2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 #include "bonding_priv.h"
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 static int packets_per_slave = 1;
114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
116 module_param(max_bonds, int, 0);
117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118 module_param(tx_queues, int, 0);
119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126 module_param(miimon, int, 0);
127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128 module_param(updelay, int, 0);
129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130 module_param(downdelay, int, 0);
131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
133 module_param(use_carrier, int, 0);
134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136 module_param(mode, charp, 0);
137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141 module_param(primary, charp, 0);
142 MODULE_PARM_DESC(primary, "Primary network device to use");
143 module_param(primary_reselect, charp, 0);
144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
149 "2 for only on active slave "
151 module_param(lacp_rate, charp, 0);
152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154 module_param(ad_select, charp, 0);
155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
158 module_param(min_links, int, 0);
159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166 module_param(arp_interval, int, 0);
167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168 module_param_array(arp_ip_target, charp, NULL, 0);
169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170 module_param(arp_validate, charp, 0);
171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174 module_param(arp_all_targets, charp, 0);
175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac, charp, 0);
177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180 module_param(all_slaves_active, int, 0);
181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184 module_param(resend_igmp, int, 0);
185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
187 module_param(packets_per_slave, int, 0);
188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191 module_param(lp_interval, uint, 0);
192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
196 /*----------------------------- Global variables ----------------------------*/
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
202 unsigned int bond_net_id __read_mostly;
204 /*-------------------------- Forward declarations ---------------------------*/
206 static int bond_init(struct net_device *bond_dev);
207 static void bond_uninit(struct net_device *bond_dev);
208 static void bond_get_stats(struct net_device *bond_dev,
209 struct rtnl_link_stats64 *stats);
210 static void bond_slave_arr_handler(struct work_struct *work);
211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
213 static void bond_netdev_notify_work(struct work_struct *work);
215 /*---------------------------- General routines -----------------------------*/
217 const char *bond_mode_name(int mode)
219 static const char *names[] = {
220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
224 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
225 [BOND_MODE_TLB] = "transmit load balancing",
226 [BOND_MODE_ALB] = "adaptive load balancing",
229 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
235 /*---------------------------------- VLAN -----------------------------------*/
238 * bond_dev_queue_xmit - Prepare skb for xmit.
240 * @bond: bond device that got this skb for tx.
241 * @skb: hw accel VLAN tagged skb to transmit
242 * @slave_dev: slave that is supposed to xmit this skbuff
244 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
245 struct net_device *slave_dev)
247 skb->dev = slave_dev;
249 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
250 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
251 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
253 if (unlikely(netpoll_tx_running(bond->dev)))
254 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
259 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
260 * We don't protect the slave list iteration with a lock because:
261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
281 __be16 proto, u16 vid)
283 struct bonding *bond = netdev_priv(bond_dev);
284 struct slave *slave, *rollback_slave;
285 struct list_head *iter;
288 bond_for_each_slave(bond, slave, iter) {
289 res = vlan_vid_add(slave->dev, proto, vid);
297 /* unwind to the slave that failed */
298 bond_for_each_slave(bond, rollback_slave, iter) {
299 if (rollback_slave == slave)
302 vlan_vid_del(rollback_slave->dev, proto, vid);
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
314 __be16 proto, u16 vid)
316 struct bonding *bond = netdev_priv(bond_dev);
317 struct list_head *iter;
320 bond_for_each_slave(bond, slave, iter)
321 vlan_vid_del(slave->dev, proto, vid);
323 if (bond_is_lb(bond))
324 bond_alb_clear_vlan(bond, vid);
329 /*------------------------------- Link status -------------------------------*/
331 /* Set the carrier state for the master according to the state of its
332 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
333 * do special 802.3ad magic.
335 * Returns zero if carrier state does not change, nonzero if it does.
337 int bond_set_carrier(struct bonding *bond)
339 struct list_head *iter;
342 if (!bond_has_slaves(bond))
345 if (BOND_MODE(bond) == BOND_MODE_8023AD)
346 return bond_3ad_set_carrier(bond);
348 bond_for_each_slave(bond, slave, iter) {
349 if (slave->link == BOND_LINK_UP) {
350 if (!netif_carrier_ok(bond->dev)) {
351 netif_carrier_on(bond->dev);
359 if (netif_carrier_ok(bond->dev)) {
360 netif_carrier_off(bond->dev);
366 /* Get link speed and duplex from the slave's base driver
367 * using ethtool. If for some reason the call fails or the
368 * values are invalid, set speed and duplex to -1,
369 * and return. Return 1 if speed or duplex settings are
370 * UNKNOWN; 0 otherwise.
372 static int bond_update_speed_duplex(struct slave *slave)
374 struct net_device *slave_dev = slave->dev;
375 struct ethtool_link_ksettings ecmd;
378 slave->speed = SPEED_UNKNOWN;
379 slave->duplex = DUPLEX_UNKNOWN;
381 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
384 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
386 switch (ecmd.base.duplex) {
394 slave->speed = ecmd.base.speed;
395 slave->duplex = ecmd.base.duplex;
400 const char *bond_slave_link_status(s8 link)
416 /* if <dev> supports MII link status reporting, check its link status.
418 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
419 * depending upon the setting of the use_carrier parameter.
421 * Return either BMSR_LSTATUS, meaning that the link is up (or we
422 * can't tell and just pretend it is), or 0, meaning that the link is
425 * If reporting is non-zero, instead of faking link up, return -1 if
426 * both ETHTOOL and MII ioctls fail (meaning the device does not
427 * support them). If use_carrier is set, return whatever it says.
428 * It'd be nice if there was a good way to tell if a driver supports
429 * netif_carrier, but there really isn't.
431 static int bond_check_dev_link(struct bonding *bond,
432 struct net_device *slave_dev, int reporting)
434 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
435 int (*ioctl)(struct net_device *, struct ifreq *, int);
437 struct mii_ioctl_data *mii;
439 if (!reporting && !netif_running(slave_dev))
442 if (bond->params.use_carrier)
443 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
445 /* Try to get link status using Ethtool first. */
446 if (slave_dev->ethtool_ops->get_link)
447 return slave_dev->ethtool_ops->get_link(slave_dev) ?
450 /* Ethtool can't be used, fallback to MII ioctls. */
451 ioctl = slave_ops->ndo_do_ioctl;
453 /* TODO: set pointer to correct ioctl on a per team member
454 * bases to make this more efficient. that is, once
455 * we determine the correct ioctl, we will always
456 * call it and not the others for that team
460 /* We cannot assume that SIOCGMIIPHY will also read a
461 * register; not all network drivers (e.g., e100)
465 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
466 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
468 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
469 mii->reg_num = MII_BMSR;
470 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
471 return mii->val_out & BMSR_LSTATUS;
475 /* If reporting, report that either there's no dev->do_ioctl,
476 * or both SIOCGMIIREG and get_link failed (meaning that we
477 * cannot report link status). If not reporting, pretend
480 return reporting ? -1 : BMSR_LSTATUS;
483 /*----------------------------- Multicast list ------------------------------*/
485 /* Push the promiscuity flag down to appropriate slaves */
486 static int bond_set_promiscuity(struct bonding *bond, int inc)
488 struct list_head *iter;
491 if (bond_uses_primary(bond)) {
492 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
495 err = dev_set_promiscuity(curr_active->dev, inc);
499 bond_for_each_slave(bond, slave, iter) {
500 err = dev_set_promiscuity(slave->dev, inc);
508 /* Push the allmulti flag down to all slaves */
509 static int bond_set_allmulti(struct bonding *bond, int inc)
511 struct list_head *iter;
514 if (bond_uses_primary(bond)) {
515 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
518 err = dev_set_allmulti(curr_active->dev, inc);
522 bond_for_each_slave(bond, slave, iter) {
523 err = dev_set_allmulti(slave->dev, inc);
531 /* Retrieve the list of registered multicast addresses for the bonding
532 * device and retransmit an IGMP JOIN request to the current active
535 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
537 struct bonding *bond = container_of(work, struct bonding,
540 if (!rtnl_trylock()) {
541 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
544 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
546 if (bond->igmp_retrans > 1) {
547 bond->igmp_retrans--;
548 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
553 /* Flush bond's hardware addresses from slave */
554 static void bond_hw_addr_flush(struct net_device *bond_dev,
555 struct net_device *slave_dev)
557 struct bonding *bond = netdev_priv(bond_dev);
559 dev_uc_unsync(slave_dev, bond_dev);
560 dev_mc_unsync(slave_dev, bond_dev);
562 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
563 /* del lacpdu mc addr from mc list */
564 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
566 dev_mc_del(slave_dev, lacpdu_multicast);
570 /*--------------------------- Active slave change ---------------------------*/
572 /* Update the hardware address list and promisc/allmulti for the new and
573 * old active slaves (if any). Modes that are not using primary keep all
574 * slaves up date at all times; only the modes that use primary need to call
575 * this function to swap these settings during a failover.
577 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
578 struct slave *old_active)
581 if (bond->dev->flags & IFF_PROMISC)
582 dev_set_promiscuity(old_active->dev, -1);
584 if (bond->dev->flags & IFF_ALLMULTI)
585 dev_set_allmulti(old_active->dev, -1);
587 bond_hw_addr_flush(bond->dev, old_active->dev);
591 /* FIXME: Signal errors upstream. */
592 if (bond->dev->flags & IFF_PROMISC)
593 dev_set_promiscuity(new_active->dev, 1);
595 if (bond->dev->flags & IFF_ALLMULTI)
596 dev_set_allmulti(new_active->dev, 1);
598 netif_addr_lock_bh(bond->dev);
599 dev_uc_sync(new_active->dev, bond->dev);
600 dev_mc_sync(new_active->dev, bond->dev);
601 netif_addr_unlock_bh(bond->dev);
606 * bond_set_dev_addr - clone slave's address to bond
607 * @bond_dev: bond net device
608 * @slave_dev: slave net device
610 * Should be called with RTNL held.
612 static int bond_set_dev_addr(struct net_device *bond_dev,
613 struct net_device *slave_dev)
617 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
618 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
619 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
623 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
624 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
625 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
629 static struct slave *bond_get_old_active(struct bonding *bond,
630 struct slave *new_active)
633 struct list_head *iter;
635 bond_for_each_slave(bond, slave, iter) {
636 if (slave == new_active)
639 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
646 /* bond_do_fail_over_mac
648 * Perform special MAC address swapping for fail_over_mac settings
652 static void bond_do_fail_over_mac(struct bonding *bond,
653 struct slave *new_active,
654 struct slave *old_active)
656 u8 tmp_mac[MAX_ADDR_LEN];
657 struct sockaddr_storage ss;
660 switch (bond->params.fail_over_mac) {
661 case BOND_FOM_ACTIVE:
663 rv = bond_set_dev_addr(bond->dev, new_active->dev);
665 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
666 -rv, bond->dev->name);
669 case BOND_FOM_FOLLOW:
670 /* if new_active && old_active, swap them
671 * if just old_active, do nothing (going to no active slave)
672 * if just new_active, set new_active to bond's MAC
678 old_active = bond_get_old_active(bond, new_active);
681 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
682 new_active->dev->addr_len);
683 bond_hw_addr_copy(ss.__data,
684 old_active->dev->dev_addr,
685 old_active->dev->addr_len);
686 ss.ss_family = new_active->dev->type;
688 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
689 bond->dev->addr_len);
690 ss.ss_family = bond->dev->type;
693 rv = dev_set_mac_address(new_active->dev,
694 (struct sockaddr *)&ss, NULL);
696 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
697 -rv, new_active->dev->name);
704 bond_hw_addr_copy(ss.__data, tmp_mac,
705 new_active->dev->addr_len);
706 ss.ss_family = old_active->dev->type;
708 rv = dev_set_mac_address(old_active->dev,
709 (struct sockaddr *)&ss, NULL);
711 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
712 -rv, new_active->dev->name);
716 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
717 bond->params.fail_over_mac);
723 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
725 struct slave *prim = rtnl_dereference(bond->primary_slave);
726 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
728 if (!prim || prim->link != BOND_LINK_UP) {
729 if (!curr || curr->link != BOND_LINK_UP)
734 if (bond->force_primary) {
735 bond->force_primary = false;
739 if (!curr || curr->link != BOND_LINK_UP)
742 /* At this point, prim and curr are both up */
743 switch (bond->params.primary_reselect) {
744 case BOND_PRI_RESELECT_ALWAYS:
746 case BOND_PRI_RESELECT_BETTER:
747 if (prim->speed < curr->speed)
749 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
752 case BOND_PRI_RESELECT_FAILURE:
755 netdev_err(bond->dev, "impossible primary_reselect %d\n",
756 bond->params.primary_reselect);
762 * bond_find_best_slave - select the best available slave to be the active one
763 * @bond: our bonding struct
765 static struct slave *bond_find_best_slave(struct bonding *bond)
767 struct slave *slave, *bestslave = NULL;
768 struct list_head *iter;
769 int mintime = bond->params.updelay;
771 slave = bond_choose_primary_or_current(bond);
775 bond_for_each_slave(bond, slave, iter) {
776 if (slave->link == BOND_LINK_UP)
778 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
779 slave->delay < mintime) {
780 mintime = slave->delay;
788 static bool bond_should_notify_peers(struct bonding *bond)
793 slave = rcu_dereference(bond->curr_active_slave);
796 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
797 slave ? slave->dev->name : "NULL");
799 if (!slave || !bond->send_peer_notif ||
800 !netif_carrier_ok(bond->dev) ||
801 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
808 * change_active_interface - change the active slave into the specified one
809 * @bond: our bonding struct
810 * @new: the new slave to make the active one
812 * Set the new slave to the bond's settings and unset them on the old
814 * Setting include flags, mc-list, promiscuity, allmulti, etc.
