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, "803.ad 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 int bond_net_id __read_mostly;
204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
205 static int arp_ip_count;
206 static int bond_mode = BOND_MODE_ROUNDROBIN;
207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
208 static int lacp_fast;
210 /*-------------------------- Forward declarations ---------------------------*/
212 static int bond_init(struct net_device *bond_dev);
213 static void bond_uninit(struct net_device *bond_dev);
214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
215 struct rtnl_link_stats64 *stats);
216 static void bond_slave_arr_handler(struct work_struct *work);
218 /*---------------------------- General routines -----------------------------*/
220 const char *bond_mode_name(int mode)
222 static const char *names[] = {
223 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
224 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
225 [BOND_MODE_XOR] = "load balancing (xor)",
226 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
227 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
228 [BOND_MODE_TLB] = "transmit load balancing",
229 [BOND_MODE_ALB] = "adaptive load balancing",
232 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
238 /*---------------------------------- VLAN -----------------------------------*/
241 * bond_dev_queue_xmit - Prepare skb for xmit.
243 * @bond: bond device that got this skb for tx.
244 * @skb: hw accel VLAN tagged skb to transmit
245 * @slave_dev: slave that is supposed to xmit this skbuff
247 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
248 struct net_device *slave_dev)
250 skb->dev = slave_dev;
252 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
253 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
254 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
256 if (unlikely(netpoll_tx_running(bond->dev)))
257 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
262 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
263 * We don't protect the slave list iteration with a lock because:
264 * a. This operation is performed in IOCTL context,
265 * b. The operation is protected by the RTNL semaphore in the 8021q code,
266 * c. Holding a lock with BH disabled while directly calling a base driver
267 * entry point is generally a BAD idea.
269 * The design of synchronization/protection for this operation in the 8021q
270 * module is good for one or more VLAN devices over a single physical device
271 * and cannot be extended for a teaming solution like bonding, so there is a
272 * potential race condition here where a net device from the vlan group might
273 * be referenced (either by a base driver or the 8021q code) while it is being
274 * removed from the system. However, it turns out we're not making matters
275 * worse, and if it works for regular VLAN usage it will work here too.
279 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
280 * @bond_dev: bonding net device that got called
281 * @vid: vlan id being added
283 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
284 __be16 proto, u16 vid)
286 struct bonding *bond = netdev_priv(bond_dev);
287 struct slave *slave, *rollback_slave;
288 struct list_head *iter;
291 bond_for_each_slave(bond, slave, iter) {
292 res = vlan_vid_add(slave->dev, proto, vid);
300 /* unwind to the slave that failed */
301 bond_for_each_slave(bond, rollback_slave, iter) {
302 if (rollback_slave == slave)
305 vlan_vid_del(rollback_slave->dev, proto, vid);
312 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
313 * @bond_dev: bonding net device that got called
314 * @vid: vlan id being removed
316 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
317 __be16 proto, u16 vid)
319 struct bonding *bond = netdev_priv(bond_dev);
320 struct list_head *iter;
323 bond_for_each_slave(bond, slave, iter)
324 vlan_vid_del(slave->dev, proto, vid);
326 if (bond_is_lb(bond))
327 bond_alb_clear_vlan(bond, vid);
332 /*------------------------------- Link status -------------------------------*/
334 /* Set the carrier state for the master according to the state of its
335 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
336 * do special 802.3ad magic.
338 * Returns zero if carrier state does not change, nonzero if it does.
340 int bond_set_carrier(struct bonding *bond)
342 struct list_head *iter;
345 if (!bond_has_slaves(bond))
348 if (BOND_MODE(bond) == BOND_MODE_8023AD)
349 return bond_3ad_set_carrier(bond);
351 bond_for_each_slave(bond, slave, iter) {
352 if (slave->link == BOND_LINK_UP) {
353 if (!netif_carrier_ok(bond->dev)) {
354 netif_carrier_on(bond->dev);
362 if (netif_carrier_ok(bond->dev)) {
363 netif_carrier_off(bond->dev);
369 /* Get link speed and duplex from the slave's base driver
370 * using ethtool. If for some reason the call fails or the
371 * values are invalid, set speed and duplex to -1,
374 static void bond_update_speed_duplex(struct slave *slave)
376 struct net_device *slave_dev = slave->dev;
377 struct ethtool_cmd ecmd;
381 slave->speed = SPEED_UNKNOWN;
382 slave->duplex = DUPLEX_UNKNOWN;
384 res = __ethtool_get_settings(slave_dev, &ecmd);
388 slave_speed = ethtool_cmd_speed(&ecmd);
389 if (slave_speed == 0 || slave_speed == ((__u32) -1))
392 switch (ecmd.duplex) {
400 slave->speed = slave_speed;
401 slave->duplex = ecmd.duplex;
406 const char *bond_slave_link_status(s8 link)
422 /* if <dev> supports MII link status reporting, check its link status.
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
425 * depending upon the setting of the use_carrier parameter.
427 * Return either BMSR_LSTATUS, meaning that the link is up (or we
428 * can't tell and just pretend it is), or 0, meaning that the link is
431 * If reporting is non-zero, instead of faking link up, return -1 if
432 * both ETHTOOL and MII ioctls fail (meaning the device does not
433 * support them). If use_carrier is set, return whatever it says.
434 * It'd be nice if there was a good way to tell if a driver supports
435 * netif_carrier, but there really isn't.
437 static int bond_check_dev_link(struct bonding *bond,
438 struct net_device *slave_dev, int reporting)
440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441 int (*ioctl)(struct net_device *, struct ifreq *, int);
443 struct mii_ioctl_data *mii;
445 if (!reporting && !netif_running(slave_dev))
448 if (bond->params.use_carrier)
449 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
451 /* Try to get link status using Ethtool first. */
452 if (slave_dev->ethtool_ops->get_link)
453 return slave_dev->ethtool_ops->get_link(slave_dev) ?
456 /* Ethtool can't be used, fallback to MII ioctls. */
457 ioctl = slave_ops->ndo_do_ioctl;
459 /* TODO: set pointer to correct ioctl on a per team member
460 * bases to make this more efficient. that is, once
461 * we determine the correct ioctl, we will always
462 * call it and not the others for that team
466 /* We cannot assume that SIOCGMIIPHY will also read a
467 * register; not all network drivers (e.g., e100)
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
474 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
475 mii->reg_num = MII_BMSR;
476 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
477 return mii->val_out & BMSR_LSTATUS;
481 /* If reporting, report that either there's no dev->do_ioctl,
482 * or both SIOCGMIIREG and get_link failed (meaning that we
483 * cannot report link status). If not reporting, pretend
486 return reporting ? -1 : BMSR_LSTATUS;
489 /*----------------------------- Multicast list ------------------------------*/
491 /* Push the promiscuity flag down to appropriate slaves */
492 static int bond_set_promiscuity(struct bonding *bond, int inc)
494 struct list_head *iter;
497 if (bond_uses_primary(bond)) {
498 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
501 err = dev_set_promiscuity(curr_active->dev, inc);
505 bond_for_each_slave(bond, slave, iter) {
506 err = dev_set_promiscuity(slave->dev, inc);
514 /* Push the allmulti flag down to all slaves */
515 static int bond_set_allmulti(struct bonding *bond, int inc)
517 struct list_head *iter;
520 if (bond_uses_primary(bond)) {
521 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
524 err = dev_set_allmulti(curr_active->dev, inc);
528 bond_for_each_slave(bond, slave, iter) {
529 err = dev_set_allmulti(slave->dev, inc);
537 /* Retrieve the list of registered multicast addresses for the bonding
538 * device and retransmit an IGMP JOIN request to the current active
541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
543 struct bonding *bond = container_of(work, struct bonding,
546 if (!rtnl_trylock()) {
547 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
550 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
552 if (bond->igmp_retrans > 1) {
553 bond->igmp_retrans--;
554 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
559 /* Flush bond's hardware addresses from slave */
560 static void bond_hw_addr_flush(struct net_device *bond_dev,
561 struct net_device *slave_dev)
563 struct bonding *bond = netdev_priv(bond_dev);
565 dev_uc_unsync(slave_dev, bond_dev);
566 dev_mc_unsync(slave_dev, bond_dev);
568 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
569 /* del lacpdu mc addr from mc list */
570 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
572 dev_mc_del(slave_dev, lacpdu_multicast);
576 /*--------------------------- Active slave change ---------------------------*/
578 /* Update the hardware address list and promisc/allmulti for the new and
579 * old active slaves (if any). Modes that are not using primary keep all
580 * slaves up date at all times; only the modes that use primary need to call
581 * this function to swap these settings during a failover.
583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
584 struct slave *old_active)
587 if (bond->dev->flags & IFF_PROMISC)
588 dev_set_promiscuity(old_active->dev, -1);
590 if (bond->dev->flags & IFF_ALLMULTI)
591 dev_set_allmulti(old_active->dev, -1);
593 bond_hw_addr_flush(bond->dev, old_active->dev);
597 /* FIXME: Signal errors upstream. */
598 if (bond->dev->flags & IFF_PROMISC)
599 dev_set_promiscuity(new_active->dev, 1);
601 if (bond->dev->flags & IFF_ALLMULTI)
602 dev_set_allmulti(new_active->dev, 1);
604 netif_addr_lock_bh(bond->dev);
605 dev_uc_sync(new_active->dev, bond->dev);
606 dev_mc_sync(new_active->dev, bond->dev);
607 netif_addr_unlock_bh(bond->dev);
612 * bond_set_dev_addr - clone slave's address to bond
613 * @bond_dev: bond net device
614 * @slave_dev: slave net device
616 * Should be called with RTNL held.
618 static void bond_set_dev_addr(struct net_device *bond_dev,
619 struct net_device *slave_dev)
621 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
622 bond_dev, slave_dev, slave_dev->addr_len);
623 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
624 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
625 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
628 /* bond_do_fail_over_mac
630 * Perform special MAC address swapping for fail_over_mac settings
634 static void bond_do_fail_over_mac(struct bonding *bond,
635 struct slave *new_active,
636 struct slave *old_active)
638 u8 tmp_mac[ETH_ALEN];
639 struct sockaddr saddr;
642 switch (bond->params.fail_over_mac) {
643 case BOND_FOM_ACTIVE:
645 bond_set_dev_addr(bond->dev, new_active->dev);
647 case BOND_FOM_FOLLOW:
648 /* if new_active && old_active, swap them
649 * if just old_active, do nothing (going to no active slave)
650 * if just new_active, set new_active to bond's MAC
656 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
657 ether_addr_copy(saddr.sa_data,
658 old_active->dev->dev_addr);
659 saddr.sa_family = new_active->dev->type;
661 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
662 saddr.sa_family = bond->dev->type;
665 rv = dev_set_mac_address(new_active->dev, &saddr);
667 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
668 -rv, new_active->dev->name);
675 ether_addr_copy(saddr.sa_data, tmp_mac);
676 saddr.sa_family = old_active->dev->type;
678 rv = dev_set_mac_address(old_active->dev, &saddr);
680 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
681 -rv, new_active->dev->name);
685 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
686 bond->params.fail_over_mac);
692 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
694 struct slave *prim = rtnl_dereference(bond->primary_slave);
695 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
697 if (!prim || prim->link != BOND_LINK_UP) {
698 if (!curr || curr->link != BOND_LINK_UP)
703 if (bond->force_primary) {
704 bond->force_primary = false;
708 if (!curr || curr->link != BOND_LINK_UP)
711 /* At this point, prim and curr are both up */
712 switch (bond->params.primary_reselect) {
713 case BOND_PRI_RESELECT_ALWAYS:
715 case BOND_PRI_RESELECT_BETTER:
716 if (prim->speed < curr->speed)
718 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
721 case BOND_PRI_RESELECT_FAILURE:
724 netdev_err(bond->dev, "impossible primary_reselect %d\n",
725 bond->params.primary_reselect);
731 * bond_find_best_slave - select the best available slave to be the active one
732 * @bond: our bonding struct
734 static struct slave *bond_find_best_slave(struct bonding *bond)
736 struct slave *slave, *bestslave = NULL;
737 struct list_head *iter;
738 int mintime = bond->params.updelay;
740 slave = bond_choose_primary_or_current(bond);
744 bond_for_each_slave(bond, slave, iter) {
745 if (slave->link == BOND_LINK_UP)
747 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
748 slave->delay < mintime) {
749 mintime = slave->delay;
757 static bool bond_should_notify_peers(struct bonding *bond)
762 slave = rcu_dereference(bond->curr_active_slave);
765 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
766 slave ? slave->dev->name : "NULL");
768 if (!slave || !bond->send_peer_notif ||
769 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
776 * change_active_interface - change the active slave into the specified one
777 * @bond: our bonding struct
778 * @new: the new slave to make the active one
780 * Set the new slave to the bond's settings and unset them on the old
782 * Setting include flags, mc-list, promiscuity, allmulti, etc.
784 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
785 * because it is apparently the best available slave we have, even though its
786 * updelay hasn't timed out yet.
788 * Caller must hold RTNL.
