2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/bonding.h>
81 #include <net/bond_3ad.h>
82 #include <net/bond_alb.h>
84 #include "bonding_priv.h"
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
96 static int use_carrier = 1;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *arp_all_targets;
108 static char *fail_over_mac;
109 static int all_slaves_active;
110 static struct bond_params bonding_defaults;
111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 static int packets_per_slave = 1;
113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125 module_param(miimon, int, 0);
126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127 module_param(updelay, int, 0);
128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129 module_param(downdelay, int, 0);
130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 module_param(use_carrier, int, 0);
133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135 module_param(mode, charp, 0);
136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140 module_param(primary, charp, 0);
141 MODULE_PARM_DESC(primary, "Primary network device to use");
142 module_param(primary_reselect, charp, 0);
143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
148 "2 for only on active slave "
150 module_param(lacp_rate, charp, 0);
151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153 module_param(ad_select, charp, 0);
154 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
157 module_param(min_links, int, 0);
158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160 module_param(xmit_hash_policy, charp, 0);
161 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165 module_param(arp_interval, int, 0);
166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167 module_param_array(arp_ip_target, charp, NULL, 0);
168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169 module_param(arp_validate, charp, 0);
170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173 module_param(arp_all_targets, charp, 0);
174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac, charp, 0);
176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179 module_param(all_slaves_active, int, 0);
180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183 module_param(resend_igmp, int, 0);
184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 module_param(packets_per_slave, int, 0);
187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190 module_param(lp_interval, uint, 0);
191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
195 /*----------------------------- Global variables ----------------------------*/
197 #ifdef CONFIG_NET_POLL_CONTROLLER
198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
201 unsigned int bond_net_id __read_mostly;
203 /*-------------------------- Forward declarations ---------------------------*/
205 static int bond_init(struct net_device *bond_dev);
206 static void bond_uninit(struct net_device *bond_dev);
207 static void bond_get_stats(struct net_device *bond_dev,
208 struct rtnl_link_stats64 *stats);
209 static void bond_slave_arr_handler(struct work_struct *work);
210 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
212 static void bond_netdev_notify_work(struct work_struct *work);
214 /*---------------------------- General routines -----------------------------*/
216 const char *bond_mode_name(int mode)
218 static const char *names[] = {
219 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
223 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
224 [BOND_MODE_TLB] = "transmit load balancing",
225 [BOND_MODE_ALB] = "adaptive load balancing",
228 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
234 /*---------------------------------- VLAN -----------------------------------*/
237 * bond_dev_queue_xmit - Prepare skb for xmit.
239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
244 struct net_device *slave_dev)
246 skb->dev = slave_dev;
248 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
249 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
250 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
252 if (unlikely(netpoll_tx_running(bond->dev)))
253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
258 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
259 * We don't protect the slave list iteration with a lock because:
260 * a. This operation is performed in IOCTL context,
261 * b. The operation is protected by the RTNL semaphore in the 8021q code,
262 * c. Holding a lock with BH disabled while directly calling a base driver
263 * entry point is generally a BAD idea.
265 * The design of synchronization/protection for this operation in the 8021q
266 * module is good for one or more VLAN devices over a single physical device
267 * and cannot be extended for a teaming solution like bonding, so there is a
268 * potential race condition here where a net device from the vlan group might
269 * be referenced (either by a base driver or the 8021q code) while it is being
270 * removed from the system. However, it turns out we're not making matters
271 * worse, and if it works for regular VLAN usage it will work here too.
275 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
276 * @bond_dev: bonding net device that got called
277 * @vid: vlan id being added
279 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
280 __be16 proto, u16 vid)
282 struct bonding *bond = netdev_priv(bond_dev);
283 struct slave *slave, *rollback_slave;
284 struct list_head *iter;
287 bond_for_each_slave(bond, slave, iter) {
288 res = vlan_vid_add(slave->dev, proto, vid);
296 /* unwind to the slave that failed */
297 bond_for_each_slave(bond, rollback_slave, iter) {
298 if (rollback_slave == slave)
301 vlan_vid_del(rollback_slave->dev, proto, vid);
308 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
309 * @bond_dev: bonding net device that got called
310 * @vid: vlan id being removed
312 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
313 __be16 proto, u16 vid)
315 struct bonding *bond = netdev_priv(bond_dev);
316 struct list_head *iter;
319 bond_for_each_slave(bond, slave, iter)
320 vlan_vid_del(slave->dev, proto, vid);
322 if (bond_is_lb(bond))
323 bond_alb_clear_vlan(bond, vid);
328 /*------------------------------- Link status -------------------------------*/
330 /* Set the carrier state for the master according to the state of its
331 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
332 * do special 802.3ad magic.
334 * Returns zero if carrier state does not change, nonzero if it does.
336 int bond_set_carrier(struct bonding *bond)
338 struct list_head *iter;
341 if (!bond_has_slaves(bond))
344 if (BOND_MODE(bond) == BOND_MODE_8023AD)
345 return bond_3ad_set_carrier(bond);
347 bond_for_each_slave(bond, slave, iter) {
348 if (slave->link == BOND_LINK_UP) {
349 if (!netif_carrier_ok(bond->dev)) {
350 netif_carrier_on(bond->dev);
358 if (netif_carrier_ok(bond->dev)) {
359 netif_carrier_off(bond->dev);
365 /* Get link speed and duplex from the slave's base driver
366 * using ethtool. If for some reason the call fails or the
367 * values are invalid, set speed and duplex to -1,
368 * and return. Return 1 if speed or duplex settings are
369 * UNKNOWN; 0 otherwise.
371 static int bond_update_speed_duplex(struct slave *slave)
373 struct net_device *slave_dev = slave->dev;
374 struct ethtool_link_ksettings ecmd;
377 slave->speed = SPEED_UNKNOWN;
378 slave->duplex = DUPLEX_UNKNOWN;
380 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
383 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
385 switch (ecmd.base.duplex) {
393 slave->speed = ecmd.base.speed;
394 slave->duplex = ecmd.base.duplex;
399 const char *bond_slave_link_status(s8 link)
415 /* if <dev> supports MII link status reporting, check its link status.
417 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
418 * depending upon the setting of the use_carrier parameter.
420 * Return either BMSR_LSTATUS, meaning that the link is up (or we
421 * can't tell and just pretend it is), or 0, meaning that the link is
424 * If reporting is non-zero, instead of faking link up, return -1 if
425 * both ETHTOOL and MII ioctls fail (meaning the device does not
426 * support them). If use_carrier is set, return whatever it says.
427 * It'd be nice if there was a good way to tell if a driver supports
428 * netif_carrier, but there really isn't.
430 static int bond_check_dev_link(struct bonding *bond,
431 struct net_device *slave_dev, int reporting)
433 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
434 int (*ioctl)(struct net_device *, struct ifreq *, int);
436 struct mii_ioctl_data *mii;
438 if (!reporting && !netif_running(slave_dev))
441 if (bond->params.use_carrier)
442 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
444 /* Try to get link status using Ethtool first. */
445 if (slave_dev->ethtool_ops->get_link)
446 return slave_dev->ethtool_ops->get_link(slave_dev) ?
449 /* Ethtool can't be used, fallback to MII ioctls. */
450 ioctl = slave_ops->ndo_do_ioctl;
452 /* TODO: set pointer to correct ioctl on a per team member
453 * bases to make this more efficient. that is, once
454 * we determine the correct ioctl, we will always
455 * call it and not the others for that team
459 /* We cannot assume that SIOCGMIIPHY will also read a
460 * register; not all network drivers (e.g., e100)
464 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
465 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
467 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
468 mii->reg_num = MII_BMSR;
469 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
470 return mii->val_out & BMSR_LSTATUS;
474 /* If reporting, report that either there's no dev->do_ioctl,
475 * or both SIOCGMIIREG and get_link failed (meaning that we
476 * cannot report link status). If not reporting, pretend
479 return reporting ? -1 : BMSR_LSTATUS;
482 /*----------------------------- Multicast list ------------------------------*/
484 /* Push the promiscuity flag down to appropriate slaves */
485 static int bond_set_promiscuity(struct bonding *bond, int inc)
487 struct list_head *iter;
490 if (bond_uses_primary(bond)) {
491 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
494 err = dev_set_promiscuity(curr_active->dev, inc);
498 bond_for_each_slave(bond, slave, iter) {
499 err = dev_set_promiscuity(slave->dev, inc);
507 /* Push the allmulti flag down to all slaves */
508 static int bond_set_allmulti(struct bonding *bond, int inc)
510 struct list_head *iter;
513 if (bond_uses_primary(bond)) {
514 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
517 err = dev_set_allmulti(curr_active->dev, inc);
521 bond_for_each_slave(bond, slave, iter) {
522 err = dev_set_allmulti(slave->dev, inc);
530 /* Retrieve the list of registered multicast addresses for the bonding
531 * device and retransmit an IGMP JOIN request to the current active
534 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
536 struct bonding *bond = container_of(work, struct bonding,
539 if (!rtnl_trylock()) {
540 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
543 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
545 if (bond->igmp_retrans > 1) {
546 bond->igmp_retrans--;
547 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
552 /* Flush bond's hardware addresses from slave */
553 static void bond_hw_addr_flush(struct net_device *bond_dev,
554 struct net_device *slave_dev)
556 struct bonding *bond = netdev_priv(bond_dev);
558 dev_uc_unsync(slave_dev, bond_dev);
559 dev_mc_unsync(slave_dev, bond_dev);
561 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
562 /* del lacpdu mc addr from mc list */
563 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
565 dev_mc_del(slave_dev, lacpdu_multicast);
569 /*--------------------------- Active slave change ---------------------------*/
571 /* Update the hardware address list and promisc/allmulti for the new and
572 * old active slaves (if any). Modes that are not using primary keep all
573 * slaves up date at all times; only the modes that use primary need to call
574 * this function to swap these settings during a failover.
576 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
577 struct slave *old_active)
580 if (bond->dev->flags & IFF_PROMISC)
581 dev_set_promiscuity(old_active->dev, -1);
583 if (bond->dev->flags & IFF_ALLMULTI)
584 dev_set_allmulti(old_active->dev, -1);
586 bond_hw_addr_flush(bond->dev, old_active->dev);
590 /* FIXME: Signal errors upstream. */
591 if (bond->dev->flags & IFF_PROMISC)
592 dev_set_promiscuity(new_active->dev, 1);
594 if (bond->dev->flags & IFF_ALLMULTI)
595 dev_set_allmulti(new_active->dev, 1);
597 netif_addr_lock_bh(bond->dev);
598 dev_uc_sync(new_active->dev, bond->dev);
599 dev_mc_sync(new_active->dev, bond->dev);
600 netif_addr_unlock_bh(bond->dev);
605 * bond_set_dev_addr - clone slave's address to bond
606 * @bond_dev: bond net device
607 * @slave_dev: slave net device
609 * Should be called with RTNL held.
611 static int bond_set_dev_addr(struct net_device *bond_dev,
612 struct net_device *slave_dev)
616 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
617 bond_dev, slave_dev, slave_dev->addr_len);
618 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
622 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
623 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
624 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
628 static struct slave *bond_get_old_active(struct bonding *bond,
629 struct slave *new_active)
632 struct list_head *iter;
634 bond_for_each_slave(bond, slave, iter) {
635 if (slave == new_active)
638 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
645 /* bond_do_fail_over_mac
647 * Perform special MAC address swapping for fail_over_mac settings
651 static void bond_do_fail_over_mac(struct bonding *bond,
652 struct slave *new_active,
653 struct slave *old_active)
655 u8 tmp_mac[MAX_ADDR_LEN];
656 struct sockaddr_storage ss;
659 switch (bond->params.fail_over_mac) {
660 case BOND_FOM_ACTIVE:
662 rv = bond_set_dev_addr(bond->dev, new_active->dev);
664 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
668 case BOND_FOM_FOLLOW:
669 /* if new_active && old_active, swap them
670 * if just old_active, do nothing (going to no active slave)
671 * if just new_active, set new_active to bond's MAC
677 old_active = bond_get_old_active(bond, new_active);
680 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
681 new_active->dev->addr_len);
682 bond_hw_addr_copy(ss.__data,
683 old_active->dev->dev_addr,
684 old_active->dev->addr_len);
685 ss.ss_family = new_active->dev->type;
687 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
688 bond->dev->addr_len);
689 ss.ss_family = bond->dev->type;
692 rv = dev_set_mac_address(new_active->dev,
693 (struct sockaddr *)&ss, NULL);
695 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
703 bond_hw_addr_copy(ss.__data, tmp_mac,
704 new_active->dev->addr_len);
705 ss.ss_family = old_active->dev->type;
707 rv = dev_set_mac_address(old_active->dev,
708 (struct sockaddr *)&ss, NULL);
710 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
715 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
716 bond->params.fail_over_mac);
722 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
724 struct slave *prim = rtnl_dereference(bond->primary_slave);
725 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
727 if (!prim || prim->link != BOND_LINK_UP) {
728 if (!curr || curr->link != BOND_LINK_UP)
733 if (bond->force_primary) {
734 bond->force_primary = false;
738 if (!curr || curr->link != BOND_LINK_UP)
741 /* At this point, prim and curr are both up */
742 switch (bond->params.primary_reselect) {
743 case BOND_PRI_RESELECT_ALWAYS:
745 case BOND_PRI_RESELECT_BETTER:
746 if (prim->speed < curr->speed)
748 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
751 case BOND_PRI_RESELECT_FAILURE:
754 netdev_err(bond->dev, "impossible primary_reselect %d\n",
755 bond->params.primary_reselect);
761 * bond_find_best_slave - select the best available slave to be the active one
762 * @bond: our bonding struct
764 static struct slave *bond_find_best_slave(struct bonding *bond)
766 struct slave *slave, *bestslave = NULL;
767 struct list_head *iter;
768 int mintime = bond->params.updelay;
770 slave = bond_choose_primary_or_current(bond);
774 bond_for_each_slave(bond, slave, iter) {
775 if (slave->link == BOND_LINK_UP)
777 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
778 slave->delay < mintime) {
779 mintime = slave->delay;
787 static bool bond_should_notify_peers(struct bonding *bond)
792 slave = rcu_dereference(bond->curr_active_slave);
795 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
796 slave ? slave->dev->name : "NULL");
798 if (!slave || !bond->send_peer_notif ||
799 bond->send_peer_notif %
800 max(1, bond->params.peer_notif_delay) != 0 ||
801 !netif_carrier_ok(bond->dev) ||
802 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
809 * change_active_interface - change the active slave into the specified one
810 * @bond: our bonding struct
811 * @new: the new slave to make the active one
813 * Set the new slave to the bond's settings and unset them on the old
815 * Setting include flags, mc-list, promiscuity, allmulti, etc.
