1841 slave = bond_get_slave_by_dev(bond, slave_dev);
1843 /* not a slave of this bond */
1844 netdev_info(bond_dev, "%s not enslaved\n",
1846 unblock_netpoll_tx();
1850 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1852 bond_sysfs_slave_del(slave);
1854 /* recompute stats just before removing the slave */
1855 bond_get_stats(bond->dev, &bond->bond_stats);
1857 bond_upper_dev_unlink(bond, slave);
1858 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1859 * for this slave anymore.
1861 netdev_rx_handler_unregister(slave_dev);
1863 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1864 bond_3ad_unbind_slave(slave);
1866 if (bond_mode_uses_xmit_hash(bond))
1867 bond_update_slave_arr(bond, slave);
1869 netdev_info(bond_dev, "Releasing %s interface %s\n",
1870 bond_is_active_slave(slave) ? "active" : "backup",
1873 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1875 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1877 if (!all && (!bond->params.fail_over_mac ||
1878 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1879 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1880 bond_has_slaves(bond))
1881 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",
1882 slave_dev->name, slave->perm_hwaddr,
1883 bond_dev->name, slave_dev->name);
1886 if (rtnl_dereference(bond->primary_slave) == slave)
1887 RCU_INIT_POINTER(bond->primary_slave, NULL);
1889 if (oldcurrent == slave)
1890 bond_change_active_slave(bond, NULL);
1892 if (bond_is_lb(bond)) {
1893 /* Must be called only after the slave has been
1894 * detached from the list and the curr_active_slave
1895 * has been cleared (if our_slave == old_current),
1896 * but before a new active slave is selected.
1898 bond_alb_deinit_slave(bond, slave);
1902 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1903 } else if (oldcurrent == slave) {
1904 /* Note that we hold RTNL over this sequence, so there
1905 * is no concern that another slave add/remove event
1908 bond_select_active_slave(bond);
1911 if (!bond_has_slaves(bond)) {
1912 bond_set_carrier(bond);
1913 eth_hw_addr_random(bond_dev);
1916 unblock_netpoll_tx();
1920 if (!bond_has_slaves(bond)) {
1921 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1922 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1925 bond_compute_features(bond);
1926 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1927 (old_features & NETIF_F_VLAN_CHALLENGED))
1928 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1929 slave_dev->name, bond_dev->name);
1931 vlan_vids_del_by_dev(slave_dev, bond_dev);
1933 /* If the mode uses primary, then this case was handled above by
1934 * bond_change_active_slave(..., NULL)
1936 if (!bond_uses_primary(bond)) {
1937 /* unset promiscuity level from slave
1938 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1939 * of the IFF_PROMISC flag in the bond_dev, but we need the
1940 * value of that flag before that change, as that was the value
1941 * when this slave was attached, so we cache at the start of the
1942 * function and use it here. Same goes for ALLMULTI below
1944 if (old_flags & IFF_PROMISC)
1945 dev_set_promiscuity(slave_dev, -1);
1947 /* unset allmulti level from slave */
1948 if (old_flags & IFF_ALLMULTI)
1949 dev_set_allmulti(slave_dev, -1);
1951 bond_hw_addr_flush(bond_dev, slave_dev);
1954 slave_disable_netpoll(slave);
1956 /* close slave before restoring its mac address */
1957 dev_close(slave_dev);
1959 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1960 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1961 /* restore original ("permanent") mac address */
1962 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
1963 slave->dev->addr_len);
1964 ss.ss_family = slave_dev->type;
1965 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1968 dev_set_mtu(slave_dev, slave->original_mtu);
1970 slave_dev->priv_flags &= ~IFF_BONDING;
1972 bond_free_slave(slave);
1977 /* A wrapper used because of ndo_del_link */
1978 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1980 return __bond_release_one(bond_dev, slave_dev, false);
1983 /* First release a slave and then destroy the bond if no more slaves are left.
1984 * Must be under rtnl_lock when this function is called.
1986 static int bond_release_and_destroy(struct net_device *bond_dev,
1987 struct net_device *slave_dev)
1989 struct bonding *bond = netdev_priv(bond_dev);
1992 ret = bond_release(bond_dev, slave_dev);
1993 if (ret == 0 && !bond_has_slaves(bond)) {
1994 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1995 netdev_info(bond_dev, "Destroying bond %s\n",
1997 bond_remove_proc_entry(bond);
1998 unregister_netdevice(bond_dev);
2003 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2005 struct bonding *bond = netdev_priv(bond_dev);
2006 bond_fill_ifbond(bond, info);
2009 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2011 struct bonding *bond = netdev_priv(bond_dev);
2012 struct list_head *iter;
2013 int i = 0, res = -ENODEV;
2014 struct slave *slave;
2016 bond_for_each_slave(bond, slave, iter) {
2017 if (i++ == (int)info->slave_id) {
2019 bond_fill_ifslave(slave, info);
2027 /*-------------------------------- Monitoring -------------------------------*/
2029 /* called with rcu_read_lock() */
2030 static int bond_miimon_inspect(struct bonding *bond)
2032 int link_state, commit = 0;
2033 struct list_head *iter;
2034 struct slave *slave;
2035 bool ignore_updelay;
2037 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2039 bond_for_each_slave_rcu(bond, slave, iter) {
2040 slave->new_link = BOND_LINK_NOCHANGE;
2042 link_state = bond_check_dev_link(bond, slave->dev, 0);
2044 switch (slave->link) {
2049 bond_propose_link_state(slave, BOND_LINK_FAIL);
2050 slave->delay = bond->params.downdelay;
2052 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2054 BOND_MODE_ACTIVEBACKUP) ?
2055 (bond_is_active_slave(slave) ?
2056 "active " : "backup ") : "",
2058 bond->params.downdelay * bond->params.miimon);
2061 case BOND_LINK_FAIL:
2063 /* recovered before downdelay expired */
2064 bond_propose_link_state(slave, BOND_LINK_UP);
2065 slave->last_link_up = jiffies;
2066 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2067 (bond->params.downdelay - slave->delay) *
2068 bond->params.miimon,
2074 if (slave->delay <= 0) {
2075 slave->new_link = BOND_LINK_DOWN;
2083 case BOND_LINK_DOWN:
2087 bond_propose_link_state(slave, BOND_LINK_BACK);
2088 slave->delay = bond->params.updelay;
2091 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2093 ignore_updelay ? 0 :
2094 bond->params.updelay *
2095 bond->params.miimon);
2098 case BOND_LINK_BACK:
2100 bond_propose_link_state(slave, BOND_LINK_DOWN);
2101 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2102 (bond->params.updelay - slave->delay) *
2103 bond->params.miimon,
2112 if (slave->delay <= 0) {
2113 slave->new_link = BOND_LINK_UP;
2115 ignore_updelay = false;
2127 static void bond_miimon_commit(struct bonding *bond)
2129 struct list_head *iter;
2130 struct slave *slave, *primary;
2132 bond_for_each_slave(bond, slave, iter) {
2133 switch (slave->new_link) {
2134 case BOND_LINK_NOCHANGE:
2138 if (bond_update_speed_duplex(slave)) {
2139 slave->link = BOND_LINK_DOWN;
2140 netdev_warn(bond->dev,
2141 "failed to get link speed/duplex for %s\n",
2145 bond_set_slave_link_state(slave, BOND_LINK_UP,
2146 BOND_SLAVE_NOTIFY_NOW);
2147 slave->last_link_up = jiffies;
2149 primary = rtnl_dereference(bond->primary_slave);
2150 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2151 /* prevent it from being the active one */
2152 bond_set_backup_slave(slave);
2153 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2154 /* make it immediately active */
2155 bond_set_active_slave(slave);
2156 } else if (slave != primary) {
2157 /* prevent it from being the active one */
2158 bond_set_backup_slave(slave);
2161 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2163 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2164 slave->duplex ? "full" : "half");
2166 /* notify ad that the link status has changed */
2167 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2168 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2170 if (bond_is_lb(bond))
2171 bond_alb_handle_link_change(bond, slave,
2174 if (BOND_MODE(bond) == BOND_MODE_XOR)
2175 bond_update_slave_arr(bond, NULL);
2177 if (!bond->curr_active_slave || slave == primary)
2182 case BOND_LINK_DOWN:
2183 if (slave->link_failure_count < UINT_MAX)
2184 slave->link_failure_count++;
2186 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2187 BOND_SLAVE_NOTIFY_NOW);
2189 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2190 BOND_MODE(bond) == BOND_MODE_8023AD)
2191 bond_set_slave_inactive_flags(slave,
2192 BOND_SLAVE_NOTIFY_NOW);
2194 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2197 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2198 bond_3ad_handle_link_change(slave,
2201 if (bond_is_lb(bond))
2202 bond_alb_handle_link_change(bond, slave,
2205 if (BOND_MODE(bond) == BOND_MODE_XOR)
2206 bond_update_slave_arr(bond, NULL);
2208 if (slave == rcu_access_pointer(bond->curr_active_slave))
2214 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2215 slave->new_link, slave->dev->name);
2216 slave->new_link = BOND_LINK_NOCHANGE;
2223 bond_select_active_slave(bond);
2224 unblock_netpoll_tx();
2227 bond_set_carrier(bond);
2232 * Really a wrapper that splits the mii monitor into two phases: an
2233 * inspection, then (if inspection indicates something needs to be done)
2234 * an acquisition of appropriate locks followed by a commit phase to
2235 * implement whatever link state changes are indicated.