816 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
817 * because it is apparently the best available slave we have, even though its
818 * updelay hasn't timed out yet.
820 * Caller must hold RTNL.
822 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
824 struct slave *old_active;
828 old_active = rtnl_dereference(bond->curr_active_slave);
830 if (old_active == new_active)
834 new_active->last_link_up = jiffies;
836 if (new_active->link == BOND_LINK_BACK) {
837 if (bond_uses_primary(bond)) {
838 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
839 new_active->dev->name,
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 netdev_info(bond->dev, "making interface %s the new active one\n",
855 new_active->dev->name);
860 if (bond_uses_primary(bond))
861 bond_hw_addr_swap(bond, new_active, old_active);
863 if (bond_is_lb(bond)) {
864 bond_alb_handle_active_change(bond, new_active);
866 bond_set_slave_inactive_flags(old_active,
867 BOND_SLAVE_NOTIFY_NOW);
869 bond_set_slave_active_flags(new_active,
870 BOND_SLAVE_NOTIFY_NOW);
872 rcu_assign_pointer(bond->curr_active_slave, new_active);
875 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
877 bond_set_slave_inactive_flags(old_active,
878 BOND_SLAVE_NOTIFY_NOW);
881 bool should_notify_peers = false;
883 bond_set_slave_active_flags(new_active,
884 BOND_SLAVE_NOTIFY_NOW);
886 if (bond->params.fail_over_mac)
887 bond_do_fail_over_mac(bond, new_active,
890 if (netif_running(bond->dev)) {
891 bond->send_peer_notif =
892 bond->params.num_peer_notif;
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 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
904 /* resend IGMP joins since active slave has changed or
905 * all were sent on curr_active_slave.
906 * resend only if bond is brought up with the affected
907 * bonding modes and the retransmission is enabled
909 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
910 ((bond_uses_primary(bond) && new_active) ||
911 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
912 bond->igmp_retrans = bond->params.resend_igmp;
913 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
918 * bond_select_active_slave - select a new active slave, if needed
919 * @bond: our bonding struct
921 * This functions should be called when one of the following occurs:
922 * - The old curr_active_slave has been released or lost its link.
923 * - The primary_slave has got its link back.
924 * - A slave has got its link back and there's no old curr_active_slave.
926 * Caller must hold RTNL.
928 void bond_select_active_slave(struct bonding *bond)
930 struct slave *best_slave;
935 best_slave = bond_find_best_slave(bond);
936 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
937 bond_change_active_slave(bond, best_slave);
938 rv = bond_set_carrier(bond);
942 if (netif_carrier_ok(bond->dev))
943 netdev_info(bond->dev, "first active interface up!\n");
945 netdev_info(bond->dev, "now running without any active interface!\n");
949 #ifdef CONFIG_NET_POLL_CONTROLLER
950 static inline int slave_enable_netpoll(struct slave *slave)
955 np = kzalloc(sizeof(*np), GFP_KERNEL);
960 err = __netpoll_setup(np, slave->dev);
969 static inline void slave_disable_netpoll(struct slave *slave)
971 struct netpoll *np = slave->np;
981 static void bond_poll_controller(struct net_device *bond_dev)
983 struct bonding *bond = netdev_priv(bond_dev);
984 struct slave *slave = NULL;
985 struct list_head *iter;
986 struct ad_info ad_info;
988 if (BOND_MODE(bond) == BOND_MODE_8023AD)
989 if (bond_3ad_get_active_agg_info(bond, &ad_info))
992 bond_for_each_slave_rcu(bond, slave, iter) {
993 if (!bond_slave_is_up(slave))
996 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
997 struct aggregator *agg =
998 SLAVE_AD_INFO(slave)->port.aggregator;
1001 agg->aggregator_identifier != ad_info.aggregator_id)
1005 netpoll_poll_dev(slave->dev);
1009 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1011 struct bonding *bond = netdev_priv(bond_dev);
1012 struct list_head *iter;
1013 struct slave *slave;
1015 bond_for_each_slave(bond, slave, iter)
1016 if (bond_slave_is_up(slave))
1017 slave_disable_netpoll(slave);
1020 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1022 struct bonding *bond = netdev_priv(dev);
1023 struct list_head *iter;
1024 struct slave *slave;
1027 bond_for_each_slave(bond, slave, iter) {
1028 err = slave_enable_netpoll(slave);
1030 bond_netpoll_cleanup(dev);
1037 static inline int slave_enable_netpoll(struct slave *slave)
1041 static inline void slave_disable_netpoll(struct slave *slave)
1044 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1049 /*---------------------------------- IOCTL ----------------------------------*/
1051 static netdev_features_t bond_fix_features(struct net_device *dev,
1052 netdev_features_t features)
1054 struct bonding *bond = netdev_priv(dev);
1055 struct list_head *iter;
1056 netdev_features_t mask;
1057 struct slave *slave;
1061 features &= ~NETIF_F_ONE_FOR_ALL;
1062 features |= NETIF_F_ALL_FOR_ALL;
1064 bond_for_each_slave(bond, slave, iter) {
1065 features = netdev_increment_features(features,
1066 slave->dev->features,
1069 features = netdev_add_tso_features(features, mask);
1074 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1075 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1076 NETIF_F_HIGHDMA | NETIF_F_LRO)
1078 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1079 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1081 static void bond_compute_features(struct bonding *bond)
1083 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1084 IFF_XMIT_DST_RELEASE_PERM;
1085 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1086 netdev_features_t enc_features = BOND_ENC_FEATURES;
1087 struct net_device *bond_dev = bond->dev;
1088 struct list_head *iter;
1089 struct slave *slave;
1090 unsigned short max_hard_header_len = ETH_HLEN;
1091 unsigned int gso_max_size = GSO_MAX_SIZE;
1092 u16 gso_max_segs = GSO_MAX_SEGS;
1094 if (!bond_has_slaves(bond))
1096 vlan_features &= NETIF_F_ALL_FOR_ALL;
1098 bond_for_each_slave(bond, slave, iter) {
1099 vlan_features = netdev_increment_features(vlan_features,
1100 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1102 enc_features = netdev_increment_features(enc_features,
1103 slave->dev->hw_enc_features,
1105 dst_release_flag &= slave->dev->priv_flags;
1106 if (slave->dev->hard_header_len > max_hard_header_len)
1107 max_hard_header_len = slave->dev->hard_header_len;
1109 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1110 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1112 bond_dev->hard_header_len = max_hard_header_len;
1115 bond_dev->vlan_features = vlan_features;
1116 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1118 bond_dev->gso_max_segs = gso_max_segs;
1119 netif_set_gso_max_size(bond_dev, gso_max_size);
1121 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1122 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1123 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1124 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1126 netdev_change_features(bond_dev);
1129 static void bond_setup_by_slave(struct net_device *bond_dev,
1130 struct net_device *slave_dev)
1132 bond_dev->header_ops = slave_dev->header_ops;
1134 bond_dev->type = slave_dev->type;
1135 bond_dev->hard_header_len = slave_dev->hard_header_len;
1136 bond_dev->addr_len = slave_dev->addr_len;
1138 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1139 slave_dev->addr_len);
1142 /* On bonding slaves other than the currently active slave, suppress
1143 * duplicates except for alb non-mcast/bcast.
1145 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1146 struct slave *slave,
1147 struct bonding *bond)
1149 if (bond_is_slave_inactive(slave)) {
1150 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1151 skb->pkt_type != PACKET_BROADCAST &&
1152 skb->pkt_type != PACKET_MULTICAST)
1159 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1161 struct sk_buff *skb = *pskb;
1162 struct slave *slave;
1163 struct bonding *bond;
1164 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1166 int ret = RX_HANDLER_ANOTHER;
1168 skb = skb_share_check(skb, GFP_ATOMIC);
1170 return RX_HANDLER_CONSUMED;
1174 slave = bond_slave_get_rcu(skb->dev);
1177 recv_probe = READ_ONCE(bond->recv_probe);
1179 ret = recv_probe(skb, bond, slave);
1180 if (ret == RX_HANDLER_CONSUMED) {
1186 /* Link-local multicast packets should be passed to the
1187 * stack on the link they arrive as well as pass them to the
1188 * bond-master device. These packets are mostly usable when
1189 * stack receives it with the link on which they arrive
1190 * (e.g. LLDP) they also must be available on master. Some of
1191 * the use cases include (but are not limited to): LLDP agents
1192 * that must be able to operate both on enslaved interfaces as
1193 * well as on bonds themselves; linux bridges that must be able
1194 * to process/pass BPDUs from attached bonds when any kind of
1195 * STP version is enabled on the network.
1197 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) {
1198 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1201 nskb->dev = bond->dev;
1202 nskb->queue_mapping = 0;
1205 return RX_HANDLER_PASS;
1207 if (bond_should_deliver_exact_match(skb, slave, bond))
1208 return RX_HANDLER_EXACT;
1210 skb->dev = bond->dev;
1212 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1213 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1214 skb->pkt_type == PACKET_HOST) {
1216 if (unlikely(skb_cow_head(skb,
1217 skb->data - skb_mac_header(skb)))) {
1219 return RX_HANDLER_CONSUMED;
1221 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1222 bond->dev->addr_len);
1228 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1230 switch (BOND_MODE(bond)) {
1231 case BOND_MODE_ROUNDROBIN:
1232 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1233 case BOND_MODE_ACTIVEBACKUP:
1234 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1235 case BOND_MODE_BROADCAST:
1236 return NETDEV_LAG_TX_TYPE_BROADCAST;
1238 case BOND_MODE_8023AD:
1239 return NETDEV_LAG_TX_TYPE_HASH;
1241 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1245 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1246 enum netdev_lag_tx_type type)
1248 if (type != NETDEV_LAG_TX_TYPE_HASH)
1249 return NETDEV_LAG_HASH_NONE;
1251 switch (bond->params.xmit_policy) {
1252 case BOND_XMIT_POLICY_LAYER2:
1253 return NETDEV_LAG_HASH_L2;
1254 case BOND_XMIT_POLICY_LAYER34:
1255 return NETDEV_LAG_HASH_L34;
1256 case BOND_XMIT_POLICY_LAYER23:
1257 return NETDEV_LAG_HASH_L23;
1258 case BOND_XMIT_POLICY_ENCAP23:
1259 return NETDEV_LAG_HASH_E23;
1260 case BOND_XMIT_POLICY_ENCAP34:
1261 return NETDEV_LAG_HASH_E34;
1263 return NETDEV_LAG_HASH_UNKNOWN;
1267 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1268 struct netlink_ext_ack *extack)
1270 struct netdev_lag_upper_info lag_upper_info;
1271 enum netdev_lag_tx_type type;
1273 type = bond_lag_tx_type(bond);
1274 lag_upper_info.tx_type = type;
1275 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1277 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1278 &lag_upper_info, extack);
1281 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1283 netdev_upper_dev_unlink(slave->dev, bond->dev);
1284 slave->dev->flags &= ~IFF_SLAVE;
1287 static struct slave *bond_alloc_slave(struct bonding *bond)
1289 struct slave *slave = NULL;
1291 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1295 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1296 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1298 if (!SLAVE_AD_INFO(slave)) {
1303 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1308 static void bond_free_slave(struct slave *slave)
1310 struct bonding *bond = bond_get_bond_by_slave(slave);
1312 cancel_delayed_work_sync(&slave->notify_work);
1313 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1314 kfree(SLAVE_AD_INFO(slave));
1319 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1321 info->bond_mode = BOND_MODE(bond);
1322 info->miimon = bond->params.miimon;
1323 info->num_slaves = bond->slave_cnt;
1326 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1328 strcpy(info->slave_name, slave->dev->name);
1329 info->link = slave->link;
1330 info->state = bond_slave_state(slave);
1331 info->link_failure_count = slave->link_failure_count;
1334 static void bond_netdev_notify_work(struct work_struct *_work)
1336 struct slave *slave = container_of(_work, struct slave,
1339 if (rtnl_trylock()) {
1340 struct netdev_bonding_info binfo;
1342 bond_fill_ifslave(slave, &binfo.slave);
1343 bond_fill_ifbond(slave->bond, &binfo.master);
1344 netdev_bonding_info_change(slave->dev, &binfo);
1347 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1351 void bond_queue_slave_event(struct slave *slave)
1353 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1356 void bond_lower_state_changed(struct slave *slave)
1358 struct netdev_lag_lower_state_info info;
1360 info.link_up = slave->link == BOND_LINK_UP ||
1361 slave->link == BOND_LINK_FAIL;
1362 info.tx_enabled = bond_is_active_slave(slave);
1363 netdev_lower_state_changed(slave->dev, &info);
1366 /* enslave device <slave> to bond device <master> */
1367 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1368 struct netlink_ext_ack *extack)
1370 struct bonding *bond = netdev_priv(bond_dev);
1371 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1372 struct slave *new_slave = NULL, *prev_slave;
1373 struct sockaddr_storage ss;
1377 if (!bond->params.use_carrier &&
1378 slave_dev->ethtool_ops->get_link == NULL &&
1379 slave_ops->ndo_do_ioctl == NULL) {
1380 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1384 /* already in-use? */
1385 if (netdev_is_rx_handler_busy(slave_dev)) {
1386 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1387 netdev_err(bond_dev,
1388 "Error: Device is in use and cannot be enslaved\n");
1392 if (bond_dev == slave_dev) {
1393 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1394 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1398 /* vlan challenged mutual exclusion */
1399 /* no need to lock since we're protected by rtnl_lock */
1400 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1401 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1403 if (vlan_uses_dev(bond_dev)) {
1404 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1405 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1406 slave_dev->name, bond_dev->name);
1409 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1410 slave_dev->name, slave_dev->name,
1414 netdev_dbg(bond_dev, "%s 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 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1430 /* set bonding device ether type by slave - bonding netdevices are
1431 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1432 * there is a need to override some of the type dependent attribs/funcs.