817 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
818 * because it is apparently the best available slave we have, even though its
819 * updelay hasn't timed out yet.
821 * Caller must hold RTNL.
823 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
825 struct slave *old_active;
829 old_active = rtnl_dereference(bond->curr_active_slave);
831 if (old_active == new_active)
835 new_active->last_link_up = jiffies;
837 if (new_active->link == BOND_LINK_BACK) {
838 if (bond_uses_primary(bond)) {
839 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
840 (bond->params.updelay - new_active->delay) * bond->params.miimon);
843 new_active->delay = 0;
844 bond_set_slave_link_state(new_active, BOND_LINK_UP,
845 BOND_SLAVE_NOTIFY_NOW);
847 if (BOND_MODE(bond) == BOND_MODE_8023AD)
848 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
850 if (bond_is_lb(bond))
851 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
853 if (bond_uses_primary(bond)) {
854 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
859 if (bond_uses_primary(bond))
860 bond_hw_addr_swap(bond, new_active, old_active);
862 if (bond_is_lb(bond)) {
863 bond_alb_handle_active_change(bond, new_active);
865 bond_set_slave_inactive_flags(old_active,
866 BOND_SLAVE_NOTIFY_NOW);
868 bond_set_slave_active_flags(new_active,
869 BOND_SLAVE_NOTIFY_NOW);
871 rcu_assign_pointer(bond->curr_active_slave, new_active);
874 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
876 bond_set_slave_inactive_flags(old_active,
877 BOND_SLAVE_NOTIFY_NOW);
880 bool should_notify_peers = false;
882 bond_set_slave_active_flags(new_active,
883 BOND_SLAVE_NOTIFY_NOW);
885 if (bond->params.fail_over_mac)
886 bond_do_fail_over_mac(bond, new_active,
889 if (netif_running(bond->dev)) {
890 bond->send_peer_notif =
891 bond->params.num_peer_notif *
892 max(1, bond->params.peer_notif_delay);
893 should_notify_peers =
894 bond_should_notify_peers(bond);
897 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
898 if (should_notify_peers) {
899 bond->send_peer_notif--;
900 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
906 /* resend IGMP joins since active slave has changed or
907 * all were sent on curr_active_slave.
908 * resend only if bond is brought up with the affected
909 * bonding modes and the retransmission is enabled
911 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
912 ((bond_uses_primary(bond) && new_active) ||
913 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
914 bond->igmp_retrans = bond->params.resend_igmp;
915 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
920 * bond_select_active_slave - select a new active slave, if needed
921 * @bond: our bonding struct
923 * This functions should be called when one of the following occurs:
924 * - The old curr_active_slave has been released or lost its link.
925 * - The primary_slave has got its link back.
926 * - A slave has got its link back and there's no old curr_active_slave.
928 * Caller must hold RTNL.
930 void bond_select_active_slave(struct bonding *bond)
932 struct slave *best_slave;
937 best_slave = bond_find_best_slave(bond);
938 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
939 bond_change_active_slave(bond, best_slave);
940 rv = bond_set_carrier(bond);
944 if (netif_carrier_ok(bond->dev))
945 netdev_info(bond->dev, "active interface up!\n");
947 netdev_info(bond->dev, "now running without any active interface!\n");
951 #ifdef CONFIG_NET_POLL_CONTROLLER
952 static inline int slave_enable_netpoll(struct slave *slave)
957 np = kzalloc(sizeof(*np), GFP_KERNEL);
962 err = __netpoll_setup(np, slave->dev);
971 static inline void slave_disable_netpoll(struct slave *slave)
973 struct netpoll *np = slave->np;
983 static void bond_poll_controller(struct net_device *bond_dev)
985 struct bonding *bond = netdev_priv(bond_dev);
986 struct slave *slave = NULL;
987 struct list_head *iter;
988 struct ad_info ad_info;
990 if (BOND_MODE(bond) == BOND_MODE_8023AD)
991 if (bond_3ad_get_active_agg_info(bond, &ad_info))
994 bond_for_each_slave_rcu(bond, slave, iter) {
995 if (!bond_slave_is_up(slave))
998 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
999 struct aggregator *agg =
1000 SLAVE_AD_INFO(slave)->port.aggregator;
1003 agg->aggregator_identifier != ad_info.aggregator_id)
1007 netpoll_poll_dev(slave->dev);
1011 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1013 struct bonding *bond = netdev_priv(bond_dev);
1014 struct list_head *iter;
1015 struct slave *slave;
1017 bond_for_each_slave(bond, slave, iter)
1018 if (bond_slave_is_up(slave))
1019 slave_disable_netpoll(slave);
1022 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1024 struct bonding *bond = netdev_priv(dev);
1025 struct list_head *iter;
1026 struct slave *slave;
1029 bond_for_each_slave(bond, slave, iter) {
1030 err = slave_enable_netpoll(slave);
1032 bond_netpoll_cleanup(dev);
1039 static inline int slave_enable_netpoll(struct slave *slave)
1043 static inline void slave_disable_netpoll(struct slave *slave)
1046 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1051 /*---------------------------------- IOCTL ----------------------------------*/
1053 static netdev_features_t bond_fix_features(struct net_device *dev,
1054 netdev_features_t features)
1056 struct bonding *bond = netdev_priv(dev);
1057 struct list_head *iter;
1058 netdev_features_t mask;
1059 struct slave *slave;
1063 features &= ~NETIF_F_ONE_FOR_ALL;
1064 features |= NETIF_F_ALL_FOR_ALL;
1066 bond_for_each_slave(bond, slave, iter) {
1067 features = netdev_increment_features(features,
1068 slave->dev->features,
1071 features = netdev_add_tso_features(features, mask);
1076 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1077 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1078 NETIF_F_HIGHDMA | NETIF_F_LRO)
1080 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1081 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1083 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1086 static void bond_compute_features(struct bonding *bond)
1088 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1089 IFF_XMIT_DST_RELEASE_PERM;
1090 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1091 netdev_features_t enc_features = BOND_ENC_FEATURES;
1092 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1093 struct net_device *bond_dev = bond->dev;
1094 struct list_head *iter;
1095 struct slave *slave;
1096 unsigned short max_hard_header_len = ETH_HLEN;
1097 unsigned int gso_max_size = GSO_MAX_SIZE;
1098 u16 gso_max_segs = GSO_MAX_SEGS;
1100 if (!bond_has_slaves(bond))
1102 vlan_features &= NETIF_F_ALL_FOR_ALL;
1103 mpls_features &= NETIF_F_ALL_FOR_ALL;
1105 bond_for_each_slave(bond, slave, iter) {
1106 vlan_features = netdev_increment_features(vlan_features,
1107 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1109 enc_features = netdev_increment_features(enc_features,
1110 slave->dev->hw_enc_features,
1113 mpls_features = netdev_increment_features(mpls_features,
1114 slave->dev->mpls_features,
1115 BOND_MPLS_FEATURES);
1117 dst_release_flag &= slave->dev->priv_flags;
1118 if (slave->dev->hard_header_len > max_hard_header_len)
1119 max_hard_header_len = slave->dev->hard_header_len;
1121 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1122 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1124 bond_dev->hard_header_len = max_hard_header_len;
1127 bond_dev->vlan_features = vlan_features;
1128 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1130 bond_dev->mpls_features = mpls_features;
1131 bond_dev->gso_max_segs = gso_max_segs;
1132 netif_set_gso_max_size(bond_dev, gso_max_size);
1134 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1135 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1136 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1137 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1139 netdev_change_features(bond_dev);
1142 static void bond_setup_by_slave(struct net_device *bond_dev,
1143 struct net_device *slave_dev)
1145 bond_dev->header_ops = slave_dev->header_ops;
1147 bond_dev->type = slave_dev->type;
1148 bond_dev->hard_header_len = slave_dev->hard_header_len;
1149 bond_dev->addr_len = slave_dev->addr_len;
1151 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1152 slave_dev->addr_len);
1155 /* On bonding slaves other than the currently active slave, suppress
1156 * duplicates except for alb non-mcast/bcast.
1158 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1159 struct slave *slave,
1160 struct bonding *bond)
1162 if (bond_is_slave_inactive(slave)) {
1163 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1164 skb->pkt_type != PACKET_BROADCAST &&
1165 skb->pkt_type != PACKET_MULTICAST)
1172 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1174 struct sk_buff *skb = *pskb;
1175 struct slave *slave;
1176 struct bonding *bond;
1177 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1179 int ret = RX_HANDLER_ANOTHER;
1181 skb = skb_share_check(skb, GFP_ATOMIC);
1183 return RX_HANDLER_CONSUMED;
1187 slave = bond_slave_get_rcu(skb->dev);
1190 recv_probe = READ_ONCE(bond->recv_probe);
1192 ret = recv_probe(skb, bond, slave);
1193 if (ret == RX_HANDLER_CONSUMED) {
1200 * For packets determined by bond_should_deliver_exact_match() call to
1201 * be suppressed we want to make an exception for link-local packets.
1202 * This is necessary for e.g. LLDP daemons to be able to monitor
1203 * inactive slave links without being forced to bind to them
1206 * At the same time, packets that are passed to the bonding master
1207 * (including link-local ones) can have their originating interface
1208 * determined via PACKET_ORIGDEV socket option.
1210 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1211 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1212 return RX_HANDLER_PASS;
1213 return RX_HANDLER_EXACT;
1216 skb->dev = bond->dev;
1218 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1219 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1220 skb->pkt_type == PACKET_HOST) {
1222 if (unlikely(skb_cow_head(skb,
1223 skb->data - skb_mac_header(skb)))) {
1225 return RX_HANDLER_CONSUMED;
1227 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1228 bond->dev->addr_len);
1234 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1236 switch (BOND_MODE(bond)) {
1237 case BOND_MODE_ROUNDROBIN:
1238 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1239 case BOND_MODE_ACTIVEBACKUP:
1240 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1241 case BOND_MODE_BROADCAST:
1242 return NETDEV_LAG_TX_TYPE_BROADCAST;
1244 case BOND_MODE_8023AD:
1245 return NETDEV_LAG_TX_TYPE_HASH;
1247 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1251 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1252 enum netdev_lag_tx_type type)
1254 if (type != NETDEV_LAG_TX_TYPE_HASH)
1255 return NETDEV_LAG_HASH_NONE;
1257 switch (bond->params.xmit_policy) {
1258 case BOND_XMIT_POLICY_LAYER2:
1259 return NETDEV_LAG_HASH_L2;
1260 case BOND_XMIT_POLICY_LAYER34:
1261 return NETDEV_LAG_HASH_L34;
1262 case BOND_XMIT_POLICY_LAYER23:
1263 return NETDEV_LAG_HASH_L23;
1264 case BOND_XMIT_POLICY_ENCAP23:
1265 return NETDEV_LAG_HASH_E23;
1266 case BOND_XMIT_POLICY_ENCAP34:
1267 return NETDEV_LAG_HASH_E34;
1269 return NETDEV_LAG_HASH_UNKNOWN;
1273 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1274 struct netlink_ext_ack *extack)
1276 struct netdev_lag_upper_info lag_upper_info;
1277 enum netdev_lag_tx_type type;
1279 type = bond_lag_tx_type(bond);
1280 lag_upper_info.tx_type = type;
1281 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1283 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1284 &lag_upper_info, extack);
1287 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1289 netdev_upper_dev_unlink(slave->dev, bond->dev);
1290 slave->dev->flags &= ~IFF_SLAVE;
1293 static struct slave *bond_alloc_slave(struct bonding *bond)
1295 struct slave *slave = NULL;
1297 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1301 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1302 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1304 if (!SLAVE_AD_INFO(slave)) {
1309 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1314 static void bond_free_slave(struct slave *slave)
1316 struct bonding *bond = bond_get_bond_by_slave(slave);
1318 cancel_delayed_work_sync(&slave->notify_work);
1319 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1320 kfree(SLAVE_AD_INFO(slave));
1325 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1327 info->bond_mode = BOND_MODE(bond);
1328 info->miimon = bond->params.miimon;
1329 info->num_slaves = bond->slave_cnt;
1332 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1334 strcpy(info->slave_name, slave->dev->name);
1335 info->link = slave->link;
1336 info->state = bond_slave_state(slave);
1337 info->link_failure_count = slave->link_failure_count;
1340 static void bond_netdev_notify_work(struct work_struct *_work)
1342 struct slave *slave = container_of(_work, struct slave,
1345 if (rtnl_trylock()) {
1346 struct netdev_bonding_info binfo;
1348 bond_fill_ifslave(slave, &binfo.slave);
1349 bond_fill_ifbond(slave->bond, &binfo.master);
1350 netdev_bonding_info_change(slave->dev, &binfo);
1353 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1357 void bond_queue_slave_event(struct slave *slave)
1359 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1362 void bond_lower_state_changed(struct slave *slave)
1364 struct netdev_lag_lower_state_info info;
1366 info.link_up = slave->link == BOND_LINK_UP ||
1367 slave->link == BOND_LINK_FAIL;
1368 info.tx_enabled = bond_is_active_slave(slave);
1369 netdev_lower_state_changed(slave->dev, &info);
1372 /* enslave device <slave> to bond device <master> */
1373 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1374 struct netlink_ext_ack *extack)
1376 struct bonding *bond = netdev_priv(bond_dev);
1377 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1378 struct slave *new_slave = NULL, *prev_slave;
1379 struct sockaddr_storage ss;
1383 if (!bond->params.use_carrier &&
1384 slave_dev->ethtool_ops->get_link == NULL &&
1385 slave_ops->ndo_do_ioctl == NULL) {
1386 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1389 /* already in-use? */
1390 if (netdev_is_rx_handler_busy(slave_dev)) {
1391 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1392 slave_err(bond_dev, slave_dev,
1393 "Error: Device is in use and cannot be enslaved\n");
1397 if (bond_dev == slave_dev) {
1398 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1399 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1403 /* vlan challenged mutual exclusion */
1404 /* no need to lock since we're protected by rtnl_lock */
1405 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1406 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1407 if (vlan_uses_dev(bond_dev)) {
1408 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1409 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1412 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1415 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1418 /* Old ifenslave binaries are no longer supported. These can
1419 * be identified with moderate accuracy by the state of the slave:
1420 * the current ifenslave will set the interface down prior to
1421 * enslaving it; the old ifenslave will not.