790 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
792 struct slave *old_active;
796 old_active = rtnl_dereference(bond->curr_active_slave);
798 if (old_active == new_active)
802 new_active->last_link_up = jiffies;
804 if (new_active->link == BOND_LINK_BACK) {
805 if (bond_uses_primary(bond)) {
806 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
807 new_active->dev->name,
808 (bond->params.updelay - new_active->delay) * bond->params.miimon);
811 new_active->delay = 0;
812 bond_set_slave_link_state(new_active, BOND_LINK_UP);
814 if (BOND_MODE(bond) == BOND_MODE_8023AD)
815 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
817 if (bond_is_lb(bond))
818 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
820 if (bond_uses_primary(bond)) {
821 netdev_info(bond->dev, "making interface %s the new active one\n",
822 new_active->dev->name);
827 if (bond_uses_primary(bond))
828 bond_hw_addr_swap(bond, new_active, old_active);
830 if (bond_is_lb(bond)) {
831 bond_alb_handle_active_change(bond, new_active);
833 bond_set_slave_inactive_flags(old_active,
834 BOND_SLAVE_NOTIFY_NOW);
836 bond_set_slave_active_flags(new_active,
837 BOND_SLAVE_NOTIFY_NOW);
839 rcu_assign_pointer(bond->curr_active_slave, new_active);
842 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
844 bond_set_slave_inactive_flags(old_active,
845 BOND_SLAVE_NOTIFY_NOW);
848 bool should_notify_peers = false;
850 bond_set_slave_active_flags(new_active,
851 BOND_SLAVE_NOTIFY_NOW);
853 if (bond->params.fail_over_mac)
854 bond_do_fail_over_mac(bond, new_active,
857 if (netif_running(bond->dev)) {
858 bond->send_peer_notif =
859 bond->params.num_peer_notif;
860 should_notify_peers =
861 bond_should_notify_peers(bond);
864 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
865 if (should_notify_peers)
866 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
871 /* resend IGMP joins since active slave has changed or
872 * all were sent on curr_active_slave.
873 * resend only if bond is brought up with the affected
874 * bonding modes and the retransmission is enabled
876 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
877 ((bond_uses_primary(bond) && new_active) ||
878 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
879 bond->igmp_retrans = bond->params.resend_igmp;
880 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
885 * bond_select_active_slave - select a new active slave, if needed
886 * @bond: our bonding struct
888 * This functions should be called when one of the following occurs:
889 * - The old curr_active_slave has been released or lost its link.
890 * - The primary_slave has got its link back.
891 * - A slave has got its link back and there's no old curr_active_slave.
893 * Caller must hold RTNL.
895 void bond_select_active_slave(struct bonding *bond)
897 struct slave *best_slave;
902 best_slave = bond_find_best_slave(bond);
903 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
904 bond_change_active_slave(bond, best_slave);
905 rv = bond_set_carrier(bond);
909 if (netif_carrier_ok(bond->dev)) {
910 netdev_info(bond->dev, "first active interface up!\n");
912 netdev_info(bond->dev, "now running without any active interface!\n");
917 #ifdef CONFIG_NET_POLL_CONTROLLER
918 static inline int slave_enable_netpoll(struct slave *slave)
923 np = kzalloc(sizeof(*np), GFP_KERNEL);
928 err = __netpoll_setup(np, slave->dev);
937 static inline void slave_disable_netpoll(struct slave *slave)
939 struct netpoll *np = slave->np;
945 __netpoll_free_async(np);
948 static void bond_poll_controller(struct net_device *bond_dev)
950 struct bonding *bond = netdev_priv(bond_dev);
951 struct slave *slave = NULL;
952 struct list_head *iter;
953 struct ad_info ad_info;
954 struct netpoll_info *ni;
955 const struct net_device_ops *ops;
957 if (BOND_MODE(bond) == BOND_MODE_8023AD)
958 if (bond_3ad_get_active_agg_info(bond, &ad_info))
962 bond_for_each_slave_rcu(bond, slave, iter) {
963 ops = slave->dev->netdev_ops;
964 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
967 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
968 struct aggregator *agg =
969 SLAVE_AD_INFO(slave)->port.aggregator;
972 agg->aggregator_identifier != ad_info.aggregator_id)
976 ni = rcu_dereference_bh(slave->dev->npinfo);
977 if (down_trylock(&ni->dev_lock))
979 ops->ndo_poll_controller(slave->dev);
982 rcu_read_unlock_bh();
985 static void bond_netpoll_cleanup(struct net_device *bond_dev)
987 struct bonding *bond = netdev_priv(bond_dev);
988 struct list_head *iter;
991 bond_for_each_slave(bond, slave, iter)
992 if (bond_slave_is_up(slave))
993 slave_disable_netpoll(slave);
996 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
998 struct bonding *bond = netdev_priv(dev);
999 struct list_head *iter;
1000 struct slave *slave;
1003 bond_for_each_slave(bond, slave, iter) {
1004 err = slave_enable_netpoll(slave);
1006 bond_netpoll_cleanup(dev);
1013 static inline int slave_enable_netpoll(struct slave *slave)
1017 static inline void slave_disable_netpoll(struct slave *slave)
1020 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1025 /*---------------------------------- IOCTL ----------------------------------*/
1027 static netdev_features_t bond_fix_features(struct net_device *dev,
1028 netdev_features_t features)
1030 struct bonding *bond = netdev_priv(dev);
1031 struct list_head *iter;
1032 netdev_features_t mask;
1033 struct slave *slave;
1037 features &= ~NETIF_F_ONE_FOR_ALL;
1038 features |= NETIF_F_ALL_FOR_ALL;
1040 bond_for_each_slave(bond, slave, iter) {
1041 features = netdev_increment_features(features,
1042 slave->dev->features,
1045 features = netdev_add_tso_features(features, mask);
1050 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1051 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1052 NETIF_F_HIGHDMA | NETIF_F_LRO)
1054 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1057 static void bond_compute_features(struct bonding *bond)
1059 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1060 IFF_XMIT_DST_RELEASE_PERM;
1061 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1062 netdev_features_t enc_features = BOND_ENC_FEATURES;
1063 struct net_device *bond_dev = bond->dev;
1064 struct list_head *iter;
1065 struct slave *slave;
1066 unsigned short max_hard_header_len = ETH_HLEN;
1067 unsigned int gso_max_size = GSO_MAX_SIZE;
1068 u16 gso_max_segs = GSO_MAX_SEGS;
1070 if (!bond_has_slaves(bond))
1072 vlan_features &= NETIF_F_ALL_FOR_ALL;
1074 bond_for_each_slave(bond, slave, iter) {
1075 vlan_features = netdev_increment_features(vlan_features,
1076 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1078 enc_features = netdev_increment_features(enc_features,
1079 slave->dev->hw_enc_features,
1081 dst_release_flag &= slave->dev->priv_flags;
1082 if (slave->dev->hard_header_len > max_hard_header_len)
1083 max_hard_header_len = slave->dev->hard_header_len;
1085 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1086 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1090 bond_dev->vlan_features = vlan_features;
1091 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1092 bond_dev->hard_header_len = max_hard_header_len;
1093 bond_dev->gso_max_segs = gso_max_segs;
1094 netif_set_gso_max_size(bond_dev, gso_max_size);
1096 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1097 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1098 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1099 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1101 netdev_change_features(bond_dev);
1104 static void bond_setup_by_slave(struct net_device *bond_dev,
1105 struct net_device *slave_dev)
1107 bond_dev->header_ops = slave_dev->header_ops;
1109 bond_dev->type = slave_dev->type;
1110 bond_dev->hard_header_len = slave_dev->hard_header_len;
1111 bond_dev->addr_len = slave_dev->addr_len;
1113 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1114 slave_dev->addr_len);
1117 /* On bonding slaves other than the currently active slave, suppress
1118 * duplicates except for alb non-mcast/bcast.
1120 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1121 struct slave *slave,
1122 struct bonding *bond)
1124 if (bond_is_slave_inactive(slave)) {
1125 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1126 skb->pkt_type != PACKET_BROADCAST &&
1127 skb->pkt_type != PACKET_MULTICAST)
1134 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1136 struct sk_buff *skb = *pskb;
1137 struct slave *slave;
1138 struct bonding *bond;
1139 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1141 int ret = RX_HANDLER_ANOTHER;
1143 skb = skb_share_check(skb, GFP_ATOMIC);
1145 return RX_HANDLER_CONSUMED;
1149 slave = bond_slave_get_rcu(skb->dev);
1152 recv_probe = ACCESS_ONCE(bond->recv_probe);
1154 ret = recv_probe(skb, bond, slave);
1155 if (ret == RX_HANDLER_CONSUMED) {
1161 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1162 return RX_HANDLER_EXACT;
1165 skb->dev = bond->dev;
1167 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1168 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1169 skb->pkt_type == PACKET_HOST) {
1171 if (unlikely(skb_cow_head(skb,
1172 skb->data - skb_mac_header(skb)))) {
1174 return RX_HANDLER_CONSUMED;
1176 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1182 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1183 struct net_device *slave_dev,
1184 struct slave *slave)
1188 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1191 slave_dev->flags |= IFF_SLAVE;
1192 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1196 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1197 struct net_device *slave_dev)
1199 netdev_upper_dev_unlink(slave_dev, bond_dev);
1200 slave_dev->flags &= ~IFF_SLAVE;
1201 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1204 static struct slave *bond_alloc_slave(struct bonding *bond)
1206 struct slave *slave = NULL;
1208 slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1212 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1213 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1215 if (!SLAVE_AD_INFO(slave)) {
1223 static void bond_free_slave(struct slave *slave)
1225 struct bonding *bond = bond_get_bond_by_slave(slave);
1227 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1228 kfree(SLAVE_AD_INFO(slave));
1233 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1235 info->bond_mode = BOND_MODE(bond);
1236 info->miimon = bond->params.miimon;
1237 info->num_slaves = bond->slave_cnt;
1240 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1242 strcpy(info->slave_name, slave->dev->name);
1243 info->link = slave->link;
1244 info->state = bond_slave_state(slave);
1245 info->link_failure_count = slave->link_failure_count;
1248 static void bond_netdev_notify(struct net_device *dev,
1249 struct netdev_bonding_info *info)
1252 netdev_bonding_info_change(dev, info);
1256 static void bond_netdev_notify_work(struct work_struct *_work)
1258 struct netdev_notify_work *w =
1259 container_of(_work, struct netdev_notify_work, work.work);
1261 bond_netdev_notify(w->dev, &w->bonding_info);
1266 void bond_queue_slave_event(struct slave *slave)
1268 struct bonding *bond = slave->bond;
1269 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1274 dev_hold(slave->dev);
1275 nnw->dev = slave->dev;
1276 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1277 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1278 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1280 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1283 /* enslave device <slave> to bond device <master> */
1284 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1286 struct bonding *bond = netdev_priv(bond_dev);
1287 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1288 struct slave *new_slave = NULL, *prev_slave;
1289 struct sockaddr addr;
1293 if (!bond->params.use_carrier &&
1294 slave_dev->ethtool_ops->get_link == NULL &&
1295 slave_ops->ndo_do_ioctl == NULL) {
1296 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1300 /* already enslaved */
1301 if (slave_dev->flags & IFF_SLAVE) {
1302 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1306 if (bond_dev == slave_dev) {
1307 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1311 /* vlan challenged mutual exclusion */
1312 /* no need to lock since we're protected by rtnl_lock */
1313 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1314 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1316 if (vlan_uses_dev(bond_dev)) {
1317 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1318 slave_dev->name, bond_dev->name);
1321 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1322 slave_dev->name, slave_dev->name,
1326 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1330 /* Old ifenslave binaries are no longer supported. These can
1331 * be identified with moderate accuracy by the state of the slave:
1332 * the current ifenslave will set the interface down prior to
1333 * enslaving it; the old ifenslave will not.
1335 if ((slave_dev->flags & IFF_UP)) {
1336 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1339 goto err_undo_flags;
1342 /* set bonding device ether type by slave - bonding netdevices are
1343 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1344 * there is a need to override some of the type dependent attribs/funcs.
1346 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1347 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1349 if (!bond_has_slaves(bond)) {
1350 if (bond_dev->type != slave_dev->type) {
1351 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1352 bond_dev->type, slave_dev->type);
1354 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1356 res = notifier_to_errno(res);
1358 netdev_err(bond_dev, "refused to change device type\n");
1360 goto err_undo_flags;
1363 /* Flush unicast and multicast addresses */
1364 dev_uc_flush(bond_dev);
1365 dev_mc_flush(bond_dev);
1367 if (slave_dev->type != ARPHRD_ETHER)
1368 bond_setup_by_slave(bond_dev, slave_dev);
1370 ether_setup(bond_dev);
1371 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1374 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1377 } else if (bond_dev->type != slave_dev->type) {
1378 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1379 slave_dev->name, slave_dev->type, bond_dev->type);
1381 goto err_undo_flags;
1384 if (slave_ops->ndo_set_mac_address == NULL) {
1385 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1386 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1387 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1388 if (!bond_has_slaves(bond)) {
1389 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1390 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1392 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");
1394 goto err_undo_flags;
1399 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1401 /* If this is the first slave, then we need to set the master's hardware
1402 * address to be the same as the slave's.