1434 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1435 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1437 if (!bond_has_slaves(bond)) {
1438 if (bond_dev->type != slave_dev->type) {
1439 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1440 bond_dev->type, slave_dev->type);
1442 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1444 res = notifier_to_errno(res);
1446 netdev_err(bond_dev, "refused to change device type\n");
1450 /* Flush unicast and multicast addresses */
1451 dev_uc_flush(bond_dev);
1452 dev_mc_flush(bond_dev);
1454 if (slave_dev->type != ARPHRD_ETHER)
1455 bond_setup_by_slave(bond_dev, slave_dev);
1457 ether_setup(bond_dev);
1458 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1461 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1464 } else if (bond_dev->type != slave_dev->type) {
1465 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1466 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1467 slave_dev->name, slave_dev->type, bond_dev->type);
1471 if (slave_dev->type == ARPHRD_INFINIBAND &&
1472 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1473 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1474 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1477 goto err_undo_flags;
1480 if (!slave_ops->ndo_set_mac_address ||
1481 slave_dev->type == ARPHRD_INFINIBAND) {
1482 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1483 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1484 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1485 if (!bond_has_slaves(bond)) {
1486 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1487 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1489 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1490 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1492 goto err_undo_flags;
1497 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1499 /* If this is the first slave, then we need to set the master's hardware
1500 * address to be the same as the slave's.
1502 if (!bond_has_slaves(bond) &&
1503 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1504 res = bond_set_dev_addr(bond->dev, slave_dev);
1506 goto err_undo_flags;
1509 new_slave = bond_alloc_slave(bond);
1512 goto err_undo_flags;
1515 new_slave->bond = bond;
1516 new_slave->dev = slave_dev;
1517 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1518 * is set via sysfs or module option if desired.
1520 new_slave->queue_id = 0;
1522 /* Save slave's original mtu and then set it to match the bond */
1523 new_slave->original_mtu = slave_dev->mtu;
1524 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1526 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1530 /* Save slave's original ("permanent") mac address for modes
1531 * that need it, and for restoring it upon release, and then
1532 * set it to the master's address
1534 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1535 slave_dev->addr_len);
1537 if (!bond->params.fail_over_mac ||
1538 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1539 /* Set slave to master's mac address. The application already
1540 * set the master's mac address to that of the first slave
1542 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1543 ss.ss_family = slave_dev->type;
1544 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1547 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1548 goto err_restore_mtu;
1552 /* set slave flag before open to prevent IPv6 addrconf */
1553 slave_dev->flags |= IFF_SLAVE;
1555 /* open the slave since the application closed it */
1556 res = dev_open(slave_dev, extack);
1558 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1559 goto err_restore_mac;
1562 slave_dev->priv_flags |= IFF_BONDING;
1563 /* initialize slave stats */
1564 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1566 if (bond_is_lb(bond)) {
1567 /* bond_alb_init_slave() must be called before all other stages since
1568 * it might fail and we do not want to have to undo everything
1570 res = bond_alb_init_slave(bond, new_slave);
1575 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1577 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1582 prev_slave = bond_last_slave(bond);
1584 new_slave->delay = 0;
1585 new_slave->link_failure_count = 0;
1587 if (bond_update_speed_duplex(new_slave) &&
1588 bond_needs_speed_duplex(bond))
1589 new_slave->link = BOND_LINK_DOWN;
1591 new_slave->last_rx = jiffies -
1592 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1593 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1594 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1596 if (bond->params.miimon && !bond->params.use_carrier) {
1597 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1599 if ((link_reporting == -1) && !bond->params.arp_interval) {
1600 /* miimon is set but a bonded network driver
1601 * does not support ETHTOOL/MII and
1602 * arp_interval is not set. Note: if
1603 * use_carrier is enabled, we will never go
1604 * here (because netif_carrier is always
1605 * supported); thus, we don't need to change
1606 * the messages for netif_carrier.
1608 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1610 } else if (link_reporting == -1) {
1611 /* unable get link status using mii/ethtool */
1612 netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1617 /* check for initial state */
1618 new_slave->link = BOND_LINK_NOCHANGE;
1619 if (bond->params.miimon) {
1620 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1621 if (bond->params.updelay) {
1622 bond_set_slave_link_state(new_slave,
1624 BOND_SLAVE_NOTIFY_NOW);
1625 new_slave->delay = bond->params.updelay;
1627 bond_set_slave_link_state(new_slave,
1629 BOND_SLAVE_NOTIFY_NOW);
1632 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1633 BOND_SLAVE_NOTIFY_NOW);
1635 } else if (bond->params.arp_interval) {
1636 bond_set_slave_link_state(new_slave,
1637 (netif_carrier_ok(slave_dev) ?
1638 BOND_LINK_UP : BOND_LINK_DOWN),
1639 BOND_SLAVE_NOTIFY_NOW);
1641 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1642 BOND_SLAVE_NOTIFY_NOW);
1645 if (new_slave->link != BOND_LINK_DOWN)
1646 new_slave->last_link_up = jiffies;
1647 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1648 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1649 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1651 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1652 /* if there is a primary slave, remember it */
1653 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1654 rcu_assign_pointer(bond->primary_slave, new_slave);
1655 bond->force_primary = true;
1659 switch (BOND_MODE(bond)) {
1660 case BOND_MODE_ACTIVEBACKUP:
1661 bond_set_slave_inactive_flags(new_slave,
1662 BOND_SLAVE_NOTIFY_NOW);
1664 case BOND_MODE_8023AD:
1665 /* in 802.3ad mode, the internal mechanism
1666 * will activate the slaves in the selected
1669 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1670 /* if this is the first slave */
1672 SLAVE_AD_INFO(new_slave)->id = 1;
1673 /* Initialize AD with the number of times that the AD timer is called in 1 second
1674 * can be called only after the mac address of the bond is set
1676 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1678 SLAVE_AD_INFO(new_slave)->id =
1679 SLAVE_AD_INFO(prev_slave)->id + 1;
1682 bond_3ad_bind_slave(new_slave);
1686 bond_set_active_slave(new_slave);
1687 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1690 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1692 /* always active in trunk mode */
1693 bond_set_active_slave(new_slave);
1695 /* In trunking mode there is little meaning to curr_active_slave
1696 * anyway (it holds no special properties of the bond device),
1697 * so we can change it without calling change_active_interface()
1699 if (!rcu_access_pointer(bond->curr_active_slave) &&
1700 new_slave->link == BOND_LINK_UP)
1701 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1704 } /* switch(bond_mode) */
1706 #ifdef CONFIG_NET_POLL_CONTROLLER
1707 if (bond->dev->npinfo) {
1708 if (slave_enable_netpoll(new_slave)) {
1709 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1716 if (!(bond_dev->features & NETIF_F_LRO))
1717 dev_disable_lro(slave_dev);
1719 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1722 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1726 res = bond_master_upper_dev_link(bond, new_slave, extack);
1728 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1729 goto err_unregister;
1732 res = bond_sysfs_slave_add(new_slave);
1734 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1735 goto err_upper_unlink;
1738 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1740 /* If the mode uses primary, then the following is handled by
1741 * bond_change_active_slave().
1743 if (!bond_uses_primary(bond)) {
1744 /* set promiscuity level to new slave */
1745 if (bond_dev->flags & IFF_PROMISC) {
1746 res = dev_set_promiscuity(slave_dev, 1);
1751 /* set allmulti level to new slave */
1752 if (bond_dev->flags & IFF_ALLMULTI) {
1753 res = dev_set_allmulti(slave_dev, 1);
1755 if (bond_dev->flags & IFF_PROMISC)
1756 dev_set_promiscuity(slave_dev, -1);
1761 netif_addr_lock_bh(bond_dev);
1762 dev_mc_sync_multiple(slave_dev, bond_dev);
1763 dev_uc_sync_multiple(slave_dev, bond_dev);
1764 netif_addr_unlock_bh(bond_dev);
1766 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1767 /* add lacpdu mc addr to mc list */
1768 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1770 dev_mc_add(slave_dev, lacpdu_multicast);
1775 bond_compute_features(bond);
1776 bond_set_carrier(bond);
1778 if (bond_uses_primary(bond)) {
1780 bond_select_active_slave(bond);
1781 unblock_netpoll_tx();
1784 if (bond_mode_can_use_xmit_hash(bond))
1785 bond_update_slave_arr(bond, NULL);
1788 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1790 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1791 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1793 /* enslave is successful */
1794 bond_queue_slave_event(new_slave);
1797 /* Undo stages on error */
1799 bond_sysfs_slave_del(new_slave);
1802 bond_upper_dev_unlink(bond, new_slave);
1805 netdev_rx_handler_unregister(slave_dev);
1808 vlan_vids_del_by_dev(slave_dev, bond_dev);
1809 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1810 RCU_INIT_POINTER(bond->primary_slave, NULL);
1811 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1813 bond_change_active_slave(bond, NULL);
1814 bond_select_active_slave(bond);
1815 unblock_netpoll_tx();
1817 /* either primary_slave or curr_active_slave might've changed */
1819 slave_disable_netpoll(new_slave);
1822 slave_dev->priv_flags &= ~IFF_BONDING;
1823 dev_close(slave_dev);
1826 slave_dev->flags &= ~IFF_SLAVE;
1827 if (!bond->params.fail_over_mac ||
1828 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1829 /* XXX TODO - fom follow mode needs to change master's
1830 * MAC if this slave's MAC is in use by the bond, or at
1831 * least print a warning.
1833 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1834 new_slave->dev->addr_len);
1835 ss.ss_family = slave_dev->type;
1836 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
1840 dev_set_mtu(slave_dev, new_slave->original_mtu);
1843 bond_free_slave(new_slave);
1846 /* Enslave of first slave has failed and we need to fix master's mac */
1847 if (!bond_has_slaves(bond)) {
1848 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1849 slave_dev->dev_addr))
1850 eth_hw_addr_random(bond_dev);
1851 if (bond_dev->type != ARPHRD_ETHER) {
1852 dev_close(bond_dev);
1853 ether_setup(bond_dev);
1854 bond_dev->flags |= IFF_MASTER;
1855 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1862 /* Try to release the slave device <slave> from the bond device <master>
1863 * It is legal to access curr_active_slave without a lock because all the function
1864 * is RTNL-locked. If "all" is true it means that the function is being called
1865 * while destroying a bond interface and all slaves are being released.
1867 * The rules for slave state should be:
1868 * for Active/Backup:
1869 * Active stays on all backups go down
1870 * for Bonded connections:
1871 * The first up interface should be left on and all others downed.
1873 static int __bond_release_one(struct net_device *bond_dev,
1874 struct net_device *slave_dev,
1875 bool all, bool unregister)
1877 struct bonding *bond = netdev_priv(bond_dev);
1878 struct slave *slave, *oldcurrent;
1879 struct sockaddr_storage ss;
1880 int old_flags = bond_dev->flags;
1881 netdev_features_t old_features = bond_dev->features;
1883 /* slave is not a slave or master is not master of this slave */
1884 if (!(slave_dev->flags & IFF_SLAVE) ||
1885 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1886 netdev_dbg(bond_dev, "cannot release %s\n",
1893 slave = bond_get_slave_by_dev(bond, slave_dev);
1895 /* not a slave of this bond */
1896 netdev_info(bond_dev, "%s not enslaved\n",
1898 unblock_netpoll_tx();
1902 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1904 bond_sysfs_slave_del(slave);
1906 /* recompute stats just before removing the slave */
1907 bond_get_stats(bond->dev, &bond->bond_stats);
1909 bond_upper_dev_unlink(bond, slave);
1910 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1911 * for this slave anymore.
1913 netdev_rx_handler_unregister(slave_dev);
1915 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1916 bond_3ad_unbind_slave(slave);
1918 if (bond_mode_can_use_xmit_hash(bond))
1919 bond_update_slave_arr(bond, slave);
1921 netdev_info(bond_dev, "Releasing %s interface %s\n",
1922 bond_is_active_slave(slave) ? "active" : "backup",
1925 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1927 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1929 if (!all && (!bond->params.fail_over_mac ||
1930 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1931 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1932 bond_has_slaves(bond))
1933 netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1934 slave_dev->name, slave->perm_hwaddr,
1935 bond_dev->name, slave_dev->name);
1938 if (rtnl_dereference(bond->primary_slave) == slave)
1939 RCU_INIT_POINTER(bond->primary_slave, NULL);
1941 if (oldcurrent == slave)
1942 bond_change_active_slave(bond, NULL);
1944 if (bond_is_lb(bond)) {
1945 /* Must be called only after the slave has been
1946 * detached from the list and the curr_active_slave
1947 * has been cleared (if our_slave == old_current),
1948 * but before a new active slave is selected.