2237 static void bond_mii_monitor(struct work_struct *work)
2239 struct bonding *bond = container_of(work, struct bonding,
2241 bool should_notify_peers = false;
2242 unsigned long delay;
2243 struct slave *slave;
2244 struct list_head *iter;
2246 delay = msecs_to_jiffies(bond->params.miimon);
2248 if (!bond_has_slaves(bond))
2253 should_notify_peers = bond_should_notify_peers(bond);
2255 if (bond_miimon_inspect(bond)) {
2258 /* Race avoidance with bond_close cancel of workqueue */
2259 if (!rtnl_trylock()) {
2261 should_notify_peers = false;
2265 bond_for_each_slave(bond, slave, iter) {
2266 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2268 bond_miimon_commit(bond);
2270 rtnl_unlock(); /* might sleep, hold no other locks */
2275 if (bond->params.miimon)
2276 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2278 if (should_notify_peers) {
2279 if (!rtnl_trylock())
2281 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2286 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2288 __be32 ip = *((__be32 *)data);
2290 return ip == bond_confirm_addr(upper, 0, ip);
2293 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2297 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2301 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2308 /* We go to the (large) trouble of VLAN tagging ARP frames because
2309 * switches in VLAN mode (especially if ports are configured as
2310 * "native" to a VLAN) might not pass non-tagged frames.
2312 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2313 __be32 dest_ip, __be32 src_ip,
2314 struct bond_vlan_tag *tags)
2316 struct sk_buff *skb;
2317 struct bond_vlan_tag *outer_tag = tags;
2319 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2320 arp_op, slave_dev->name, &dest_ip, &src_ip);
2322 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2323 NULL, slave_dev->dev_addr, NULL);
2326 net_err_ratelimited("ARP packet allocation failed\n");
2330 if (!tags || tags->vlan_proto == VLAN_N_VID)
2335 /* Go through all the tags backwards and add them to the packet */
2336 while (tags->vlan_proto != VLAN_N_VID) {
2337 if (!tags->vlan_id) {
2342 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2343 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2344 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2347 net_err_ratelimited("failed to insert inner VLAN tag\n");
2353 /* Set the outer tag */
2354 if (outer_tag->vlan_id) {
2355 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2356 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2357 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2358 outer_tag->vlan_id);
2365 /* Validate the device path between the @start_dev and the @end_dev.
2366 * The path is valid if the @end_dev is reachable through device
2368 * When the path is validated, collect any vlan information in the
2371 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2372 struct net_device *end_dev,
2375 struct bond_vlan_tag *tags;
2376 struct net_device *upper;
2377 struct list_head *iter;
2379 if (start_dev == end_dev) {
2380 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2382 return ERR_PTR(-ENOMEM);
2383 tags[level].vlan_proto = VLAN_N_VID;
2387 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2388 tags = bond_verify_device_path(upper, end_dev, level + 1);
2389 if (IS_ERR_OR_NULL(tags)) {
2394 if (is_vlan_dev(upper)) {
2395 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2396 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2405 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2408 struct bond_vlan_tag *tags;
2409 __be32 *targets = bond->params.arp_targets, addr;
2412 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2413 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2416 /* Find out through which dev should the packet go */
2417 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2420 /* there's no route to target - try to send arp
2421 * probe to generate any traffic (arp_validate=0)
2423 if (bond->params.arp_validate)
2424 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2427 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2432 /* bond device itself */
2433 if (rt->dst.dev == bond->dev)
2437 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2440 if (!IS_ERR_OR_NULL(tags))
2443 /* Not our device - skip */
2444 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2445 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2451 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2453 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2459 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2463 if (!sip || !bond_has_this_ip(bond, tip)) {
2464 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2469 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2471 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2475 slave->last_rx = jiffies;
2476 slave->target_last_arp_rx[i] = jiffies;
2479 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2480 struct slave *slave)
2482 struct arphdr *arp = (struct arphdr *)skb->data;
2483 struct slave *curr_active_slave, *curr_arp_slave;
2484 unsigned char *arp_ptr;
2486 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2488 if (!slave_do_arp_validate(bond, slave)) {
2489 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2490 !slave_do_arp_validate_only(bond))
2491 slave->last_rx = jiffies;
2492 return RX_HANDLER_ANOTHER;
2493 } else if (!is_arp) {
2494 return RX_HANDLER_ANOTHER;
2497 alen = arp_hdr_len(bond->dev);
2499 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2502 if (alen > skb_headlen(skb)) {
2503 arp = kmalloc(alen, GFP_ATOMIC);
2506 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2510 if (arp->ar_hln != bond->dev->addr_len ||
2511 skb->pkt_type == PACKET_OTHERHOST ||
2512 skb->pkt_type == PACKET_LOOPBACK ||
2513 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2514 arp->ar_pro != htons(ETH_P_IP) ||
2518 arp_ptr = (unsigned char *)(arp + 1);
2519 arp_ptr += bond->dev->addr_len;
2520 memcpy(&sip, arp_ptr, 4);
2521 arp_ptr += 4 + bond->dev->addr_len;
2522 memcpy(&tip, arp_ptr, 4);
2524 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2525 slave->dev->name, bond_slave_state(slave),
2526 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2529 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2530 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2532 /* We 'trust' the received ARP enough to validate it if:
2534 * (a) the slave receiving the ARP is active (which includes the
2535 * current ARP slave, if any), or
2537 * (b) the receiving slave isn't active, but there is a currently
2538 * active slave and it received valid arp reply(s) after it became
2539 * the currently active slave, or
2541 * (c) there is an ARP slave that sent an ARP during the prior ARP
2542 * interval, and we receive an ARP reply on any slave. We accept
2543 * these because switch FDB update delays may deliver the ARP
2544 * reply to a slave other than the sender of the ARP request.
2546 * Note: for (b), backup slaves are receiving the broadcast ARP
2547 * request, not a reply. This request passes from the sending
2548 * slave through the L2 switch(es) to the receiving slave. Since
2549 * this is checking the request, sip/tip are swapped for
2552 * This is done to avoid endless looping when we can't reach the
2553 * arp_ip_target and fool ourselves with our own arp requests.
2555 if (bond_is_active_slave(slave))
2556 bond_validate_arp(bond, slave, sip, tip);
2557 else if (curr_active_slave &&
2558 time_after(slave_last_rx(bond, curr_active_slave),
2559 curr_active_slave->last_link_up))
2560 bond_validate_arp(bond, slave, tip, sip);
2561 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2562 bond_time_in_interval(bond,
2563 dev_trans_start(curr_arp_slave->dev), 1))
2564 bond_validate_arp(bond, slave, sip, tip);
2567 if (arp != (struct arphdr *)skb->data)
2569 return RX_HANDLER_ANOTHER;
2572 /* function to verify if we're in the arp_interval timeslice, returns true if
2573 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2574 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2576 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2579 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2581 return time_in_range(jiffies,
2582 last_act - delta_in_ticks,
2583 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2586 /* This function is called regularly to monitor each slave's link
2587 * ensuring that traffic is being sent and received when arp monitoring
2588 * is used in load-balancing mode. if the adapter has been dormant, then an
2589 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2590 * arp monitoring in active backup mode.
2592 static void bond_loadbalance_arp_mon(struct bonding *bond)
2594 struct slave *slave, *oldcurrent;
2595 struct list_head *iter;
2596 int do_failover = 0, slave_state_changed = 0;
2598 if (!bond_has_slaves(bond))
2603 oldcurrent = rcu_dereference(bond->curr_active_slave);
2604 /* see if any of the previous devices are up now (i.e. they have
2605 * xmt and rcv traffic). the curr_active_slave does not come into
2606 * the picture unless it is null. also, slave->last_link_up is not
2607 * needed here because we send an arp on each slave and give a slave
2608 * as long as it needs to get the tx/rx within the delta.