1404 if (!bond_has_slaves(bond) &&
1405 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1406 bond_set_dev_addr(bond->dev, slave_dev);
1408 new_slave = bond_alloc_slave(bond);
1411 goto err_undo_flags;
1414 new_slave->bond = bond;
1415 new_slave->dev = slave_dev;
1416 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1417 * is set via sysfs or module option if desired.
1419 new_slave->queue_id = 0;
1421 /* Save slave's original mtu and then set it to match the bond */
1422 new_slave->original_mtu = slave_dev->mtu;
1423 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1425 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1429 /* Save slave's original ("permanent") mac address for modes
1430 * that need it, and for restoring it upon release, and then
1431 * set it to the master's address
1433 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1435 if (!bond->params.fail_over_mac ||
1436 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1437 /* Set slave to master's mac address. The application already
1438 * set the master's mac address to that of the first slave
1440 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1441 addr.sa_family = slave_dev->type;
1442 res = dev_set_mac_address(slave_dev, &addr);
1444 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1445 goto err_restore_mtu;
1449 /* open the slave since the application closed it */
1450 res = dev_open(slave_dev);
1452 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1453 goto err_restore_mac;
1456 slave_dev->priv_flags |= IFF_BONDING;
1457 /* initialize slave stats */
1458 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1460 if (bond_is_lb(bond)) {
1461 /* bond_alb_init_slave() must be called before all other stages since
1462 * it might fail and we do not want to have to undo everything
1464 res = bond_alb_init_slave(bond, new_slave);
1469 /* If the mode uses primary, then the following is handled by
1470 * bond_change_active_slave().
1472 if (!bond_uses_primary(bond)) {
1473 /* set promiscuity level to new slave */
1474 if (bond_dev->flags & IFF_PROMISC) {
1475 res = dev_set_promiscuity(slave_dev, 1);
1480 /* set allmulti level to new slave */
1481 if (bond_dev->flags & IFF_ALLMULTI) {
1482 res = dev_set_allmulti(slave_dev, 1);
1487 netif_addr_lock_bh(bond_dev);
1489 dev_mc_sync_multiple(slave_dev, bond_dev);
1490 dev_uc_sync_multiple(slave_dev, bond_dev);
1492 netif_addr_unlock_bh(bond_dev);
1495 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1496 /* add lacpdu mc addr to mc list */
1497 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1499 dev_mc_add(slave_dev, lacpdu_multicast);
1502 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1504 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1509 prev_slave = bond_last_slave(bond);
1511 new_slave->delay = 0;
1512 new_slave->link_failure_count = 0;
1514 bond_update_speed_duplex(new_slave);
1516 new_slave->last_rx = jiffies -
1517 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1518 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1519 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1521 if (bond->params.miimon && !bond->params.use_carrier) {
1522 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1524 if ((link_reporting == -1) && !bond->params.arp_interval) {
1525 /* miimon is set but a bonded network driver
1526 * does not support ETHTOOL/MII and
1527 * arp_interval is not set. Note: if
1528 * use_carrier is enabled, we will never go
1529 * here (because netif_carrier is always
1530 * supported); thus, we don't need to change
1531 * the messages for netif_carrier.
1533 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",
1535 } else if (link_reporting == -1) {
1536 /* unable get link status using mii/ethtool */
1537 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",
1542 /* check for initial state */
1543 if (bond->params.miimon) {
1544 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1545 if (bond->params.updelay) {
1546 bond_set_slave_link_state(new_slave,
1548 new_slave->delay = bond->params.updelay;
1550 bond_set_slave_link_state(new_slave,
1554 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1556 } else if (bond->params.arp_interval) {
1557 bond_set_slave_link_state(new_slave,
1558 (netif_carrier_ok(slave_dev) ?
1559 BOND_LINK_UP : BOND_LINK_DOWN));
1561 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1564 if (new_slave->link != BOND_LINK_DOWN)
1565 new_slave->last_link_up = jiffies;
1566 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1567 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1568 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1570 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1571 /* if there is a primary slave, remember it */
1572 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1573 rcu_assign_pointer(bond->primary_slave, new_slave);
1574 bond->force_primary = true;
1578 switch (BOND_MODE(bond)) {
1579 case BOND_MODE_ACTIVEBACKUP:
1580 bond_set_slave_inactive_flags(new_slave,
1581 BOND_SLAVE_NOTIFY_NOW);
1583 case BOND_MODE_8023AD:
1584 /* in 802.3ad mode, the internal mechanism
1585 * will activate the slaves in the selected
1588 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1589 /* if this is the first slave */
1591 SLAVE_AD_INFO(new_slave)->id = 1;
1592 /* Initialize AD with the number of times that the AD timer is called in 1 second
1593 * can be called only after the mac address of the bond is set
1595 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1597 SLAVE_AD_INFO(new_slave)->id =
1598 SLAVE_AD_INFO(prev_slave)->id + 1;
1601 bond_3ad_bind_slave(new_slave);
1605 bond_set_active_slave(new_slave);
1606 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1609 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1611 /* always active in trunk mode */
1612 bond_set_active_slave(new_slave);
1614 /* In trunking mode there is little meaning to curr_active_slave
1615 * anyway (it holds no special properties of the bond device),
1616 * so we can change it without calling change_active_interface()
1618 if (!rcu_access_pointer(bond->curr_active_slave) &&
1619 new_slave->link == BOND_LINK_UP)
1620 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1623 } /* switch(bond_mode) */
1625 #ifdef CONFIG_NET_POLL_CONTROLLER
1626 slave_dev->npinfo = bond->dev->npinfo;
1627 if (slave_dev->npinfo) {
1628 if (slave_enable_netpoll(new_slave)) {
1629 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1636 if (!(bond_dev->features & NETIF_F_LRO))
1637 dev_disable_lro(slave_dev);
1639 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1642 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1646 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1648 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1649 goto err_unregister;
1652 res = bond_sysfs_slave_add(new_slave);
1654 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1655 goto err_upper_unlink;
1659 bond_compute_features(bond);
1660 bond_set_carrier(bond);
1662 if (bond_uses_primary(bond)) {
1664 bond_select_active_slave(bond);
1665 unblock_netpoll_tx();
1668 if (bond_mode_uses_xmit_hash(bond))
1669 bond_update_slave_arr(bond, NULL);
1671 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1673 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1674 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1676 /* enslave is successful */
1677 bond_queue_slave_event(new_slave);
1680 /* Undo stages on error */
1682 bond_upper_dev_unlink(bond_dev, slave_dev);
1685 netdev_rx_handler_unregister(slave_dev);
1688 if (!bond_uses_primary(bond))
1689 bond_hw_addr_flush(bond_dev, slave_dev);
1691 vlan_vids_del_by_dev(slave_dev, bond_dev);
1692 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1693 RCU_INIT_POINTER(bond->primary_slave, NULL);
1694 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1696 bond_change_active_slave(bond, NULL);
1697 bond_select_active_slave(bond);
1698 unblock_netpoll_tx();
1700 /* either primary_slave or curr_active_slave might've changed */
1702 slave_disable_netpoll(new_slave);
1705 slave_dev->priv_flags &= ~IFF_BONDING;
1706 dev_close(slave_dev);
1709 if (!bond->params.fail_over_mac ||
1710 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1711 /* XXX TODO - fom follow mode needs to change master's
1712 * MAC if this slave's MAC is in use by the bond, or at
1713 * least print a warning.
1715 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1716 addr.sa_family = slave_dev->type;
1717 dev_set_mac_address(slave_dev, &addr);
1721 dev_set_mtu(slave_dev, new_slave->original_mtu);
1724 bond_free_slave(new_slave);
1727 /* Enslave of first slave has failed and we need to fix master's mac */
1728 if (!bond_has_slaves(bond) &&
1729 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1730 eth_hw_addr_random(bond_dev);
1735 /* Try to release the slave device <slave> from the bond device <master>
1736 * It is legal to access curr_active_slave without a lock because all the function
1737 * is RTNL-locked. If "all" is true it means that the function is being called
1738 * while destroying a bond interface and all slaves are being released.
1740 * The rules for slave state should be:
1741 * for Active/Backup:
1742 * Active stays on all backups go down
1743 * for Bonded connections:
1744 * The first up interface should be left on and all others downed.
1746 static int __bond_release_one(struct net_device *bond_dev,
1747 struct net_device *slave_dev,
1750 struct bonding *bond = netdev_priv(bond_dev);
1751 struct slave *slave, *oldcurrent;
1752 struct sockaddr addr;
1753 int old_flags = bond_dev->flags;
1754 netdev_features_t old_features = bond_dev->features;
1756 /* slave is not a slave or master is not master of this slave */
1757 if (!(slave_dev->flags & IFF_SLAVE) ||
1758 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1759 netdev_dbg(bond_dev, "cannot release %s\n",
1766 slave = bond_get_slave_by_dev(bond, slave_dev);
1768 /* not a slave of this bond */
1769 netdev_info(bond_dev, "%s not enslaved\n",
1771 unblock_netpoll_tx();
1775 bond_sysfs_slave_del(slave);
1777 /* recompute stats just before removing the slave */
1778 bond_get_stats(bond->dev, &bond->bond_stats);
1780 bond_upper_dev_unlink(bond_dev, slave_dev);
1781 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1782 * for this slave anymore.
1784 netdev_rx_handler_unregister(slave_dev);
1786 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1787 bond_3ad_unbind_slave(slave);
1789 if (bond_mode_uses_xmit_hash(bond))
1790 bond_update_slave_arr(bond, slave);
1792 netdev_info(bond_dev, "Releasing %s interface %s\n",
1793 bond_is_active_slave(slave) ? "active" : "backup",
1796 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1798 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1800 if (!all && (!bond->params.fail_over_mac ||
1801 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1802 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1803 bond_has_slaves(bond))
1804 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",
1805 slave_dev->name, slave->perm_hwaddr,
1806 bond_dev->name, slave_dev->name);
1809 if (rtnl_dereference(bond->primary_slave) == slave)
1810 RCU_INIT_POINTER(bond->primary_slave, NULL);
1812 if (oldcurrent == slave)
1813 bond_change_active_slave(bond, NULL);
1815 if (bond_is_lb(bond)) {
1816 /* Must be called only after the slave has been
1817 * detached from the list and the curr_active_slave
1818 * has been cleared (if our_slave == old_current),
1819 * but before a new active slave is selected.
1821 bond_alb_deinit_slave(bond, slave);
1825 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1826 } else if (oldcurrent == slave) {
1827 /* Note that we hold RTNL over this sequence, so there
1828 * is no concern that another slave add/remove event
1831 bond_select_active_slave(bond);
1834 if (!bond_has_slaves(bond)) {
1835 bond_set_carrier(bond);
1836 eth_hw_addr_random(bond_dev);
1839 unblock_netpoll_tx();
1843 if (!bond_has_slaves(bond)) {
1844 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1845 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1848 bond_compute_features(bond);
1849 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1850 (old_features & NETIF_F_VLAN_CHALLENGED))
1851 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1852 slave_dev->name, bond_dev->name);
1854 vlan_vids_del_by_dev(slave_dev, bond_dev);
1856 /* If the mode uses primary, then this case was handled above by
1857 * bond_change_active_slave(..., NULL)
1859 if (!bond_uses_primary(bond)) {
1860 /* unset promiscuity level from slave
1861 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1862 * of the IFF_PROMISC flag in the bond_dev, but we need the
1863 * value of that flag before that change, as that was the value
1864 * when this slave was attached, so we cache at the start of the
1865 * function and use it here. Same goes for ALLMULTI below
1867 if (old_flags & IFF_PROMISC)
1868 dev_set_promiscuity(slave_dev, -1);
1870 /* unset allmulti level from slave */
1871 if (old_flags & IFF_ALLMULTI)
1872 dev_set_allmulti(slave_dev, -1);
1874 bond_hw_addr_flush(bond_dev, slave_dev);
1877 slave_disable_netpoll(slave);
1879 /* close slave before restoring its mac address */
1880 dev_close(slave_dev);
1882 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1883 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1884 /* restore original ("permanent") mac address */
1885 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1886 addr.sa_family = slave_dev->type;
1887 dev_set_mac_address(slave_dev, &addr);
1890 dev_set_mtu(slave_dev, slave->original_mtu);
1892 slave_dev->priv_flags &= ~IFF_BONDING;
1894 bond_free_slave(slave);
1899 /* A wrapper used because of ndo_del_link */
1900 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1902 return __bond_release_one(bond_dev, slave_dev, false);
1905 /* First release a slave and then destroy the bond if no more slaves are left.
1906 * Must be under rtnl_lock when this function is called.