1950 bond_alb_deinit_slave(bond, slave);
1954 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1955 } else if (oldcurrent == slave) {
1956 /* Note that we hold RTNL over this sequence, so there
1957 * is no concern that another slave add/remove event
1960 bond_select_active_slave(bond);
1963 if (!bond_has_slaves(bond)) {
1964 bond_set_carrier(bond);
1965 eth_hw_addr_random(bond_dev);
1966 bond->nest_level = SINGLE_DEPTH_NESTING;
1968 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1971 unblock_netpoll_tx();
1975 if (!bond_has_slaves(bond)) {
1976 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1977 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1980 bond_compute_features(bond);
1981 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1982 (old_features & NETIF_F_VLAN_CHALLENGED))
1983 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1984 slave_dev->name, bond_dev->name);
1986 vlan_vids_del_by_dev(slave_dev, bond_dev);
1988 /* If the mode uses primary, then this case was handled above by
1989 * bond_change_active_slave(..., NULL)
1991 if (!bond_uses_primary(bond)) {
1992 /* unset promiscuity level from slave
1993 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1994 * of the IFF_PROMISC flag in the bond_dev, but we need the
1995 * value of that flag before that change, as that was the value
1996 * when this slave was attached, so we cache at the start of the
1997 * function and use it here. Same goes for ALLMULTI below
1999 if (old_flags & IFF_PROMISC)
2000 dev_set_promiscuity(slave_dev, -1);
2002 /* unset allmulti level from slave */
2003 if (old_flags & IFF_ALLMULTI)
2004 dev_set_allmulti(slave_dev, -1);
2006 bond_hw_addr_flush(bond_dev, slave_dev);
2009 slave_disable_netpoll(slave);
2011 /* close slave before restoring its mac address */
2012 dev_close(slave_dev);
2014 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2015 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2016 /* restore original ("permanent") mac address */
2017 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2018 slave->dev->addr_len);
2019 ss.ss_family = slave_dev->type;
2020 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2024 __dev_set_mtu(slave_dev, slave->original_mtu);
2026 dev_set_mtu(slave_dev, slave->original_mtu);
2028 slave_dev->priv_flags &= ~IFF_BONDING;
2030 bond_free_slave(slave);
2035 /* A wrapper used because of ndo_del_link */
2036 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2038 return __bond_release_one(bond_dev, slave_dev, false, false);
2041 /* First release a slave and then destroy the bond if no more slaves are left.
2042 * Must be under rtnl_lock when this function is called.
2044 static int bond_release_and_destroy(struct net_device *bond_dev,
2045 struct net_device *slave_dev)
2047 struct bonding *bond = netdev_priv(bond_dev);
2050 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2051 if (ret == 0 && !bond_has_slaves(bond)) {
2052 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2053 netdev_info(bond_dev, "Destroying bond %s\n",
2055 bond_remove_proc_entry(bond);
2056 unregister_netdevice(bond_dev);
2061 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2063 struct bonding *bond = netdev_priv(bond_dev);
2064 bond_fill_ifbond(bond, info);
2067 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2069 struct bonding *bond = netdev_priv(bond_dev);
2070 struct list_head *iter;
2071 int i = 0, res = -ENODEV;
2072 struct slave *slave;
2074 bond_for_each_slave(bond, slave, iter) {
2075 if (i++ == (int)info->slave_id) {
2077 bond_fill_ifslave(slave, info);
2085 /*-------------------------------- Monitoring -------------------------------*/
2087 /* called with rcu_read_lock() */
2088 static int bond_miimon_inspect(struct bonding *bond)
2090 int link_state, commit = 0;
2091 struct list_head *iter;
2092 struct slave *slave;
2093 bool ignore_updelay;
2095 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2097 bond_for_each_slave_rcu(bond, slave, iter) {
2098 slave->new_link = BOND_LINK_NOCHANGE;
2099 slave->link_new_state = slave->link;
2101 link_state = bond_check_dev_link(bond, slave->dev, 0);
2103 switch (slave->link) {
2108 bond_propose_link_state(slave, BOND_LINK_FAIL);
2110 slave->delay = bond->params.downdelay;
2112 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2114 BOND_MODE_ACTIVEBACKUP) ?
2115 (bond_is_active_slave(slave) ?
2116 "active " : "backup ") : "",
2118 bond->params.downdelay * bond->params.miimon);
2121 case BOND_LINK_FAIL:
2123 /* recovered before downdelay expired */
2124 bond_propose_link_state(slave, BOND_LINK_UP);
2125 slave->last_link_up = jiffies;
2126 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2127 (bond->params.downdelay - slave->delay) *
2128 bond->params.miimon,
2134 if (slave->delay <= 0) {
2135 slave->new_link = BOND_LINK_DOWN;
2143 case BOND_LINK_DOWN:
2147 bond_propose_link_state(slave, BOND_LINK_BACK);
2149 slave->delay = bond->params.updelay;
2152 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2154 ignore_updelay ? 0 :
2155 bond->params.updelay *
2156 bond->params.miimon);
2159 case BOND_LINK_BACK:
2161 bond_propose_link_state(slave, BOND_LINK_DOWN);
2162 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2163 (bond->params.updelay - slave->delay) *
2164 bond->params.miimon,
2173 if (slave->delay <= 0) {
2174 slave->new_link = BOND_LINK_UP;
2176 ignore_updelay = false;
2188 static void bond_miimon_link_change(struct bonding *bond,
2189 struct slave *slave,
2192 switch (BOND_MODE(bond)) {
2193 case BOND_MODE_8023AD:
2194 bond_3ad_handle_link_change(slave, link);
2198 bond_alb_handle_link_change(bond, slave, link);
2201 bond_update_slave_arr(bond, NULL);
2206 static void bond_miimon_commit(struct bonding *bond)
2208 struct list_head *iter;
2209 struct slave *slave, *primary;
2211 bond_for_each_slave(bond, slave, iter) {
2212 switch (slave->new_link) {
2213 case BOND_LINK_NOCHANGE:
2217 if (bond_update_speed_duplex(slave) &&
2218 bond_needs_speed_duplex(bond)) {
2219 slave->link = BOND_LINK_DOWN;
2220 if (net_ratelimit())
2221 netdev_warn(bond->dev,
2222 "failed to get link speed/duplex for %s\n",
2226 bond_set_slave_link_state(slave, BOND_LINK_UP,
2227 BOND_SLAVE_NOTIFY_NOW);
2228 slave->last_link_up = jiffies;
2230 primary = rtnl_dereference(bond->primary_slave);
2231 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2232 /* prevent it from being the active one */
2233 bond_set_backup_slave(slave);
2234 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2235 /* make it immediately active */
2236 bond_set_active_slave(slave);
2237 } else if (slave != primary) {
2238 /* prevent it from being the active one */
2239 bond_set_backup_slave(slave);
2242 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2244 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2245 slave->duplex ? "full" : "half");
2247 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2249 if (!bond->curr_active_slave || slave == primary)
2254 case BOND_LINK_DOWN:
2255 if (slave->link_failure_count < UINT_MAX)
2256 slave->link_failure_count++;
2258 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2259 BOND_SLAVE_NOTIFY_NOW);
2261 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2262 BOND_MODE(bond) == BOND_MODE_8023AD)
2263 bond_set_slave_inactive_flags(slave,
2264 BOND_SLAVE_NOTIFY_NOW);
2266 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2269 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2271 if (slave == rcu_access_pointer(bond->curr_active_slave))
2277 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2278 slave->new_link, slave->dev->name);
2279 slave->new_link = BOND_LINK_NOCHANGE;
2286 bond_select_active_slave(bond);
2287 unblock_netpoll_tx();
2290 bond_set_carrier(bond);
2295 * Really a wrapper that splits the mii monitor into two phases: an
2296 * inspection, then (if inspection indicates something needs to be done)
2297 * an acquisition of appropriate locks followed by a commit phase to
2298 * implement whatever link state changes are indicated.
2300 static void bond_mii_monitor(struct work_struct *work)
2302 struct bonding *bond = container_of(work, struct bonding,
2304 bool should_notify_peers = false;
2305 unsigned long delay;
2306 struct slave *slave;
2307 struct list_head *iter;
2309 delay = msecs_to_jiffies(bond->params.miimon);
2311 if (!bond_has_slaves(bond))
2316 should_notify_peers = bond_should_notify_peers(bond);
2318 if (bond_miimon_inspect(bond)) {
2321 /* Race avoidance with bond_close cancel of workqueue */
2322 if (!rtnl_trylock()) {
2324 should_notify_peers = false;
2328 bond_for_each_slave(bond, slave, iter) {
2329 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2331 bond_miimon_commit(bond);
2333 rtnl_unlock(); /* might sleep, hold no other locks */
2338 if (bond->params.miimon)
2339 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2341 if (should_notify_peers) {
2342 if (!rtnl_trylock())
2344 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2349 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2351 __be32 ip = *((__be32 *)data);
2353 return ip == bond_confirm_addr(upper, 0, ip);
2356 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2360 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2364 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2371 /* We go to the (large) trouble of VLAN tagging ARP frames because
2372 * switches in VLAN mode (especially if ports are configured as
2373 * "native" to a VLAN) might not pass non-tagged frames.
2375 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2376 __be32 dest_ip, __be32 src_ip,
2377 struct bond_vlan_tag *tags)
2379 struct sk_buff *skb;
2380 struct bond_vlan_tag *outer_tag = tags;
2382 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2383 arp_op, slave_dev->name, &dest_ip, &src_ip);
2385 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2386 NULL, slave_dev->dev_addr, NULL);
2389 net_err_ratelimited("ARP packet allocation failed\n");
2393 if (!tags || tags->vlan_proto == VLAN_N_VID)
2398 /* Go through all the tags backwards and add them to the packet */
2399 while (tags->vlan_proto != VLAN_N_VID) {
2400 if (!tags->vlan_id) {
2405 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2406 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2407 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2410 net_err_ratelimited("failed to insert inner VLAN tag\n");
2416 /* Set the outer tag */
2417 if (outer_tag->vlan_id) {
2418 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2419 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2420 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2421 outer_tag->vlan_id);
2428 /* Validate the device path between the @start_dev and the @end_dev.
2429 * The path is valid if the @end_dev is reachable through device
2431 * When the path is validated, collect any vlan information in the
2434 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2435 struct net_device *end_dev,
2438 struct bond_vlan_tag *tags;
2439 struct net_device *upper;
2440 struct list_head *iter;
2442 if (start_dev == end_dev) {
2443 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2445 return ERR_PTR(-ENOMEM);
2446 tags[level].vlan_proto = VLAN_N_VID;
2450 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2451 tags = bond_verify_device_path(upper, end_dev, level + 1);
2452 if (IS_ERR_OR_NULL(tags)) {
2457 if (is_vlan_dev(upper)) {
2458 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2459 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2468 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2471 struct bond_vlan_tag *tags;
2472 __be32 *targets = bond->params.arp_targets, addr;
2475 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2476 netdev_dbg(bond->dev, "basa: target %pI4\n", &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->dev, 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 netdev_dbg(bond->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->dev, ARPOP_REQUEST, targets[i],
2522 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2526 if (!sip || !bond_has_this_ip(bond, tip)) {
2527 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2532 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2534 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2538 slave->last_rx = jiffies;
2539 slave->target_last_arp_rx[i] = jiffies;
2542 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2543 struct slave *slave)
2545 struct arphdr *arp = (struct arphdr *)skb->data;
2546 struct slave *curr_active_slave, *curr_arp_slave;
2547 unsigned char *arp_ptr;
2549 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2552 if (!slave_do_arp_validate(bond, slave)) {
2553 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2554 !slave_do_arp_validate_only(bond))
2555 slave->last_rx = jiffies;
2556 return RX_HANDLER_ANOTHER;
2557 } else if (!is_arp) {
2558 return RX_HANDLER_ANOTHER;
2561 alen = arp_hdr_len(bond->dev);
2563 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2566 if (alen > skb_headlen(skb)) {
2567 arp = kmalloc(alen, GFP_ATOMIC);
2570 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2574 if (arp->ar_hln != bond->dev->addr_len ||
2575 skb->pkt_type == PACKET_OTHERHOST ||
2576 skb->pkt_type == PACKET_LOOPBACK ||
2577 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2578 arp->ar_pro != htons(ETH_P_IP) ||
2582 arp_ptr = (unsigned char *)(arp + 1);
2583 arp_ptr += bond->dev->addr_len;
2584 memcpy(&sip, arp_ptr, 4);
2585 arp_ptr += 4 + bond->dev->addr_len;
2586 memcpy(&tip, arp_ptr, 4);
2588 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2589 slave->dev->name, bond_slave_state(slave),
2590 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2593 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2594 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2596 /* We 'trust' the received ARP enough to validate it if:
2598 * (a) the slave receiving the ARP is active (which includes the
2599 * current ARP slave, if any), or
2601 * (b) the receiving slave isn't active, but there is a currently
2602 * active slave and it received valid arp reply(s) after it became
2603 * the currently active slave, or
2605 * (c) there is an ARP slave that sent an ARP during the prior ARP
2606 * interval, and we receive an ARP reply on any slave. We accept
2607 * these because switch FDB update delays may deliver the ARP
2608 * reply to a slave other than the sender of the ARP request.
2610 * Note: for (b), backup slaves are receiving the broadcast ARP
2611 * request, not a reply. This request passes from the sending
2612 * slave through the L2 switch(es) to the receiving slave. Since
2613 * this is checking the request, sip/tip are swapped for
2616 * This is done to avoid endless looping when we can't reach the
2617 * arp_ip_target and fool ourselves with our own arp requests.
2619 if (bond_is_active_slave(slave))
2620 bond_validate_arp(bond, slave, sip, tip);
2621 else if (curr_active_slave &&
2622 time_after(slave_last_rx(bond, curr_active_slave),
2623 curr_active_slave->last_link_up))
2624 bond_validate_arp(bond, slave, tip, sip);
2625 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2626 bond_time_in_interval(bond,
2627 dev_trans_start(curr_arp_slave->dev), 1))
2628 bond_validate_arp(bond, slave, sip, tip);
2631 if (arp != (struct arphdr *)skb->data)
2633 return RX_HANDLER_ANOTHER;
2636 /* function to verify if we're in the arp_interval timeslice, returns true if
2637 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2638 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2640 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2643 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2645 return time_in_range(jiffies,
2646 last_act - delta_in_ticks,
2647 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2650 /* This function is called regularly to monitor each slave's link
2651 * ensuring that traffic is being sent and received when arp monitoring
2652 * is used in load-balancing mode. if the adapter has been dormant, then an
2653 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2654 * arp monitoring in active backup mode.