2609 * TODO: what about up/down delay in arp mode? it wasn't here before
2612 bond_for_each_slave_rcu(bond, slave, iter) {
2613 unsigned long trans_start = dev_trans_start(slave->dev);
2615 slave->new_link = BOND_LINK_NOCHANGE;
2617 if (slave->link != BOND_LINK_UP) {
2618 if (bond_time_in_interval(bond, trans_start, 1) &&
2619 bond_time_in_interval(bond, slave->last_rx, 1)) {
2621 slave->new_link = BOND_LINK_UP;
2622 slave_state_changed = 1;
2624 /* primary_slave has no meaning in round-robin
2625 * mode. the window of a slave being up and
2626 * curr_active_slave being null after enslaving
2630 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2634 netdev_info(bond->dev, "interface %s is now up\n",
2639 /* slave->link == BOND_LINK_UP */
2641 /* not all switches will respond to an arp request
2642 * when the source ip is 0, so don't take the link down
2643 * if we don't know our ip yet
2645 if (!bond_time_in_interval(bond, trans_start, 2) ||
2646 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2648 slave->new_link = BOND_LINK_DOWN;
2649 slave_state_changed = 1;
2651 if (slave->link_failure_count < UINT_MAX)
2652 slave->link_failure_count++;
2654 netdev_info(bond->dev, "interface %s is now down\n",
2657 if (slave == oldcurrent)
2662 /* note: if switch is in round-robin mode, all links
2663 * must tx arp to ensure all links rx an arp - otherwise
2664 * links may oscillate or not come up at all; if switch is
2665 * in something like xor mode, there is nothing we can
2666 * do - all replies will be rx'ed on same link causing slaves
2667 * to be unstable during low/no traffic periods
2669 if (bond_slave_is_up(slave))
2670 bond_arp_send_all(bond, slave);
2675 if (do_failover || slave_state_changed) {
2676 if (!rtnl_trylock())
2679 bond_for_each_slave(bond, slave, iter) {
2680 if (slave->new_link != BOND_LINK_NOCHANGE)
2681 slave->link = slave->new_link;
2684 if (slave_state_changed) {
2685 bond_slave_state_change(bond);
2686 if (BOND_MODE(bond) == BOND_MODE_XOR)
2687 bond_update_slave_arr(bond, NULL);
2691 bond_select_active_slave(bond);
2692 unblock_netpoll_tx();
2698 if (bond->params.arp_interval)
2699 queue_delayed_work(bond->wq, &bond->arp_work,
2700 msecs_to_jiffies(bond->params.arp_interval));
2703 /* Called to inspect slaves for active-backup mode ARP monitor link state
2704 * changes. Sets new_link in slaves to specify what action should take
2705 * place for the slave. Returns 0 if no changes are found, >0 if changes
2706 * to link states must be committed.
2708 * Called with rcu_read_lock held.
2710 static int bond_ab_arp_inspect(struct bonding *bond)
2712 unsigned long trans_start, last_rx;
2713 struct list_head *iter;
2714 struct slave *slave;
2717 bond_for_each_slave_rcu(bond, slave, iter) {
2718 slave->new_link = BOND_LINK_NOCHANGE;
2719 last_rx = slave_last_rx(bond, slave);
2721 if (slave->link != BOND_LINK_UP) {
2722 if (bond_time_in_interval(bond, last_rx, 1)) {
2723 slave->new_link = BOND_LINK_UP;
2729 /* Give slaves 2*delta after being enslaved or made
2730 * active. This avoids bouncing, as the last receive
2731 * times need a full ARP monitor cycle to be updated.
2733 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2736 /* Backup slave is down if:
2737 * - No current_arp_slave AND
2738 * - more than 3*delta since last receive AND
2739 * - the bond has an IP address
2741 * Note: a non-null current_arp_slave indicates
2742 * the curr_active_slave went down and we are
2743 * searching for a new one; under this condition
2744 * we only take the curr_active_slave down - this
2745 * gives each slave a chance to tx/rx traffic
2746 * before being taken out
2748 if (!bond_is_active_slave(slave) &&
2749 !rcu_access_pointer(bond->current_arp_slave) &&
2750 !bond_time_in_interval(bond, last_rx, 3)) {
2751 slave->new_link = BOND_LINK_DOWN;
2755 /* Active slave is down if:
2756 * - more than 2*delta since transmitting OR
2757 * - (more than 2*delta since receive AND
2758 * the bond has an IP address)
2760 trans_start = dev_trans_start(slave->dev);
2761 if (bond_is_active_slave(slave) &&
2762 (!bond_time_in_interval(bond, trans_start, 2) ||
2763 !bond_time_in_interval(bond, last_rx, 2))) {
2764 slave->new_link = BOND_LINK_DOWN;
2772 /* Called to commit link state changes noted by inspection step of
2773 * active-backup mode ARP monitor.
2775 * Called with RTNL hold.
2777 static void bond_ab_arp_commit(struct bonding *bond)
2779 unsigned long trans_start;
2780 struct list_head *iter;
2781 struct slave *slave;
2783 bond_for_each_slave(bond, slave, iter) {
2784 switch (slave->new_link) {
2785 case BOND_LINK_NOCHANGE:
2789 trans_start = dev_trans_start(slave->dev);
2790 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2791 (!rtnl_dereference(bond->curr_active_slave) &&
2792 bond_time_in_interval(bond, trans_start, 1))) {
2793 struct slave *current_arp_slave;
2795 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2796 bond_set_slave_link_state(slave, BOND_LINK_UP,
2797 BOND_SLAVE_NOTIFY_NOW);
2798 if (current_arp_slave) {
2799 bond_set_slave_inactive_flags(
2801 BOND_SLAVE_NOTIFY_NOW);
2802 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2805 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2808 if (!rtnl_dereference(bond->curr_active_slave) ||
2809 slave == rtnl_dereference(bond->primary_slave))
2816 case BOND_LINK_DOWN:
2817 if (slave->link_failure_count < UINT_MAX)
2818 slave->link_failure_count++;
2820 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2821 BOND_SLAVE_NOTIFY_NOW);
2822 bond_set_slave_inactive_flags(slave,
2823 BOND_SLAVE_NOTIFY_NOW);
2825 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2828 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2829 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2836 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2837 slave->new_link, slave->dev->name);
2843 bond_select_active_slave(bond);
2844 unblock_netpoll_tx();
2847 bond_set_carrier(bond);
2850 /* Send ARP probes for active-backup mode ARP monitor.
2852 * Called with rcu_read_lock held.
2854 static bool bond_ab_arp_probe(struct bonding *bond)
2856 struct slave *slave, *before = NULL, *new_slave = NULL,
2857 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2858 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2859 struct list_head *iter;
2861 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2863 if (curr_arp_slave && curr_active_slave)
2864 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2865 curr_arp_slave->dev->name,
2866 curr_active_slave->dev->name);
2868 if (curr_active_slave) {
2869 bond_arp_send_all(bond, curr_active_slave);
2870 return should_notify_rtnl;
2873 /* if we don't have a curr_active_slave, search for the next available
2874 * backup slave from the current_arp_slave and make it the candidate
2875 * for becoming the curr_active_slave
2878 if (!curr_arp_slave) {
2879 curr_arp_slave = bond_first_slave_rcu(bond);
2880 if (!curr_arp_slave)
2881 return should_notify_rtnl;
2884 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2886 bond_for_each_slave_rcu(bond, slave, iter) {
2887 if (!found && !before && bond_slave_is_up(slave))
2890 if (found && !new_slave && bond_slave_is_up(slave))
2892 /* if the link state is up at this point, we
2893 * mark it down - this can happen if we have
2894 * simultaneous link failures and
2895 * reselect_active_interface doesn't make this
2896 * one the current slave so it is still marked
2897 * up when it is actually down
2899 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2900 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2901 BOND_SLAVE_NOTIFY_LATER);
2902 if (slave->link_failure_count < UINT_MAX)
2903 slave->link_failure_count++;
2905 bond_set_slave_inactive_flags(slave,
2906 BOND_SLAVE_NOTIFY_LATER);
2908 netdev_info(bond->dev, "backup interface %s is now down\n",
2911 if (slave == curr_arp_slave)
2915 if (!new_slave && before)
2921 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2922 BOND_SLAVE_NOTIFY_LATER);
2923 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2924 bond_arp_send_all(bond, new_slave);
2925 new_slave->last_link_up = jiffies;
2926 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2929 bond_for_each_slave_rcu(bond, slave, iter) {
2930 if (slave->should_notify || slave->should_notify_link) {
2931 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2935 return should_notify_rtnl;
2938 static void bond_activebackup_arp_mon(struct bonding *bond)
2940 bool should_notify_peers = false;
2941 bool should_notify_rtnl = false;
2944 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2946 if (!bond_has_slaves(bond))
2951 should_notify_peers = bond_should_notify_peers(bond);
2953 if (bond_ab_arp_inspect(bond)) {
2956 /* Race avoidance with bond_close flush of workqueue */
2957 if (!rtnl_trylock()) {
2959 should_notify_peers = false;
2963 bond_ab_arp_commit(bond);
2969 should_notify_rtnl = bond_ab_arp_probe(bond);
2973 if (bond->params.arp_interval)
2974 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2976 if (should_notify_peers || should_notify_rtnl) {
2977 if (!rtnl_trylock())
2980 if (should_notify_peers)
2981 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2983 if (should_notify_rtnl) {
2984 bond_slave_state_notify(bond);
2985 bond_slave_link_notify(bond);
2992 static void bond_arp_monitor(struct work_struct *work)
2994 struct bonding *bond = container_of(work, struct bonding,
2997 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
2998 bond_activebackup_arp_mon(bond);
3000 bond_loadbalance_arp_mon(bond);
3003 /*-------------------------- netdev event handling --------------------------*/
3005 /* Change device name */
3006 static int bond_event_changename(struct bonding *bond)
3008 bond_remove_proc_entry(bond);
3009 bond_create_proc_entry(bond);
3011 bond_debug_reregister(bond);
3016 static int bond_master_netdev_event(unsigned long event,
3017 struct net_device *bond_dev)
3019 struct bonding *event_bond = netdev_priv(bond_dev);
3022 case NETDEV_CHANGENAME:
3023 return bond_event_changename(event_bond);
3024 case NETDEV_UNREGISTER:
3025 bond_remove_proc_entry(event_bond);
3027 case NETDEV_REGISTER:
3028 bond_create_proc_entry(event_bond);
3030 case NETDEV_NOTIFY_PEERS:
3031 if (event_bond->send_peer_notif)
3032 event_bond->send_peer_notif--;
3041 static int bond_slave_netdev_event(unsigned long event,
3042 struct net_device *slave_dev)
3044 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3045 struct bonding *bond;
3046 struct net_device *bond_dev;
3048 /* A netdev event can be generated while enslaving a device
3049 * before netdev_rx_handler_register is called in which case
3050 * slave will be NULL
3054 bond_dev = slave->bond->dev;
3056 primary = rtnl_dereference(bond->primary_slave);
3059 case NETDEV_UNREGISTER:
3060 if (bond_dev->type != ARPHRD_ETHER)
3061 bond_release_and_destroy(bond_dev, slave_dev);
3063 bond_release(bond_dev, slave_dev);
3067 bond_update_speed_duplex(slave);
3068 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3069 bond_3ad_adapter_speed_duplex_changed(slave);
3072 /* Refresh slave-array if applicable!