1423 if (slave_dev->flags & IFF_UP) {
1424 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1425 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1429 /* set bonding device ether type by slave - bonding netdevices are
1430 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1431 * there is a need to override some of the type dependent attribs/funcs.
1433 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1434 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1436 if (!bond_has_slaves(bond)) {
1437 if (bond_dev->type != slave_dev->type) {
1438 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1439 bond_dev->type, slave_dev->type);
1441 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1443 res = notifier_to_errno(res);
1445 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1449 /* Flush unicast and multicast addresses */
1450 dev_uc_flush(bond_dev);
1451 dev_mc_flush(bond_dev);
1453 if (slave_dev->type != ARPHRD_ETHER)
1454 bond_setup_by_slave(bond_dev, slave_dev);
1456 ether_setup(bond_dev);
1457 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1460 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1463 } else if (bond_dev->type != slave_dev->type) {
1464 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1465 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1466 slave_dev->type, bond_dev->type);
1470 if (slave_dev->type == ARPHRD_INFINIBAND &&
1471 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1472 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1473 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1476 goto err_undo_flags;
1479 if (!slave_ops->ndo_set_mac_address ||
1480 slave_dev->type == ARPHRD_INFINIBAND) {
1481 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1482 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1483 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1484 if (!bond_has_slaves(bond)) {
1485 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1486 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1488 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1489 slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1491 goto err_undo_flags;
1496 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1498 /* If this is the first slave, then we need to set the master's hardware
1499 * address to be the same as the slave's.
1501 if (!bond_has_slaves(bond) &&
1502 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1503 res = bond_set_dev_addr(bond->dev, slave_dev);
1505 goto err_undo_flags;
1508 new_slave = bond_alloc_slave(bond);
1511 goto err_undo_flags;
1514 new_slave->bond = bond;
1515 new_slave->dev = slave_dev;
1516 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1517 * is set via sysfs or module option if desired.
1519 new_slave->queue_id = 0;
1521 /* Save slave's original mtu and then set it to match the bond */
1522 new_slave->original_mtu = slave_dev->mtu;
1523 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1525 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1529 /* Save slave's original ("permanent") mac address for modes
1530 * that need it, and for restoring it upon release, and then
1531 * set it to the master's address
1533 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1534 slave_dev->addr_len);
1536 if (!bond->params.fail_over_mac ||
1537 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1538 /* Set slave to master's mac address. The application already
1539 * set the master's mac address to that of the first slave
1541 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1542 ss.ss_family = slave_dev->type;
1543 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1546 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1547 goto err_restore_mtu;
1551 /* set slave flag before open to prevent IPv6 addrconf */
1552 slave_dev->flags |= IFF_SLAVE;
1554 /* open the slave since the application closed it */
1555 res = dev_open(slave_dev, extack);
1557 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1558 goto err_restore_mac;
1561 slave_dev->priv_flags |= IFF_BONDING;
1562 /* initialize slave stats */
1563 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1565 if (bond_is_lb(bond)) {
1566 /* bond_alb_init_slave() must be called before all other stages since
1567 * it might fail and we do not want to have to undo everything
1569 res = bond_alb_init_slave(bond, new_slave);
1574 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1576 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1580 prev_slave = bond_last_slave(bond);
1582 new_slave->delay = 0;
1583 new_slave->link_failure_count = 0;
1585 if (bond_update_speed_duplex(new_slave) &&
1586 bond_needs_speed_duplex(bond))
1587 new_slave->link = BOND_LINK_DOWN;
1589 new_slave->last_rx = jiffies -
1590 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1591 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1592 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1594 if (bond->params.miimon && !bond->params.use_carrier) {
1595 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1597 if ((link_reporting == -1) && !bond->params.arp_interval) {
1598 /* miimon is set but a bonded network driver
1599 * does not support ETHTOOL/MII and
1600 * arp_interval is not set. Note: if
1601 * use_carrier is enabled, we will never go
1602 * here (because netif_carrier is always
1603 * supported); thus, we don't need to change
1604 * the messages for netif_carrier.
1606 slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n");
1607 } else if (link_reporting == -1) {
1608 /* unable get link status using mii/ethtool */
1609 slave_warn(bond_dev, slave_dev, "can't get link status from slave; 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");
1613 /* check for initial state */
1614 new_slave->link = BOND_LINK_NOCHANGE;
1615 if (bond->params.miimon) {
1616 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1617 if (bond->params.updelay) {
1618 bond_set_slave_link_state(new_slave,
1620 BOND_SLAVE_NOTIFY_NOW);
1621 new_slave->delay = bond->params.updelay;
1623 bond_set_slave_link_state(new_slave,
1625 BOND_SLAVE_NOTIFY_NOW);
1628 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1629 BOND_SLAVE_NOTIFY_NOW);
1631 } else if (bond->params.arp_interval) {
1632 bond_set_slave_link_state(new_slave,
1633 (netif_carrier_ok(slave_dev) ?
1634 BOND_LINK_UP : BOND_LINK_DOWN),
1635 BOND_SLAVE_NOTIFY_NOW);
1637 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1638 BOND_SLAVE_NOTIFY_NOW);
1641 if (new_slave->link != BOND_LINK_DOWN)
1642 new_slave->last_link_up = jiffies;
1643 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1644 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1645 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1647 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1648 /* if there is a primary slave, remember it */
1649 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1650 rcu_assign_pointer(bond->primary_slave, new_slave);
1651 bond->force_primary = true;
1655 switch (BOND_MODE(bond)) {
1656 case BOND_MODE_ACTIVEBACKUP:
1657 bond_set_slave_inactive_flags(new_slave,
1658 BOND_SLAVE_NOTIFY_NOW);
1660 case BOND_MODE_8023AD:
1661 /* in 802.3ad mode, the internal mechanism
1662 * will activate the slaves in the selected
1665 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1666 /* if this is the first slave */
1668 SLAVE_AD_INFO(new_slave)->id = 1;
1669 /* Initialize AD with the number of times that the AD timer is called in 1 second
1670 * can be called only after the mac address of the bond is set
1672 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1674 SLAVE_AD_INFO(new_slave)->id =
1675 SLAVE_AD_INFO(prev_slave)->id + 1;
1678 bond_3ad_bind_slave(new_slave);
1682 bond_set_active_slave(new_slave);
1683 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1686 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
1688 /* always active in trunk mode */
1689 bond_set_active_slave(new_slave);
1691 /* In trunking mode there is little meaning to curr_active_slave
1692 * anyway (it holds no special properties of the bond device),
1693 * so we can change it without calling change_active_interface()
1695 if (!rcu_access_pointer(bond->curr_active_slave) &&
1696 new_slave->link == BOND_LINK_UP)
1697 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1700 } /* switch(bond_mode) */
1702 #ifdef CONFIG_NET_POLL_CONTROLLER
1703 if (bond->dev->npinfo) {
1704 if (slave_enable_netpoll(new_slave)) {
1705 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1712 if (!(bond_dev->features & NETIF_F_LRO))
1713 dev_disable_lro(slave_dev);
1715 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1718 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
1722 res = bond_master_upper_dev_link(bond, new_slave, extack);
1724 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1725 goto err_unregister;
1728 res = bond_sysfs_slave_add(new_slave);
1730 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1731 goto err_upper_unlink;
1734 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1736 /* If the mode uses primary, then the following is handled by
1737 * bond_change_active_slave().
1739 if (!bond_uses_primary(bond)) {
1740 /* set promiscuity level to new slave */
1741 if (bond_dev->flags & IFF_PROMISC) {
1742 res = dev_set_promiscuity(slave_dev, 1);
1747 /* set allmulti level to new slave */
1748 if (bond_dev->flags & IFF_ALLMULTI) {
1749 res = dev_set_allmulti(slave_dev, 1);
1751 if (bond_dev->flags & IFF_PROMISC)
1752 dev_set_promiscuity(slave_dev, -1);
1757 netif_addr_lock_bh(bond_dev);
1758 dev_mc_sync_multiple(slave_dev, bond_dev);
1759 dev_uc_sync_multiple(slave_dev, bond_dev);
1760 netif_addr_unlock_bh(bond_dev);
1762 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1763 /* add lacpdu mc addr to mc list */
1764 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1766 dev_mc_add(slave_dev, lacpdu_multicast);
1771 bond_compute_features(bond);
1772 bond_set_carrier(bond);
1774 if (bond_uses_primary(bond)) {
1776 bond_select_active_slave(bond);
1777 unblock_netpoll_tx();
1780 if (bond_mode_can_use_xmit_hash(bond))
1781 bond_update_slave_arr(bond, NULL);
1784 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
1785 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1786 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1788 /* enslave is successful */
1789 bond_queue_slave_event(new_slave);
1792 /* Undo stages on error */
1794 bond_sysfs_slave_del(new_slave);
1797 bond_upper_dev_unlink(bond, new_slave);
1800 netdev_rx_handler_unregister(slave_dev);
1803 vlan_vids_del_by_dev(slave_dev, bond_dev);
1804 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1805 RCU_INIT_POINTER(bond->primary_slave, NULL);
1806 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1808 bond_change_active_slave(bond, NULL);
1809 bond_select_active_slave(bond);
1810 unblock_netpoll_tx();
1812 /* either primary_slave or curr_active_slave might've changed */
1814 slave_disable_netpoll(new_slave);
1817 slave_dev->priv_flags &= ~IFF_BONDING;
1818 dev_close(slave_dev);
1821 slave_dev->flags &= ~IFF_SLAVE;
1822 if (!bond->params.fail_over_mac ||
1823 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1824 /* XXX TODO - fom follow mode needs to change master's
1825 * MAC if this slave's MAC is in use by the bond, or at
1826 * least print a warning.
1828 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1829 new_slave->dev->addr_len);
1830 ss.ss_family = slave_dev->type;
1831 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
1835 dev_set_mtu(slave_dev, new_slave->original_mtu);
1838 bond_free_slave(new_slave);
1841 /* Enslave of first slave has failed and we need to fix master's mac */
1842 if (!bond_has_slaves(bond)) {
1843 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1844 slave_dev->dev_addr))
1845 eth_hw_addr_random(bond_dev);
1846 if (bond_dev->type != ARPHRD_ETHER) {
1847 dev_close(bond_dev);
1848 ether_setup(bond_dev);
1849 bond_dev->flags |= IFF_MASTER;
1850 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1857 /* Try to release the slave device <slave> from the bond device <master>
1858 * It is legal to access curr_active_slave without a lock because all the function
1859 * is RTNL-locked. If "all" is true it means that the function is being called
1860 * while destroying a bond interface and all slaves are being released.
1862 * The rules for slave state should be:
1863 * for Active/Backup:
1864 * Active stays on all backups go down
1865 * for Bonded connections:
1866 * The first up interface should be left on and all others downed.
1868 static int __bond_release_one(struct net_device *bond_dev,
1869 struct net_device *slave_dev,
1870 bool all, bool unregister)
1872 struct bonding *bond = netdev_priv(bond_dev);
1873 struct slave *slave, *oldcurrent;
1874 struct sockaddr_storage ss;
1875 int old_flags = bond_dev->flags;
1876 netdev_features_t old_features = bond_dev->features;
1878 /* slave is not a slave or master is not master of this slave */
1879 if (!(slave_dev->flags & IFF_SLAVE) ||
1880 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1881 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
1887 slave = bond_get_slave_by_dev(bond, slave_dev);
1889 /* not a slave of this bond */
1890 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
1891 unblock_netpoll_tx();
1895 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1897 bond_sysfs_slave_del(slave);
1899 /* recompute stats just before removing the slave */
1900 bond_get_stats(bond->dev, &bond->bond_stats);
1902 bond_upper_dev_unlink(bond, slave);
1903 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1904 * for this slave anymore.
1906 netdev_rx_handler_unregister(slave_dev);
1908 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1909 bond_3ad_unbind_slave(slave);
1911 if (bond_mode_can_use_xmit_hash(bond))
1912 bond_update_slave_arr(bond, slave);
1914 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
1915 bond_is_active_slave(slave) ? "active" : "backup");
1917 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1919 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1921 if (!all && (!bond->params.fail_over_mac ||
1922 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1923 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1924 bond_has_slaves(bond))
1925 slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n",
1926 slave->perm_hwaddr);
1929 if (rtnl_dereference(bond->primary_slave) == slave)
1930 RCU_INIT_POINTER(bond->primary_slave, NULL);
1932 if (oldcurrent == slave)
1933 bond_change_active_slave(bond, NULL);
1935 if (bond_is_lb(bond)) {
1936 /* Must be called only after the slave has been
1937 * detached from the list and the curr_active_slave
1938 * has been cleared (if our_slave == old_current),
1939 * but before a new active slave is selected.