2656 static void bond_loadbalance_arp_mon(struct bonding *bond)
2658 struct slave *slave, *oldcurrent;
2659 struct list_head *iter;
2660 int do_failover = 0, slave_state_changed = 0;
2662 if (!bond_has_slaves(bond))
2667 oldcurrent = rcu_dereference(bond->curr_active_slave);
2668 /* see if any of the previous devices are up now (i.e. they have
2669 * xmt and rcv traffic). the curr_active_slave does not come into
2670 * the picture unless it is null. also, slave->last_link_up is not
2671 * needed here because we send an arp on each slave and give a slave
2672 * as long as it needs to get the tx/rx within the delta.
2673 * TODO: what about up/down delay in arp mode? it wasn't here before
2676 bond_for_each_slave_rcu(bond, slave, iter) {
2677 unsigned long trans_start = dev_trans_start(slave->dev);
2679 slave->new_link = BOND_LINK_NOCHANGE;
2681 if (slave->link != BOND_LINK_UP) {
2682 if (bond_time_in_interval(bond, trans_start, 1) &&
2683 bond_time_in_interval(bond, slave->last_rx, 1)) {
2685 slave->new_link = BOND_LINK_UP;
2686 slave_state_changed = 1;
2688 /* primary_slave has no meaning in round-robin
2689 * mode. the window of a slave being up and
2690 * curr_active_slave being null after enslaving
2694 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2698 netdev_info(bond->dev, "interface %s is now up\n",
2703 /* slave->link == BOND_LINK_UP */
2705 /* not all switches will respond to an arp request
2706 * when the source ip is 0, so don't take the link down
2707 * if we don't know our ip yet
2709 if (!bond_time_in_interval(bond, trans_start, 2) ||
2710 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2712 slave->new_link = BOND_LINK_DOWN;
2713 slave_state_changed = 1;
2715 if (slave->link_failure_count < UINT_MAX)
2716 slave->link_failure_count++;
2718 netdev_info(bond->dev, "interface %s is now down\n",
2721 if (slave == oldcurrent)
2726 /* note: if switch is in round-robin mode, all links
2727 * must tx arp to ensure all links rx an arp - otherwise
2728 * links may oscillate or not come up at all; if switch is
2729 * in something like xor mode, there is nothing we can
2730 * do - all replies will be rx'ed on same link causing slaves
2731 * to be unstable during low/no traffic periods
2733 if (bond_slave_is_up(slave))
2734 bond_arp_send_all(bond, slave);
2739 if (do_failover || slave_state_changed) {
2740 if (!rtnl_trylock())
2743 bond_for_each_slave(bond, slave, iter) {
2744 if (slave->new_link != BOND_LINK_NOCHANGE)
2745 slave->link = slave->new_link;
2748 if (slave_state_changed) {
2749 bond_slave_state_change(bond);
2750 if (BOND_MODE(bond) == BOND_MODE_XOR)
2751 bond_update_slave_arr(bond, NULL);
2755 bond_select_active_slave(bond);
2756 unblock_netpoll_tx();
2762 if (bond->params.arp_interval)
2763 queue_delayed_work(bond->wq, &bond->arp_work,
2764 msecs_to_jiffies(bond->params.arp_interval));
2767 /* Called to inspect slaves for active-backup mode ARP monitor link state
2768 * changes. Sets new_link in slaves to specify what action should take
2769 * place for the slave. Returns 0 if no changes are found, >0 if changes
2770 * to link states must be committed.
2772 * Called with rcu_read_lock held.
2774 static int bond_ab_arp_inspect(struct bonding *bond)
2776 unsigned long trans_start, last_rx;
2777 struct list_head *iter;
2778 struct slave *slave;
2781 bond_for_each_slave_rcu(bond, slave, iter) {
2782 slave->new_link = BOND_LINK_NOCHANGE;
2783 last_rx = slave_last_rx(bond, slave);
2785 if (slave->link != BOND_LINK_UP) {
2786 if (bond_time_in_interval(bond, last_rx, 1)) {
2787 slave->new_link = BOND_LINK_UP;
2793 /* Give slaves 2*delta after being enslaved or made
2794 * active. This avoids bouncing, as the last receive
2795 * times need a full ARP monitor cycle to be updated.
2797 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2800 /* Backup slave is down if:
2801 * - No current_arp_slave AND
2802 * - more than 3*delta since last receive AND
2803 * - the bond has an IP address
2805 * Note: a non-null current_arp_slave indicates
2806 * the curr_active_slave went down and we are
2807 * searching for a new one; under this condition
2808 * we only take the curr_active_slave down - this
2809 * gives each slave a chance to tx/rx traffic
2810 * before being taken out
2812 if (!bond_is_active_slave(slave) &&
2813 !rcu_access_pointer(bond->current_arp_slave) &&
2814 !bond_time_in_interval(bond, last_rx, 3)) {
2815 slave->new_link = BOND_LINK_DOWN;
2819 /* Active slave is down if:
2820 * - more than 2*delta since transmitting OR
2821 * - (more than 2*delta since receive AND
2822 * the bond has an IP address)
2824 trans_start = dev_trans_start(slave->dev);
2825 if (bond_is_active_slave(slave) &&
2826 (!bond_time_in_interval(bond, trans_start, 2) ||
2827 !bond_time_in_interval(bond, last_rx, 2))) {
2828 slave->new_link = BOND_LINK_DOWN;
2836 /* Called to commit link state changes noted by inspection step of
2837 * active-backup mode ARP monitor.
2839 * Called with RTNL hold.
2841 static void bond_ab_arp_commit(struct bonding *bond)
2843 unsigned long trans_start;
2844 struct list_head *iter;
2845 struct slave *slave;
2847 bond_for_each_slave(bond, slave, iter) {
2848 switch (slave->new_link) {
2849 case BOND_LINK_NOCHANGE:
2853 trans_start = dev_trans_start(slave->dev);
2854 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2855 (!rtnl_dereference(bond->curr_active_slave) &&
2856 bond_time_in_interval(bond, trans_start, 1))) {
2857 struct slave *current_arp_slave;
2859 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2860 bond_set_slave_link_state(slave, BOND_LINK_UP,
2861 BOND_SLAVE_NOTIFY_NOW);
2862 if (current_arp_slave) {
2863 bond_set_slave_inactive_flags(
2865 BOND_SLAVE_NOTIFY_NOW);
2866 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2869 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2872 if (!rtnl_dereference(bond->curr_active_slave) ||
2873 slave == rtnl_dereference(bond->primary_slave))
2880 case BOND_LINK_DOWN:
2881 if (slave->link_failure_count < UINT_MAX)
2882 slave->link_failure_count++;
2884 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2885 BOND_SLAVE_NOTIFY_NOW);
2886 bond_set_slave_inactive_flags(slave,
2887 BOND_SLAVE_NOTIFY_NOW);
2889 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2892 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2893 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2900 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2901 slave->new_link, slave->dev->name);
2907 bond_select_active_slave(bond);
2908 unblock_netpoll_tx();
2911 bond_set_carrier(bond);
2914 /* Send ARP probes for active-backup mode ARP monitor.
2916 * Called with rcu_read_lock held.
2918 static bool bond_ab_arp_probe(struct bonding *bond)
2920 struct slave *slave, *before = NULL, *new_slave = NULL,
2921 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2922 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2923 struct list_head *iter;
2925 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2927 if (curr_arp_slave && curr_active_slave)
2928 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2929 curr_arp_slave->dev->name,
2930 curr_active_slave->dev->name);
2932 if (curr_active_slave) {
2933 bond_arp_send_all(bond, curr_active_slave);
2934 return should_notify_rtnl;
2937 /* if we don't have a curr_active_slave, search for the next available
2938 * backup slave from the current_arp_slave and make it the candidate
2939 * for becoming the curr_active_slave
2942 if (!curr_arp_slave) {
2943 curr_arp_slave = bond_first_slave_rcu(bond);
2944 if (!curr_arp_slave)
2945 return should_notify_rtnl;
2948 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2950 bond_for_each_slave_rcu(bond, slave, iter) {
2951 if (!found && !before && bond_slave_is_up(slave))
2954 if (found && !new_slave && bond_slave_is_up(slave))
2956 /* if the link state is up at this point, we
2957 * mark it down - this can happen if we have
2958 * simultaneous link failures and
2959 * reselect_active_interface doesn't make this
2960 * one the current slave so it is still marked
2961 * up when it is actually down
2963 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2964 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2965 BOND_SLAVE_NOTIFY_LATER);
2966 if (slave->link_failure_count < UINT_MAX)
2967 slave->link_failure_count++;
2969 bond_set_slave_inactive_flags(slave,
2970 BOND_SLAVE_NOTIFY_LATER);
2972 netdev_info(bond->dev, "backup interface %s is now down\n",
2975 if (slave == curr_arp_slave)
2979 if (!new_slave && before)
2985 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2986 BOND_SLAVE_NOTIFY_LATER);
2987 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2988 bond_arp_send_all(bond, new_slave);
2989 new_slave->last_link_up = jiffies;
2990 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2993 bond_for_each_slave_rcu(bond, slave, iter) {
2994 if (slave->should_notify || slave->should_notify_link) {
2995 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2999 return should_notify_rtnl;
3002 static void bond_activebackup_arp_mon(struct bonding *bond)
3004 bool should_notify_peers = false;
3005 bool should_notify_rtnl = false;
3008 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3010 if (!bond_has_slaves(bond))
3015 should_notify_peers = bond_should_notify_peers(bond);
3017 if (bond_ab_arp_inspect(bond)) {
3020 /* Race avoidance with bond_close flush of workqueue */
3021 if (!rtnl_trylock()) {
3023 should_notify_peers = false;
3027 bond_ab_arp_commit(bond);
3033 should_notify_rtnl = bond_ab_arp_probe(bond);
3037 if (bond->params.arp_interval)
3038 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3040 if (should_notify_peers || should_notify_rtnl) {
3041 if (!rtnl_trylock())
3044 if (should_notify_peers)
3045 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3047 if (should_notify_rtnl) {
3048 bond_slave_state_notify(bond);
3049 bond_slave_link_notify(bond);
3056 static void bond_arp_monitor(struct work_struct *work)
3058 struct bonding *bond = container_of(work, struct bonding,
3061 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3062 bond_activebackup_arp_mon(bond);
3064 bond_loadbalance_arp_mon(bond);
3067 /*-------------------------- netdev event handling --------------------------*/
3069 /* Change device name */
3070 static int bond_event_changename(struct bonding *bond)
3072 bond_remove_proc_entry(bond);
3073 bond_create_proc_entry(bond);
3075 bond_debug_reregister(bond);
3080 static int bond_master_netdev_event(unsigned long event,
3081 struct net_device *bond_dev)
3083 struct bonding *event_bond = netdev_priv(bond_dev);
3086 case NETDEV_CHANGENAME:
3087 return bond_event_changename(event_bond);
3088 case NETDEV_UNREGISTER:
3089 bond_remove_proc_entry(event_bond);
3091 case NETDEV_REGISTER:
3092 bond_create_proc_entry(event_bond);
3094 case NETDEV_NOTIFY_PEERS:
3095 if (event_bond->send_peer_notif)
3096 event_bond->send_peer_notif--;
3105 static int bond_slave_netdev_event(unsigned long event,
3106 struct net_device *slave_dev)
3108 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3109 struct bonding *bond;
3110 struct net_device *bond_dev;
3112 /* A netdev event can be generated while enslaving a device
3113 * before netdev_rx_handler_register is called in which case
3114 * slave will be NULL
3118 bond_dev = slave->bond->dev;
3120 primary = rtnl_dereference(bond->primary_slave);
3123 case NETDEV_UNREGISTER:
3124 if (bond_dev->type != ARPHRD_ETHER)
3125 bond_release_and_destroy(bond_dev, slave_dev);
3127 __bond_release_one(bond_dev, slave_dev, false, true);
3131 /* For 802.3ad mode only:
3132 * Getting invalid Speed/Duplex values here will put slave
3133 * in weird state. So mark it as link-fail for the time
3134 * being and let link-monitoring (miimon) set it right when
3135 * correct speeds/duplex are available.
3137 if (bond_update_speed_duplex(slave) &&
3138 BOND_MODE(bond) == BOND_MODE_8023AD)
3139 slave->link = BOND_LINK_FAIL;
3141 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3142 bond_3ad_adapter_speed_duplex_changed(slave);
3145 /* Refresh slave-array if applicable!
3146 * If the setup does not use miimon or arpmon (mode-specific!),
3147 * then these events will not cause the slave-array to be
3148 * refreshed. This will cause xmit to use a slave that is not
3149 * usable. Avoid such situation by refeshing the array at these
3150 * events. If these (miimon/arpmon) parameters are configured
3151 * then array gets refreshed twice and that should be fine!
3153 if (bond_mode_can_use_xmit_hash(bond))
3154 bond_update_slave_arr(bond, NULL);
3156 case NETDEV_CHANGEMTU:
3157 /* TODO: Should slaves be allowed to
3158 * independently alter their MTU? For
3159 * an active-backup bond, slaves need
3160 * not be the same type of device, so
3161 * MTUs may vary. For other modes,
3162 * slaves arguably should have the
3163 * same MTUs. To do this, we'd need to
3164 * take over the slave's change_mtu
3165 * function for the duration of their
3169 case NETDEV_CHANGENAME:
3170 /* we don't care if we don't have primary set */
3171 if (!bond_uses_primary(bond) ||
3172 !bond->params.primary[0])
3175 if (slave == primary) {
3176 /* slave's name changed - he's no longer primary */
3177 RCU_INIT_POINTER(bond->primary_slave, NULL);
3178 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3179 /* we have a new primary slave */
3180 rcu_assign_pointer(bond->primary_slave, slave);
3181 } else { /* we didn't change primary - exit */
3185 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3186 primary ? slave_dev->name : "none");
3189 bond_select_active_slave(bond);
3190 unblock_netpoll_tx();
3192 case NETDEV_FEAT_CHANGE:
3193 bond_compute_features(bond);
3195 case NETDEV_RESEND_IGMP:
3196 /* Propagate to master device */
3197 call_netdevice_notifiers(event, slave->bond->dev);
3206 /* bond_netdev_event: handle netdev notifier chain events.