3073 * If the setup does not use miimon or arpmon (mode-specific!),
3074 * then these events will not cause the slave-array to be
3075 * refreshed. This will cause xmit to use a slave that is not
3076 * usable. Avoid such situation by refeshing the array at these
3077 * events. If these (miimon/arpmon) parameters are configured
3078 * then array gets refreshed twice and that should be fine!
3080 if (bond_mode_uses_xmit_hash(bond))
3081 bond_update_slave_arr(bond, NULL);
3083 case NETDEV_CHANGEMTU:
3084 /* TODO: Should slaves be allowed to
3085 * independently alter their MTU? For
3086 * an active-backup bond, slaves need
3087 * not be the same type of device, so
3088 * MTUs may vary. For other modes,
3089 * slaves arguably should have the
3090 * same MTUs. To do this, we'd need to
3091 * take over the slave's change_mtu
3092 * function for the duration of their
3096 case NETDEV_CHANGENAME:
3097 /* we don't care if we don't have primary set */
3098 if (!bond_uses_primary(bond) ||
3099 !bond->params.primary[0])
3102 if (slave == primary) {
3103 /* slave's name changed - he's no longer primary */
3104 RCU_INIT_POINTER(bond->primary_slave, NULL);
3105 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3106 /* we have a new primary slave */
3107 rcu_assign_pointer(bond->primary_slave, slave);
3108 } else { /* we didn't change primary - exit */
3112 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3113 primary ? slave_dev->name : "none");
3116 bond_select_active_slave(bond);
3117 unblock_netpoll_tx();
3119 case NETDEV_FEAT_CHANGE:
3120 bond_compute_features(bond);
3122 case NETDEV_RESEND_IGMP:
3123 /* Propagate to master device */
3124 call_netdevice_notifiers(event, slave->bond->dev);
3133 /* bond_netdev_event: handle netdev notifier chain events.
3135 * This function receives events for the netdev chain. The caller (an
3136 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3137 * locks for us to safely manipulate the slave devices (RTNL lock,
3140 static int bond_netdev_event(struct notifier_block *this,
3141 unsigned long event, void *ptr)
3143 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3145 netdev_dbg(event_dev, "event: %lx\n", event);
3147 if (!(event_dev->priv_flags & IFF_BONDING))
3150 if (event_dev->flags & IFF_MASTER) {
3151 netdev_dbg(event_dev, "IFF_MASTER\n");
3152 return bond_master_netdev_event(event, event_dev);
3155 if (event_dev->flags & IFF_SLAVE) {
3156 netdev_dbg(event_dev, "IFF_SLAVE\n");
3157 return bond_slave_netdev_event(event, event_dev);
3163 static struct notifier_block bond_netdev_notifier = {
3164 .notifier_call = bond_netdev_event,
3167 /*---------------------------- Hashing Policies -----------------------------*/
3169 /* L2 hash helper */
3170 static inline u32 bond_eth_hash(struct sk_buff *skb)
3172 struct ethhdr *ep, hdr_tmp;
3174 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3176 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3180 /* Extract the appropriate headers based on bond's xmit policy */
3181 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3182 struct flow_keys *fk)
3184 const struct ipv6hdr *iph6;
3185 const struct iphdr *iph;
3186 int noff, proto = -1;
3188 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3189 return skb_flow_dissect_flow_keys(skb, fk, 0);
3191 fk->ports.ports = 0;
3192 noff = skb_network_offset(skb);
3193 if (skb->protocol == htons(ETH_P_IP)) {
3194 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3197 iph_to_flow_copy_v4addrs(fk, iph);
3198 noff += iph->ihl << 2;
3199 if (!ip_is_fragment(iph))
3200 proto = iph->protocol;
3201 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3202 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3204 iph6 = ipv6_hdr(skb);
3205 iph_to_flow_copy_v6addrs(fk, iph6);
3206 noff += sizeof(*iph6);
3207 proto = iph6->nexthdr;
3211 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3212 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3218 * bond_xmit_hash - generate a hash value based on the xmit policy
3219 * @bond: bonding device
3220 * @skb: buffer to use for headers
3222 * This function will extract the necessary headers from the skb buffer and use
3223 * them to generate a hash based on the xmit_policy set in the bonding device
3225 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3227 struct flow_keys flow;
3230 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3234 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3235 !bond_flow_dissect(bond, skb, &flow))
3236 return bond_eth_hash(skb);
3238 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3239 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3240 hash = bond_eth_hash(skb);
3242 hash = (__force u32)flow.ports.ports;
3243 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3244 (__force u32)flow_get_u32_src(&flow);
3245 hash ^= (hash >> 16);
3246 hash ^= (hash >> 8);
3251 /*-------------------------- Device entry points ----------------------------*/
3253 void bond_work_init_all(struct bonding *bond)
3255 INIT_DELAYED_WORK(&bond->mcast_work,
3256 bond_resend_igmp_join_requests_delayed);
3257 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3258 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3259 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3260 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3261 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3264 static void bond_work_cancel_all(struct bonding *bond)
3266 cancel_delayed_work_sync(&bond->mii_work);
3267 cancel_delayed_work_sync(&bond->arp_work);
3268 cancel_delayed_work_sync(&bond->alb_work);
3269 cancel_delayed_work_sync(&bond->ad_work);
3270 cancel_delayed_work_sync(&bond->mcast_work);
3271 cancel_delayed_work_sync(&bond->slave_arr_work);
3274 static int bond_open(struct net_device *bond_dev)
3276 struct bonding *bond = netdev_priv(bond_dev);
3277 struct list_head *iter;
3278 struct slave *slave;
3280 /* reset slave->backup and slave->inactive */
3281 if (bond_has_slaves(bond)) {
3282 bond_for_each_slave(bond, slave, iter) {
3283 if (bond_uses_primary(bond) &&
3284 slave != rcu_access_pointer(bond->curr_active_slave)) {
3285 bond_set_slave_inactive_flags(slave,
3286 BOND_SLAVE_NOTIFY_NOW);
3287 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3288 bond_set_slave_active_flags(slave,
3289 BOND_SLAVE_NOTIFY_NOW);
3294 if (bond_is_lb(bond)) {
3295 /* bond_alb_initialize must be called before the timer
3298 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3300 if (bond->params.tlb_dynamic_lb)
3301 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3304 if (bond->params.miimon) /* link check interval, in milliseconds. */
3305 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3307 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3308 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3309 bond->recv_probe = bond_arp_rcv;
3312 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3313 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3314 /* register to receive LACPDUs */
3315 bond->recv_probe = bond_3ad_lacpdu_recv;
3316 bond_3ad_initiate_agg_selection(bond, 1);
3319 if (bond_mode_uses_xmit_hash(bond))
3320 bond_update_slave_arr(bond, NULL);
3325 static int bond_close(struct net_device *bond_dev)
3327 struct bonding *bond = netdev_priv(bond_dev);
3329 bond_work_cancel_all(bond);
3330 bond->send_peer_notif = 0;
3331 if (bond_is_lb(bond))
3332 bond_alb_deinitialize(bond);
3333 bond->recv_probe = NULL;
3338 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3339 * that some drivers can provide 32bit values only.