1941 bond_alb_deinit_slave(bond, slave);
1945 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1946 } else if (oldcurrent == slave) {
1947 /* Note that we hold RTNL over this sequence, so there
1948 * is no concern that another slave add/remove event
1951 bond_select_active_slave(bond);
1954 if (!bond_has_slaves(bond)) {
1955 bond_set_carrier(bond);
1956 eth_hw_addr_random(bond_dev);
1957 bond->nest_level = SINGLE_DEPTH_NESTING;
1959 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1962 unblock_netpoll_tx();
1966 if (!bond_has_slaves(bond)) {
1967 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1968 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1971 bond_compute_features(bond);
1972 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1973 (old_features & NETIF_F_VLAN_CHALLENGED))
1974 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
1976 vlan_vids_del_by_dev(slave_dev, bond_dev);
1978 /* If the mode uses primary, then this case was handled above by
1979 * bond_change_active_slave(..., NULL)
1981 if (!bond_uses_primary(bond)) {
1982 /* unset promiscuity level from slave
1983 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1984 * of the IFF_PROMISC flag in the bond_dev, but we need the
1985 * value of that flag before that change, as that was the value
1986 * when this slave was attached, so we cache at the start of the
1987 * function and use it here. Same goes for ALLMULTI below
1989 if (old_flags & IFF_PROMISC)
1990 dev_set_promiscuity(slave_dev, -1);
1992 /* unset allmulti level from slave */
1993 if (old_flags & IFF_ALLMULTI)
1994 dev_set_allmulti(slave_dev, -1);
1996 bond_hw_addr_flush(bond_dev, slave_dev);
1999 slave_disable_netpoll(slave);
2001 /* close slave before restoring its mac address */
2002 dev_close(slave_dev);
2004 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2005 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2006 /* restore original ("permanent") mac address */
2007 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2008 slave->dev->addr_len);
2009 ss.ss_family = slave_dev->type;
2010 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2014 __dev_set_mtu(slave_dev, slave->original_mtu);
2016 dev_set_mtu(slave_dev, slave->original_mtu);
2018 slave_dev->priv_flags &= ~IFF_BONDING;
2020 bond_free_slave(slave);
2025 /* A wrapper used because of ndo_del_link */
2026 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2028 return __bond_release_one(bond_dev, slave_dev, false, false);
2031 /* First release a slave and then destroy the bond if no more slaves are left.
2032 * Must be under rtnl_lock when this function is called.
2034 static int bond_release_and_destroy(struct net_device *bond_dev,
2035 struct net_device *slave_dev)
2037 struct bonding *bond = netdev_priv(bond_dev);
2040 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2041 if (ret == 0 && !bond_has_slaves(bond)) {
2042 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2043 netdev_info(bond_dev, "Destroying bond\n");
2044 bond_remove_proc_entry(bond);
2045 unregister_netdevice(bond_dev);
2050 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2052 struct bonding *bond = netdev_priv(bond_dev);
2053 bond_fill_ifbond(bond, info);
2056 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2058 struct bonding *bond = netdev_priv(bond_dev);
2059 struct list_head *iter;
2060 int i = 0, res = -ENODEV;
2061 struct slave *slave;
2063 bond_for_each_slave(bond, slave, iter) {
2064 if (i++ == (int)info->slave_id) {
2066 bond_fill_ifslave(slave, info);
2074 /*-------------------------------- Monitoring -------------------------------*/
2076 /* called with rcu_read_lock() */
2077 static int bond_miimon_inspect(struct bonding *bond)
2079 int link_state, commit = 0;
2080 struct list_head *iter;
2081 struct slave *slave;
2082 bool ignore_updelay;
2084 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2086 bond_for_each_slave_rcu(bond, slave, iter) {
2087 slave->new_link = BOND_LINK_NOCHANGE;
2088 slave->link_new_state = slave->link;
2090 link_state = bond_check_dev_link(bond, slave->dev, 0);
2092 switch (slave->link) {
2097 bond_propose_link_state(slave, BOND_LINK_FAIL);
2099 slave->delay = bond->params.downdelay;
2101 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2103 BOND_MODE_ACTIVEBACKUP) ?
2104 (bond_is_active_slave(slave) ?
2105 "active " : "backup ") : "",
2106 bond->params.downdelay * bond->params.miimon);
2109 case BOND_LINK_FAIL:
2111 /* recovered before downdelay expired */
2112 bond_propose_link_state(slave, BOND_LINK_UP);
2113 slave->last_link_up = jiffies;
2114 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2115 (bond->params.downdelay - slave->delay) *
2116 bond->params.miimon);
2121 if (slave->delay <= 0) {
2122 slave->new_link = BOND_LINK_DOWN;
2130 case BOND_LINK_DOWN:
2134 bond_propose_link_state(slave, BOND_LINK_BACK);
2136 slave->delay = bond->params.updelay;
2139 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2140 ignore_updelay ? 0 :
2141 bond->params.updelay *
2142 bond->params.miimon);
2145 case BOND_LINK_BACK:
2147 bond_propose_link_state(slave, BOND_LINK_DOWN);
2148 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2149 (bond->params.updelay - slave->delay) *
2150 bond->params.miimon);
2158 if (slave->delay <= 0) {
2159 slave->new_link = BOND_LINK_UP;
2161 ignore_updelay = false;
2173 static void bond_miimon_link_change(struct bonding *bond,
2174 struct slave *slave,
2177 switch (BOND_MODE(bond)) {
2178 case BOND_MODE_8023AD:
2179 bond_3ad_handle_link_change(slave, link);
2183 bond_alb_handle_link_change(bond, slave, link);
2186 bond_update_slave_arr(bond, NULL);
2191 static void bond_miimon_commit(struct bonding *bond)
2193 struct list_head *iter;
2194 struct slave *slave, *primary;
2196 bond_for_each_slave(bond, slave, iter) {
2197 switch (slave->new_link) {
2198 case BOND_LINK_NOCHANGE:
2199 /* For 802.3ad mode, check current slave speed and
2200 * duplex again in case its port was disabled after
2201 * invalid speed/duplex reporting but recovered before
2202 * link monitoring could make a decision on the actual
2205 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2206 slave->link == BOND_LINK_UP)
2207 bond_3ad_adapter_speed_duplex_changed(slave);
2211 if (bond_update_speed_duplex(slave) &&
2212 bond_needs_speed_duplex(bond)) {
2213 slave->link = BOND_LINK_DOWN;
2214 if (net_ratelimit())
2215 slave_warn(bond->dev, slave->dev,
2216 "failed to get link speed/duplex\n");
2219 bond_set_slave_link_state(slave, BOND_LINK_UP,
2220 BOND_SLAVE_NOTIFY_NOW);
2221 slave->last_link_up = jiffies;
2223 primary = rtnl_dereference(bond->primary_slave);
2224 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2225 /* prevent it from being the active one */
2226 bond_set_backup_slave(slave);
2227 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2228 /* make it immediately active */
2229 bond_set_active_slave(slave);
2230 } else if (slave != primary) {
2231 /* prevent it from being the active one */
2232 bond_set_backup_slave(slave);
2235 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2236 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2237 slave->duplex ? "full" : "half");
2239 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2241 if (!bond->curr_active_slave || slave == primary)
2246 case BOND_LINK_DOWN:
2247 if (slave->link_failure_count < UINT_MAX)
2248 slave->link_failure_count++;
2250 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2251 BOND_SLAVE_NOTIFY_NOW);
2253 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2254 BOND_MODE(bond) == BOND_MODE_8023AD)
2255 bond_set_slave_inactive_flags(slave,
2256 BOND_SLAVE_NOTIFY_NOW);
2258 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2260 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2262 if (slave == rcu_access_pointer(bond->curr_active_slave))
2268 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2270 slave->new_link = BOND_LINK_NOCHANGE;
2277 bond_select_active_slave(bond);
2278 unblock_netpoll_tx();
2281 bond_set_carrier(bond);
2286 * Really a wrapper that splits the mii monitor into two phases: an
2287 * inspection, then (if inspection indicates something needs to be done)
2288 * an acquisition of appropriate locks followed by a commit phase to
2289 * implement whatever link state changes are indicated.
2291 static void bond_mii_monitor(struct work_struct *work)
2293 struct bonding *bond = container_of(work, struct bonding,
2295 bool should_notify_peers = false;
2297 unsigned long delay;
2298 struct slave *slave;
2299 struct list_head *iter;
2301 delay = msecs_to_jiffies(bond->params.miimon);
2303 if (!bond_has_slaves(bond))
2307 should_notify_peers = bond_should_notify_peers(bond);
2308 commit = !!bond_miimon_inspect(bond);
2309 if (bond->send_peer_notif) {
2311 if (rtnl_trylock()) {
2312 bond->send_peer_notif--;
2320 /* Race avoidance with bond_close cancel of workqueue */
2321 if (!rtnl_trylock()) {
2323 should_notify_peers = false;
2327 bond_for_each_slave(bond, slave, iter) {
2328 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2330 bond_miimon_commit(bond);
2332 rtnl_unlock(); /* might sleep, hold no other locks */
2336 if (bond->params.miimon)
2337 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2339 if (should_notify_peers) {
2340 if (!rtnl_trylock())
2342 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2347 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2349 __be32 ip = *((__be32 *)data);
2351 return ip == bond_confirm_addr(upper, 0, ip);
2354 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2358 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2362 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2369 /* We go to the (large) trouble of VLAN tagging ARP frames because
2370 * switches in VLAN mode (especially if ports are configured as
2371 * "native" to a VLAN) might not pass non-tagged frames.
2373 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2374 __be32 src_ip, struct bond_vlan_tag *tags)
2376 struct sk_buff *skb;
2377 struct bond_vlan_tag *outer_tag = tags;
2378 struct net_device *slave_dev = slave->dev;
2379 struct net_device *bond_dev = slave->bond->dev;
2381 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2382 arp_op, &dest_ip, &src_ip);
2384 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2385 NULL, slave_dev->dev_addr, NULL);
2388 net_err_ratelimited("ARP packet allocation failed\n");
2392 if (!tags || tags->vlan_proto == VLAN_N_VID)
2397 /* Go through all the tags backwards and add them to the packet */
2398 while (tags->vlan_proto != VLAN_N_VID) {
2399 if (!tags->vlan_id) {
2404 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2405 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2406 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2409 net_err_ratelimited("failed to insert inner VLAN tag\n");
2415 /* Set the outer tag */
2416 if (outer_tag->vlan_id) {
2417 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2418 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2419 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2420 outer_tag->vlan_id);
2427 /* Validate the device path between the @start_dev and the @end_dev.
2428 * The path is valid if the @end_dev is reachable through device
2430 * When the path is validated, collect any vlan information in the
2433 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2434 struct net_device *end_dev,
2437 struct bond_vlan_tag *tags;
2438 struct net_device *upper;
2439 struct list_head *iter;
2441 if (start_dev == end_dev) {
2442 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2444 return ERR_PTR(-ENOMEM);
2445 tags[level].vlan_proto = VLAN_N_VID;
2449 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2450 tags = bond_verify_device_path(upper, end_dev, level + 1);
2451 if (IS_ERR_OR_NULL(tags)) {
2456 if (is_vlan_dev(upper)) {
2457 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2458 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2467 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2470 struct bond_vlan_tag *tags;
2471 __be32 *targets = bond->params.arp_targets, addr;
2474 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2475 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2476 __func__, &targets[i]);
2479 /* Find out through which dev should the packet go */
2480 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2483 /* there's no route to target - try to send arp
2484 * probe to generate any traffic (arp_validate=0)
2486 if (bond->params.arp_validate)
2487 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2490 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2495 /* bond device itself */
2496 if (rt->dst.dev == bond->dev)
2500 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2503 if (!IS_ERR_OR_NULL(tags))
2506 /* Not our device - skip */
2507 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2508 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2514 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2516 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2521 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2525 if (!sip || !bond_has_this_ip(bond, tip)) {
2526 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2527 __func__, &sip, &tip);
2531 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2533 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2537 slave->last_rx = jiffies;
2538 slave->target_last_arp_rx[i] = jiffies;
2541 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2542 struct slave *slave)
2544 struct arphdr *arp = (struct arphdr *)skb->data;
2545 struct slave *curr_active_slave, *curr_arp_slave;
2546 unsigned char *arp_ptr;
2548 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2551 if (!slave_do_arp_validate(bond, slave)) {
2552 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2553 !slave_do_arp_validate_only(bond))
2554 slave->last_rx = jiffies;
2555 return RX_HANDLER_ANOTHER;
2556 } else if (!is_arp) {
2557 return RX_HANDLER_ANOTHER;
2560 alen = arp_hdr_len(bond->dev);
2562 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2563 __func__, skb->dev->name);
2565 if (alen > skb_headlen(skb)) {
2566 arp = kmalloc(alen, GFP_ATOMIC);
2569 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2573 if (arp->ar_hln != bond->dev->addr_len ||
2574 skb->pkt_type == PACKET_OTHERHOST ||
2575 skb->pkt_type == PACKET_LOOPBACK ||
2576 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2577 arp->ar_pro != htons(ETH_P_IP) ||
2581 arp_ptr = (unsigned char *)(arp + 1);
2582 arp_ptr += bond->dev->addr_len;
2583 memcpy(&sip, arp_ptr, 4);
2584 arp_ptr += 4 + bond->dev->addr_len;
2585 memcpy(&tip, arp_ptr, 4);
2587 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2588 __func__, slave->dev->name, bond_slave_state(slave),
2589 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2592 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2593 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2595 /* We 'trust' the received ARP enough to validate it if:
2597 * (a) the slave receiving the ARP is active (which includes the
2598 * current ARP slave, if any), or
2600 * (b) the receiving slave isn't active, but there is a currently
2601 * active slave and it received valid arp reply(s) after it became
2602 * the currently active slave, or
2604 * (c) there is an ARP slave that sent an ARP during the prior ARP
2605 * interval, and we receive an ARP reply on any slave. We accept
2606 * these because switch FDB update delays may deliver the ARP
2607 * reply to a slave other than the sender of the ARP request.
2609 * Note: for (b), backup slaves are receiving the broadcast ARP
2610 * request, not a reply. This request passes from the sending
2611 * slave through the L2 switch(es) to the receiving slave. Since
2612 * this is checking the request, sip/tip are swapped for
2615 * This is done to avoid endless looping when we can't reach the
2616 * arp_ip_target and fool ourselves with our own arp requests.