1908 static int bond_release_and_destroy(struct net_device *bond_dev,
1909 struct net_device *slave_dev)
1911 struct bonding *bond = netdev_priv(bond_dev);
1914 ret = bond_release(bond_dev, slave_dev);
1915 if (ret == 0 && !bond_has_slaves(bond)) {
1916 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1917 netdev_info(bond_dev, "Destroying bond %s\n",
1919 bond_remove_proc_entry(bond);
1920 unregister_netdevice(bond_dev);
1925 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1927 struct bonding *bond = netdev_priv(bond_dev);
1928 bond_fill_ifbond(bond, info);
1932 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1934 struct bonding *bond = netdev_priv(bond_dev);
1935 struct list_head *iter;
1936 int i = 0, res = -ENODEV;
1937 struct slave *slave;
1939 bond_for_each_slave(bond, slave, iter) {
1940 if (i++ == (int)info->slave_id) {
1942 bond_fill_ifslave(slave, info);
1950 /*-------------------------------- Monitoring -------------------------------*/
1952 /* called with rcu_read_lock() */
1953 static int bond_miimon_inspect(struct bonding *bond)
1955 int link_state, commit = 0;
1956 struct list_head *iter;
1957 struct slave *slave;
1958 bool ignore_updelay;
1960 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1962 bond_for_each_slave_rcu(bond, slave, iter) {
1963 slave->new_link = BOND_LINK_NOCHANGE;
1965 link_state = bond_check_dev_link(bond, slave->dev, 0);
1967 switch (slave->link) {
1972 bond_set_slave_link_state(slave, BOND_LINK_FAIL);
1973 slave->delay = bond->params.downdelay;
1975 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1977 BOND_MODE_ACTIVEBACKUP) ?
1978 (bond_is_active_slave(slave) ?
1979 "active " : "backup ") : "",
1981 bond->params.downdelay * bond->params.miimon);
1984 case BOND_LINK_FAIL:
1986 /* recovered before downdelay expired */
1987 bond_set_slave_link_state(slave, BOND_LINK_UP);
1988 slave->last_link_up = jiffies;
1989 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1990 (bond->params.downdelay - slave->delay) *
1991 bond->params.miimon,
1996 if (slave->delay <= 0) {
1997 slave->new_link = BOND_LINK_DOWN;
2005 case BOND_LINK_DOWN:
2009 bond_set_slave_link_state(slave, BOND_LINK_BACK);
2010 slave->delay = bond->params.updelay;
2013 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2015 ignore_updelay ? 0 :
2016 bond->params.updelay *
2017 bond->params.miimon);
2020 case BOND_LINK_BACK:
2022 bond_set_slave_link_state(slave,
2024 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2025 (bond->params.updelay - slave->delay) *
2026 bond->params.miimon,
2035 if (slave->delay <= 0) {
2036 slave->new_link = BOND_LINK_UP;
2038 ignore_updelay = false;
2050 static void bond_miimon_commit(struct bonding *bond)
2052 struct list_head *iter;
2053 struct slave *slave, *primary;
2055 bond_for_each_slave(bond, slave, iter) {
2056 switch (slave->new_link) {
2057 case BOND_LINK_NOCHANGE:
2061 bond_set_slave_link_state(slave, BOND_LINK_UP);
2062 slave->last_link_up = jiffies;
2064 primary = rtnl_dereference(bond->primary_slave);
2065 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2066 /* prevent it from being the active one */
2067 bond_set_backup_slave(slave);
2068 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2069 /* make it immediately active */
2070 bond_set_active_slave(slave);
2071 } else if (slave != primary) {
2072 /* prevent it from being the active one */
2073 bond_set_backup_slave(slave);
2076 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2078 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2079 slave->duplex ? "full" : "half");
2081 /* notify ad that the link status has changed */
2082 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2083 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2085 if (bond_is_lb(bond))
2086 bond_alb_handle_link_change(bond, slave,
2089 if (BOND_MODE(bond) == BOND_MODE_XOR)
2090 bond_update_slave_arr(bond, NULL);
2092 if (!bond->curr_active_slave || slave == primary)
2097 case BOND_LINK_DOWN:
2098 if (slave->link_failure_count < UINT_MAX)
2099 slave->link_failure_count++;
2101 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2103 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2104 BOND_MODE(bond) == BOND_MODE_8023AD)
2105 bond_set_slave_inactive_flags(slave,
2106 BOND_SLAVE_NOTIFY_NOW);
2108 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2111 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2112 bond_3ad_handle_link_change(slave,
2115 if (bond_is_lb(bond))
2116 bond_alb_handle_link_change(bond, slave,
2119 if (BOND_MODE(bond) == BOND_MODE_XOR)
2120 bond_update_slave_arr(bond, NULL);
2122 if (slave == rcu_access_pointer(bond->curr_active_slave))
2128 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2129 slave->new_link, slave->dev->name);
2130 slave->new_link = BOND_LINK_NOCHANGE;
2137 bond_select_active_slave(bond);
2138 unblock_netpoll_tx();
2141 bond_set_carrier(bond);
2146 * Really a wrapper that splits the mii monitor into two phases: an
2147 * inspection, then (if inspection indicates something needs to be done)
2148 * an acquisition of appropriate locks followed by a commit phase to
2149 * implement whatever link state changes are indicated.
2151 static void bond_mii_monitor(struct work_struct *work)
2153 struct bonding *bond = container_of(work, struct bonding,
2155 bool should_notify_peers = false;
2156 unsigned long delay;
2158 delay = msecs_to_jiffies(bond->params.miimon);
2160 if (!bond_has_slaves(bond))
2165 should_notify_peers = bond_should_notify_peers(bond);
2167 if (bond_miimon_inspect(bond)) {
2170 /* Race avoidance with bond_close cancel of workqueue */
2171 if (!rtnl_trylock()) {
2173 should_notify_peers = false;
2177 bond_miimon_commit(bond);
2179 rtnl_unlock(); /* might sleep, hold no other locks */
2184 if (bond->params.miimon)
2185 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2187 if (should_notify_peers) {
2188 if (!rtnl_trylock())
2190 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2195 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2197 struct net_device *upper;
2198 struct list_head *iter;
2201 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2205 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2206 if (ip == bond_confirm_addr(upper, 0, ip)) {
2216 /* We go to the (large) trouble of VLAN tagging ARP frames because
2217 * switches in VLAN mode (especially if ports are configured as
2218 * "native" to a VLAN) might not pass non-tagged frames.
2220 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2221 __be32 dest_ip, __be32 src_ip,
2222 struct bond_vlan_tag *tags)
2224 struct sk_buff *skb;
2225 struct bond_vlan_tag *outer_tag = tags;
2227 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2228 arp_op, slave_dev->name, &dest_ip, &src_ip);
2230 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2231 NULL, slave_dev->dev_addr, NULL);
2234 net_err_ratelimited("ARP packet allocation failed\n");
2238 if (!tags || tags->vlan_proto == VLAN_N_VID)
2243 /* Go through all the tags backwards and add them to the packet */
2244 while (tags->vlan_proto != VLAN_N_VID) {
2245 if (!tags->vlan_id) {
2250 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2251 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2252 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2255 net_err_ratelimited("failed to insert inner VLAN tag\n");
2261 /* Set the outer tag */
2262 if (outer_tag->vlan_id) {
2263 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2264 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2265 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2266 outer_tag->vlan_id);
2273 /* Validate the device path between the @start_dev and the @end_dev.
2274 * The path is valid if the @end_dev is reachable through device
2276 * When the path is validated, collect any vlan information in the
2279 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2280 struct net_device *end_dev,
2283 struct bond_vlan_tag *tags;
2284 struct net_device *upper;
2285 struct list_head *iter;
2287 if (start_dev == end_dev) {
2288 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2290 return ERR_PTR(-ENOMEM);
2291 tags[level].vlan_proto = VLAN_N_VID;
2295 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2296 tags = bond_verify_device_path(upper, end_dev, level + 1);
2297 if (IS_ERR_OR_NULL(tags)) {
2302 if (is_vlan_dev(upper)) {
2303 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2304 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2313 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2316 struct bond_vlan_tag *tags;
2317 __be32 *targets = bond->params.arp_targets, addr;
2320 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2321 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2324 /* Find out through which dev should the packet go */
2325 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2328 /* there's no route to target - try to send arp
2329 * probe to generate any traffic (arp_validate=0)
2331 if (bond->params.arp_validate)
2332 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2335 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2340 /* bond device itself */
2341 if (rt->dst.dev == bond->dev)
2345 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2348 if (!IS_ERR_OR_NULL(tags))
2351 /* Not our device - skip */
2352 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2353 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2359 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2361 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2367 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2371 if (!sip || !bond_has_this_ip(bond, tip)) {
2372 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2377 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2379 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2383 slave->last_rx = jiffies;
2384 slave->target_last_arp_rx[i] = jiffies;
2387 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2388 struct slave *slave)
2390 struct arphdr *arp = (struct arphdr *)skb->data;
2391 struct slave *curr_active_slave;
2392 unsigned char *arp_ptr;
2394 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2396 if (!slave_do_arp_validate(bond, slave)) {
2397 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2398 !slave_do_arp_validate_only(bond))
2399 slave->last_rx = jiffies;
2400 return RX_HANDLER_ANOTHER;
2401 } else if (!is_arp) {
2402 return RX_HANDLER_ANOTHER;
2405 alen = arp_hdr_len(bond->dev);
2407 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2410 if (alen > skb_headlen(skb)) {
2411 arp = kmalloc(alen, GFP_ATOMIC);
2414 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2418 if (arp->ar_hln != bond->dev->addr_len ||
2419 skb->pkt_type == PACKET_OTHERHOST ||
2420 skb->pkt_type == PACKET_LOOPBACK ||
2421 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2422 arp->ar_pro != htons(ETH_P_IP) ||
2426 arp_ptr = (unsigned char *)(arp + 1);
2427 arp_ptr += bond->dev->addr_len;
2428 memcpy(&sip, arp_ptr, 4);
2429 arp_ptr += 4 + bond->dev->addr_len;
2430 memcpy(&tip, arp_ptr, 4);
2432 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2433 slave->dev->name, bond_slave_state(slave),
2434 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2437 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2439 /* Backup slaves won't see the ARP reply, but do come through
2440 * here for each ARP probe (so we swap the sip/tip to validate
2441 * the probe). In a "redundant switch, common router" type of
2442 * configuration, the ARP probe will (hopefully) travel from
2443 * the active, through one switch, the router, then the other
2444 * switch before reaching the backup.
2446 * We 'trust' the arp requests if there is an active slave and
2447 * it received valid arp reply(s) after it became active. This
2448 * is done to avoid endless looping when we can't reach the
2449 * arp_ip_target and fool ourselves with our own arp requests.
2452 if (bond_is_active_slave(slave))
2453 bond_validate_arp(bond, slave, sip, tip);
2454 else if (curr_active_slave &&
2455 time_after(slave_last_rx(bond, curr_active_slave),
2456 curr_active_slave->last_link_up))
2457 bond_validate_arp(bond, slave, tip, sip);
2460 if (arp != (struct arphdr *)skb->data)
2462 return RX_HANDLER_ANOTHER;
2465 /* function to verify if we're in the arp_interval timeslice, returns true if
2466 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2467 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2469 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2472 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2474 return time_in_range(jiffies,
2475 last_act - delta_in_ticks,
2476 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2479 /* This function is called regularly to monitor each slave's link
2480 * ensuring that traffic is being sent and received when arp monitoring
2481 * is used in load-balancing mode. if the adapter has been dormant, then an
2482 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2483 * arp monitoring in active backup mode.
2485 static void bond_loadbalance_arp_mon(struct work_struct *work)
2487 struct bonding *bond = container_of(work, struct bonding,
2489 struct slave *slave, *oldcurrent;
2490 struct list_head *iter;
2491 int do_failover = 0, slave_state_changed = 0;
2493 if (!bond_has_slaves(bond))
2498 oldcurrent = rcu_dereference(bond->curr_active_slave);
2499 /* see if any of the previous devices are up now (i.e. they have
2500 * xmt and rcv traffic). the curr_active_slave does not come into
2501 * the picture unless it is null. also, slave->last_link_up is not
2502 * needed here because we send an arp on each slave and give a slave
2503 * as long as it needs to get the tx/rx within the delta.