3341 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3342 const struct rtnl_link_stats64 *_new,
3343 const struct rtnl_link_stats64 *_old)
3345 const u64 *new = (const u64 *)_new;
3346 const u64 *old = (const u64 *)_old;
3347 u64 *res = (u64 *)_res;
3350 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3353 s64 delta = nv - ov;
3355 /* detects if this particular field is 32bit only */
3356 if (((nv | ov) >> 32) == 0)
3357 delta = (s64)(s32)((u32)nv - (u32)ov);
3359 /* filter anomalies, some drivers reset their stats
3360 * at down/up events.
3367 static void bond_get_stats(struct net_device *bond_dev,
3368 struct rtnl_link_stats64 *stats)
3370 struct bonding *bond = netdev_priv(bond_dev);
3371 struct rtnl_link_stats64 temp;
3372 struct list_head *iter;
3373 struct slave *slave;
3375 spin_lock(&bond->stats_lock);
3376 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3379 bond_for_each_slave_rcu(bond, slave, iter) {
3380 const struct rtnl_link_stats64 *new =
3381 dev_get_stats(slave->dev, &temp);
3383 bond_fold_stats(stats, new, &slave->slave_stats);
3385 /* save off the slave stats for the next run */
3386 memcpy(&slave->slave_stats, new, sizeof(*new));
3390 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3391 spin_unlock(&bond->stats_lock);
3394 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3396 struct bonding *bond = netdev_priv(bond_dev);
3397 struct net_device *slave_dev = NULL;
3398 struct ifbond k_binfo;
3399 struct ifbond __user *u_binfo = NULL;
3400 struct ifslave k_sinfo;
3401 struct ifslave __user *u_sinfo = NULL;
3402 struct mii_ioctl_data *mii = NULL;
3403 struct bond_opt_value newval;
3407 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3418 /* We do this again just in case we were called by SIOCGMIIREG
3419 * instead of SIOCGMIIPHY.
3425 if (mii->reg_num == 1) {
3427 if (netif_carrier_ok(bond->dev))
3428 mii->val_out = BMSR_LSTATUS;
3432 case BOND_INFO_QUERY_OLD:
3433 case SIOCBONDINFOQUERY:
3434 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3436 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3439 bond_info_query(bond_dev, &k_binfo);
3440 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3444 case BOND_SLAVE_INFO_QUERY_OLD:
3445 case SIOCBONDSLAVEINFOQUERY:
3446 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3448 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3451 res = bond_slave_info_query(bond_dev, &k_sinfo);
3453 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3461 net = dev_net(bond_dev);
3463 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3466 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3468 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3473 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3475 case BOND_ENSLAVE_OLD:
3476 case SIOCBONDENSLAVE:
3477 res = bond_enslave(bond_dev, slave_dev);
3479 case BOND_RELEASE_OLD:
3480 case SIOCBONDRELEASE:
3481 res = bond_release(bond_dev, slave_dev);
3483 case BOND_SETHWADDR_OLD:
3484 case SIOCBONDSETHWADDR:
3485 bond_set_dev_addr(bond_dev, slave_dev);
3488 case BOND_CHANGE_ACTIVE_OLD:
3489 case SIOCBONDCHANGEACTIVE:
3490 bond_opt_initstr(&newval, slave_dev->name);
3491 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3501 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3503 struct bonding *bond = netdev_priv(bond_dev);
3505 if (change & IFF_PROMISC)
3506 bond_set_promiscuity(bond,
3507 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3509 if (change & IFF_ALLMULTI)
3510 bond_set_allmulti(bond,
3511 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3514 static void bond_set_rx_mode(struct net_device *bond_dev)
3516 struct bonding *bond = netdev_priv(bond_dev);
3517 struct list_head *iter;
3518 struct slave *slave;
3521 if (bond_uses_primary(bond)) {
3522 slave = rcu_dereference(bond->curr_active_slave);
3524 dev_uc_sync(slave->dev, bond_dev);
3525 dev_mc_sync(slave->dev, bond_dev);
3528 bond_for_each_slave_rcu(bond, slave, iter) {
3529 dev_uc_sync_multiple(slave->dev, bond_dev);
3530 dev_mc_sync_multiple(slave->dev, bond_dev);
3536 static int bond_neigh_init(struct neighbour *n)
3538 struct bonding *bond = netdev_priv(n->dev);
3539 const struct net_device_ops *slave_ops;
3540 struct neigh_parms parms;
3541 struct slave *slave;
3544 slave = bond_first_slave(bond);
3547 slave_ops = slave->dev->netdev_ops;
3548 if (!slave_ops->ndo_neigh_setup)
3551 parms.neigh_setup = NULL;
3552 parms.neigh_cleanup = NULL;
3553 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3557 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3558 * after the last slave has been detached. Assumes that all slaves
3559 * utilize the same neigh_cleanup (true at this writing as only user
3562 n->parms->neigh_cleanup = parms.neigh_cleanup;
3564 if (!parms.neigh_setup)
3567 return parms.neigh_setup(n);
3570 /* The bonding ndo_neigh_setup is called at init time beofre any
3571 * slave exists. So we must declare proxy setup function which will
3572 * be used at run time to resolve the actual slave neigh param setup.
3574 * It's also called by master devices (such as vlans) to setup their
3575 * underlying devices. In that case - do nothing, we're already set up from
3578 static int bond_neigh_setup(struct net_device *dev,
3579 struct neigh_parms *parms)
3581 /* modify only our neigh_parms */
3582 if (parms->dev == dev)
3583 parms->neigh_setup = bond_neigh_init;
3588 /* Change the MTU of all of a master's slaves to match the master */
3589 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3591 struct bonding *bond = netdev_priv(bond_dev);
3592 struct slave *slave, *rollback_slave;
3593 struct list_head *iter;
3596 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3598 bond_for_each_slave(bond, slave, iter) {
3599 netdev_dbg(bond_dev, "s %p c_m %p\n",
3600 slave, slave->dev->netdev_ops->ndo_change_mtu);
3602 res = dev_set_mtu(slave->dev, new_mtu);
3605 /* If we failed to set the slave's mtu to the new value
3606 * we must abort the operation even in ACTIVE_BACKUP
3607 * mode, because if we allow the backup slaves to have
3608 * different mtu values than the active slave we'll
3609 * need to change their mtu when doing a failover. That
3610 * means changing their mtu from timer context, which
3611 * is probably not a good idea.
3613 netdev_dbg(bond_dev, "err %d %s\n", res,
3619 bond_dev->mtu = new_mtu;
3624 /* unwind from head to the slave that failed */
3625 bond_for_each_slave(bond, rollback_slave, iter) {
3628 if (rollback_slave == slave)
3631 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3633 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3634 tmp_res, rollback_slave->dev->name);
3641 /* Change HW address
3643 * Note that many devices must be down to change the HW address, and
3644 * downing the master releases all slaves. We can make bonds full of
3645 * bonding devices to test this, however.
3647 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3649 struct bonding *bond = netdev_priv(bond_dev);
3650 struct slave *slave, *rollback_slave;
3651 struct sockaddr_storage *ss = addr, tmp_ss;
3652 struct list_head *iter;
3655 if (BOND_MODE(bond) == BOND_MODE_ALB)
3656 return bond_alb_set_mac_address(bond_dev, addr);
3659 netdev_dbg(bond_dev, "bond=%p\n", bond);
3661 /* If fail_over_mac is enabled, do nothing and return success.
3662 * Returning an error causes ifenslave to fail.
3664 if (bond->params.fail_over_mac &&
3665 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3668 if (!is_valid_ether_addr(ss->__data))
3669 return -EADDRNOTAVAIL;
3671 bond_for_each_slave(bond, slave, iter) {
3672 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3673 res = dev_set_mac_address(slave->dev, addr);
3675 /* TODO: consider downing the slave
3677 * User should expect communications
3678 * breakage anyway until ARP finish
3681 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3687 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3691 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3692 tmp_ss.ss_family = bond_dev->type;
3694 /* unwind from head to the slave that failed */
3695 bond_for_each_slave(bond, rollback_slave, iter) {
3698 if (rollback_slave == slave)
3701 tmp_res = dev_set_mac_address(rollback_slave->dev,
3702 (struct sockaddr *)&tmp_ss);
3704 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3705 tmp_res, rollback_slave->dev->name);
3713 * bond_xmit_slave_id - transmit skb through slave with slave_id
3714 * @bond: bonding device that is transmitting
3715 * @skb: buffer to transmit
3716 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3718 * This function tries to transmit through slave with slave_id but in case
3719 * it fails, it tries to find the first available slave for transmission.
3720 * The skb is consumed in all cases, thus the function is void.