2618 if (bond_is_active_slave(slave))
2619 bond_validate_arp(bond, slave, sip, tip);
2620 else if (curr_active_slave &&
2621 time_after(slave_last_rx(bond, curr_active_slave),
2622 curr_active_slave->last_link_up))
2623 bond_validate_arp(bond, slave, tip, sip);
2624 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2625 bond_time_in_interval(bond,
2626 dev_trans_start(curr_arp_slave->dev), 1))
2627 bond_validate_arp(bond, slave, sip, tip);
2630 if (arp != (struct arphdr *)skb->data)
2632 return RX_HANDLER_ANOTHER;
2635 /* function to verify if we're in the arp_interval timeslice, returns true if
2636 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2637 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2639 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2642 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2644 return time_in_range(jiffies,
2645 last_act - delta_in_ticks,
2646 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2649 /* This function is called regularly to monitor each slave's link
2650 * ensuring that traffic is being sent and received when arp monitoring
2651 * is used in load-balancing mode. if the adapter has been dormant, then an
2652 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2653 * arp monitoring in active backup mode.
2655 static void bond_loadbalance_arp_mon(struct bonding *bond)
2657 struct slave *slave, *oldcurrent;
2658 struct list_head *iter;
2659 int do_failover = 0, slave_state_changed = 0;
2661 if (!bond_has_slaves(bond))
2666 oldcurrent = rcu_dereference(bond->curr_active_slave);
2667 /* see if any of the previous devices are up now (i.e. they have
2668 * xmt and rcv traffic). the curr_active_slave does not come into
2669 * the picture unless it is null. also, slave->last_link_up is not
2670 * needed here because we send an arp on each slave and give a slave
2671 * as long as it needs to get the tx/rx within the delta.
2672 * TODO: what about up/down delay in arp mode? it wasn't here before
2675 bond_for_each_slave_rcu(bond, slave, iter) {
2676 unsigned long trans_start = dev_trans_start(slave->dev);
2678 slave->new_link = BOND_LINK_NOCHANGE;
2680 if (slave->link != BOND_LINK_UP) {
2681 if (bond_time_in_interval(bond, trans_start, 1) &&
2682 bond_time_in_interval(bond, slave->last_rx, 1)) {
2684 slave->new_link = BOND_LINK_UP;
2685 slave_state_changed = 1;
2687 /* primary_slave has no meaning in round-robin
2688 * mode. the window of a slave being up and
2689 * curr_active_slave being null after enslaving
2693 slave_info(bond->dev, slave->dev, "link status definitely up\n");
2696 slave_info(bond->dev, slave->dev, "interface is now up\n");
2700 /* slave->link == BOND_LINK_UP */
2702 /* not all switches will respond to an arp request
2703 * when the source ip is 0, so don't take the link down
2704 * if we don't know our ip yet
2706 if (!bond_time_in_interval(bond, trans_start, 2) ||
2707 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2709 slave->new_link = BOND_LINK_DOWN;
2710 slave_state_changed = 1;
2712 if (slave->link_failure_count < UINT_MAX)
2713 slave->link_failure_count++;
2715 slave_info(bond->dev, slave->dev, "interface is now down\n");
2717 if (slave == oldcurrent)
2722 /* note: if switch is in round-robin mode, all links
2723 * must tx arp to ensure all links rx an arp - otherwise
2724 * links may oscillate or not come up at all; if switch is
2725 * in something like xor mode, there is nothing we can
2726 * do - all replies will be rx'ed on same link causing slaves
2727 * to be unstable during low/no traffic periods
2729 if (bond_slave_is_up(slave))
2730 bond_arp_send_all(bond, slave);
2735 if (do_failover || slave_state_changed) {
2736 if (!rtnl_trylock())
2739 bond_for_each_slave(bond, slave, iter) {
2740 if (slave->new_link != BOND_LINK_NOCHANGE)
2741 slave->link = slave->new_link;
2744 if (slave_state_changed) {
2745 bond_slave_state_change(bond);
2746 if (BOND_MODE(bond) == BOND_MODE_XOR)
2747 bond_update_slave_arr(bond, NULL);
2751 bond_select_active_slave(bond);
2752 unblock_netpoll_tx();
2758 if (bond->params.arp_interval)
2759 queue_delayed_work(bond->wq, &bond->arp_work,
2760 msecs_to_jiffies(bond->params.arp_interval));
2763 /* Called to inspect slaves for active-backup mode ARP monitor link state
2764 * changes. Sets new_link in slaves to specify what action should take
2765 * place for the slave. Returns 0 if no changes are found, >0 if changes
2766 * to link states must be committed.
2768 * Called with rcu_read_lock held.
2770 static int bond_ab_arp_inspect(struct bonding *bond)
2772 unsigned long trans_start, last_rx;
2773 struct list_head *iter;
2774 struct slave *slave;
2777 bond_for_each_slave_rcu(bond, slave, iter) {
2778 slave->new_link = BOND_LINK_NOCHANGE;
2779 last_rx = slave_last_rx(bond, slave);
2781 if (slave->link != BOND_LINK_UP) {
2782 if (bond_time_in_interval(bond, last_rx, 1)) {
2783 slave->new_link = BOND_LINK_UP;
2789 /* Give slaves 2*delta after being enslaved or made
2790 * active. This avoids bouncing, as the last receive
2791 * times need a full ARP monitor cycle to be updated.
2793 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2796 /* Backup slave is down if:
2797 * - No current_arp_slave AND
2798 * - more than 3*delta since last receive AND
2799 * - the bond has an IP address
2801 * Note: a non-null current_arp_slave indicates
2802 * the curr_active_slave went down and we are
2803 * searching for a new one; under this condition
2804 * we only take the curr_active_slave down - this
2805 * gives each slave a chance to tx/rx traffic
2806 * before being taken out
2808 if (!bond_is_active_slave(slave) &&
2809 !rcu_access_pointer(bond->current_arp_slave) &&
2810 !bond_time_in_interval(bond, last_rx, 3)) {
2811 slave->new_link = BOND_LINK_DOWN;
2815 /* Active slave is down if:
2816 * - more than 2*delta since transmitting OR
2817 * - (more than 2*delta since receive AND
2818 * the bond has an IP address)
2820 trans_start = dev_trans_start(slave->dev);
2821 if (bond_is_active_slave(slave) &&
2822 (!bond_time_in_interval(bond, trans_start, 2) ||
2823 !bond_time_in_interval(bond, last_rx, 2))) {
2824 slave->new_link = BOND_LINK_DOWN;
2832 /* Called to commit link state changes noted by inspection step of
2833 * active-backup mode ARP monitor.
2835 * Called with RTNL hold.
2837 static void bond_ab_arp_commit(struct bonding *bond)
2839 unsigned long trans_start;
2840 struct list_head *iter;
2841 struct slave *slave;
2843 bond_for_each_slave(bond, slave, iter) {
2844 switch (slave->new_link) {
2845 case BOND_LINK_NOCHANGE:
2849 trans_start = dev_trans_start(slave->dev);
2850 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2851 (!rtnl_dereference(bond->curr_active_slave) &&
2852 bond_time_in_interval(bond, trans_start, 1))) {
2853 struct slave *current_arp_slave;
2855 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2856 bond_set_slave_link_state(slave, BOND_LINK_UP,
2857 BOND_SLAVE_NOTIFY_NOW);
2858 if (current_arp_slave) {
2859 bond_set_slave_inactive_flags(
2861 BOND_SLAVE_NOTIFY_NOW);
2862 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2865 slave_info(bond->dev, slave->dev, "link status definitely up\n");
2867 if (!rtnl_dereference(bond->curr_active_slave) ||
2868 slave == rtnl_dereference(bond->primary_slave))
2875 case BOND_LINK_DOWN:
2876 if (slave->link_failure_count < UINT_MAX)
2877 slave->link_failure_count++;
2879 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2880 BOND_SLAVE_NOTIFY_NOW);
2881 bond_set_slave_inactive_flags(slave,
2882 BOND_SLAVE_NOTIFY_NOW);
2884 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2886 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2887 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2894 slave_err(bond->dev, slave->dev, "impossible: new_link %d on slave\n",
2901 bond_select_active_slave(bond);
2902 unblock_netpoll_tx();
2905 bond_set_carrier(bond);
2908 /* Send ARP probes for active-backup mode ARP monitor.
2910 * Called with rcu_read_lock held.
2912 static bool bond_ab_arp_probe(struct bonding *bond)
2914 struct slave *slave, *before = NULL, *new_slave = NULL,
2915 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2916 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2917 struct list_head *iter;
2919 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2921 if (curr_arp_slave && curr_active_slave)
2922 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2923 curr_arp_slave->dev->name,
2924 curr_active_slave->dev->name);
2926 if (curr_active_slave) {
2927 bond_arp_send_all(bond, curr_active_slave);
2928 return should_notify_rtnl;
2931 /* if we don't have a curr_active_slave, search for the next available
2932 * backup slave from the current_arp_slave and make it the candidate
2933 * for becoming the curr_active_slave
2936 if (!curr_arp_slave) {
2937 curr_arp_slave = bond_first_slave_rcu(bond);
2938 if (!curr_arp_slave)
2939 return should_notify_rtnl;
2942 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2944 bond_for_each_slave_rcu(bond, slave, iter) {
2945 if (!found && !before && bond_slave_is_up(slave))
2948 if (found && !new_slave && bond_slave_is_up(slave))
2950 /* if the link state is up at this point, we
2951 * mark it down - this can happen if we have
2952 * simultaneous link failures and
2953 * reselect_active_interface doesn't make this
2954 * one the current slave so it is still marked
2955 * up when it is actually down
2957 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2958 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2959 BOND_SLAVE_NOTIFY_LATER);
2960 if (slave->link_failure_count < UINT_MAX)
2961 slave->link_failure_count++;
2963 bond_set_slave_inactive_flags(slave,
2964 BOND_SLAVE_NOTIFY_LATER);
2966 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
2968 if (slave == curr_arp_slave)
2972 if (!new_slave && before)
2978 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2979 BOND_SLAVE_NOTIFY_LATER);
2980 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2981 bond_arp_send_all(bond, new_slave);
2982 new_slave->last_link_up = jiffies;
2983 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2986 bond_for_each_slave_rcu(bond, slave, iter) {
2987 if (slave->should_notify || slave->should_notify_link) {
2988 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2992 return should_notify_rtnl;
2995 static void bond_activebackup_arp_mon(struct bonding *bond)
2997 bool should_notify_peers = false;
2998 bool should_notify_rtnl = false;
3001 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3003 if (!bond_has_slaves(bond))
3008 should_notify_peers = bond_should_notify_peers(bond);
3010 if (bond_ab_arp_inspect(bond)) {
3013 /* Race avoidance with bond_close flush of workqueue */
3014 if (!rtnl_trylock()) {
3016 should_notify_peers = false;
3020 bond_ab_arp_commit(bond);
3026 should_notify_rtnl = bond_ab_arp_probe(bond);
3030 if (bond->params.arp_interval)
3031 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3033 if (should_notify_peers || should_notify_rtnl) {
3034 if (!rtnl_trylock())
3037 if (should_notify_peers)
3038 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3040 if (should_notify_rtnl) {
3041 bond_slave_state_notify(bond);
3042 bond_slave_link_notify(bond);
3049 static void bond_arp_monitor(struct work_struct *work)
3051 struct bonding *bond = container_of(work, struct bonding,
3054 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3055 bond_activebackup_arp_mon(bond);
3057 bond_loadbalance_arp_mon(bond);
3060 /*-------------------------- netdev event handling --------------------------*/
3062 /* Change device name */
3063 static int bond_event_changename(struct bonding *bond)
3065 bond_remove_proc_entry(bond);
3066 bond_create_proc_entry(bond);
3068 bond_debug_reregister(bond);
3073 static int bond_master_netdev_event(unsigned long event,
3074 struct net_device *bond_dev)
3076 struct bonding *event_bond = netdev_priv(bond_dev);
3078 netdev_dbg(bond_dev, "%s called\n", __func__);
3081 case NETDEV_CHANGENAME:
3082 return bond_event_changename(event_bond);
3083 case NETDEV_UNREGISTER:
3084 bond_remove_proc_entry(event_bond);
3086 case NETDEV_REGISTER:
3087 bond_create_proc_entry(event_bond);
3096 static int bond_slave_netdev_event(unsigned long event,
3097 struct net_device *slave_dev)
3099 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3100 struct bonding *bond;
3101 struct net_device *bond_dev;
3103 /* A netdev event can be generated while enslaving a device
3104 * before netdev_rx_handler_register is called in which case
3105 * slave will be NULL
3108 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3112 bond_dev = slave->bond->dev;
3114 primary = rtnl_dereference(bond->primary_slave);
3116 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3119 case NETDEV_UNREGISTER:
3120 if (bond_dev->type != ARPHRD_ETHER)
3121 bond_release_and_destroy(bond_dev, slave_dev);
3123 __bond_release_one(bond_dev, slave_dev, false, true);
3127 /* For 802.3ad mode only:
3128 * Getting invalid Speed/Duplex values here will put slave
3129 * in weird state. Mark it as link-fail if the link was
3130 * previously up or link-down if it hasn't yet come up, and
3131 * let link-monitoring (miimon) set it right when correct
3132 * speeds/duplex are available.
3134 if (bond_update_speed_duplex(slave) &&
3135 BOND_MODE(bond) == BOND_MODE_8023AD) {
3136 if (slave->last_link_up)
3137 slave->link = BOND_LINK_FAIL;
3139 slave->link = BOND_LINK_DOWN;
3142 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3143 bond_3ad_adapter_speed_duplex_changed(slave);
3146 /* Refresh slave-array if applicable!
3147 * If the setup does not use miimon or arpmon (mode-specific!),
3148 * then these events will not cause the slave-array to be
3149 * refreshed. This will cause xmit to use a slave that is not
3150 * usable. Avoid such situation by refeshing the array at these
3151 * events. If these (miimon/arpmon) parameters are configured
3152 * then array gets refreshed twice and that should be fine!