2504 * TODO: what about up/down delay in arp mode? it wasn't here before
2507 bond_for_each_slave_rcu(bond, slave, iter) {
2508 unsigned long trans_start = dev_trans_start(slave->dev);
2510 if (slave->link != BOND_LINK_UP) {
2511 if (bond_time_in_interval(bond, trans_start, 1) &&
2512 bond_time_in_interval(bond, slave->last_rx, 1)) {
2514 slave->link = BOND_LINK_UP;
2515 slave_state_changed = 1;
2517 /* primary_slave has no meaning in round-robin
2518 * mode. the window of a slave being up and
2519 * curr_active_slave being null after enslaving
2523 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2527 netdev_info(bond->dev, "interface %s is now up\n",
2532 /* slave->link == BOND_LINK_UP */
2534 /* not all switches will respond to an arp request
2535 * when the source ip is 0, so don't take the link down
2536 * if we don't know our ip yet
2538 if (!bond_time_in_interval(bond, trans_start, 2) ||
2539 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2541 slave->link = BOND_LINK_DOWN;
2542 slave_state_changed = 1;
2544 if (slave->link_failure_count < UINT_MAX)
2545 slave->link_failure_count++;
2547 netdev_info(bond->dev, "interface %s is now down\n",
2550 if (slave == oldcurrent)
2555 /* note: if switch is in round-robin mode, all links
2556 * must tx arp to ensure all links rx an arp - otherwise
2557 * links may oscillate or not come up at all; if switch is
2558 * in something like xor mode, there is nothing we can
2559 * do - all replies will be rx'ed on same link causing slaves
2560 * to be unstable during low/no traffic periods
2562 if (bond_slave_is_up(slave))
2563 bond_arp_send_all(bond, slave);
2568 if (do_failover || slave_state_changed) {
2569 if (!rtnl_trylock())
2572 if (slave_state_changed) {
2573 bond_slave_state_change(bond);
2574 if (BOND_MODE(bond) == BOND_MODE_XOR)
2575 bond_update_slave_arr(bond, NULL);
2579 bond_select_active_slave(bond);
2580 unblock_netpoll_tx();
2586 if (bond->params.arp_interval)
2587 queue_delayed_work(bond->wq, &bond->arp_work,
2588 msecs_to_jiffies(bond->params.arp_interval));
2591 /* Called to inspect slaves for active-backup mode ARP monitor link state
2592 * changes. Sets new_link in slaves to specify what action should take
2593 * place for the slave. Returns 0 if no changes are found, >0 if changes
2594 * to link states must be committed.
2596 * Called with rcu_read_lock held.
2598 static int bond_ab_arp_inspect(struct bonding *bond)
2600 unsigned long trans_start, last_rx;
2601 struct list_head *iter;
2602 struct slave *slave;
2605 bond_for_each_slave_rcu(bond, slave, iter) {
2606 slave->new_link = BOND_LINK_NOCHANGE;
2607 last_rx = slave_last_rx(bond, slave);
2609 if (slave->link != BOND_LINK_UP) {
2610 if (bond_time_in_interval(bond, last_rx, 1)) {
2611 slave->new_link = BOND_LINK_UP;
2617 /* Give slaves 2*delta after being enslaved or made
2618 * active. This avoids bouncing, as the last receive
2619 * times need a full ARP monitor cycle to be updated.
2621 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2624 /* Backup slave is down if:
2625 * - No current_arp_slave AND
2626 * - more than 3*delta since last receive AND
2627 * - the bond has an IP address
2629 * Note: a non-null current_arp_slave indicates
2630 * the curr_active_slave went down and we are
2631 * searching for a new one; under this condition
2632 * we only take the curr_active_slave down - this
2633 * gives each slave a chance to tx/rx traffic
2634 * before being taken out
2636 if (!bond_is_active_slave(slave) &&
2637 !rcu_access_pointer(bond->current_arp_slave) &&
2638 !bond_time_in_interval(bond, last_rx, 3)) {
2639 slave->new_link = BOND_LINK_DOWN;
2643 /* Active slave is down if:
2644 * - more than 2*delta since transmitting OR
2645 * - (more than 2*delta since receive AND
2646 * the bond has an IP address)
2648 trans_start = dev_trans_start(slave->dev);
2649 if (bond_is_active_slave(slave) &&
2650 (!bond_time_in_interval(bond, trans_start, 2) ||
2651 !bond_time_in_interval(bond, last_rx, 2))) {
2652 slave->new_link = BOND_LINK_DOWN;
2660 /* Called to commit link state changes noted by inspection step of
2661 * active-backup mode ARP monitor.
2663 * Called with RTNL hold.
2665 static void bond_ab_arp_commit(struct bonding *bond)
2667 unsigned long trans_start;
2668 struct list_head *iter;
2669 struct slave *slave;
2671 bond_for_each_slave(bond, slave, iter) {
2672 switch (slave->new_link) {
2673 case BOND_LINK_NOCHANGE:
2677 trans_start = dev_trans_start(slave->dev);
2678 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2679 (!rtnl_dereference(bond->curr_active_slave) &&
2680 bond_time_in_interval(bond, trans_start, 1))) {
2681 struct slave *current_arp_slave;
2683 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2684 bond_set_slave_link_state(slave, BOND_LINK_UP);
2685 if (current_arp_slave) {
2686 bond_set_slave_inactive_flags(
2688 BOND_SLAVE_NOTIFY_NOW);
2689 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2692 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2695 if (!rtnl_dereference(bond->curr_active_slave) ||
2696 slave == rtnl_dereference(bond->primary_slave))
2703 case BOND_LINK_DOWN:
2704 if (slave->link_failure_count < UINT_MAX)
2705 slave->link_failure_count++;
2707 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2708 bond_set_slave_inactive_flags(slave,
2709 BOND_SLAVE_NOTIFY_NOW);
2711 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2714 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2715 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2722 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2723 slave->new_link, slave->dev->name);
2729 bond_select_active_slave(bond);
2730 unblock_netpoll_tx();
2733 bond_set_carrier(bond);
2736 /* Send ARP probes for active-backup mode ARP monitor.
2738 * Called with rcu_read_lock held.
2740 static bool bond_ab_arp_probe(struct bonding *bond)
2742 struct slave *slave, *before = NULL, *new_slave = NULL,
2743 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2744 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2745 struct list_head *iter;
2747 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2749 if (curr_arp_slave && curr_active_slave)
2750 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2751 curr_arp_slave->dev->name,
2752 curr_active_slave->dev->name);
2754 if (curr_active_slave) {
2755 bond_arp_send_all(bond, curr_active_slave);
2756 return should_notify_rtnl;
2759 /* if we don't have a curr_active_slave, search for the next available
2760 * backup slave from the current_arp_slave and make it the candidate
2761 * for becoming the curr_active_slave
2764 if (!curr_arp_slave) {
2765 curr_arp_slave = bond_first_slave_rcu(bond);
2766 if (!curr_arp_slave)
2767 return should_notify_rtnl;
2770 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2772 bond_for_each_slave_rcu(bond, slave, iter) {
2773 if (!found && !before && bond_slave_is_up(slave))
2776 if (found && !new_slave && bond_slave_is_up(slave))
2778 /* if the link state is up at this point, we
2779 * mark it down - this can happen if we have
2780 * simultaneous link failures and
2781 * reselect_active_interface doesn't make this
2782 * one the current slave so it is still marked
2783 * up when it is actually down
2785 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2786 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2787 if (slave->link_failure_count < UINT_MAX)
2788 slave->link_failure_count++;
2790 bond_set_slave_inactive_flags(slave,
2791 BOND_SLAVE_NOTIFY_LATER);
2793 netdev_info(bond->dev, "backup interface %s is now down\n",
2796 if (slave == curr_arp_slave)
2800 if (!new_slave && before)
2806 bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2807 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2808 bond_arp_send_all(bond, new_slave);
2809 new_slave->last_link_up = jiffies;
2810 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2813 bond_for_each_slave_rcu(bond, slave, iter) {
2814 if (slave->should_notify) {
2815 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2819 return should_notify_rtnl;
2822 static void bond_activebackup_arp_mon(struct work_struct *work)
2824 struct bonding *bond = container_of(work, struct bonding,
2826 bool should_notify_peers = false;
2827 bool should_notify_rtnl = false;
2830 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2832 if (!bond_has_slaves(bond))
2837 should_notify_peers = bond_should_notify_peers(bond);
2839 if (bond_ab_arp_inspect(bond)) {
2842 /* Race avoidance with bond_close flush of workqueue */
2843 if (!rtnl_trylock()) {
2845 should_notify_peers = false;
2849 bond_ab_arp_commit(bond);
2855 should_notify_rtnl = bond_ab_arp_probe(bond);
2859 if (bond->params.arp_interval)
2860 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2862 if (should_notify_peers || should_notify_rtnl) {
2863 if (!rtnl_trylock())
2866 if (should_notify_peers)
2867 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2869 if (should_notify_rtnl)
2870 bond_slave_state_notify(bond);
2876 /*-------------------------- netdev event handling --------------------------*/
2878 /* Change device name */
2879 static int bond_event_changename(struct bonding *bond)
2881 bond_remove_proc_entry(bond);
2882 bond_create_proc_entry(bond);
2884 bond_debug_reregister(bond);
2889 static int bond_master_netdev_event(unsigned long event,
2890 struct net_device *bond_dev)
2892 struct bonding *event_bond = netdev_priv(bond_dev);
2895 case NETDEV_CHANGENAME:
2896 return bond_event_changename(event_bond);
2897 case NETDEV_UNREGISTER:
2898 bond_remove_proc_entry(event_bond);
2900 case NETDEV_REGISTER:
2901 bond_create_proc_entry(event_bond);
2903 case NETDEV_NOTIFY_PEERS:
2904 if (event_bond->send_peer_notif)
2905 event_bond->send_peer_notif--;
2914 static int bond_slave_netdev_event(unsigned long event,
2915 struct net_device *slave_dev)
2917 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2918 struct bonding *bond;
2919 struct net_device *bond_dev;
2923 /* A netdev event can be generated while enslaving a device
2924 * before netdev_rx_handler_register is called in which case
2925 * slave will be NULL
2929 bond_dev = slave->bond->dev;
2931 primary = rtnl_dereference(bond->primary_slave);
2934 case NETDEV_UNREGISTER:
2935 if (bond_dev->type != ARPHRD_ETHER)
2936 bond_release_and_destroy(bond_dev, slave_dev);
2938 bond_release(bond_dev, slave_dev);
2942 old_speed = slave->speed;
2943 old_duplex = slave->duplex;
2945 bond_update_speed_duplex(slave);
2947 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2948 if (old_speed != slave->speed)
2949 bond_3ad_adapter_speed_changed(slave);
2950 if (old_duplex != slave->duplex)
2951 bond_3ad_adapter_duplex_changed(slave);
2955 /* Refresh slave-array if applicable!
2956 * If the setup does not use miimon or arpmon (mode-specific!),
2957 * then these events will not cause the slave-array to be
2958 * refreshed. This will cause xmit to use a slave that is not
2959 * usable. Avoid such situation by refeshing the array at these
2960 * events. If these (miimon/arpmon) parameters are configured
2961 * then array gets refreshed twice and that should be fine!
2963 if (bond_mode_uses_xmit_hash(bond))
2964 bond_update_slave_arr(bond, NULL);
2966 case NETDEV_CHANGEMTU:
2967 /* TODO: Should slaves be allowed to
2968 * independently alter their MTU? For
2969 * an active-backup bond, slaves need
2970 * not be the same type of device, so
2971 * MTUs may vary. For other modes,
2972 * slaves arguably should have the
2973 * same MTUs. To do this, we'd need to
2974 * take over the slave's change_mtu
2975 * function for the duration of their
2979 case NETDEV_CHANGENAME:
2980 /* we don't care if we don't have primary set */
2981 if (!bond_uses_primary(bond) ||
2982 !bond->params.primary[0])
2985 if (slave == primary) {
2986 /* slave's name changed - he's no longer primary */
2987 RCU_INIT_POINTER(bond->primary_slave, NULL);
2988 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2989 /* we have a new primary slave */
2990 rcu_assign_pointer(bond->primary_slave, slave);
2991 } else { /* we didn't change primary - exit */
2995 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2996 primary ? slave_dev->name : "none");
2999 bond_select_active_slave(bond);
3000 unblock_netpoll_tx();
3002 case NETDEV_FEAT_CHANGE:
3003 bond_compute_features(bond);
3005 case NETDEV_RESEND_IGMP:
3006 /* Propagate to master device */
3007 call_netdevice_notifiers(event, slave->bond->dev);
3016 /* bond_netdev_event: handle netdev notifier chain events.