3722 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3724 struct list_head *iter;
3725 struct slave *slave;
3728 /* Here we start from the slave with slave_id */
3729 bond_for_each_slave_rcu(bond, slave, iter) {
3731 if (bond_slave_can_tx(slave)) {
3732 bond_dev_queue_xmit(bond, skb, slave->dev);
3738 /* Here we start from the first slave up to slave_id */
3740 bond_for_each_slave_rcu(bond, slave, iter) {
3743 if (bond_slave_can_tx(slave)) {
3744 bond_dev_queue_xmit(bond, skb, slave->dev);
3748 /* no slave that can tx has been found */
3749 bond_tx_drop(bond->dev, skb);
3753 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3754 * @bond: bonding device to use
3756 * Based on the value of the bonding device's packets_per_slave parameter
3757 * this function generates a slave id, which is usually used as the next
3758 * slave to transmit through.
3760 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3763 struct reciprocal_value reciprocal_packets_per_slave;
3764 int packets_per_slave = bond->params.packets_per_slave;
3766 switch (packets_per_slave) {
3768 slave_id = prandom_u32();
3771 slave_id = bond->rr_tx_counter;
3774 reciprocal_packets_per_slave =
3775 bond->params.reciprocal_packets_per_slave;
3776 slave_id = reciprocal_divide(bond->rr_tx_counter,
3777 reciprocal_packets_per_slave);
3780 bond->rr_tx_counter++;
3785 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3787 struct bonding *bond = netdev_priv(bond_dev);
3788 struct iphdr *iph = ip_hdr(skb);
3789 struct slave *slave;
3792 /* Start with the curr_active_slave that joined the bond as the
3793 * default for sending IGMP traffic. For failover purposes one
3794 * needs to maintain some consistency for the interface that will
3795 * send the join/membership reports. The curr_active_slave found
3796 * will send all of this type of traffic.
3798 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3799 slave = rcu_dereference(bond->curr_active_slave);
3801 bond_dev_queue_xmit(bond, skb, slave->dev);
3803 bond_xmit_slave_id(bond, skb, 0);
3805 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3807 if (likely(slave_cnt)) {
3808 slave_id = bond_rr_gen_slave_id(bond);
3809 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3811 bond_tx_drop(bond_dev, skb);
3815 return NETDEV_TX_OK;
3818 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3819 * the bond has a usable interface.
3821 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3823 struct bonding *bond = netdev_priv(bond_dev);
3824 struct slave *slave;
3826 slave = rcu_dereference(bond->curr_active_slave);
3828 bond_dev_queue_xmit(bond, skb, slave->dev);
3830 bond_tx_drop(bond_dev, skb);
3832 return NETDEV_TX_OK;
3835 /* Use this to update slave_array when (a) it's not appropriate to update
3836 * slave_array right away (note that update_slave_array() may sleep)
3837 * and / or (b) RTNL is not held.
3839 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3841 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3844 /* Slave array work handler. Holds only RTNL */
3845 static void bond_slave_arr_handler(struct work_struct *work)
3847 struct bonding *bond = container_of(work, struct bonding,
3848 slave_arr_work.work);
3851 if (!rtnl_trylock())
3854 ret = bond_update_slave_arr(bond, NULL);
3857 pr_warn_ratelimited("Failed to update slave array from WT\n");
3863 bond_slave_arr_work_rearm(bond, 1);
3866 /* Build the usable slaves array in control path for modes that use xmit-hash
3867 * to determine the slave interface -
3868 * (a) BOND_MODE_8023AD
3870 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3872 * The caller is expected to hold RTNL only and NO other lock!
3874 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3876 struct slave *slave;
3877 struct list_head *iter;
3878 struct bond_up_slave *new_arr, *old_arr;
3882 #ifdef CONFIG_LOCKDEP
3883 WARN_ON(lockdep_is_held(&bond->mode_lock));
3886 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3890 pr_err("Failed to build slave-array.\n");
3893 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3894 struct ad_info ad_info;
3896 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3897 pr_debug("bond_3ad_get_active_agg_info failed\n");
3898 kfree_rcu(new_arr, rcu);
3899 /* No active aggragator means it's not safe to use
3900 * the previous array.
3902 old_arr = rtnl_dereference(bond->slave_arr);
3904 RCU_INIT_POINTER(bond->slave_arr, NULL);
3905 kfree_rcu(old_arr, rcu);
3909 agg_id = ad_info.aggregator_id;
3911 bond_for_each_slave(bond, slave, iter) {
3912 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3913 struct aggregator *agg;
3915 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3916 if (!agg || agg->aggregator_identifier != agg_id)
3919 if (!bond_slave_can_tx(slave))
3921 if (skipslave == slave)
3923 new_arr->arr[new_arr->count++] = slave;
3926 old_arr = rtnl_dereference(bond->slave_arr);
3927 rcu_assign_pointer(bond->slave_arr, new_arr);
3929 kfree_rcu(old_arr, rcu);
3931 if (ret != 0 && skipslave) {
3934 /* Rare situation where caller has asked to skip a specific
3935 * slave but allocation failed (most likely!). BTW this is
3936 * only possible when the call is initiated from
3937 * __bond_release_one(). In this situation; overwrite the
3938 * skipslave entry in the array with the last entry from the
3939 * array to avoid a situation where the xmit path may choose
3940 * this to-be-skipped slave to send a packet out.
3942 old_arr = rtnl_dereference(bond->slave_arr);
3943 for (idx = 0; idx < old_arr->count; idx++) {
3944 if (skipslave == old_arr->arr[idx]) {
3946 old_arr->arr[old_arr->count-1];
3955 /* Use this Xmit function for 3AD as well as XOR modes. The current
3956 * usable slave array is formed in the control path. The xmit function
3957 * just calculates hash and sends the packet out.
3959 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3961 struct bonding *bond = netdev_priv(dev);
3962 struct slave *slave;
3963 struct bond_up_slave *slaves;
3966 slaves = rcu_dereference(bond->slave_arr);
3967 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3968 if (likely(count)) {
3969 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3970 bond_dev_queue_xmit(bond, skb, slave->dev);
3972 bond_tx_drop(dev, skb);
3975 return NETDEV_TX_OK;
3978 /* in broadcast mode, we send everything to all usable interfaces. */
3979 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3981 struct bonding *bond = netdev_priv(bond_dev);
3982 struct slave *slave = NULL;
3983 struct list_head *iter;
3985 bond_for_each_slave_rcu(bond, slave, iter) {
3986 if (bond_is_last_slave(bond, slave))
3988 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3989 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3992 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3993 bond_dev->name, __func__);
3996 bond_dev_queue_xmit(bond, skb2, slave->dev);
3999 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4000 bond_dev_queue_xmit(bond, skb, slave->dev);
4002 bond_tx_drop(bond_dev, skb);
4004 return NETDEV_TX_OK;
4007 /*------------------------- Device initialization ---------------------------*/
4009 /* Lookup the slave that corresponds to a qid */
4010 static inline int bond_slave_override(struct bonding *bond,
4011 struct sk_buff *skb)
4013 struct slave *slave = NULL;
4014 struct list_head *iter;
4016 if (!skb->queue_mapping)
4019 /* Find out if any slaves have the same mapping as this skb. */
4020 bond_for_each_slave_rcu(bond, slave, iter) {
4021 if (slave->queue_id == skb->queue_mapping) {
4022 if (bond_slave_is_up(slave) &&
4023 slave->link == BOND_LINK_UP) {
4024 bond_dev_queue_xmit(bond, skb, slave->dev);
4027 /* If the slave isn't UP, use default transmit policy. */
4036 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4037 void *accel_priv, select_queue_fallback_t fallback)
4039 /* This helper function exists to help dev_pick_tx get the correct
4040 * destination queue. Using a helper function skips a call to
4041 * skb_tx_hash and will put the skbs in the queue we expect on their
4042 * way down to the bonding driver.
4044 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4046 /* Save the original txq to restore before passing to the driver */
4047 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4049 if (unlikely(txq >= dev->real_num_tx_queues)) {
4051 txq -= dev->real_num_tx_queues;
4052 } while (txq >= dev->real_num_tx_queues);
4057 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4059 struct bonding *bond = netdev_priv(dev);
4061 if (bond_should_override_tx_queue(bond) &&
4062 !bond_slave_override(bond, skb))
4063 return NETDEV_TX_OK;
4065 switch (BOND_MODE(bond)) {
4066 case BOND_MODE_ROUNDROBIN:
4067 return bond_xmit_roundrobin(skb, dev);
4068 case BOND_MODE_ACTIVEBACKUP:
4069 return bond_xmit_activebackup(skb, dev);
4070 case BOND_MODE_8023AD:
4072 return bond_3ad_xor_xmit(skb, dev);
4073 case BOND_MODE_BROADCAST:
4074 return bond_xmit_broadcast(skb, dev);
4076 return bond_alb_xmit(skb, dev);
4078 return bond_tlb_xmit(skb, dev);
4080 /* Should never happen, mode already checked */
4081 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4083 bond_tx_drop(dev, skb);
4084 return NETDEV_TX_OK;
4088 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4090 struct bonding *bond = netdev_priv(dev);
4091 netdev_tx_t ret = NETDEV_TX_OK;
4093 /* If we risk deadlock from transmitting this in the
4094 * netpoll path, tell netpoll to queue the frame for later tx
4096 if (unlikely(is_netpoll_tx_blocked(dev)))
4097 return NETDEV_TX_BUSY;
4100 if (bond_has_slaves(bond))
4101 ret = __bond_start_xmit(skb, dev);
4103 bond_tx_drop(dev, skb);
4109 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4110 struct ethtool_link_ksettings *cmd)
4112 struct bonding *bond = netdev_priv(bond_dev);
4113 unsigned long speed = 0;
4114 struct list_head *iter;
4115 struct slave *slave;
4117 cmd->base.duplex = DUPLEX_UNKNOWN;
4118 cmd->base.port = PORT_OTHER;
4120 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4121 * do not need to check mode. Though link speed might not represent
4122 * the true receive or transmit bandwidth (not all modes are symmetric)
4123 * this is an accurate maximum.