3154 if (bond_mode_can_use_xmit_hash(bond))
3155 bond_update_slave_arr(bond, NULL);
3157 case NETDEV_CHANGEMTU:
3158 /* TODO: Should slaves be allowed to
3159 * independently alter their MTU? For
3160 * an active-backup bond, slaves need
3161 * not be the same type of device, so
3162 * MTUs may vary. For other modes,
3163 * slaves arguably should have the
3164 * same MTUs. To do this, we'd need to
3165 * take over the slave's change_mtu
3166 * function for the duration of their
3170 case NETDEV_CHANGENAME:
3171 /* we don't care if we don't have primary set */
3172 if (!bond_uses_primary(bond) ||
3173 !bond->params.primary[0])
3176 if (slave == primary) {
3177 /* slave's name changed - he's no longer primary */
3178 RCU_INIT_POINTER(bond->primary_slave, NULL);
3179 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3180 /* we have a new primary slave */
3181 rcu_assign_pointer(bond->primary_slave, slave);
3182 } else { /* we didn't change primary - exit */
3186 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3187 primary ? slave_dev->name : "none");
3190 bond_select_active_slave(bond);
3191 unblock_netpoll_tx();
3193 case NETDEV_FEAT_CHANGE:
3194 bond_compute_features(bond);
3196 case NETDEV_RESEND_IGMP:
3197 /* Propagate to master device */
3198 call_netdevice_notifiers(event, slave->bond->dev);
3207 /* bond_netdev_event: handle netdev notifier chain events.
3209 * This function receives events for the netdev chain. The caller (an
3210 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3211 * locks for us to safely manipulate the slave devices (RTNL lock,
3214 static int bond_netdev_event(struct notifier_block *this,
3215 unsigned long event, void *ptr)
3217 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3219 netdev_dbg(event_dev, "%s received %s\n",
3220 __func__, netdev_cmd_to_name(event));
3222 if (!(event_dev->priv_flags & IFF_BONDING))
3225 if (event_dev->flags & IFF_MASTER) {
3228 ret = bond_master_netdev_event(event, event_dev);
3229 if (ret != NOTIFY_DONE)
3233 if (event_dev->flags & IFF_SLAVE)
3234 return bond_slave_netdev_event(event, event_dev);
3239 static struct notifier_block bond_netdev_notifier = {
3240 .notifier_call = bond_netdev_event,
3243 /*---------------------------- Hashing Policies -----------------------------*/
3245 /* L2 hash helper */
3246 static inline u32 bond_eth_hash(struct sk_buff *skb)
3248 struct ethhdr *ep, hdr_tmp;
3250 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3252 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3256 /* Extract the appropriate headers based on bond's xmit policy */
3257 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3258 struct flow_keys *fk)
3260 const struct ipv6hdr *iph6;
3261 const struct iphdr *iph;
3262 int noff, proto = -1;
3264 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3265 return skb_flow_dissect_flow_keys(skb, fk, 0);
3267 fk->ports.ports = 0;
3268 noff = skb_network_offset(skb);
3269 if (skb->protocol == htons(ETH_P_IP)) {
3270 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3273 iph_to_flow_copy_v4addrs(fk, iph);
3274 noff += iph->ihl << 2;
3275 if (!ip_is_fragment(iph))
3276 proto = iph->protocol;
3277 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3278 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3280 iph6 = ipv6_hdr(skb);
3281 iph_to_flow_copy_v6addrs(fk, iph6);
3282 noff += sizeof(*iph6);
3283 proto = iph6->nexthdr;
3287 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3288 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3294 * bond_xmit_hash - generate a hash value based on the xmit policy
3295 * @bond: bonding device
3296 * @skb: buffer to use for headers
3298 * This function will extract the necessary headers from the skb buffer and use
3299 * them to generate a hash based on the xmit_policy set in the bonding device
3301 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3303 struct flow_keys flow;
3306 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3310 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3311 !bond_flow_dissect(bond, skb, &flow))
3312 return bond_eth_hash(skb);
3314 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3315 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3316 hash = bond_eth_hash(skb);
3318 hash = (__force u32)flow.ports.ports;
3319 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3320 (__force u32)flow_get_u32_src(&flow);
3321 hash ^= (hash >> 16);
3322 hash ^= (hash >> 8);
3327 /*-------------------------- Device entry points ----------------------------*/
3329 void bond_work_init_all(struct bonding *bond)
3331 INIT_DELAYED_WORK(&bond->mcast_work,
3332 bond_resend_igmp_join_requests_delayed);
3333 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3334 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3335 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3336 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3337 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3340 static void bond_work_cancel_all(struct bonding *bond)
3342 cancel_delayed_work_sync(&bond->mii_work);
3343 cancel_delayed_work_sync(&bond->arp_work);
3344 cancel_delayed_work_sync(&bond->alb_work);
3345 cancel_delayed_work_sync(&bond->ad_work);
3346 cancel_delayed_work_sync(&bond->mcast_work);
3347 cancel_delayed_work_sync(&bond->slave_arr_work);
3350 static int bond_open(struct net_device *bond_dev)
3352 struct bonding *bond = netdev_priv(bond_dev);
3353 struct list_head *iter;
3354 struct slave *slave;
3356 /* reset slave->backup and slave->inactive */
3357 if (bond_has_slaves(bond)) {
3358 bond_for_each_slave(bond, slave, iter) {
3359 if (bond_uses_primary(bond) &&
3360 slave != rcu_access_pointer(bond->curr_active_slave)) {
3361 bond_set_slave_inactive_flags(slave,
3362 BOND_SLAVE_NOTIFY_NOW);
3363 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3364 bond_set_slave_active_flags(slave,
3365 BOND_SLAVE_NOTIFY_NOW);
3370 if (bond_is_lb(bond)) {
3371 /* bond_alb_initialize must be called before the timer
3374 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3376 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3377 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3380 if (bond->params.miimon) /* link check interval, in milliseconds. */
3381 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3383 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3384 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3385 bond->recv_probe = bond_arp_rcv;
3388 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3389 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3390 /* register to receive LACPDUs */
3391 bond->recv_probe = bond_3ad_lacpdu_recv;
3392 bond_3ad_initiate_agg_selection(bond, 1);
3395 if (bond_mode_can_use_xmit_hash(bond))
3396 bond_update_slave_arr(bond, NULL);
3401 static int bond_close(struct net_device *bond_dev)
3403 struct bonding *bond = netdev_priv(bond_dev);
3405 bond_work_cancel_all(bond);
3406 bond->send_peer_notif = 0;
3407 if (bond_is_lb(bond))
3408 bond_alb_deinitialize(bond);
3409 bond->recv_probe = NULL;
3414 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3415 * that some drivers can provide 32bit values only.
3417 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3418 const struct rtnl_link_stats64 *_new,
3419 const struct rtnl_link_stats64 *_old)
3421 const u64 *new = (const u64 *)_new;
3422 const u64 *old = (const u64 *)_old;
3423 u64 *res = (u64 *)_res;
3426 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3429 s64 delta = nv - ov;
3431 /* detects if this particular field is 32bit only */
3432 if (((nv | ov) >> 32) == 0)
3433 delta = (s64)(s32)((u32)nv - (u32)ov);
3435 /* filter anomalies, some drivers reset their stats
3436 * at down/up events.
3443 static int bond_get_nest_level(struct net_device *bond_dev)
3445 struct bonding *bond = netdev_priv(bond_dev);
3447 return bond->nest_level;
3450 static void bond_get_stats(struct net_device *bond_dev,
3451 struct rtnl_link_stats64 *stats)
3453 struct bonding *bond = netdev_priv(bond_dev);
3454 struct rtnl_link_stats64 temp;
3455 struct list_head *iter;
3456 struct slave *slave;
3458 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3459 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3462 bond_for_each_slave_rcu(bond, slave, iter) {
3463 const struct rtnl_link_stats64 *new =
3464 dev_get_stats(slave->dev, &temp);
3466 bond_fold_stats(stats, new, &slave->slave_stats);
3468 /* save off the slave stats for the next run */
3469 memcpy(&slave->slave_stats, new, sizeof(*new));
3473 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3474 spin_unlock(&bond->stats_lock);
3477 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3479 struct bonding *bond = netdev_priv(bond_dev);
3480 struct net_device *slave_dev = NULL;
3481 struct ifbond k_binfo;
3482 struct ifbond __user *u_binfo = NULL;
3483 struct ifslave k_sinfo;
3484 struct ifslave __user *u_sinfo = NULL;
3485 struct mii_ioctl_data *mii = NULL;
3486 struct bond_opt_value newval;
3490 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3501 /* We do this again just in case we were called by SIOCGMIIREG
3502 * instead of SIOCGMIIPHY.
3508 if (mii->reg_num == 1) {
3510 if (netif_carrier_ok(bond->dev))
3511 mii->val_out = BMSR_LSTATUS;
3515 case BOND_INFO_QUERY_OLD:
3516 case SIOCBONDINFOQUERY:
3517 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3519 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3522 bond_info_query(bond_dev, &k_binfo);
3523 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3527 case BOND_SLAVE_INFO_QUERY_OLD:
3528 case SIOCBONDSLAVEINFOQUERY:
3529 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3531 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3534 res = bond_slave_info_query(bond_dev, &k_sinfo);
3536 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3544 net = dev_net(bond_dev);
3546 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3549 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3551 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
3557 case BOND_ENSLAVE_OLD:
3558 case SIOCBONDENSLAVE:
3559 res = bond_enslave(bond_dev, slave_dev, NULL);
3561 case BOND_RELEASE_OLD:
3562 case SIOCBONDRELEASE:
3563 res = bond_release(bond_dev, slave_dev);
3565 case BOND_SETHWADDR_OLD:
3566 case SIOCBONDSETHWADDR:
3567 res = bond_set_dev_addr(bond_dev, slave_dev);
3569 case BOND_CHANGE_ACTIVE_OLD:
3570 case SIOCBONDCHANGEACTIVE:
3571 bond_opt_initstr(&newval, slave_dev->name);
3572 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3582 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3584 struct bonding *bond = netdev_priv(bond_dev);
3586 if (change & IFF_PROMISC)
3587 bond_set_promiscuity(bond,
3588 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3590 if (change & IFF_ALLMULTI)
3591 bond_set_allmulti(bond,
3592 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3595 static void bond_set_rx_mode(struct net_device *bond_dev)
3597 struct bonding *bond = netdev_priv(bond_dev);
3598 struct list_head *iter;
3599 struct slave *slave;
3602 if (bond_uses_primary(bond)) {
3603 slave = rcu_dereference(bond->curr_active_slave);
3605 dev_uc_sync(slave->dev, bond_dev);
3606 dev_mc_sync(slave->dev, bond_dev);
3609 bond_for_each_slave_rcu(bond, slave, iter) {
3610 dev_uc_sync_multiple(slave->dev, bond_dev);
3611 dev_mc_sync_multiple(slave->dev, bond_dev);
3617 static int bond_neigh_init(struct neighbour *n)
3619 struct bonding *bond = netdev_priv(n->dev);
3620 const struct net_device_ops *slave_ops;
3621 struct neigh_parms parms;
3622 struct slave *slave;
3625 slave = bond_first_slave(bond);
3628 slave_ops = slave->dev->netdev_ops;
3629 if (!slave_ops->ndo_neigh_setup)
3632 parms.neigh_setup = NULL;
3633 parms.neigh_cleanup = NULL;
3634 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3638 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3639 * after the last slave has been detached. Assumes that all slaves
3640 * utilize the same neigh_cleanup (true at this writing as only user
3643 n->parms->neigh_cleanup = parms.neigh_cleanup;
3645 if (!parms.neigh_setup)
3648 return parms.neigh_setup(n);
3651 /* The bonding ndo_neigh_setup is called at init time beofre any
3652 * slave exists. So we must declare proxy setup function which will
3653 * be used at run time to resolve the actual slave neigh param setup.
3655 * It's also called by master devices (such as vlans) to setup their
3656 * underlying devices. In that case - do nothing, we're already set up from
3659 static int bond_neigh_setup(struct net_device *dev,
3660 struct neigh_parms *parms)
3662 /* modify only our neigh_parms */
3663 if (parms->dev == dev)
3664 parms->neigh_setup = bond_neigh_init;
3669 /* Change the MTU of all of a master's slaves to match the master */
3670 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3672 struct bonding *bond = netdev_priv(bond_dev);
3673 struct slave *slave, *rollback_slave;
3674 struct list_head *iter;
3677 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3679 bond_for_each_slave(bond, slave, iter) {
3680 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
3681 slave, slave->dev->netdev_ops->ndo_change_mtu);
3683 res = dev_set_mtu(slave->dev, new_mtu);
3686 /* If we failed to set the slave's mtu to the new value
3687 * we must abort the operation even in ACTIVE_BACKUP
3688 * mode, because if we allow the backup slaves to have
3689 * different mtu values than the active slave we'll
3690 * need to change their mtu when doing a failover. That
3691 * means changing their mtu from timer context, which
3692 * is probably not a good idea.
3694 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
3700 bond_dev->mtu = new_mtu;
3705 /* unwind from head to the slave that failed */
3706 bond_for_each_slave(bond, rollback_slave, iter) {
3709 if (rollback_slave == slave)
3712 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3714 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
3721 /* Change HW address
3723 * Note that many devices must be down to change the HW address, and
3724 * downing the master releases all slaves. We can make bonds full of
3725 * bonding devices to test this, however.
3727 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3729 struct bonding *bond = netdev_priv(bond_dev);
3730 struct slave *slave, *rollback_slave;
3731 struct sockaddr_storage *ss = addr, tmp_ss;
3732 struct list_head *iter;
3735 if (BOND_MODE(bond) == BOND_MODE_ALB)
3736 return bond_alb_set_mac_address(bond_dev, addr);
3739 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
3741 /* If fail_over_mac is enabled, do nothing and return success.
3742 * Returning an error causes ifenslave to fail.