3208 * This function receives events for the netdev chain. The caller (an
3209 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3210 * locks for us to safely manipulate the slave devices (RTNL lock,
3213 static int bond_netdev_event(struct notifier_block *this,
3214 unsigned long event, void *ptr)
3216 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3218 netdev_dbg(event_dev, "event: %lx\n", event);
3220 if (!(event_dev->priv_flags & IFF_BONDING))
3223 if (event_dev->flags & IFF_MASTER) {
3224 netdev_dbg(event_dev, "IFF_MASTER\n");
3225 return bond_master_netdev_event(event, event_dev);
3228 if (event_dev->flags & IFF_SLAVE) {
3229 netdev_dbg(event_dev, "IFF_SLAVE\n");
3230 return bond_slave_netdev_event(event, event_dev);
3236 static struct notifier_block bond_netdev_notifier = {
3237 .notifier_call = bond_netdev_event,
3240 /*---------------------------- Hashing Policies -----------------------------*/
3242 /* L2 hash helper */
3243 static inline u32 bond_eth_hash(struct sk_buff *skb)
3245 struct ethhdr *ep, hdr_tmp;
3247 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3249 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3253 /* Extract the appropriate headers based on bond's xmit policy */
3254 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3255 struct flow_keys *fk)
3257 const struct ipv6hdr *iph6;
3258 const struct iphdr *iph;
3259 int noff, proto = -1;
3261 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3262 return skb_flow_dissect_flow_keys(skb, fk, 0);
3264 fk->ports.ports = 0;
3265 noff = skb_network_offset(skb);
3266 if (skb->protocol == htons(ETH_P_IP)) {
3267 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3270 iph_to_flow_copy_v4addrs(fk, iph);
3271 noff += iph->ihl << 2;
3272 if (!ip_is_fragment(iph))
3273 proto = iph->protocol;
3274 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3275 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3277 iph6 = ipv6_hdr(skb);
3278 iph_to_flow_copy_v6addrs(fk, iph6);
3279 noff += sizeof(*iph6);
3280 proto = iph6->nexthdr;
3284 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3285 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3291 * bond_xmit_hash - generate a hash value based on the xmit policy
3292 * @bond: bonding device
3293 * @skb: buffer to use for headers
3295 * This function will extract the necessary headers from the skb buffer and use
3296 * them to generate a hash based on the xmit_policy set in the bonding device
3298 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3300 struct flow_keys flow;
3303 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3307 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3308 !bond_flow_dissect(bond, skb, &flow))
3309 return bond_eth_hash(skb);
3311 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3312 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3313 hash = bond_eth_hash(skb);
3315 hash = (__force u32)flow.ports.ports;
3316 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3317 (__force u32)flow_get_u32_src(&flow);
3318 hash ^= (hash >> 16);
3319 hash ^= (hash >> 8);
3324 /*-------------------------- Device entry points ----------------------------*/
3326 void bond_work_init_all(struct bonding *bond)
3328 INIT_DELAYED_WORK(&bond->mcast_work,
3329 bond_resend_igmp_join_requests_delayed);
3330 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3331 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3332 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3333 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3334 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3337 static void bond_work_cancel_all(struct bonding *bond)
3339 cancel_delayed_work_sync(&bond->mii_work);
3340 cancel_delayed_work_sync(&bond->arp_work);
3341 cancel_delayed_work_sync(&bond->alb_work);
3342 cancel_delayed_work_sync(&bond->ad_work);
3343 cancel_delayed_work_sync(&bond->mcast_work);
3344 cancel_delayed_work_sync(&bond->slave_arr_work);
3347 static int bond_open(struct net_device *bond_dev)
3349 struct bonding *bond = netdev_priv(bond_dev);
3350 struct list_head *iter;
3351 struct slave *slave;
3353 /* reset slave->backup and slave->inactive */
3354 if (bond_has_slaves(bond)) {
3355 bond_for_each_slave(bond, slave, iter) {
3356 if (bond_uses_primary(bond) &&
3357 slave != rcu_access_pointer(bond->curr_active_slave)) {
3358 bond_set_slave_inactive_flags(slave,
3359 BOND_SLAVE_NOTIFY_NOW);
3360 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3361 bond_set_slave_active_flags(slave,
3362 BOND_SLAVE_NOTIFY_NOW);
3367 if (bond_is_lb(bond)) {
3368 /* bond_alb_initialize must be called before the timer
3371 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3373 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3374 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3377 if (bond->params.miimon) /* link check interval, in milliseconds. */
3378 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3380 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3381 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3382 bond->recv_probe = bond_arp_rcv;
3385 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3386 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3387 /* register to receive LACPDUs */
3388 bond->recv_probe = bond_3ad_lacpdu_recv;
3389 bond_3ad_initiate_agg_selection(bond, 1);
3392 if (bond_mode_can_use_xmit_hash(bond))
3393 bond_update_slave_arr(bond, NULL);
3398 static int bond_close(struct net_device *bond_dev)
3400 struct bonding *bond = netdev_priv(bond_dev);
3402 bond_work_cancel_all(bond);
3403 bond->send_peer_notif = 0;
3404 if (bond_is_lb(bond))
3405 bond_alb_deinitialize(bond);
3406 bond->recv_probe = NULL;
3411 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3412 * that some drivers can provide 32bit values only.
3414 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3415 const struct rtnl_link_stats64 *_new,
3416 const struct rtnl_link_stats64 *_old)
3418 const u64 *new = (const u64 *)_new;
3419 const u64 *old = (const u64 *)_old;
3420 u64 *res = (u64 *)_res;
3423 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3426 s64 delta = nv - ov;
3428 /* detects if this particular field is 32bit only */
3429 if (((nv | ov) >> 32) == 0)
3430 delta = (s64)(s32)((u32)nv - (u32)ov);
3432 /* filter anomalies, some drivers reset their stats
3433 * at down/up events.
3440 static int bond_get_nest_level(struct net_device *bond_dev)
3442 struct bonding *bond = netdev_priv(bond_dev);
3444 return bond->nest_level;
3447 static void bond_get_stats(struct net_device *bond_dev,
3448 struct rtnl_link_stats64 *stats)
3450 struct bonding *bond = netdev_priv(bond_dev);
3451 struct rtnl_link_stats64 temp;
3452 struct list_head *iter;
3453 struct slave *slave;
3455 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3456 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3459 bond_for_each_slave_rcu(bond, slave, iter) {
3460 const struct rtnl_link_stats64 *new =
3461 dev_get_stats(slave->dev, &temp);
3463 bond_fold_stats(stats, new, &slave->slave_stats);
3465 /* save off the slave stats for the next run */
3466 memcpy(&slave->slave_stats, new, sizeof(*new));
3470 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3471 spin_unlock(&bond->stats_lock);
3474 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3476 struct bonding *bond = netdev_priv(bond_dev);
3477 struct net_device *slave_dev = NULL;
3478 struct ifbond k_binfo;
3479 struct ifbond __user *u_binfo = NULL;
3480 struct ifslave k_sinfo;
3481 struct ifslave __user *u_sinfo = NULL;
3482 struct mii_ioctl_data *mii = NULL;
3483 struct bond_opt_value newval;
3487 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3498 /* We do this again just in case we were called by SIOCGMIIREG
3499 * instead of SIOCGMIIPHY.
3505 if (mii->reg_num == 1) {
3507 if (netif_carrier_ok(bond->dev))
3508 mii->val_out = BMSR_LSTATUS;
3512 case BOND_INFO_QUERY_OLD:
3513 case SIOCBONDINFOQUERY:
3514 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3516 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3519 bond_info_query(bond_dev, &k_binfo);
3520 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3524 case BOND_SLAVE_INFO_QUERY_OLD:
3525 case SIOCBONDSLAVEINFOQUERY:
3526 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3528 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3531 res = bond_slave_info_query(bond_dev, &k_sinfo);
3533 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3541 net = dev_net(bond_dev);
3543 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3546 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3548 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3553 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3555 case BOND_ENSLAVE_OLD:
3556 case SIOCBONDENSLAVE:
3557 res = bond_enslave(bond_dev, slave_dev, NULL);
3559 case BOND_RELEASE_OLD:
3560 case SIOCBONDRELEASE:
3561 res = bond_release(bond_dev, slave_dev);
3563 case BOND_SETHWADDR_OLD:
3564 case SIOCBONDSETHWADDR:
3565 res = bond_set_dev_addr(bond_dev, slave_dev);
3567 case BOND_CHANGE_ACTIVE_OLD:
3568 case SIOCBONDCHANGEACTIVE:
3569 bond_opt_initstr(&newval, slave_dev->name);
3570 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3580 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3582 struct bonding *bond = netdev_priv(bond_dev);
3584 if (change & IFF_PROMISC)
3585 bond_set_promiscuity(bond,
3586 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3588 if (change & IFF_ALLMULTI)
3589 bond_set_allmulti(bond,
3590 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3593 static void bond_set_rx_mode(struct net_device *bond_dev)
3595 struct bonding *bond = netdev_priv(bond_dev);
3596 struct list_head *iter;
3597 struct slave *slave;
3600 if (bond_uses_primary(bond)) {
3601 slave = rcu_dereference(bond->curr_active_slave);
3603 dev_uc_sync(slave->dev, bond_dev);
3604 dev_mc_sync(slave->dev, bond_dev);
3607 bond_for_each_slave_rcu(bond, slave, iter) {
3608 dev_uc_sync_multiple(slave->dev, bond_dev);
3609 dev_mc_sync_multiple(slave->dev, bond_dev);
3615 static int bond_neigh_init(struct neighbour *n)
3617 struct bonding *bond = netdev_priv(n->dev);
3618 const struct net_device_ops *slave_ops;
3619 struct neigh_parms parms;
3620 struct slave *slave;
3623 slave = bond_first_slave(bond);
3626 slave_ops = slave->dev->netdev_ops;
3627 if (!slave_ops->ndo_neigh_setup)
3630 parms.neigh_setup = NULL;
3631 parms.neigh_cleanup = NULL;
3632 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3636 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3637 * after the last slave has been detached. Assumes that all slaves
3638 * utilize the same neigh_cleanup (true at this writing as only user
3641 n->parms->neigh_cleanup = parms.neigh_cleanup;
3643 if (!parms.neigh_setup)
3646 return parms.neigh_setup(n);
3649 /* The bonding ndo_neigh_setup is called at init time beofre any
3650 * slave exists. So we must declare proxy setup function which will
3651 * be used at run time to resolve the actual slave neigh param setup.
3653 * It's also called by master devices (such as vlans) to setup their
3654 * underlying devices. In that case - do nothing, we're already set up from
3657 static int bond_neigh_setup(struct net_device *dev,
3658 struct neigh_parms *parms)
3660 /* modify only our neigh_parms */
3661 if (parms->dev == dev)
3662 parms->neigh_setup = bond_neigh_init;
3667 /* Change the MTU of all of a master's slaves to match the master */
3668 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3670 struct bonding *bond = netdev_priv(bond_dev);
3671 struct slave *slave, *rollback_slave;
3672 struct list_head *iter;
3675 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3677 bond_for_each_slave(bond, slave, iter) {
3678 netdev_dbg(bond_dev, "s %p c_m %p\n",
3679 slave, slave->dev->netdev_ops->ndo_change_mtu);
3681 res = dev_set_mtu(slave->dev, new_mtu);
3684 /* If we failed to set the slave's mtu to the new value
3685 * we must abort the operation even in ACTIVE_BACKUP
3686 * mode, because if we allow the backup slaves to have
3687 * different mtu values than the active slave we'll
3688 * need to change their mtu when doing a failover. That
3689 * means changing their mtu from timer context, which
3690 * is probably not a good idea.
3692 netdev_dbg(bond_dev, "err %d %s\n", res,
3698 bond_dev->mtu = new_mtu;
3703 /* unwind from head to the slave that failed */
3704 bond_for_each_slave(bond, rollback_slave, iter) {
3707 if (rollback_slave == slave)
3710 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3712 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3713 tmp_res, rollback_slave->dev->name);
3720 /* Change HW address
3722 * Note that many devices must be down to change the HW address, and
3723 * downing the master releases all slaves. We can make bonds full of
3724 * bonding devices to test this, however.
3726 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3728 struct bonding *bond = netdev_priv(bond_dev);
3729 struct slave *slave, *rollback_slave;
3730 struct sockaddr_storage *ss = addr, tmp_ss;
3731 struct list_head *iter;
3734 if (BOND_MODE(bond) == BOND_MODE_ALB)
3735 return bond_alb_set_mac_address(bond_dev, addr);
3738 netdev_dbg(bond_dev, "bond=%p\n", bond);
3740 /* If fail_over_mac is enabled, do nothing and return success.
3741 * Returning an error causes ifenslave to fail.
3743 if (bond->params.fail_over_mac &&
3744 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3747 if (!is_valid_ether_addr(ss->__data))
3748 return -EADDRNOTAVAIL;
3750 bond_for_each_slave(bond, slave, iter) {
3751 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3752 res = dev_set_mac_address(slave->dev, addr, NULL);
3754 /* TODO: consider downing the slave
3756 * User should expect communications
3757 * breakage anyway until ARP finish
3760 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3766 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3770 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3771 tmp_ss.ss_family = bond_dev->type;
3773 /* unwind from head to the slave that failed */
3774 bond_for_each_slave(bond, rollback_slave, iter) {
3777 if (rollback_slave == slave)
3780 tmp_res = dev_set_mac_address(rollback_slave->dev,
3781 (struct sockaddr *)&tmp_ss, NULL);
3783 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3784 tmp_res, rollback_slave->dev->name);
3792 * bond_xmit_slave_id - transmit skb through slave with slave_id
3793 * @bond: bonding device that is transmitting
3794 * @skb: buffer to transmit
3795 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3797 * This function tries to transmit through slave with slave_id but in case
3798 * it fails, it tries to find the first available slave for transmission.