4125 bond_for_each_slave(bond, slave, iter) {
4126 if (bond_slave_can_tx(slave)) {
4127 if (slave->speed != SPEED_UNKNOWN)
4128 speed += slave->speed;
4129 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4130 slave->duplex != DUPLEX_UNKNOWN)
4131 cmd->base.duplex = slave->duplex;
4134 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4139 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4140 struct ethtool_drvinfo *drvinfo)
4142 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4143 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4144 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4148 static const struct ethtool_ops bond_ethtool_ops = {
4149 .get_drvinfo = bond_ethtool_get_drvinfo,
4150 .get_link = ethtool_op_get_link,
4151 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4154 static const struct net_device_ops bond_netdev_ops = {
4155 .ndo_init = bond_init,
4156 .ndo_uninit = bond_uninit,
4157 .ndo_open = bond_open,
4158 .ndo_stop = bond_close,
4159 .ndo_start_xmit = bond_start_xmit,
4160 .ndo_select_queue = bond_select_queue,
4161 .ndo_get_stats64 = bond_get_stats,
4162 .ndo_do_ioctl = bond_do_ioctl,
4163 .ndo_change_rx_flags = bond_change_rx_flags,
4164 .ndo_set_rx_mode = bond_set_rx_mode,
4165 .ndo_change_mtu = bond_change_mtu,
4166 .ndo_set_mac_address = bond_set_mac_address,
4167 .ndo_neigh_setup = bond_neigh_setup,
4168 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4169 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4170 #ifdef CONFIG_NET_POLL_CONTROLLER
4171 .ndo_netpoll_setup = bond_netpoll_setup,
4172 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4173 .ndo_poll_controller = bond_poll_controller,
4175 .ndo_add_slave = bond_enslave,
4176 .ndo_del_slave = bond_release,
4177 .ndo_fix_features = bond_fix_features,
4178 .ndo_features_check = passthru_features_check,
4181 static const struct device_type bond_type = {
4185 static void bond_destructor(struct net_device *bond_dev)
4187 struct bonding *bond = netdev_priv(bond_dev);
4189 destroy_workqueue(bond->wq);
4192 void bond_setup(struct net_device *bond_dev)
4194 struct bonding *bond = netdev_priv(bond_dev);
4196 spin_lock_init(&bond->mode_lock);
4197 spin_lock_init(&bond->stats_lock);
4198 bond->params = bonding_defaults;
4200 /* Initialize pointers */
4201 bond->dev = bond_dev;
4203 /* Initialize the device entry points */
4204 ether_setup(bond_dev);
4205 bond_dev->max_mtu = ETH_MAX_MTU;
4206 bond_dev->netdev_ops = &bond_netdev_ops;
4207 bond_dev->ethtool_ops = &bond_ethtool_ops;
4209 bond_dev->needs_free_netdev = true;
4210 bond_dev->priv_destructor = bond_destructor;
4212 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4214 /* Initialize the device options */
4215 bond_dev->flags |= IFF_MASTER;
4216 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4217 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4219 /* don't acquire bond device's netif_tx_lock when transmitting */
4220 bond_dev->features |= NETIF_F_LLTX;
4222 /* By default, we declare the bond to be fully
4223 * VLAN hardware accelerated capable. Special
4224 * care is taken in the various xmit functions
4225 * when there are slaves that are not hw accel
4229 /* Don't allow bond devices to change network namespaces. */
4230 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4232 bond_dev->hw_features = BOND_VLAN_FEATURES |
4233 NETIF_F_HW_VLAN_CTAG_TX |
4234 NETIF_F_HW_VLAN_CTAG_RX |
4235 NETIF_F_HW_VLAN_CTAG_FILTER;
4237 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4238 bond_dev->features |= bond_dev->hw_features;
4241 /* Destroy a bonding device.
4242 * Must be under rtnl_lock when this function is called.
4244 static void bond_uninit(struct net_device *bond_dev)
4246 struct bonding *bond = netdev_priv(bond_dev);
4247 struct list_head *iter;
4248 struct slave *slave;
4249 struct bond_up_slave *arr;
4251 bond_netpoll_cleanup(bond_dev);
4253 /* Release the bonded slaves */
4254 bond_for_each_slave(bond, slave, iter)
4255 __bond_release_one(bond_dev, slave->dev, true);
4256 netdev_info(bond_dev, "Released all slaves\n");
4258 arr = rtnl_dereference(bond->slave_arr);
4260 RCU_INIT_POINTER(bond->slave_arr, NULL);
4261 kfree_rcu(arr, rcu);
4264 list_del(&bond->bond_list);
4266 bond_debug_unregister(bond);
4269 /*------------------------- Module initialization ---------------------------*/
4271 static int bond_check_params(struct bond_params *params)
4273 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4274 struct bond_opt_value newval;
4275 const struct bond_opt_value *valptr;
4276 int arp_all_targets_value = 0;
4277 u16 ad_actor_sys_prio = 0;
4278 u16 ad_user_port_key = 0;
4279 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4281 int bond_mode = BOND_MODE_ROUNDROBIN;
4282 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4284 int tlb_dynamic_lb = 0;
4286 /* Convert string parameters. */
4288 bond_opt_initstr(&newval, mode);
4289 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4291 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4294 bond_mode = valptr->value;
4297 if (xmit_hash_policy) {
4298 if ((bond_mode != BOND_MODE_XOR) &&
4299 (bond_mode != BOND_MODE_8023AD) &&
4300 (bond_mode != BOND_MODE_TLB)) {
4301 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4302 bond_mode_name(bond_mode));
4304 bond_opt_initstr(&newval, xmit_hash_policy);
4305 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4308 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4312 xmit_hashtype = valptr->value;
4317 if (bond_mode != BOND_MODE_8023AD) {
4318 pr_info("lacp_rate param is irrelevant in mode %s\n",
4319 bond_mode_name(bond_mode));
4321 bond_opt_initstr(&newval, lacp_rate);
4322 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4325 pr_err("Error: Invalid lacp rate \"%s\"\n",
4329 lacp_fast = valptr->value;
4334 bond_opt_initstr(&newval, ad_select);
4335 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4338 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4341 params->ad_select = valptr->value;
4342 if (bond_mode != BOND_MODE_8023AD)
4343 pr_warn("ad_select param only affects 802.3ad mode\n");
4345 params->ad_select = BOND_AD_STABLE;
4348 if (max_bonds < 0) {
4349 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4350 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4351 max_bonds = BOND_DEFAULT_MAX_BONDS;
4355 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4361 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4366 if (downdelay < 0) {
4367 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4368 downdelay, INT_MAX);
4372 if ((use_carrier != 0) && (use_carrier != 1)) {
4373 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4378 if (num_peer_notif < 0 || num_peer_notif > 255) {
4379 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4384 /* reset values for 802.3ad/TLB/ALB */
4385 if (!bond_mode_uses_arp(bond_mode)) {
4387 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");
4388 pr_warn("Forcing miimon to 100msec\n");
4389 miimon = BOND_DEFAULT_MIIMON;
4393 if (tx_queues < 1 || tx_queues > 255) {
4394 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4395 tx_queues, BOND_DEFAULT_TX_QUEUES);
4396 tx_queues = BOND_DEFAULT_TX_QUEUES;
4399 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4400 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4402 all_slaves_active = 0;
4405 if (resend_igmp < 0 || resend_igmp > 255) {
4406 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4407 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4408 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4411 bond_opt_initval(&newval, packets_per_slave);
4412 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4413 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4414 packets_per_slave, USHRT_MAX);
4415 packets_per_slave = 1;
4418 if (bond_mode == BOND_MODE_ALB) {
4419 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",
4424 if (updelay || downdelay) {
4425 /* just warn the user the up/down delay will have
4426 * no effect since miimon is zero...