3744 if (bond->params.fail_over_mac &&
3745 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3748 if (!is_valid_ether_addr(ss->__data))
3749 return -EADDRNOTAVAIL;
3751 bond_for_each_slave(bond, slave, iter) {
3752 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
3754 res = dev_set_mac_address(slave->dev, addr, NULL);
3756 /* TODO: consider downing the slave
3758 * User should expect communications
3759 * breakage anyway until ARP finish
3762 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
3769 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3773 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3774 tmp_ss.ss_family = bond_dev->type;
3776 /* unwind from head to the slave that failed */
3777 bond_for_each_slave(bond, rollback_slave, iter) {
3780 if (rollback_slave == slave)
3783 tmp_res = dev_set_mac_address(rollback_slave->dev,
3784 (struct sockaddr *)&tmp_ss, NULL);
3786 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
3795 * bond_xmit_slave_id - transmit skb through slave with slave_id
3796 * @bond: bonding device that is transmitting
3797 * @skb: buffer to transmit
3798 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3800 * This function tries to transmit through slave with slave_id but in case
3801 * it fails, it tries to find the first available slave for transmission.
3802 * The skb is consumed in all cases, thus the function is void.
3804 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3806 struct list_head *iter;
3807 struct slave *slave;
3810 /* Here we start from the slave with slave_id */
3811 bond_for_each_slave_rcu(bond, slave, iter) {
3813 if (bond_slave_can_tx(slave)) {
3814 bond_dev_queue_xmit(bond, skb, slave->dev);
3820 /* Here we start from the first slave up to slave_id */
3822 bond_for_each_slave_rcu(bond, slave, iter) {
3825 if (bond_slave_can_tx(slave)) {
3826 bond_dev_queue_xmit(bond, skb, slave->dev);
3830 /* no slave that can tx has been found */
3831 bond_tx_drop(bond->dev, skb);
3835 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3836 * @bond: bonding device to use
3838 * Based on the value of the bonding device's packets_per_slave parameter
3839 * this function generates a slave id, which is usually used as the next
3840 * slave to transmit through.
3842 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3845 struct reciprocal_value reciprocal_packets_per_slave;
3846 int packets_per_slave = bond->params.packets_per_slave;
3848 switch (packets_per_slave) {
3850 slave_id = prandom_u32();
3853 slave_id = bond->rr_tx_counter;
3856 reciprocal_packets_per_slave =
3857 bond->params.reciprocal_packets_per_slave;
3858 slave_id = reciprocal_divide(bond->rr_tx_counter,
3859 reciprocal_packets_per_slave);
3862 bond->rr_tx_counter++;
3867 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3868 struct net_device *bond_dev)
3870 struct bonding *bond = netdev_priv(bond_dev);
3871 struct slave *slave;
3875 /* Start with the curr_active_slave that joined the bond as the
3876 * default for sending IGMP traffic. For failover purposes one
3877 * needs to maintain some consistency for the interface that will
3878 * send the join/membership reports. The curr_active_slave found
3879 * will send all of this type of traffic.
3881 if (skb->protocol == htons(ETH_P_IP)) {
3882 int noff = skb_network_offset(skb);
3885 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3889 if (iph->protocol == IPPROTO_IGMP) {
3890 slave = rcu_dereference(bond->curr_active_slave);
3892 bond_dev_queue_xmit(bond, skb, slave->dev);
3894 bond_xmit_slave_id(bond, skb, 0);
3895 return NETDEV_TX_OK;
3900 slave_cnt = READ_ONCE(bond->slave_cnt);
3901 if (likely(slave_cnt)) {
3902 slave_id = bond_rr_gen_slave_id(bond);
3903 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3905 bond_tx_drop(bond_dev, skb);
3907 return NETDEV_TX_OK;
3910 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3911 * the bond has a usable interface.
3913 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3914 struct net_device *bond_dev)
3916 struct bonding *bond = netdev_priv(bond_dev);
3917 struct slave *slave;
3919 slave = rcu_dereference(bond->curr_active_slave);
3921 bond_dev_queue_xmit(bond, skb, slave->dev);
3923 bond_tx_drop(bond_dev, skb);
3925 return NETDEV_TX_OK;
3928 /* Use this to update slave_array when (a) it's not appropriate to update
3929 * slave_array right away (note that update_slave_array() may sleep)
3930 * and / or (b) RTNL is not held.
3932 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3934 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3937 /* Slave array work handler. Holds only RTNL */
3938 static void bond_slave_arr_handler(struct work_struct *work)
3940 struct bonding *bond = container_of(work, struct bonding,
3941 slave_arr_work.work);
3944 if (!rtnl_trylock())
3947 ret = bond_update_slave_arr(bond, NULL);
3950 pr_warn_ratelimited("Failed to update slave array from WT\n");
3956 bond_slave_arr_work_rearm(bond, 1);
3959 /* Build the usable slaves array in control path for modes that use xmit-hash
3960 * to determine the slave interface -
3961 * (a) BOND_MODE_8023AD
3963 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3965 * The caller is expected to hold RTNL only and NO other lock!
3967 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3969 struct slave *slave;
3970 struct list_head *iter;
3971 struct bond_up_slave *new_arr, *old_arr;
3975 #ifdef CONFIG_LOCKDEP
3976 WARN_ON(lockdep_is_held(&bond->mode_lock));
3979 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3983 pr_err("Failed to build slave-array.\n");
3986 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3987 struct ad_info ad_info;
3989 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3990 pr_debug("bond_3ad_get_active_agg_info failed\n");
3991 kfree_rcu(new_arr, rcu);
3992 /* No active aggragator means it's not safe to use
3993 * the previous array.
3995 old_arr = rtnl_dereference(bond->slave_arr);
3997 RCU_INIT_POINTER(bond->slave_arr, NULL);
3998 kfree_rcu(old_arr, rcu);
4002 agg_id = ad_info.aggregator_id;
4004 bond_for_each_slave(bond, slave, iter) {
4005 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4006 struct aggregator *agg;
4008 agg = SLAVE_AD_INFO(slave)->port.aggregator;
4009 if (!agg || agg->aggregator_identifier != agg_id)
4012 if (!bond_slave_can_tx(slave))
4014 if (skipslave == slave)
4017 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4020 new_arr->arr[new_arr->count++] = slave;
4023 old_arr = rtnl_dereference(bond->slave_arr);
4024 rcu_assign_pointer(bond->slave_arr, new_arr);
4026 kfree_rcu(old_arr, rcu);
4028 if (ret != 0 && skipslave) {
4031 /* Rare situation where caller has asked to skip a specific
4032 * slave but allocation failed (most likely!). BTW this is
4033 * only possible when the call is initiated from
4034 * __bond_release_one(). In this situation; overwrite the
4035 * skipslave entry in the array with the last entry from the
4036 * array to avoid a situation where the xmit path may choose
4037 * this to-be-skipped slave to send a packet out.
4039 old_arr = rtnl_dereference(bond->slave_arr);
4040 for (idx = 0; idx < old_arr->count; idx++) {
4041 if (skipslave == old_arr->arr[idx]) {
4043 old_arr->arr[old_arr->count-1];
4052 /* Use this Xmit function for 3AD as well as XOR modes. The current
4053 * usable slave array is formed in the control path. The xmit function
4054 * just calculates hash and sends the packet out.
4056 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4057 struct net_device *dev)
4059 struct bonding *bond = netdev_priv(dev);
4060 struct slave *slave;
4061 struct bond_up_slave *slaves;
4064 slaves = rcu_dereference(bond->slave_arr);
4065 count = slaves ? READ_ONCE(slaves->count) : 0;
4066 if (likely(count)) {
4067 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4068 bond_dev_queue_xmit(bond, skb, slave->dev);
4070 bond_tx_drop(dev, skb);
4073 return NETDEV_TX_OK;
4076 /* in broadcast mode, we send everything to all usable interfaces. */
4077 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4078 struct net_device *bond_dev)
4080 struct bonding *bond = netdev_priv(bond_dev);
4081 struct slave *slave = NULL;
4082 struct list_head *iter;
4084 bond_for_each_slave_rcu(bond, slave, iter) {
4085 if (bond_is_last_slave(bond, slave))
4087 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4088 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4091 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4092 bond_dev->name, __func__);
4095 bond_dev_queue_xmit(bond, skb2, slave->dev);
4098 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4099 bond_dev_queue_xmit(bond, skb, slave->dev);
4101 bond_tx_drop(bond_dev, skb);
4103 return NETDEV_TX_OK;
4106 /*------------------------- Device initialization ---------------------------*/
4108 /* Lookup the slave that corresponds to a qid */
4109 static inline int bond_slave_override(struct bonding *bond,
4110 struct sk_buff *skb)
4112 struct slave *slave = NULL;
4113 struct list_head *iter;
4115 if (!skb_rx_queue_recorded(skb))
4118 /* Find out if any slaves have the same mapping as this skb. */
4119 bond_for_each_slave_rcu(bond, slave, iter) {
4120 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4121 if (bond_slave_is_up(slave) &&
4122 slave->link == BOND_LINK_UP) {
4123 bond_dev_queue_xmit(bond, skb, slave->dev);
4126 /* If the slave isn't UP, use default transmit policy. */
4135 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4136 struct net_device *sb_dev)
4138 /* This helper function exists to help dev_pick_tx get the correct
4139 * destination queue. Using a helper function skips a call to
4140 * skb_tx_hash and will put the skbs in the queue we expect on their
4141 * way down to the bonding driver.
4143 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4145 /* Save the original txq to restore before passing to the driver */
4146 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4148 if (unlikely(txq >= dev->real_num_tx_queues)) {
4150 txq -= dev->real_num_tx_queues;
4151 } while (txq >= dev->real_num_tx_queues);
4156 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4158 struct bonding *bond = netdev_priv(dev);
4160 if (bond_should_override_tx_queue(bond) &&
4161 !bond_slave_override(bond, skb))
4162 return NETDEV_TX_OK;
4164 switch (BOND_MODE(bond)) {
4165 case BOND_MODE_ROUNDROBIN:
4166 return bond_xmit_roundrobin(skb, dev);
4167 case BOND_MODE_ACTIVEBACKUP:
4168 return bond_xmit_activebackup(skb, dev);
4169 case BOND_MODE_8023AD:
4171 return bond_3ad_xor_xmit(skb, dev);
4172 case BOND_MODE_BROADCAST:
4173 return bond_xmit_broadcast(skb, dev);
4175 return bond_alb_xmit(skb, dev);
4177 return bond_tlb_xmit(skb, dev);
4179 /* Should never happen, mode already checked */
4180 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4182 bond_tx_drop(dev, skb);
4183 return NETDEV_TX_OK;
4187 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4189 struct bonding *bond = netdev_priv(dev);
4190 netdev_tx_t ret = NETDEV_TX_OK;
4192 /* If we risk deadlock from transmitting this in the
4193 * netpoll path, tell netpoll to queue the frame for later tx
4195 if (unlikely(is_netpoll_tx_blocked(dev)))
4196 return NETDEV_TX_BUSY;
4199 if (bond_has_slaves(bond))
4200 ret = __bond_start_xmit(skb, dev);
4202 bond_tx_drop(dev, skb);
4208 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4209 struct ethtool_link_ksettings *cmd)
4211 struct bonding *bond = netdev_priv(bond_dev);
4212 unsigned long speed = 0;
4213 struct list_head *iter;
4214 struct slave *slave;
4216 cmd->base.duplex = DUPLEX_UNKNOWN;
4217 cmd->base.port = PORT_OTHER;
4219 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4220 * do not need to check mode. Though link speed might not represent
4221 * the true receive or transmit bandwidth (not all modes are symmetric)
4222 * this is an accurate maximum.
4224 bond_for_each_slave(bond, slave, iter) {
4225 if (bond_slave_can_tx(slave)) {
4226 if (slave->speed != SPEED_UNKNOWN)
4227 speed += slave->speed;
4228 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4229 slave->duplex != DUPLEX_UNKNOWN)
4230 cmd->base.duplex = slave->duplex;
4233 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4238 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4239 struct ethtool_drvinfo *drvinfo)
4241 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4242 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4243 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4247 static const struct ethtool_ops bond_ethtool_ops = {
4248 .get_drvinfo = bond_ethtool_get_drvinfo,
4249 .get_link = ethtool_op_get_link,
4250 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4253 static const struct net_device_ops bond_netdev_ops = {
4254 .ndo_init = bond_init,
4255 .ndo_uninit = bond_uninit,
4256 .ndo_open = bond_open,
4257 .ndo_stop = bond_close,
4258 .ndo_start_xmit = bond_start_xmit,
4259 .ndo_select_queue = bond_select_queue,
4260 .ndo_get_stats64 = bond_get_stats,
4261 .ndo_do_ioctl = bond_do_ioctl,
4262 .ndo_change_rx_flags = bond_change_rx_flags,
4263 .ndo_set_rx_mode = bond_set_rx_mode,
4264 .ndo_change_mtu = bond_change_mtu,
4265 .ndo_set_mac_address = bond_set_mac_address,
4266 .ndo_neigh_setup = bond_neigh_setup,
4267 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4268 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4269 .ndo_get_lock_subclass = bond_get_nest_level,
4270 #ifdef CONFIG_NET_POLL_CONTROLLER
4271 .ndo_netpoll_setup = bond_netpoll_setup,
4272 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4273 .ndo_poll_controller = bond_poll_controller,
4275 .ndo_add_slave = bond_enslave,
4276 .ndo_del_slave = bond_release,
4277 .ndo_fix_features = bond_fix_features,
4278 .ndo_features_check = passthru_features_check,
4281 static const struct device_type bond_type = {
4285 static void bond_destructor(struct net_device *bond_dev)
4287 struct bonding *bond = netdev_priv(bond_dev);
4289 destroy_workqueue(bond->wq);
4292 void bond_setup(struct net_device *bond_dev)
4294 struct bonding *bond = netdev_priv(bond_dev);
4296 spin_lock_init(&bond->mode_lock);
4297 spin_lock_init(&bond->stats_lock);
4298 bond->params = bonding_defaults;
4300 /* Initialize pointers */
4301 bond->dev = bond_dev;
4303 /* Initialize the device entry points */
4304 ether_setup(bond_dev);
4305 bond_dev->max_mtu = ETH_MAX_MTU;
4306 bond_dev->netdev_ops = &bond_netdev_ops;
4307 bond_dev->ethtool_ops = &bond_ethtool_ops;
4309 bond_dev->needs_free_netdev = true;
4310 bond_dev->priv_destructor = bond_destructor;
4312 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4314 /* Initialize the device options */
4315 bond_dev->flags |= IFF_MASTER;
4316 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4317 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4319 /* don't acquire bond device's netif_tx_lock when transmitting */
4320 bond_dev->features |= NETIF_F_LLTX;
4322 /* By default, we declare the bond to be fully
4323 * VLAN hardware accelerated capable. Special
4324 * care is taken in the various xmit functions
4325 * when there are slaves that are not hw accel
4329 /* Don't allow bond devices to change network namespaces. */
4330 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4332 bond_dev->hw_features = BOND_VLAN_FEATURES |
4333 NETIF_F_HW_VLAN_CTAG_RX |
4334 NETIF_F_HW_VLAN_CTAG_FILTER;
4336 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4337 bond_dev->features |= bond_dev->hw_features;
4338 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4341 /* Destroy a bonding device.