3018 * This function receives events for the netdev chain. The caller (an
3019 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3020 * locks for us to safely manipulate the slave devices (RTNL lock,
3023 static int bond_netdev_event(struct notifier_block *this,
3024 unsigned long event, void *ptr)
3026 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3028 netdev_dbg(event_dev, "event: %lx\n", event);
3030 if (!(event_dev->priv_flags & IFF_BONDING))
3033 if (event_dev->flags & IFF_MASTER) {
3034 netdev_dbg(event_dev, "IFF_MASTER\n");
3035 return bond_master_netdev_event(event, event_dev);
3038 if (event_dev->flags & IFF_SLAVE) {
3039 netdev_dbg(event_dev, "IFF_SLAVE\n");
3040 return bond_slave_netdev_event(event, event_dev);
3046 static struct notifier_block bond_netdev_notifier = {
3047 .notifier_call = bond_netdev_event,
3050 /*---------------------------- Hashing Policies -----------------------------*/
3052 /* L2 hash helper */
3053 static inline u32 bond_eth_hash(struct sk_buff *skb)
3055 struct ethhdr *ep, hdr_tmp;
3057 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3059 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3063 /* Extract the appropriate headers based on bond's xmit policy */
3064 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3065 struct flow_keys *fk)
3067 const struct ipv6hdr *iph6;
3068 const struct iphdr *iph;
3069 int noff, proto = -1;
3071 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3072 return skb_flow_dissect_flow_keys(skb, fk);
3074 fk->ports.ports = 0;
3075 noff = skb_network_offset(skb);
3076 if (skb->protocol == htons(ETH_P_IP)) {
3077 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3080 iph_to_flow_copy_v4addrs(fk, iph);
3081 noff += iph->ihl << 2;
3082 if (!ip_is_fragment(iph))
3083 proto = iph->protocol;
3084 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3085 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3087 iph6 = ipv6_hdr(skb);
3088 iph_to_flow_copy_v6addrs(fk, iph6);
3089 noff += sizeof(*iph6);
3090 proto = iph6->nexthdr;
3094 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3095 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3101 * bond_xmit_hash - generate a hash value based on the xmit policy
3102 * @bond: bonding device
3103 * @skb: buffer to use for headers
3105 * This function will extract the necessary headers from the skb buffer and use
3106 * them to generate a hash based on the xmit_policy set in the bonding device
3108 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3110 struct flow_keys flow;
3113 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3114 !bond_flow_dissect(bond, skb, &flow))
3115 return bond_eth_hash(skb);
3117 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3118 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3119 hash = bond_eth_hash(skb);
3121 hash = (__force u32)flow.ports.ports;
3122 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3123 (__force u32)flow_get_u32_src(&flow);
3124 hash ^= (hash >> 16);
3125 hash ^= (hash >> 8);
3130 /*-------------------------- Device entry points ----------------------------*/
3132 static void bond_work_init_all(struct bonding *bond)
3134 INIT_DELAYED_WORK(&bond->mcast_work,
3135 bond_resend_igmp_join_requests_delayed);
3136 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3137 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3138 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3139 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3141 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3142 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3143 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3146 static void bond_work_cancel_all(struct bonding *bond)
3148 cancel_delayed_work_sync(&bond->mii_work);
3149 cancel_delayed_work_sync(&bond->arp_work);
3150 cancel_delayed_work_sync(&bond->alb_work);
3151 cancel_delayed_work_sync(&bond->ad_work);
3152 cancel_delayed_work_sync(&bond->mcast_work);
3153 cancel_delayed_work_sync(&bond->slave_arr_work);
3156 static int bond_open(struct net_device *bond_dev)
3158 struct bonding *bond = netdev_priv(bond_dev);
3159 struct list_head *iter;
3160 struct slave *slave;
3162 /* reset slave->backup and slave->inactive */
3163 if (bond_has_slaves(bond)) {
3164 bond_for_each_slave(bond, slave, iter) {
3165 if (bond_uses_primary(bond) &&
3166 slave != rcu_access_pointer(bond->curr_active_slave)) {
3167 bond_set_slave_inactive_flags(slave,
3168 BOND_SLAVE_NOTIFY_NOW);
3169 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3170 bond_set_slave_active_flags(slave,
3171 BOND_SLAVE_NOTIFY_NOW);
3176 bond_work_init_all(bond);
3178 if (bond_is_lb(bond)) {
3179 /* bond_alb_initialize must be called before the timer
3182 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3184 if (bond->params.tlb_dynamic_lb)
3185 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3188 if (bond->params.miimon) /* link check interval, in milliseconds. */
3189 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3191 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3192 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3193 bond->recv_probe = bond_arp_rcv;
3196 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3197 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3198 /* register to receive LACPDUs */
3199 bond->recv_probe = bond_3ad_lacpdu_recv;
3200 bond_3ad_initiate_agg_selection(bond, 1);
3203 if (bond_mode_uses_xmit_hash(bond))
3204 bond_update_slave_arr(bond, NULL);
3209 static int bond_close(struct net_device *bond_dev)
3211 struct bonding *bond = netdev_priv(bond_dev);
3213 bond_work_cancel_all(bond);
3214 bond->send_peer_notif = 0;
3215 if (bond_is_lb(bond))
3216 bond_alb_deinitialize(bond);
3217 bond->recv_probe = NULL;
3222 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3223 struct rtnl_link_stats64 *stats)
3225 struct bonding *bond = netdev_priv(bond_dev);
3226 struct rtnl_link_stats64 temp;
3227 struct list_head *iter;
3228 struct slave *slave;
3230 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3232 bond_for_each_slave(bond, slave, iter) {
3233 const struct rtnl_link_stats64 *sstats =
3234 dev_get_stats(slave->dev, &temp);
3235 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3237 stats->rx_packets += sstats->rx_packets - pstats->rx_packets;
3238 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3239 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3240 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3242 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3243 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3244 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3245 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3247 stats->multicast += sstats->multicast - pstats->multicast;
3248 stats->collisions += sstats->collisions - pstats->collisions;
3250 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3251 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3252 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3253 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3254 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3255 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3257 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3258 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3259 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3260 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3261 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3263 /* save off the slave stats for the next run */
3264 memcpy(pstats, sstats, sizeof(*sstats));
3266 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3271 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3273 struct bonding *bond = netdev_priv(bond_dev);
3274 struct net_device *slave_dev = NULL;
3275 struct ifbond k_binfo;
3276 struct ifbond __user *u_binfo = NULL;
3277 struct ifslave k_sinfo;
3278 struct ifslave __user *u_sinfo = NULL;
3279 struct mii_ioctl_data *mii = NULL;
3280 struct bond_opt_value newval;
3284 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3295 /* We do this again just in case we were called by SIOCGMIIREG
3296 * instead of SIOCGMIIPHY.
3302 if (mii->reg_num == 1) {
3304 if (netif_carrier_ok(bond->dev))
3305 mii->val_out = BMSR_LSTATUS;
3309 case BOND_INFO_QUERY_OLD:
3310 case SIOCBONDINFOQUERY:
3311 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3313 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3316 res = bond_info_query(bond_dev, &k_binfo);
3318 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3322 case BOND_SLAVE_INFO_QUERY_OLD:
3323 case SIOCBONDSLAVEINFOQUERY:
3324 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3326 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3329 res = bond_slave_info_query(bond_dev, &k_sinfo);
3331 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3339 net = dev_net(bond_dev);
3341 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3344 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3346 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3351 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3353 case BOND_ENSLAVE_OLD:
3354 case SIOCBONDENSLAVE:
3355 res = bond_enslave(bond_dev, slave_dev);
3357 case BOND_RELEASE_OLD:
3358 case SIOCBONDRELEASE:
3359 res = bond_release(bond_dev, slave_dev);
3361 case BOND_SETHWADDR_OLD:
3362 case SIOCBONDSETHWADDR:
3363 bond_set_dev_addr(bond_dev, slave_dev);
3366 case BOND_CHANGE_ACTIVE_OLD:
3367 case SIOCBONDCHANGEACTIVE:
3368 bond_opt_initstr(&newval, slave_dev->name);
3369 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3378 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3380 struct bonding *bond = netdev_priv(bond_dev);
3382 if (change & IFF_PROMISC)
3383 bond_set_promiscuity(bond,
3384 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3386 if (change & IFF_ALLMULTI)
3387 bond_set_allmulti(bond,
3388 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3391 static void bond_set_rx_mode(struct net_device *bond_dev)
3393 struct bonding *bond = netdev_priv(bond_dev);
3394 struct list_head *iter;
3395 struct slave *slave;
3398 if (bond_uses_primary(bond)) {
3399 slave = rcu_dereference(bond->curr_active_slave);
3401 dev_uc_sync(slave->dev, bond_dev);
3402 dev_mc_sync(slave->dev, bond_dev);
3405 bond_for_each_slave_rcu(bond, slave, iter) {
3406 dev_uc_sync_multiple(slave->dev, bond_dev);
3407 dev_mc_sync_multiple(slave->dev, bond_dev);
3413 static int bond_neigh_init(struct neighbour *n)
3415 struct bonding *bond = netdev_priv(n->dev);
3416 const struct net_device_ops *slave_ops;
3417 struct neigh_parms parms;
3418 struct slave *slave;
3421 slave = bond_first_slave(bond);
3424 slave_ops = slave->dev->netdev_ops;
3425 if (!slave_ops->ndo_neigh_setup)
3428 parms.neigh_setup = NULL;
3429 parms.neigh_cleanup = NULL;
3430 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3434 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3435 * after the last slave has been detached. Assumes that all slaves
3436 * utilize the same neigh_cleanup (true at this writing as only user
3439 n->parms->neigh_cleanup = parms.neigh_cleanup;
3441 if (!parms.neigh_setup)
3444 return parms.neigh_setup(n);
3447 /* The bonding ndo_neigh_setup is called at init time beofre any
3448 * slave exists. So we must declare proxy setup function which will
3449 * be used at run time to resolve the actual slave neigh param setup.
3451 * It's also called by master devices (such as vlans) to setup their
3452 * underlying devices. In that case - do nothing, we're already set up from
3455 static int bond_neigh_setup(struct net_device *dev,
3456 struct neigh_parms *parms)
3458 /* modify only our neigh_parms */
3459 if (parms->dev == dev)
3460 parms->neigh_setup = bond_neigh_init;
3465 /* Change the MTU of all of a master's slaves to match the master */
3466 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3468 struct bonding *bond = netdev_priv(bond_dev);
3469 struct slave *slave, *rollback_slave;
3470 struct list_head *iter;
3473 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3475 bond_for_each_slave(bond, slave, iter) {
3476 netdev_dbg(bond_dev, "s %p c_m %p\n",
3477 slave, slave->dev->netdev_ops->ndo_change_mtu);
3479 res = dev_set_mtu(slave->dev, new_mtu);
3482 /* If we failed to set the slave's mtu to the new value
3483 * we must abort the operation even in ACTIVE_BACKUP
3484 * mode, because if we allow the backup slaves to have
3485 * different mtu values than the active slave we'll
3486 * need to change their mtu when doing a failover. That
3487 * means changing their mtu from timer context, which
3488 * is probably not a good idea.
3490 netdev_dbg(bond_dev, "err %d %s\n", res,
3496 bond_dev->mtu = new_mtu;
3501 /* unwind from head to the slave that failed */
3502 bond_for_each_slave(bond, rollback_slave, iter) {
3505 if (rollback_slave == slave)
3508 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3510 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3511 tmp_res, rollback_slave->dev->name);
3518 /* Change HW address
3520 * Note that many devices must be down to change the HW address, and
3521 * downing the master releases all slaves. We can make bonds full of
3522 * bonding devices to test this, however.
3524 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3526 struct bonding *bond = netdev_priv(bond_dev);
3527 struct slave *slave, *rollback_slave;
3528 struct sockaddr *sa = addr, tmp_sa;
3529 struct list_head *iter;
3532 if (BOND_MODE(bond) == BOND_MODE_ALB)
3533 return bond_alb_set_mac_address(bond_dev, addr);
3536 netdev_dbg(bond_dev, "bond=%p\n", bond);
3538 /* If fail_over_mac is enabled, do nothing and return success.
3539 * Returning an error causes ifenslave to fail.
3541 if (bond->params.fail_over_mac &&
3542 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3545 if (!is_valid_ether_addr(sa->sa_data))
3546 return -EADDRNOTAVAIL;
3548 bond_for_each_slave(bond, slave, iter) {
3549 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3550 res = dev_set_mac_address(slave->dev, addr);
3552 /* TODO: consider downing the slave
3554 * User should expect communications
3555 * breakage anyway until ARP finish
3558 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3564 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3568 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3569 tmp_sa.sa_family = bond_dev->type;
3571 /* unwind from head to the slave that failed */
3572 bond_for_each_slave(bond, rollback_slave, iter) {
3575 if (rollback_slave == slave)
3578 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3580 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3581 tmp_res, rollback_slave->dev->name);
3589 * bond_xmit_slave_id - transmit skb through slave with slave_id
3590 * @bond: bonding device that is transmitting
3591 * @skb: buffer to transmit
3592 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3594 * This function tries to transmit through slave with slave_id but in case
3595 * it fails, it tries to find the first available slave for transmission.
3596 * The skb is consumed in all cases, thus the function is void.
3598 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3600 struct list_head *iter;
3601 struct slave *slave;
3604 /* Here we start from the slave with slave_id */
3605 bond_for_each_slave_rcu(bond, slave, iter) {
3607 if (bond_slave_can_tx(slave)) {
3608 bond_dev_queue_xmit(bond, skb, slave->dev);
3614 /* Here we start from the first slave up to slave_id */
3616 bond_for_each_slave_rcu(bond, slave, iter) {
3619 if (bond_slave_can_tx(slave)) {
3620 bond_dev_queue_xmit(bond, skb, slave->dev);
3624 /* no slave that can tx has been found */
3625 bond_tx_drop(bond->dev, skb);
3629 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3630 * @bond: bonding device to use
3632 * Based on the value of the bonding device's packets_per_slave parameter
3633 * this function generates a slave id, which is usually used as the next
3634 * slave to transmit through.