4428 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",
4429 updelay, downdelay);
4432 /* don't allow arp monitoring */
4434 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4435 miimon, arp_interval);
4439 if ((updelay % miimon) != 0) {
4440 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4441 updelay, miimon, (updelay / miimon) * miimon);
4446 if ((downdelay % miimon) != 0) {
4447 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4449 (downdelay / miimon) * miimon);
4452 downdelay /= miimon;
4455 if (arp_interval < 0) {
4456 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4457 arp_interval, INT_MAX);
4461 for (arp_ip_count = 0, i = 0;
4462 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4465 /* not a complete check, but good enough to catch mistakes */
4466 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4467 !bond_is_ip_target_ok(ip)) {
4468 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4472 if (bond_get_targets_ip(arp_target, ip) == -1)
4473 arp_target[arp_ip_count++] = ip;
4475 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4480 if (arp_interval && !arp_ip_count) {
4481 /* don't allow arping if no arp_ip_target given... */
4482 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4488 if (!arp_interval) {
4489 pr_err("arp_validate requires arp_interval\n");
4493 bond_opt_initstr(&newval, arp_validate);
4494 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4497 pr_err("Error: invalid arp_validate \"%s\"\n",
4501 arp_validate_value = valptr->value;
4503 arp_validate_value = 0;
4506 if (arp_all_targets) {
4507 bond_opt_initstr(&newval, arp_all_targets);
4508 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4511 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4513 arp_all_targets_value = 0;
4515 arp_all_targets_value = valptr->value;
4520 pr_info("MII link monitoring set to %d ms\n", miimon);
4521 } else if (arp_interval) {
4522 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4523 arp_validate_value);
4524 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4525 arp_interval, valptr->string, arp_ip_count);
4527 for (i = 0; i < arp_ip_count; i++)
4528 pr_cont(" %s", arp_ip_target[i]);
4532 } else if (max_bonds) {
4533 /* miimon and arp_interval not set, we need one so things
4534 * work as expected, see bonding.txt for details
4536 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");
4539 if (primary && !bond_mode_uses_primary(bond_mode)) {
4540 /* currently, using a primary only makes sense
4541 * in active backup, TLB or ALB modes
4543 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4544 primary, bond_mode_name(bond_mode));
4548 if (primary && primary_reselect) {
4549 bond_opt_initstr(&newval, primary_reselect);
4550 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4553 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4557 primary_reselect_value = valptr->value;
4559 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4562 if (fail_over_mac) {
4563 bond_opt_initstr(&newval, fail_over_mac);
4564 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4567 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4571 fail_over_mac_value = valptr->value;
4572 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4573 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4575 fail_over_mac_value = BOND_FOM_NONE;
4578 bond_opt_initstr(&newval, "default");
4579 valptr = bond_opt_parse(
4580 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4583 pr_err("Error: No ad_actor_sys_prio default value");
4586 ad_actor_sys_prio = valptr->value;
4588 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4591 pr_err("Error: No ad_user_port_key default value");
4594 ad_user_port_key = valptr->value;
4596 if (bond_mode == BOND_MODE_TLB) {
4597 bond_opt_initstr(&newval, "default");
4598 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB),
4601 pr_err("Error: No tlb_dynamic_lb default value");
4604 tlb_dynamic_lb = valptr->value;
4607 if (lp_interval == 0) {
4608 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4609 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4610 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4613 /* fill params struct with the proper values */
4614 params->mode = bond_mode;
4615 params->xmit_policy = xmit_hashtype;
4616 params->miimon = miimon;
4617 params->num_peer_notif = num_peer_notif;
4618 params->arp_interval = arp_interval;
4619 params->arp_validate = arp_validate_value;
4620 params->arp_all_targets = arp_all_targets_value;
4621 params->updelay = updelay;
4622 params->downdelay = downdelay;
4623 params->use_carrier = use_carrier;
4624 params->lacp_fast = lacp_fast;
4625 params->primary[0] = 0;
4626 params->primary_reselect = primary_reselect_value;
4627 params->fail_over_mac = fail_over_mac_value;
4628 params->tx_queues = tx_queues;
4629 params->all_slaves_active = all_slaves_active;
4630 params->resend_igmp = resend_igmp;
4631 params->min_links = min_links;
4632 params->lp_interval = lp_interval;
4633 params->packets_per_slave = packets_per_slave;
4634 params->tlb_dynamic_lb = tlb_dynamic_lb;
4635 params->ad_actor_sys_prio = ad_actor_sys_prio;
4636 eth_zero_addr(params->ad_actor_system);
4637 params->ad_user_port_key = ad_user_port_key;
4638 if (packets_per_slave > 0) {
4639 params->reciprocal_packets_per_slave =
4640 reciprocal_value(packets_per_slave);
4642 /* reciprocal_packets_per_slave is unused if
4643 * packets_per_slave is 0 or 1, just initialize it
4645 params->reciprocal_packets_per_slave =
4646 (struct reciprocal_value) { 0 };
4650 strncpy(params->primary, primary, IFNAMSIZ);
4651 params->primary[IFNAMSIZ - 1] = 0;
4654 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4659 /* Called from registration process */
4660 static int bond_init(struct net_device *bond_dev)
4662 struct bonding *bond = netdev_priv(bond_dev);
4663 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4665 netdev_dbg(bond_dev, "Begin bond_init\n");
4667 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4671 netdev_lockdep_set_classes(bond_dev);
4673 list_add_tail(&bond->bond_list, &bn->dev_list);
4675 bond_prepare_sysfs_group(bond);
4677 bond_debug_register(bond);
4679 /* Ensure valid dev_addr */
4680 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4681 bond_dev->addr_assign_type == NET_ADDR_PERM)
4682 eth_hw_addr_random(bond_dev);
4687 unsigned int bond_get_num_tx_queues(void)
4692 /* Create a new bond based on the specified name and bonding parameters.
4693 * If name is NULL, obtain a suitable "bond%d" name for us.
4694 * Caller must NOT hold rtnl_lock; we need to release it here before we
4695 * set up our sysfs entries.
4697 int bond_create(struct net *net, const char *name)
4699 struct net_device *bond_dev;
4700 struct bonding *bond;
4701 struct alb_bond_info *bond_info;
4706 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4707 name ? name : "bond%d", NET_NAME_UNKNOWN,
4708 bond_setup, tx_queues);
4710 pr_err("%s: eek! can't alloc netdev!\n", name);
4716 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4717 * It is set to 0 by default which is wrong.
4719 bond = netdev_priv(bond_dev);
4720 bond_info = &(BOND_ALB_INFO(bond));
4721 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4723 dev_net_set(bond_dev, net);
4724 bond_dev->rtnl_link_ops = &bond_link_ops;
4726 res = register_netdevice(bond_dev);
4728 netif_carrier_off(bond_dev);
4730 bond_work_init_all(bond);
4734 free_netdev(bond_dev);
4738 static int __net_init bond_net_init(struct net *net)
4740 struct bond_net *bn = net_generic(net, bond_net_id);
4743 INIT_LIST_HEAD(&bn->dev_list);
4745 bond_create_proc_dir(bn);
4746 bond_create_sysfs(bn);
4751 static void __net_exit bond_net_exit(struct net *net)
4753 struct bond_net *bn = net_generic(net, bond_net_id);
4754 struct bonding *bond, *tmp_bond;
4757 bond_destroy_sysfs(bn);
4759 /* Kill off any bonds created after unregistering bond rtnl ops */
4761 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4762 unregister_netdevice_queue(bond->dev, &list);
4763 unregister_netdevice_many(&list);
4766 bond_destroy_proc_dir(bn);
4769 static struct pernet_operations bond_net_ops = {
4770 .init = bond_net_init,
4771 .exit = bond_net_exit,
4773 .size = sizeof(struct bond_net),
4776 static int __init bonding_init(void)
4781 pr_info("%s", bond_version);
4783 res = bond_check_params(&bonding_defaults);
4787 res = register_pernet_subsys(&bond_net_ops);
4791 res = bond_netlink_init();
4795 bond_create_debugfs();
4797 for (i = 0; i < max_bonds; i++) {
4798 res = bond_create(&init_net, NULL);
4803 register_netdevice_notifier(&bond_netdev_notifier);
4807 bond_destroy_debugfs();
4808 bond_netlink_fini();
4810 unregister_pernet_subsys(&bond_net_ops);
4815 static void __exit bonding_exit(void)
4817 unregister_netdevice_notifier(&bond_netdev_notifier);
4819 bond_destroy_debugfs();
4821 bond_netlink_fini();
4822 unregister_pernet_subsys(&bond_net_ops);
4824 #ifdef CONFIG_NET_POLL_CONTROLLER
4825 /* Make sure we don't have an imbalance on our netpoll blocking */
4826 WARN_ON(atomic_read(&netpoll_block_tx));
4830 module_init(bonding_init);
4831 module_exit(bonding_exit);
4832 MODULE_LICENSE("GPL");
4833 MODULE_VERSION(DRV_VERSION);
4834 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4835 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");