3799 * The skb is consumed in all cases, thus the function is void.
3801 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3803 struct list_head *iter;
3804 struct slave *slave;
3807 /* Here we start from the slave with slave_id */
3808 bond_for_each_slave_rcu(bond, slave, iter) {
3810 if (bond_slave_can_tx(slave)) {
3811 bond_dev_queue_xmit(bond, skb, slave->dev);
3817 /* Here we start from the first slave up to slave_id */
3819 bond_for_each_slave_rcu(bond, slave, iter) {
3822 if (bond_slave_can_tx(slave)) {
3823 bond_dev_queue_xmit(bond, skb, slave->dev);
3827 /* no slave that can tx has been found */
3828 bond_tx_drop(bond->dev, skb);
3832 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3833 * @bond: bonding device to use
3835 * Based on the value of the bonding device's packets_per_slave parameter
3836 * this function generates a slave id, which is usually used as the next
3837 * slave to transmit through.
3839 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3842 struct reciprocal_value reciprocal_packets_per_slave;
3843 int packets_per_slave = bond->params.packets_per_slave;
3845 switch (packets_per_slave) {
3847 slave_id = prandom_u32();
3850 slave_id = bond->rr_tx_counter;
3853 reciprocal_packets_per_slave =
3854 bond->params.reciprocal_packets_per_slave;
3855 slave_id = reciprocal_divide(bond->rr_tx_counter,
3856 reciprocal_packets_per_slave);
3859 bond->rr_tx_counter++;
3864 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3865 struct net_device *bond_dev)
3867 struct bonding *bond = netdev_priv(bond_dev);
3868 struct iphdr *iph = ip_hdr(skb);
3869 struct slave *slave;
3872 /* Start with the curr_active_slave that joined the bond as the
3873 * default for sending IGMP traffic. For failover purposes one
3874 * needs to maintain some consistency for the interface that will
3875 * send the join/membership reports. The curr_active_slave found
3876 * will send all of this type of traffic.
3878 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3879 slave = rcu_dereference(bond->curr_active_slave);
3881 bond_dev_queue_xmit(bond, skb, slave->dev);
3883 bond_xmit_slave_id(bond, skb, 0);
3885 int slave_cnt = READ_ONCE(bond->slave_cnt);
3887 if (likely(slave_cnt)) {
3888 slave_id = bond_rr_gen_slave_id(bond);
3889 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3891 bond_tx_drop(bond_dev, skb);
3895 return NETDEV_TX_OK;
3898 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3899 * the bond has a usable interface.
3901 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3902 struct net_device *bond_dev)
3904 struct bonding *bond = netdev_priv(bond_dev);
3905 struct slave *slave;
3907 slave = rcu_dereference(bond->curr_active_slave);
3909 bond_dev_queue_xmit(bond, skb, slave->dev);
3911 bond_tx_drop(bond_dev, skb);
3913 return NETDEV_TX_OK;
3916 /* Use this to update slave_array when (a) it's not appropriate to update
3917 * slave_array right away (note that update_slave_array() may sleep)
3918 * and / or (b) RTNL is not held.
3920 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3922 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3925 /* Slave array work handler. Holds only RTNL */
3926 static void bond_slave_arr_handler(struct work_struct *work)
3928 struct bonding *bond = container_of(work, struct bonding,
3929 slave_arr_work.work);
3932 if (!rtnl_trylock())
3935 ret = bond_update_slave_arr(bond, NULL);
3938 pr_warn_ratelimited("Failed to update slave array from WT\n");
3944 bond_slave_arr_work_rearm(bond, 1);
3947 /* Build the usable slaves array in control path for modes that use xmit-hash
3948 * to determine the slave interface -
3949 * (a) BOND_MODE_8023AD
3951 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3953 * The caller is expected to hold RTNL only and NO other lock!
3955 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3957 struct slave *slave;
3958 struct list_head *iter;
3959 struct bond_up_slave *new_arr, *old_arr;
3963 #ifdef CONFIG_LOCKDEP
3964 WARN_ON(lockdep_is_held(&bond->mode_lock));
3967 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3971 pr_err("Failed to build slave-array.\n");
3974 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3975 struct ad_info ad_info;
3977 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3978 pr_debug("bond_3ad_get_active_agg_info failed\n");
3979 kfree_rcu(new_arr, rcu);
3980 /* No active aggragator means it's not safe to use
3981 * the previous array.
3983 old_arr = rtnl_dereference(bond->slave_arr);
3985 RCU_INIT_POINTER(bond->slave_arr, NULL);
3986 kfree_rcu(old_arr, rcu);
3990 agg_id = ad_info.aggregator_id;
3992 bond_for_each_slave(bond, slave, iter) {
3993 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3994 struct aggregator *agg;
3996 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3997 if (!agg || agg->aggregator_identifier != agg_id)
4000 if (!bond_slave_can_tx(slave))
4002 if (skipslave == slave)
4005 netdev_dbg(bond->dev,
4006 "Adding slave dev %s to tx hash array[%d]\n",
4007 slave->dev->name, new_arr->count);
4009 new_arr->arr[new_arr->count++] = slave;
4012 old_arr = rtnl_dereference(bond->slave_arr);
4013 rcu_assign_pointer(bond->slave_arr, new_arr);
4015 kfree_rcu(old_arr, rcu);
4017 if (ret != 0 && skipslave) {
4020 /* Rare situation where caller has asked to skip a specific
4021 * slave but allocation failed (most likely!). BTW this is
4022 * only possible when the call is initiated from
4023 * __bond_release_one(). In this situation; overwrite the
4024 * skipslave entry in the array with the last entry from the
4025 * array to avoid a situation where the xmit path may choose
4026 * this to-be-skipped slave to send a packet out.
4028 old_arr = rtnl_dereference(bond->slave_arr);
4029 for (idx = 0; idx < old_arr->count; idx++) {
4030 if (skipslave == old_arr->arr[idx]) {
4032 old_arr->arr[old_arr->count-1];
4041 /* Use this Xmit function for 3AD as well as XOR modes. The current
4042 * usable slave array is formed in the control path. The xmit function
4043 * just calculates hash and sends the packet out.
4045 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4046 struct net_device *dev)
4048 struct bonding *bond = netdev_priv(dev);
4049 struct slave *slave;
4050 struct bond_up_slave *slaves;
4053 slaves = rcu_dereference(bond->slave_arr);
4054 count = slaves ? READ_ONCE(slaves->count) : 0;
4055 if (likely(count)) {
4056 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4057 bond_dev_queue_xmit(bond, skb, slave->dev);
4059 bond_tx_drop(dev, skb);
4062 return NETDEV_TX_OK;
4065 /* in broadcast mode, we send everything to all usable interfaces. */
4066 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4067 struct net_device *bond_dev)
4069 struct bonding *bond = netdev_priv(bond_dev);
4070 struct slave *slave = NULL;
4071 struct list_head *iter;
4073 bond_for_each_slave_rcu(bond, slave, iter) {
4074 if (bond_is_last_slave(bond, slave))
4076 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4077 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4080 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4081 bond_dev->name, __func__);
4084 bond_dev_queue_xmit(bond, skb2, slave->dev);
4087 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4088 bond_dev_queue_xmit(bond, skb, slave->dev);
4090 bond_tx_drop(bond_dev, skb);
4092 return NETDEV_TX_OK;
4095 /*------------------------- Device initialization ---------------------------*/
4097 /* Lookup the slave that corresponds to a qid */
4098 static inline int bond_slave_override(struct bonding *bond,
4099 struct sk_buff *skb)
4101 struct slave *slave = NULL;
4102 struct list_head *iter;
4104 if (!skb_rx_queue_recorded(skb))
4107 /* Find out if any slaves have the same mapping as this skb. */
4108 bond_for_each_slave_rcu(bond, slave, iter) {
4109 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4110 if (bond_slave_is_up(slave) &&
4111 slave->link == BOND_LINK_UP) {
4112 bond_dev_queue_xmit(bond, skb, slave->dev);
4115 /* If the slave isn't UP, use default transmit policy. */
4124 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4125 struct net_device *sb_dev,
4126 select_queue_fallback_t fallback)
4128 /* This helper function exists to help dev_pick_tx get the correct
4129 * destination queue. Using a helper function skips a call to
4130 * skb_tx_hash and will put the skbs in the queue we expect on their
4131 * way down to the bonding driver.
4133 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4135 /* Save the original txq to restore before passing to the driver */
4136 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4138 if (unlikely(txq >= dev->real_num_tx_queues)) {
4140 txq -= dev->real_num_tx_queues;
4141 } while (txq >= dev->real_num_tx_queues);
4146 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4148 struct bonding *bond = netdev_priv(dev);
4150 if (bond_should_override_tx_queue(bond) &&
4151 !bond_slave_override(bond, skb))
4152 return NETDEV_TX_OK;
4154 switch (BOND_MODE(bond)) {
4155 case BOND_MODE_ROUNDROBIN:
4156 return bond_xmit_roundrobin(skb, dev);
4157 case BOND_MODE_ACTIVEBACKUP:
4158 return bond_xmit_activebackup(skb, dev);
4159 case BOND_MODE_8023AD:
4161 return bond_3ad_xor_xmit(skb, dev);
4162 case BOND_MODE_BROADCAST:
4163 return bond_xmit_broadcast(skb, dev);
4165 return bond_alb_xmit(skb, dev);
4167 return bond_tlb_xmit(skb, dev);
4169 /* Should never happen, mode already checked */
4170 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4172 bond_tx_drop(dev, skb);
4173 return NETDEV_TX_OK;
4177 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4179 struct bonding *bond = netdev_priv(dev);
4180 netdev_tx_t ret = NETDEV_TX_OK;
4182 /* If we risk deadlock from transmitting this in the
4183 * netpoll path, tell netpoll to queue the frame for later tx
4185 if (unlikely(is_netpoll_tx_blocked(dev)))
4186 return NETDEV_TX_BUSY;
4189 if (bond_has_slaves(bond))
4190 ret = __bond_start_xmit(skb, dev);
4192 bond_tx_drop(dev, skb);
4198 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4199 struct ethtool_link_ksettings *cmd)
4201 struct bonding *bond = netdev_priv(bond_dev);
4202 unsigned long speed = 0;
4203 struct list_head *iter;
4204 struct slave *slave;
4206 cmd->base.duplex = DUPLEX_UNKNOWN;
4207 cmd->base.port = PORT_OTHER;
4209 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4210 * do not need to check mode. Though link speed might not represent
4211 * the true receive or transmit bandwidth (not all modes are symmetric)
4212 * this is an accurate maximum.
4214 bond_for_each_slave(bond, slave, iter) {
4215 if (bond_slave_can_tx(slave)) {
4216 if (slave->speed != SPEED_UNKNOWN)
4217 speed += slave->speed;
4218 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4219 slave->duplex != DUPLEX_UNKNOWN)
4220 cmd->base.duplex = slave->duplex;
4223 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4228 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4229 struct ethtool_drvinfo *drvinfo)
4231 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4232 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4233 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4237 static const struct ethtool_ops bond_ethtool_ops = {
4238 .get_drvinfo = bond_ethtool_get_drvinfo,
4239 .get_link = ethtool_op_get_link,
4240 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4243 static const struct net_device_ops bond_netdev_ops = {
4244 .ndo_init = bond_init,
4245 .ndo_uninit = bond_uninit,
4246 .ndo_open = bond_open,
4247 .ndo_stop = bond_close,
4248 .ndo_start_xmit = bond_start_xmit,
4249 .ndo_select_queue = bond_select_queue,
4250 .ndo_get_stats64 = bond_get_stats,
4251 .ndo_do_ioctl = bond_do_ioctl,
4252 .ndo_change_rx_flags = bond_change_rx_flags,
4253 .ndo_set_rx_mode = bond_set_rx_mode,
4254 .ndo_change_mtu = bond_change_mtu,
4255 .ndo_set_mac_address = bond_set_mac_address,
4256 .ndo_neigh_setup = bond_neigh_setup,
4257 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4258 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4259 .ndo_get_lock_subclass = bond_get_nest_level,
4260 #ifdef CONFIG_NET_POLL_CONTROLLER
4261 .ndo_netpoll_setup = bond_netpoll_setup,
4262 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4263 .ndo_poll_controller = bond_poll_controller,
4265 .ndo_add_slave = bond_enslave,
4266 .ndo_del_slave = bond_release,
4267 .ndo_fix_features = bond_fix_features,
4268 .ndo_features_check = passthru_features_check,
4271 static const struct device_type bond_type = {
4275 static void bond_destructor(struct net_device *bond_dev)
4277 struct bonding *bond = netdev_priv(bond_dev);
4279 destroy_workqueue(bond->wq);
4282 void bond_setup(struct net_device *bond_dev)
4284 struct bonding *bond = netdev_priv(bond_dev);
4286 spin_lock_init(&bond->mode_lock);
4287 spin_lock_init(&bond->stats_lock);
4288 bond->params = bonding_defaults;
4290 /* Initialize pointers */
4291 bond->dev = bond_dev;
4293 /* Initialize the device entry points */
4294 ether_setup(bond_dev);
4295 bond_dev->max_mtu = ETH_MAX_MTU;
4296 bond_dev->netdev_ops = &bond_netdev_ops;
4297 bond_dev->ethtool_ops = &bond_ethtool_ops;
4299 bond_dev->needs_free_netdev = true;
4300 bond_dev->priv_destructor = bond_destructor;
4302 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4304 /* Initialize the device options */
4305 bond_dev->flags |= IFF_MASTER;
4306 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4307 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4309 /* don't acquire bond device's netif_tx_lock when transmitting */
4310 bond_dev->features |= NETIF_F_LLTX;
4312 /* By default, we declare the bond to be fully
4313 * VLAN hardware accelerated capable. Special
4314 * care is taken in the various xmit functions
4315 * when there are slaves that are not hw accel
4319 /* Don't allow bond devices to change network namespaces. */
4320 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4322 bond_dev->hw_features = BOND_VLAN_FEATURES |
4323 NETIF_F_HW_VLAN_CTAG_TX |
4324 NETIF_F_HW_VLAN_CTAG_RX |
4325 NETIF_F_HW_VLAN_CTAG_FILTER;
4327 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4328 bond_dev->features |= bond_dev->hw_features;
4331 /* Destroy a bonding device.