3636 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3639 struct reciprocal_value reciprocal_packets_per_slave;
3640 int packets_per_slave = bond->params.packets_per_slave;
3642 switch (packets_per_slave) {
3644 slave_id = prandom_u32();
3647 slave_id = bond->rr_tx_counter;
3650 reciprocal_packets_per_slave =
3651 bond->params.reciprocal_packets_per_slave;
3652 slave_id = reciprocal_divide(bond->rr_tx_counter,
3653 reciprocal_packets_per_slave);
3656 bond->rr_tx_counter++;
3661 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3663 struct bonding *bond = netdev_priv(bond_dev);
3664 struct iphdr *iph = ip_hdr(skb);
3665 struct slave *slave;
3668 /* Start with the curr_active_slave that joined the bond as the
3669 * default for sending IGMP traffic. For failover purposes one
3670 * needs to maintain some consistency for the interface that will
3671 * send the join/membership reports. The curr_active_slave found
3672 * will send all of this type of traffic.
3674 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3675 slave = rcu_dereference(bond->curr_active_slave);
3677 bond_dev_queue_xmit(bond, skb, slave->dev);
3679 bond_xmit_slave_id(bond, skb, 0);
3681 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3683 if (likely(slave_cnt)) {
3684 slave_id = bond_rr_gen_slave_id(bond);
3685 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3687 bond_tx_drop(bond_dev, skb);
3691 return NETDEV_TX_OK;
3694 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3695 * the bond has a usable interface.
3697 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3699 struct bonding *bond = netdev_priv(bond_dev);
3700 struct slave *slave;
3702 slave = rcu_dereference(bond->curr_active_slave);
3704 bond_dev_queue_xmit(bond, skb, slave->dev);
3706 bond_tx_drop(bond_dev, skb);
3708 return NETDEV_TX_OK;
3711 /* Use this to update slave_array when (a) it's not appropriate to update
3712 * slave_array right away (note that update_slave_array() may sleep)
3713 * and / or (b) RTNL is not held.
3715 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3717 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3720 /* Slave array work handler. Holds only RTNL */
3721 static void bond_slave_arr_handler(struct work_struct *work)
3723 struct bonding *bond = container_of(work, struct bonding,
3724 slave_arr_work.work);
3727 if (!rtnl_trylock())
3730 ret = bond_update_slave_arr(bond, NULL);
3733 pr_warn_ratelimited("Failed to update slave array from WT\n");
3739 bond_slave_arr_work_rearm(bond, 1);
3742 /* Build the usable slaves array in control path for modes that use xmit-hash
3743 * to determine the slave interface -
3744 * (a) BOND_MODE_8023AD
3746 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3748 * The caller is expected to hold RTNL only and NO other lock!
3750 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3752 struct slave *slave;
3753 struct list_head *iter;
3754 struct bond_up_slave *new_arr, *old_arr;
3759 #ifdef CONFIG_LOCKDEP
3760 WARN_ON(lockdep_is_held(&bond->mode_lock));
3763 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3767 pr_err("Failed to build slave-array.\n");
3770 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3771 struct ad_info ad_info;
3773 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3774 pr_debug("bond_3ad_get_active_agg_info failed\n");
3775 kfree_rcu(new_arr, rcu);
3776 /* No active aggragator means it's not safe to use
3777 * the previous array.
3779 old_arr = rtnl_dereference(bond->slave_arr);
3781 RCU_INIT_POINTER(bond->slave_arr, NULL);
3782 kfree_rcu(old_arr, rcu);
3786 slaves_in_agg = ad_info.ports;
3787 agg_id = ad_info.aggregator_id;
3789 bond_for_each_slave(bond, slave, iter) {
3790 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3791 struct aggregator *agg;
3793 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3794 if (!agg || agg->aggregator_identifier != agg_id)
3797 if (!bond_slave_can_tx(slave))
3799 if (skipslave == slave)
3801 new_arr->arr[new_arr->count++] = slave;
3804 old_arr = rtnl_dereference(bond->slave_arr);
3805 rcu_assign_pointer(bond->slave_arr, new_arr);
3807 kfree_rcu(old_arr, rcu);
3809 if (ret != 0 && skipslave) {
3812 /* Rare situation where caller has asked to skip a specific
3813 * slave but allocation failed (most likely!). BTW this is
3814 * only possible when the call is initiated from
3815 * __bond_release_one(). In this situation; overwrite the
3816 * skipslave entry in the array with the last entry from the
3817 * array to avoid a situation where the xmit path may choose
3818 * this to-be-skipped slave to send a packet out.
3820 old_arr = rtnl_dereference(bond->slave_arr);
3821 for (idx = 0; idx < old_arr->count; idx++) {
3822 if (skipslave == old_arr->arr[idx]) {
3824 old_arr->arr[old_arr->count-1];
3833 /* Use this Xmit function for 3AD as well as XOR modes. The current
3834 * usable slave array is formed in the control path. The xmit function
3835 * just calculates hash and sends the packet out.
3837 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3839 struct bonding *bond = netdev_priv(dev);
3840 struct slave *slave;
3841 struct bond_up_slave *slaves;
3844 slaves = rcu_dereference(bond->slave_arr);
3845 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3846 if (likely(count)) {
3847 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3848 bond_dev_queue_xmit(bond, skb, slave->dev);
3850 bond_tx_drop(dev, skb);
3853 return NETDEV_TX_OK;
3856 /* in broadcast mode, we send everything to all usable interfaces. */
3857 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3859 struct bonding *bond = netdev_priv(bond_dev);
3860 struct slave *slave = NULL;
3861 struct list_head *iter;
3863 bond_for_each_slave_rcu(bond, slave, iter) {
3864 if (bond_is_last_slave(bond, slave))
3866 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3867 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3870 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3871 bond_dev->name, __func__);
3874 bond_dev_queue_xmit(bond, skb2, slave->dev);
3877 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3878 bond_dev_queue_xmit(bond, skb, slave->dev);
3880 bond_tx_drop(bond_dev, skb);
3882 return NETDEV_TX_OK;
3885 /*------------------------- Device initialization ---------------------------*/
3887 /* Lookup the slave that corresponds to a qid */
3888 static inline int bond_slave_override(struct bonding *bond,
3889 struct sk_buff *skb)
3891 struct slave *slave = NULL;
3892 struct list_head *iter;
3894 if (!skb->queue_mapping)
3897 /* Find out if any slaves have the same mapping as this skb. */
3898 bond_for_each_slave_rcu(bond, slave, iter) {
3899 if (slave->queue_id == skb->queue_mapping) {
3900 if (bond_slave_is_up(slave) &&
3901 slave->link == BOND_LINK_UP) {
3902 bond_dev_queue_xmit(bond, skb, slave->dev);
3905 /* If the slave isn't UP, use default transmit policy. */
3914 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3915 void *accel_priv, select_queue_fallback_t fallback)
3917 /* This helper function exists to help dev_pick_tx get the correct
3918 * destination queue. Using a helper function skips a call to
3919 * skb_tx_hash and will put the skbs in the queue we expect on their
3920 * way down to the bonding driver.
3922 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3924 /* Save the original txq to restore before passing to the driver */
3925 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3927 if (unlikely(txq >= dev->real_num_tx_queues)) {
3929 txq -= dev->real_num_tx_queues;
3930 } while (txq >= dev->real_num_tx_queues);
3935 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3937 struct bonding *bond = netdev_priv(dev);
3939 if (bond_should_override_tx_queue(bond) &&
3940 !bond_slave_override(bond, skb))
3941 return NETDEV_TX_OK;
3943 switch (BOND_MODE(bond)) {
3944 case BOND_MODE_ROUNDROBIN:
3945 return bond_xmit_roundrobin(skb, dev);
3946 case BOND_MODE_ACTIVEBACKUP:
3947 return bond_xmit_activebackup(skb, dev);
3948 case BOND_MODE_8023AD:
3950 return bond_3ad_xor_xmit(skb, dev);
3951 case BOND_MODE_BROADCAST:
3952 return bond_xmit_broadcast(skb, dev);
3954 return bond_alb_xmit(skb, dev);
3956 return bond_tlb_xmit(skb, dev);
3958 /* Should never happen, mode already checked */
3959 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3961 bond_tx_drop(dev, skb);
3962 return NETDEV_TX_OK;
3966 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3968 struct bonding *bond = netdev_priv(dev);
3969 netdev_tx_t ret = NETDEV_TX_OK;
3971 /* If we risk deadlock from transmitting this in the
3972 * netpoll path, tell netpoll to queue the frame for later tx
3974 if (unlikely(is_netpoll_tx_blocked(dev)))
3975 return NETDEV_TX_BUSY;
3978 if (bond_has_slaves(bond))
3979 ret = __bond_start_xmit(skb, dev);
3981 bond_tx_drop(dev, skb);
3987 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3988 struct ethtool_cmd *ecmd)
3990 struct bonding *bond = netdev_priv(bond_dev);
3991 unsigned long speed = 0;
3992 struct list_head *iter;
3993 struct slave *slave;
3995 ecmd->duplex = DUPLEX_UNKNOWN;
3996 ecmd->port = PORT_OTHER;
3998 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3999 * do not need to check mode. Though link speed might not represent
4000 * the true receive or transmit bandwidth (not all modes are symmetric)
4001 * this is an accurate maximum.
4003 bond_for_each_slave(bond, slave, iter) {
4004 if (bond_slave_can_tx(slave)) {
4005 if (slave->speed != SPEED_UNKNOWN)
4006 speed += slave->speed;
4007 if (ecmd->duplex == DUPLEX_UNKNOWN &&
4008 slave->duplex != DUPLEX_UNKNOWN)
4009 ecmd->duplex = slave->duplex;
4012 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4017 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4018 struct ethtool_drvinfo *drvinfo)
4020 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4021 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4022 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4026 static const struct ethtool_ops bond_ethtool_ops = {
4027 .get_drvinfo = bond_ethtool_get_drvinfo,
4028 .get_settings = bond_ethtool_get_settings,
4029 .get_link = ethtool_op_get_link,
4032 static const struct net_device_ops bond_netdev_ops = {
4033 .ndo_init = bond_init,
4034 .ndo_uninit = bond_uninit,
4035 .ndo_open = bond_open,
4036 .ndo_stop = bond_close,
4037 .ndo_start_xmit = bond_start_xmit,
4038 .ndo_select_queue = bond_select_queue,
4039 .ndo_get_stats64 = bond_get_stats,
4040 .ndo_do_ioctl = bond_do_ioctl,
4041 .ndo_change_rx_flags = bond_change_rx_flags,
4042 .ndo_set_rx_mode = bond_set_rx_mode,
4043 .ndo_change_mtu = bond_change_mtu,
4044 .ndo_set_mac_address = bond_set_mac_address,
4045 .ndo_neigh_setup = bond_neigh_setup,
4046 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4047 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4048 #ifdef CONFIG_NET_POLL_CONTROLLER
4049 .ndo_netpoll_setup = bond_netpoll_setup,
4050 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4051 .ndo_poll_controller = bond_poll_controller,
4053 .ndo_add_slave = bond_enslave,
4054 .ndo_del_slave = bond_release,
4055 .ndo_fix_features = bond_fix_features,
4056 .ndo_bridge_setlink = switchdev_port_bridge_setlink,
4057 .ndo_bridge_getlink = switchdev_port_bridge_getlink,
4058 .ndo_bridge_dellink = switchdev_port_bridge_dellink,
4059 .ndo_fdb_add = switchdev_port_fdb_add,
4060 .ndo_fdb_del = switchdev_port_fdb_del,
4061 .ndo_fdb_dump = switchdev_port_fdb_dump,
4062 .ndo_features_check = passthru_features_check,
4065 static const struct device_type bond_type = {
4069 static void bond_destructor(struct net_device *bond_dev)
4071 struct bonding *bond = netdev_priv(bond_dev);
4073 destroy_workqueue(bond->wq);
4074 free_netdev(bond_dev);
4077 void bond_setup(struct net_device *bond_dev)
4079 struct bonding *bond = netdev_priv(bond_dev);
4081 spin_lock_init(&bond->mode_lock);
4082 bond->params = bonding_defaults;
4084 /* Initialize pointers */
4085 bond->dev = bond_dev;
4087 /* Initialize the device entry points */
4088 ether_setup(bond_dev);
4089 bond_dev->netdev_ops = &bond_netdev_ops;
4090 bond_dev->ethtool_ops = &bond_ethtool_ops;
4092 bond_dev->destructor = bond_destructor;
4094 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4096 /* Initialize the device options */
4097 bond_dev->tx_queue_len = 0;
4098 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4099 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
4100 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4102 /* don't acquire bond device's netif_tx_lock when transmitting */
4103 bond_dev->features |= NETIF_F_LLTX;
4105 /* By default, we declare the bond to be fully
4106 * VLAN hardware accelerated capable. Special
4107 * care is taken in the various xmit functions
4108 * when there are slaves that are not hw accel
4112 /* Don't allow bond devices to change network namespaces. */
4113 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4115 bond_dev->hw_features = BOND_VLAN_FEATURES |
4116 NETIF_F_HW_VLAN_CTAG_TX |
4117 NETIF_F_HW_VLAN_CTAG_RX |
4118 NETIF_F_HW_VLAN_CTAG_FILTER;
4120 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4121 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4122 bond_dev->features |= bond_dev->hw_features;
4125 /* Destroy a bonding device.
4126 * Must be under rtnl_lock when this function is called.