4342 * Must be under rtnl_lock when this function is called.
4344 static void bond_uninit(struct net_device *bond_dev)
4346 struct bonding *bond = netdev_priv(bond_dev);
4347 struct list_head *iter;
4348 struct slave *slave;
4349 struct bond_up_slave *arr;
4351 bond_netpoll_cleanup(bond_dev);
4353 /* Release the bonded slaves */
4354 bond_for_each_slave(bond, slave, iter)
4355 __bond_release_one(bond_dev, slave->dev, true, true);
4356 netdev_info(bond_dev, "Released all slaves\n");
4358 arr = rtnl_dereference(bond->slave_arr);
4360 RCU_INIT_POINTER(bond->slave_arr, NULL);
4361 kfree_rcu(arr, rcu);
4364 list_del(&bond->bond_list);
4366 bond_debug_unregister(bond);
4369 /*------------------------- Module initialization ---------------------------*/
4371 static int bond_check_params(struct bond_params *params)
4373 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4374 struct bond_opt_value newval;
4375 const struct bond_opt_value *valptr;
4376 int arp_all_targets_value = 0;
4377 u16 ad_actor_sys_prio = 0;
4378 u16 ad_user_port_key = 0;
4379 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4381 int bond_mode = BOND_MODE_ROUNDROBIN;
4382 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4386 /* Convert string parameters. */
4388 bond_opt_initstr(&newval, mode);
4389 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4391 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4394 bond_mode = valptr->value;
4397 if (xmit_hash_policy) {
4398 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4399 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4400 bond_mode == BOND_MODE_BROADCAST) {
4401 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4402 bond_mode_name(bond_mode));
4404 bond_opt_initstr(&newval, xmit_hash_policy);
4405 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4408 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4412 xmit_hashtype = valptr->value;
4417 if (bond_mode != BOND_MODE_8023AD) {
4418 pr_info("lacp_rate param is irrelevant in mode %s\n",
4419 bond_mode_name(bond_mode));
4421 bond_opt_initstr(&newval, lacp_rate);
4422 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4425 pr_err("Error: Invalid lacp rate \"%s\"\n",
4429 lacp_fast = valptr->value;
4434 bond_opt_initstr(&newval, ad_select);
4435 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4438 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4441 params->ad_select = valptr->value;
4442 if (bond_mode != BOND_MODE_8023AD)
4443 pr_warn("ad_select param only affects 802.3ad mode\n");
4445 params->ad_select = BOND_AD_STABLE;
4448 if (max_bonds < 0) {
4449 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4450 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4451 max_bonds = BOND_DEFAULT_MAX_BONDS;
4455 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4461 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4466 if (downdelay < 0) {
4467 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4468 downdelay, INT_MAX);
4472 if ((use_carrier != 0) && (use_carrier != 1)) {
4473 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4478 if (num_peer_notif < 0 || num_peer_notif > 255) {
4479 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4484 /* reset values for 802.3ad/TLB/ALB */
4485 if (!bond_mode_uses_arp(bond_mode)) {
4487 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");
4488 pr_warn("Forcing miimon to 100msec\n");
4489 miimon = BOND_DEFAULT_MIIMON;
4493 if (tx_queues < 1 || tx_queues > 255) {
4494 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4495 tx_queues, BOND_DEFAULT_TX_QUEUES);
4496 tx_queues = BOND_DEFAULT_TX_QUEUES;
4499 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4500 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4502 all_slaves_active = 0;
4505 if (resend_igmp < 0 || resend_igmp > 255) {
4506 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4507 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4508 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4511 bond_opt_initval(&newval, packets_per_slave);
4512 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4513 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4514 packets_per_slave, USHRT_MAX);
4515 packets_per_slave = 1;
4518 if (bond_mode == BOND_MODE_ALB) {
4519 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",
4524 if (updelay || downdelay) {
4525 /* just warn the user the up/down delay will have
4526 * no effect since miimon is zero...
4528 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",
4529 updelay, downdelay);
4532 /* don't allow arp monitoring */
4534 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4535 miimon, arp_interval);
4539 if ((updelay % miimon) != 0) {
4540 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4541 updelay, miimon, (updelay / miimon) * miimon);
4546 if ((downdelay % miimon) != 0) {
4547 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4549 (downdelay / miimon) * miimon);
4552 downdelay /= miimon;
4555 if (arp_interval < 0) {
4556 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4557 arp_interval, INT_MAX);
4561 for (arp_ip_count = 0, i = 0;
4562 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4565 /* not a complete check, but good enough to catch mistakes */
4566 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4567 !bond_is_ip_target_ok(ip)) {
4568 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4572 if (bond_get_targets_ip(arp_target, ip) == -1)
4573 arp_target[arp_ip_count++] = ip;
4575 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4580 if (arp_interval && !arp_ip_count) {
4581 /* don't allow arping if no arp_ip_target given... */
4582 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4588 if (!arp_interval) {
4589 pr_err("arp_validate requires arp_interval\n");
4593 bond_opt_initstr(&newval, arp_validate);
4594 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4597 pr_err("Error: invalid arp_validate \"%s\"\n",
4601 arp_validate_value = valptr->value;
4603 arp_validate_value = 0;
4606 if (arp_all_targets) {
4607 bond_opt_initstr(&newval, arp_all_targets);
4608 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4611 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4613 arp_all_targets_value = 0;
4615 arp_all_targets_value = valptr->value;
4620 pr_info("MII link monitoring set to %d ms\n", miimon);
4621 } else if (arp_interval) {
4622 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4623 arp_validate_value);
4624 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4625 arp_interval, valptr->string, arp_ip_count);
4627 for (i = 0; i < arp_ip_count; i++)
4628 pr_cont(" %s", arp_ip_target[i]);
4632 } else if (max_bonds) {
4633 /* miimon and arp_interval not set, we need one so things
4634 * work as expected, see bonding.txt for details
4636 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");
4639 if (primary && !bond_mode_uses_primary(bond_mode)) {
4640 /* currently, using a primary only makes sense
4641 * in active backup, TLB or ALB modes
4643 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4644 primary, bond_mode_name(bond_mode));
4648 if (primary && primary_reselect) {
4649 bond_opt_initstr(&newval, primary_reselect);
4650 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4653 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4657 primary_reselect_value = valptr->value;
4659 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4662 if (fail_over_mac) {
4663 bond_opt_initstr(&newval, fail_over_mac);
4664 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4667 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4671 fail_over_mac_value = valptr->value;
4672 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4673 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4675 fail_over_mac_value = BOND_FOM_NONE;
4678 bond_opt_initstr(&newval, "default");
4679 valptr = bond_opt_parse(
4680 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4683 pr_err("Error: No ad_actor_sys_prio default value");
4686 ad_actor_sys_prio = valptr->value;
4688 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4691 pr_err("Error: No ad_user_port_key default value");
4694 ad_user_port_key = valptr->value;
4696 bond_opt_initstr(&newval, "default");
4697 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4699 pr_err("Error: No tlb_dynamic_lb default value");
4702 tlb_dynamic_lb = valptr->value;
4704 if (lp_interval == 0) {
4705 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4706 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4707 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4710 /* fill params struct with the proper values */
4711 params->mode = bond_mode;
4712 params->xmit_policy = xmit_hashtype;
4713 params->miimon = miimon;
4714 params->num_peer_notif = num_peer_notif;
4715 params->arp_interval = arp_interval;
4716 params->arp_validate = arp_validate_value;
4717 params->arp_all_targets = arp_all_targets_value;
4718 params->updelay = updelay;
4719 params->downdelay = downdelay;
4720 params->peer_notif_delay = 0;
4721 params->use_carrier = use_carrier;
4722 params->lacp_fast = lacp_fast;
4723 params->primary[0] = 0;
4724 params->primary_reselect = primary_reselect_value;
4725 params->fail_over_mac = fail_over_mac_value;
4726 params->tx_queues = tx_queues;
4727 params->all_slaves_active = all_slaves_active;
4728 params->resend_igmp = resend_igmp;
4729 params->min_links = min_links;
4730 params->lp_interval = lp_interval;
4731 params->packets_per_slave = packets_per_slave;
4732 params->tlb_dynamic_lb = tlb_dynamic_lb;
4733 params->ad_actor_sys_prio = ad_actor_sys_prio;
4734 eth_zero_addr(params->ad_actor_system);
4735 params->ad_user_port_key = ad_user_port_key;
4736 if (packets_per_slave > 0) {
4737 params->reciprocal_packets_per_slave =
4738 reciprocal_value(packets_per_slave);
4740 /* reciprocal_packets_per_slave is unused if
4741 * packets_per_slave is 0 or 1, just initialize it
4743 params->reciprocal_packets_per_slave =
4744 (struct reciprocal_value) { 0 };
4748 strncpy(params->primary, primary, IFNAMSIZ);
4749 params->primary[IFNAMSIZ - 1] = 0;
4752 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4757 /* Called from registration process */
4758 static int bond_init(struct net_device *bond_dev)
4760 struct bonding *bond = netdev_priv(bond_dev);
4761 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4763 netdev_dbg(bond_dev, "Begin bond_init\n");
4765 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4769 bond->nest_level = SINGLE_DEPTH_NESTING;
4770 netdev_lockdep_set_classes(bond_dev);
4772 list_add_tail(&bond->bond_list, &bn->dev_list);
4774 bond_prepare_sysfs_group(bond);
4776 bond_debug_register(bond);
4778 /* Ensure valid dev_addr */
4779 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4780 bond_dev->addr_assign_type == NET_ADDR_PERM)
4781 eth_hw_addr_random(bond_dev);
4786 unsigned int bond_get_num_tx_queues(void)
4791 /* Create a new bond based on the specified name and bonding parameters.
4792 * If name is NULL, obtain a suitable "bond%d" name for us.
4793 * Caller must NOT hold rtnl_lock; we need to release it here before we
4794 * set up our sysfs entries.
4796 int bond_create(struct net *net, const char *name)
4798 struct net_device *bond_dev;
4799 struct bonding *bond;
4800 struct alb_bond_info *bond_info;
4805 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4806 name ? name : "bond%d", NET_NAME_UNKNOWN,
4807 bond_setup, tx_queues);
4809 pr_err("%s: eek! can't alloc netdev!\n", name);
4815 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4816 * It is set to 0 by default which is wrong.
4818 bond = netdev_priv(bond_dev);
4819 bond_info = &(BOND_ALB_INFO(bond));
4820 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4822 dev_net_set(bond_dev, net);
4823 bond_dev->rtnl_link_ops = &bond_link_ops;
4825 res = register_netdevice(bond_dev);
4827 netif_carrier_off(bond_dev);
4829 bond_work_init_all(bond);
4833 free_netdev(bond_dev);
4837 static int __net_init bond_net_init(struct net *net)
4839 struct bond_net *bn = net_generic(net, bond_net_id);
4842 INIT_LIST_HEAD(&bn->dev_list);
4844 bond_create_proc_dir(bn);
4845 bond_create_sysfs(bn);
4850 static void __net_exit bond_net_exit(struct net *net)
4852 struct bond_net *bn = net_generic(net, bond_net_id);
4853 struct bonding *bond, *tmp_bond;
4856 bond_destroy_sysfs(bn);
4858 /* Kill off any bonds created after unregistering bond rtnl ops */
4860 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4861 unregister_netdevice_queue(bond->dev, &list);
4862 unregister_netdevice_many(&list);
4865 bond_destroy_proc_dir(bn);
4868 static struct pernet_operations bond_net_ops = {
4869 .init = bond_net_init,
4870 .exit = bond_net_exit,
4872 .size = sizeof(struct bond_net),
4875 static int __init bonding_init(void)
4880 pr_info("%s", bond_version);
4882 res = bond_check_params(&bonding_defaults);
4886 res = register_pernet_subsys(&bond_net_ops);
4890 res = bond_netlink_init();
4894 bond_create_debugfs();
4896 for (i = 0; i < max_bonds; i++) {
4897 res = bond_create(&init_net, NULL);
4902 register_netdevice_notifier(&bond_netdev_notifier);
4906 bond_destroy_debugfs();
4907 bond_netlink_fini();
4909 unregister_pernet_subsys(&bond_net_ops);
4914 static void __exit bonding_exit(void)
4916 unregister_netdevice_notifier(&bond_netdev_notifier);
4918 bond_destroy_debugfs();
4920 bond_netlink_fini();
4921 unregister_pernet_subsys(&bond_net_ops);
4923 #ifdef CONFIG_NET_POLL_CONTROLLER
4924 /* Make sure we don't have an imbalance on our netpoll blocking */
4925 WARN_ON(atomic_read(&netpoll_block_tx));
4929 module_init(bonding_init);
4930 module_exit(bonding_exit);
4931 MODULE_LICENSE("GPL");
4932 MODULE_VERSION(DRV_VERSION);
4933 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4934 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");