4332 * Must be under rtnl_lock when this function is called.
4334 static void bond_uninit(struct net_device *bond_dev)
4336 struct bonding *bond = netdev_priv(bond_dev);
4337 struct list_head *iter;
4338 struct slave *slave;
4339 struct bond_up_slave *arr;
4341 bond_netpoll_cleanup(bond_dev);
4343 /* Release the bonded slaves */
4344 bond_for_each_slave(bond, slave, iter)
4345 __bond_release_one(bond_dev, slave->dev, true, true);
4346 netdev_info(bond_dev, "Released all slaves\n");
4348 arr = rtnl_dereference(bond->slave_arr);
4350 RCU_INIT_POINTER(bond->slave_arr, NULL);
4351 kfree_rcu(arr, rcu);
4354 list_del(&bond->bond_list);
4356 bond_debug_unregister(bond);
4359 /*------------------------- Module initialization ---------------------------*/
4361 static int bond_check_params(struct bond_params *params)
4363 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4364 struct bond_opt_value newval;
4365 const struct bond_opt_value *valptr;
4366 int arp_all_targets_value = 0;
4367 u16 ad_actor_sys_prio = 0;
4368 u16 ad_user_port_key = 0;
4369 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4371 int bond_mode = BOND_MODE_ROUNDROBIN;
4372 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4376 /* Convert string parameters. */
4378 bond_opt_initstr(&newval, mode);
4379 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4381 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4384 bond_mode = valptr->value;
4387 if (xmit_hash_policy) {
4388 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4389 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4390 bond_mode == BOND_MODE_BROADCAST) {
4391 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4392 bond_mode_name(bond_mode));
4394 bond_opt_initstr(&newval, xmit_hash_policy);
4395 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4398 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4402 xmit_hashtype = valptr->value;
4407 if (bond_mode != BOND_MODE_8023AD) {
4408 pr_info("lacp_rate param is irrelevant in mode %s\n",
4409 bond_mode_name(bond_mode));
4411 bond_opt_initstr(&newval, lacp_rate);
4412 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4415 pr_err("Error: Invalid lacp rate \"%s\"\n",
4419 lacp_fast = valptr->value;
4424 bond_opt_initstr(&newval, ad_select);
4425 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4428 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4431 params->ad_select = valptr->value;
4432 if (bond_mode != BOND_MODE_8023AD)
4433 pr_warn("ad_select param only affects 802.3ad mode\n");
4435 params->ad_select = BOND_AD_STABLE;
4438 if (max_bonds < 0) {
4439 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4440 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4441 max_bonds = BOND_DEFAULT_MAX_BONDS;
4445 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4451 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4456 if (downdelay < 0) {
4457 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4458 downdelay, INT_MAX);
4462 if ((use_carrier != 0) && (use_carrier != 1)) {
4463 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4468 if (num_peer_notif < 0 || num_peer_notif > 255) {
4469 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4474 /* reset values for 802.3ad/TLB/ALB */
4475 if (!bond_mode_uses_arp(bond_mode)) {
4477 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");
4478 pr_warn("Forcing miimon to 100msec\n");
4479 miimon = BOND_DEFAULT_MIIMON;
4483 if (tx_queues < 1 || tx_queues > 255) {
4484 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4485 tx_queues, BOND_DEFAULT_TX_QUEUES);
4486 tx_queues = BOND_DEFAULT_TX_QUEUES;
4489 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4490 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4492 all_slaves_active = 0;
4495 if (resend_igmp < 0 || resend_igmp > 255) {
4496 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4497 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4498 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4501 bond_opt_initval(&newval, packets_per_slave);
4502 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4503 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4504 packets_per_slave, USHRT_MAX);
4505 packets_per_slave = 1;
4508 if (bond_mode == BOND_MODE_ALB) {
4509 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",
4514 if (updelay || downdelay) {
4515 /* just warn the user the up/down delay will have
4516 * no effect since miimon is zero...
4518 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",
4519 updelay, downdelay);
4522 /* don't allow arp monitoring */
4524 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4525 miimon, arp_interval);
4529 if ((updelay % miimon) != 0) {
4530 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4531 updelay, miimon, (updelay / miimon) * miimon);
4536 if ((downdelay % miimon) != 0) {
4537 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4539 (downdelay / miimon) * miimon);
4542 downdelay /= miimon;
4545 if (arp_interval < 0) {
4546 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4547 arp_interval, INT_MAX);
4551 for (arp_ip_count = 0, i = 0;
4552 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4555 /* not a complete check, but good enough to catch mistakes */
4556 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4557 !bond_is_ip_target_ok(ip)) {
4558 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4562 if (bond_get_targets_ip(arp_target, ip) == -1)
4563 arp_target[arp_ip_count++] = ip;
4565 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4570 if (arp_interval && !arp_ip_count) {
4571 /* don't allow arping if no arp_ip_target given... */
4572 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4578 if (!arp_interval) {
4579 pr_err("arp_validate requires arp_interval\n");
4583 bond_opt_initstr(&newval, arp_validate);
4584 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4587 pr_err("Error: invalid arp_validate \"%s\"\n",
4591 arp_validate_value = valptr->value;
4593 arp_validate_value = 0;
4596 if (arp_all_targets) {
4597 bond_opt_initstr(&newval, arp_all_targets);
4598 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4601 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4603 arp_all_targets_value = 0;
4605 arp_all_targets_value = valptr->value;
4610 pr_info("MII link monitoring set to %d ms\n", miimon);
4611 } else if (arp_interval) {
4612 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4613 arp_validate_value);
4614 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4615 arp_interval, valptr->string, arp_ip_count);
4617 for (i = 0; i < arp_ip_count; i++)
4618 pr_cont(" %s", arp_ip_target[i]);
4622 } else if (max_bonds) {
4623 /* miimon and arp_interval not set, we need one so things
4624 * work as expected, see bonding.txt for details
4626 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");
4629 if (primary && !bond_mode_uses_primary(bond_mode)) {
4630 /* currently, using a primary only makes sense
4631 * in active backup, TLB or ALB modes
4633 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4634 primary, bond_mode_name(bond_mode));
4638 if (primary && primary_reselect) {
4639 bond_opt_initstr(&newval, primary_reselect);
4640 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4643 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4647 primary_reselect_value = valptr->value;
4649 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4652 if (fail_over_mac) {
4653 bond_opt_initstr(&newval, fail_over_mac);
4654 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4657 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4661 fail_over_mac_value = valptr->value;
4662 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4663 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4665 fail_over_mac_value = BOND_FOM_NONE;
4668 bond_opt_initstr(&newval, "default");
4669 valptr = bond_opt_parse(
4670 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4673 pr_err("Error: No ad_actor_sys_prio default value");
4676 ad_actor_sys_prio = valptr->value;
4678 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4681 pr_err("Error: No ad_user_port_key default value");
4684 ad_user_port_key = valptr->value;
4686 bond_opt_initstr(&newval, "default");
4687 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4689 pr_err("Error: No tlb_dynamic_lb default value");
4692 tlb_dynamic_lb = valptr->value;
4694 if (lp_interval == 0) {
4695 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4696 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4697 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4700 /* fill params struct with the proper values */
4701 params->mode = bond_mode;
4702 params->xmit_policy = xmit_hashtype;
4703 params->miimon = miimon;
4704 params->num_peer_notif = num_peer_notif;
4705 params->arp_interval = arp_interval;
4706 params->arp_validate = arp_validate_value;
4707 params->arp_all_targets = arp_all_targets_value;
4708 params->updelay = updelay;
4709 params->downdelay = downdelay;
4710 params->use_carrier = use_carrier;
4711 params->lacp_fast = lacp_fast;
4712 params->primary[0] = 0;
4713 params->primary_reselect = primary_reselect_value;
4714 params->fail_over_mac = fail_over_mac_value;
4715 params->tx_queues = tx_queues;
4716 params->all_slaves_active = all_slaves_active;
4717 params->resend_igmp = resend_igmp;
4718 params->min_links = min_links;
4719 params->lp_interval = lp_interval;
4720 params->packets_per_slave = packets_per_slave;
4721 params->tlb_dynamic_lb = tlb_dynamic_lb;
4722 params->ad_actor_sys_prio = ad_actor_sys_prio;
4723 eth_zero_addr(params->ad_actor_system);
4724 params->ad_user_port_key = ad_user_port_key;
4725 if (packets_per_slave > 0) {
4726 params->reciprocal_packets_per_slave =
4727 reciprocal_value(packets_per_slave);
4729 /* reciprocal_packets_per_slave is unused if
4730 * packets_per_slave is 0 or 1, just initialize it
4732 params->reciprocal_packets_per_slave =
4733 (struct reciprocal_value) { 0 };
4737 strncpy(params->primary, primary, IFNAMSIZ);
4738 params->primary[IFNAMSIZ - 1] = 0;
4741 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4746 /* Called from registration process */
4747 static int bond_init(struct net_device *bond_dev)
4749 struct bonding *bond = netdev_priv(bond_dev);
4750 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4752 netdev_dbg(bond_dev, "Begin bond_init\n");
4754 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4758 bond->nest_level = SINGLE_DEPTH_NESTING;
4759 netdev_lockdep_set_classes(bond_dev);
4761 list_add_tail(&bond->bond_list, &bn->dev_list);
4763 bond_prepare_sysfs_group(bond);
4765 bond_debug_register(bond);
4767 /* Ensure valid dev_addr */
4768 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4769 bond_dev->addr_assign_type == NET_ADDR_PERM)
4770 eth_hw_addr_random(bond_dev);
4775 unsigned int bond_get_num_tx_queues(void)
4780 /* Create a new bond based on the specified name and bonding parameters.
4781 * If name is NULL, obtain a suitable "bond%d" name for us.
4782 * Caller must NOT hold rtnl_lock; we need to release it here before we
4783 * set up our sysfs entries.
4785 int bond_create(struct net *net, const char *name)
4787 struct net_device *bond_dev;
4788 struct bonding *bond;
4789 struct alb_bond_info *bond_info;
4794 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4795 name ? name : "bond%d", NET_NAME_UNKNOWN,
4796 bond_setup, tx_queues);
4798 pr_err("%s: eek! can't alloc netdev!\n", name);
4804 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4805 * It is set to 0 by default which is wrong.
4807 bond = netdev_priv(bond_dev);
4808 bond_info = &(BOND_ALB_INFO(bond));
4809 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4811 dev_net_set(bond_dev, net);
4812 bond_dev->rtnl_link_ops = &bond_link_ops;
4814 res = register_netdevice(bond_dev);
4816 netif_carrier_off(bond_dev);
4818 bond_work_init_all(bond);
4822 free_netdev(bond_dev);
4826 static int __net_init bond_net_init(struct net *net)
4828 struct bond_net *bn = net_generic(net, bond_net_id);
4831 INIT_LIST_HEAD(&bn->dev_list);
4833 bond_create_proc_dir(bn);
4834 bond_create_sysfs(bn);
4839 static void __net_exit bond_net_exit(struct net *net)
4841 struct bond_net *bn = net_generic(net, bond_net_id);
4842 struct bonding *bond, *tmp_bond;
4845 bond_destroy_sysfs(bn);
4847 /* Kill off any bonds created after unregistering bond rtnl ops */
4849 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4850 unregister_netdevice_queue(bond->dev, &list);
4851 unregister_netdevice_many(&list);
4854 bond_destroy_proc_dir(bn);
4857 static struct pernet_operations bond_net_ops = {
4858 .init = bond_net_init,
4859 .exit = bond_net_exit,
4861 .size = sizeof(struct bond_net),
4864 static int __init bonding_init(void)
4869 pr_info("%s", bond_version);
4871 res = bond_check_params(&bonding_defaults);
4875 res = register_pernet_subsys(&bond_net_ops);
4879 res = bond_netlink_init();
4883 bond_create_debugfs();
4885 for (i = 0; i < max_bonds; i++) {
4886 res = bond_create(&init_net, NULL);
4891 register_netdevice_notifier(&bond_netdev_notifier);
4895 bond_destroy_debugfs();
4896 bond_netlink_fini();
4898 unregister_pernet_subsys(&bond_net_ops);
4903 static void __exit bonding_exit(void)
4905 unregister_netdevice_notifier(&bond_netdev_notifier);
4907 bond_destroy_debugfs();
4909 bond_netlink_fini();
4910 unregister_pernet_subsys(&bond_net_ops);
4912 #ifdef CONFIG_NET_POLL_CONTROLLER
4913 /* Make sure we don't have an imbalance on our netpoll blocking */
4914 WARN_ON(atomic_read(&netpoll_block_tx));
4918 module_init(bonding_init);
4919 module_exit(bonding_exit);
4920 MODULE_LICENSE("GPL");
4921 MODULE_VERSION(DRV_VERSION);
4922 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4923 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");