4128 static void bond_uninit(struct net_device *bond_dev)
4130 struct bonding *bond = netdev_priv(bond_dev);
4131 struct list_head *iter;
4132 struct slave *slave;
4133 struct bond_up_slave *arr;
4135 bond_netpoll_cleanup(bond_dev);
4137 /* Release the bonded slaves */
4138 bond_for_each_slave(bond, slave, iter)
4139 __bond_release_one(bond_dev, slave->dev, true);
4140 netdev_info(bond_dev, "Released all slaves\n");
4142 arr = rtnl_dereference(bond->slave_arr);
4144 RCU_INIT_POINTER(bond->slave_arr, NULL);
4145 kfree_rcu(arr, rcu);
4148 list_del(&bond->bond_list);
4150 bond_debug_unregister(bond);
4153 /*------------------------- Module initialization ---------------------------*/
4155 static int bond_check_params(struct bond_params *params)
4157 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4158 struct bond_opt_value newval;
4159 const struct bond_opt_value *valptr;
4160 int arp_all_targets_value;
4161 u16 ad_actor_sys_prio = 0;
4162 u16 ad_user_port_key = 0;
4164 /* Convert string parameters. */
4166 bond_opt_initstr(&newval, mode);
4167 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4169 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4172 bond_mode = valptr->value;
4175 if (xmit_hash_policy) {
4176 if ((bond_mode != BOND_MODE_XOR) &&
4177 (bond_mode != BOND_MODE_8023AD) &&
4178 (bond_mode != BOND_MODE_TLB)) {
4179 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4180 bond_mode_name(bond_mode));
4182 bond_opt_initstr(&newval, xmit_hash_policy);
4183 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4186 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4190 xmit_hashtype = valptr->value;
4195 if (bond_mode != BOND_MODE_8023AD) {
4196 pr_info("lacp_rate param is irrelevant in mode %s\n",
4197 bond_mode_name(bond_mode));
4199 bond_opt_initstr(&newval, lacp_rate);
4200 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4203 pr_err("Error: Invalid lacp rate \"%s\"\n",
4207 lacp_fast = valptr->value;
4212 bond_opt_initstr(&newval, ad_select);
4213 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4216 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4219 params->ad_select = valptr->value;
4220 if (bond_mode != BOND_MODE_8023AD)
4221 pr_warn("ad_select param only affects 802.3ad mode\n");
4223 params->ad_select = BOND_AD_STABLE;
4226 if (max_bonds < 0) {
4227 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4228 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4229 max_bonds = BOND_DEFAULT_MAX_BONDS;
4233 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4239 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4244 if (downdelay < 0) {
4245 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4246 downdelay, INT_MAX);
4250 if ((use_carrier != 0) && (use_carrier != 1)) {
4251 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4256 if (num_peer_notif < 0 || num_peer_notif > 255) {
4257 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4262 /* reset values for 802.3ad/TLB/ALB */
4263 if (!bond_mode_uses_arp(bond_mode)) {
4265 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");
4266 pr_warn("Forcing miimon to 100msec\n");
4267 miimon = BOND_DEFAULT_MIIMON;
4271 if (tx_queues < 1 || tx_queues > 255) {
4272 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4273 tx_queues, BOND_DEFAULT_TX_QUEUES);
4274 tx_queues = BOND_DEFAULT_TX_QUEUES;
4277 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4278 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4280 all_slaves_active = 0;
4283 if (resend_igmp < 0 || resend_igmp > 255) {
4284 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4285 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4286 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4289 bond_opt_initval(&newval, packets_per_slave);
4290 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4291 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4292 packets_per_slave, USHRT_MAX);
4293 packets_per_slave = 1;
4296 if (bond_mode == BOND_MODE_ALB) {
4297 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",
4302 if (updelay || downdelay) {
4303 /* just warn the user the up/down delay will have
4304 * no effect since miimon is zero...
4306 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",
4307 updelay, downdelay);
4310 /* don't allow arp monitoring */
4312 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4313 miimon, arp_interval);
4317 if ((updelay % miimon) != 0) {
4318 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4319 updelay, miimon, (updelay / miimon) * miimon);
4324 if ((downdelay % miimon) != 0) {
4325 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4327 (downdelay / miimon) * miimon);
4330 downdelay /= miimon;
4333 if (arp_interval < 0) {
4334 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4335 arp_interval, INT_MAX);
4339 for (arp_ip_count = 0, i = 0;
4340 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4343 /* not a complete check, but good enough to catch mistakes */
4344 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4345 !bond_is_ip_target_ok(ip)) {
4346 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4350 if (bond_get_targets_ip(arp_target, ip) == -1)
4351 arp_target[arp_ip_count++] = ip;
4353 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4358 if (arp_interval && !arp_ip_count) {
4359 /* don't allow arping if no arp_ip_target given... */
4360 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4366 if (!arp_interval) {
4367 pr_err("arp_validate requires arp_interval\n");
4371 bond_opt_initstr(&newval, arp_validate);
4372 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4375 pr_err("Error: invalid arp_validate \"%s\"\n",
4379 arp_validate_value = valptr->value;
4381 arp_validate_value = 0;
4384 arp_all_targets_value = 0;
4385 if (arp_all_targets) {
4386 bond_opt_initstr(&newval, arp_all_targets);
4387 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4390 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4392 arp_all_targets_value = 0;
4394 arp_all_targets_value = valptr->value;
4399 pr_info("MII link monitoring set to %d ms\n", miimon);
4400 } else if (arp_interval) {
4401 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4402 arp_validate_value);
4403 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4404 arp_interval, valptr->string, arp_ip_count);
4406 for (i = 0; i < arp_ip_count; i++)
4407 pr_cont(" %s", arp_ip_target[i]);
4411 } else if (max_bonds) {
4412 /* miimon and arp_interval not set, we need one so things
4413 * work as expected, see bonding.txt for details
4415 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");
4418 if (primary && !bond_mode_uses_primary(bond_mode)) {
4419 /* currently, using a primary only makes sense
4420 * in active backup, TLB or ALB modes
4422 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4423 primary, bond_mode_name(bond_mode));
4427 if (primary && primary_reselect) {
4428 bond_opt_initstr(&newval, primary_reselect);
4429 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4432 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4436 primary_reselect_value = valptr->value;
4438 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4441 if (fail_over_mac) {
4442 bond_opt_initstr(&newval, fail_over_mac);
4443 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4446 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4450 fail_over_mac_value = valptr->value;
4451 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4452 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4454 fail_over_mac_value = BOND_FOM_NONE;
4457 bond_opt_initstr(&newval, "default");
4458 valptr = bond_opt_parse(
4459 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4462 pr_err("Error: No ad_actor_sys_prio default value");
4465 ad_actor_sys_prio = valptr->value;
4467 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4470 pr_err("Error: No ad_user_port_key default value");
4473 ad_user_port_key = valptr->value;
4475 if (lp_interval == 0) {
4476 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4477 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4478 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4481 /* fill params struct with the proper values */
4482 params->mode = bond_mode;
4483 params->xmit_policy = xmit_hashtype;
4484 params->miimon = miimon;
4485 params->num_peer_notif = num_peer_notif;
4486 params->arp_interval = arp_interval;
4487 params->arp_validate = arp_validate_value;
4488 params->arp_all_targets = arp_all_targets_value;
4489 params->updelay = updelay;
4490 params->downdelay = downdelay;
4491 params->use_carrier = use_carrier;
4492 params->lacp_fast = lacp_fast;
4493 params->primary[0] = 0;
4494 params->primary_reselect = primary_reselect_value;
4495 params->fail_over_mac = fail_over_mac_value;
4496 params->tx_queues = tx_queues;
4497 params->all_slaves_active = all_slaves_active;
4498 params->resend_igmp = resend_igmp;
4499 params->min_links = min_links;
4500 params->lp_interval = lp_interval;
4501 params->packets_per_slave = packets_per_slave;
4502 params->tlb_dynamic_lb = 1; /* Default value */
4503 params->ad_actor_sys_prio = ad_actor_sys_prio;
4504 eth_zero_addr(params->ad_actor_system);
4505 params->ad_user_port_key = ad_user_port_key;
4506 if (packets_per_slave > 0) {
4507 params->reciprocal_packets_per_slave =
4508 reciprocal_value(packets_per_slave);
4510 /* reciprocal_packets_per_slave is unused if
4511 * packets_per_slave is 0 or 1, just initialize it
4513 params->reciprocal_packets_per_slave =
4514 (struct reciprocal_value) { 0 };
4518 strncpy(params->primary, primary, IFNAMSIZ);
4519 params->primary[IFNAMSIZ - 1] = 0;
4522 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4527 static struct lock_class_key bonding_netdev_xmit_lock_key;
4528 static struct lock_class_key bonding_netdev_addr_lock_key;
4529 static struct lock_class_key bonding_tx_busylock_key;
4531 static void bond_set_lockdep_class_one(struct net_device *dev,
4532 struct netdev_queue *txq,
4535 lockdep_set_class(&txq->_xmit_lock,
4536 &bonding_netdev_xmit_lock_key);
4539 static void bond_set_lockdep_class(struct net_device *dev)
4541 lockdep_set_class(&dev->addr_list_lock,
4542 &bonding_netdev_addr_lock_key);
4543 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4544 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4547 /* Called from registration process */
4548 static int bond_init(struct net_device *bond_dev)
4550 struct bonding *bond = netdev_priv(bond_dev);
4551 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4553 netdev_dbg(bond_dev, "Begin bond_init\n");
4555 bond->wq = create_singlethread_workqueue(bond_dev->name);
4559 bond_set_lockdep_class(bond_dev);
4561 list_add_tail(&bond->bond_list, &bn->dev_list);
4563 bond_prepare_sysfs_group(bond);
4565 bond_debug_register(bond);
4567 /* Ensure valid dev_addr */
4568 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4569 bond_dev->addr_assign_type == NET_ADDR_PERM)
4570 eth_hw_addr_random(bond_dev);
4575 unsigned int bond_get_num_tx_queues(void)
4580 /* Create a new bond based on the specified name and bonding parameters.
4581 * If name is NULL, obtain a suitable "bond%d" name for us.
4582 * Caller must NOT hold rtnl_lock; we need to release it here before we
4583 * set up our sysfs entries.
4585 int bond_create(struct net *net, const char *name)
4587 struct net_device *bond_dev;
4588 struct bonding *bond;
4589 struct alb_bond_info *bond_info;
4594 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4595 name ? name : "bond%d", NET_NAME_UNKNOWN,
4596 bond_setup, tx_queues);
4598 pr_err("%s: eek! can't alloc netdev!\n", name);
4604 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4605 * It is set to 0 by default which is wrong.
4607 bond = netdev_priv(bond_dev);
4608 bond_info = &(BOND_ALB_INFO(bond));
4609 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4611 dev_net_set(bond_dev, net);
4612 bond_dev->rtnl_link_ops = &bond_link_ops;
4614 res = register_netdevice(bond_dev);
4616 netif_carrier_off(bond_dev);
4620 bond_destructor(bond_dev);
4624 static int __net_init bond_net_init(struct net *net)
4626 struct bond_net *bn = net_generic(net, bond_net_id);
4629 INIT_LIST_HEAD(&bn->dev_list);
4631 bond_create_proc_dir(bn);
4632 bond_create_sysfs(bn);
4637 static void __net_exit bond_net_exit(struct net *net)
4639 struct bond_net *bn = net_generic(net, bond_net_id);
4640 struct bonding *bond, *tmp_bond;
4643 bond_destroy_sysfs(bn);
4645 /* Kill off any bonds created after unregistering bond rtnl ops */
4647 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4648 unregister_netdevice_queue(bond->dev, &list);
4649 unregister_netdevice_many(&list);
4652 bond_destroy_proc_dir(bn);
4655 static struct pernet_operations bond_net_ops = {
4656 .init = bond_net_init,
4657 .exit = bond_net_exit,
4659 .size = sizeof(struct bond_net),
4662 static int __init bonding_init(void)
4667 pr_info("%s", bond_version);
4669 res = bond_check_params(&bonding_defaults);
4673 res = register_pernet_subsys(&bond_net_ops);
4677 res = bond_netlink_init();
4681 bond_create_debugfs();
4683 for (i = 0; i < max_bonds; i++) {
4684 res = bond_create(&init_net, NULL);
4689 register_netdevice_notifier(&bond_netdev_notifier);
4693 bond_destroy_debugfs();
4694 bond_netlink_fini();
4696 unregister_pernet_subsys(&bond_net_ops);
4701 static void __exit bonding_exit(void)
4703 unregister_netdevice_notifier(&bond_netdev_notifier);
4705 bond_destroy_debugfs();
4707 bond_netlink_fini();
4708 unregister_pernet_subsys(&bond_net_ops);
4710 #ifdef CONFIG_NET_POLL_CONTROLLER
4711 /* Make sure we don't have an imbalance on our netpoll blocking */
4712 WARN_ON(atomic_read(&netpoll_block_tx));
4716 module_init(bonding_init);
4717 module_exit(bonding_exit);
4718 MODULE_LICENSE("GPL");
4719 MODULE_VERSION(DRV_VERSION);
4720 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4721 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");