2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 #include "bonding_priv.h"
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 static int packets_per_slave = 1;
114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
116 module_param(max_bonds, int, 0);
117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118 module_param(tx_queues, int, 0);
119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126 module_param(miimon, int, 0);
127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128 module_param(updelay, int, 0);
129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130 module_param(downdelay, int, 0);
131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
133 module_param(use_carrier, int, 0);
134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136 module_param(mode, charp, 0);
137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141 module_param(primary, charp, 0);
142 MODULE_PARM_DESC(primary, "Primary network device to use");
143 module_param(primary_reselect, charp, 0);
144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
149 "2 for only on active slave "
151 module_param(lacp_rate, charp, 0);
152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154 module_param(ad_select, charp, 0);
155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
158 module_param(min_links, int, 0);
159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166 module_param(arp_interval, int, 0);
167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168 module_param_array(arp_ip_target, charp, NULL, 0);
169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170 module_param(arp_validate, charp, 0);
171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174 module_param(arp_all_targets, charp, 0);
175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac, charp, 0);
177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180 module_param(all_slaves_active, int, 0);
181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184 module_param(resend_igmp, int, 0);
185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
187 module_param(packets_per_slave, int, 0);
188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191 module_param(lp_interval, uint, 0);
192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
196 /*----------------------------- Global variables ----------------------------*/
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
202 unsigned int bond_net_id __read_mostly;
204 /*-------------------------- Forward declarations ---------------------------*/
206 static int bond_init(struct net_device *bond_dev);
207 static void bond_uninit(struct net_device *bond_dev);
208 static void bond_get_stats(struct net_device *bond_dev,
209 struct rtnl_link_stats64 *stats);
210 static void bond_slave_arr_handler(struct work_struct *work);
211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
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->queue_mapping = 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 void bond_set_dev_addr(struct net_device *bond_dev,
613 struct net_device *slave_dev)
615 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
616 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
617 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
618 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
619 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
622 static struct slave *bond_get_old_active(struct bonding *bond,
623 struct slave *new_active)
626 struct list_head *iter;
628 bond_for_each_slave(bond, slave, iter) {
629 if (slave == new_active)
632 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
639 /* bond_do_fail_over_mac
641 * Perform special MAC address swapping for fail_over_mac settings
645 static void bond_do_fail_over_mac(struct bonding *bond,
646 struct slave *new_active,
647 struct slave *old_active)
649 u8 tmp_mac[MAX_ADDR_LEN];
650 struct sockaddr_storage ss;
653 switch (bond->params.fail_over_mac) {
654 case BOND_FOM_ACTIVE:
656 bond_set_dev_addr(bond->dev, new_active->dev);
658 case BOND_FOM_FOLLOW:
659 /* if new_active && old_active, swap them
660 * if just old_active, do nothing (going to no active slave)
661 * if just new_active, set new_active to bond's MAC
667 old_active = bond_get_old_active(bond, new_active);
670 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
671 new_active->dev->addr_len);
672 bond_hw_addr_copy(ss.__data,
673 old_active->dev->dev_addr,
674 old_active->dev->addr_len);
675 ss.ss_family = new_active->dev->type;
677 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
678 bond->dev->addr_len);
679 ss.ss_family = bond->dev->type;
682 rv = dev_set_mac_address(new_active->dev,
683 (struct sockaddr *)&ss);
685 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
686 -rv, new_active->dev->name);
693 bond_hw_addr_copy(ss.__data, tmp_mac,
694 new_active->dev->addr_len);
695 ss.ss_family = old_active->dev->type;
697 rv = dev_set_mac_address(old_active->dev,
698 (struct sockaddr *)&ss);
700 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
701 -rv, new_active->dev->name);
705 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
706 bond->params.fail_over_mac);
712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
714 struct slave *prim = rtnl_dereference(bond->primary_slave);
715 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
717 if (!prim || prim->link != BOND_LINK_UP) {
718 if (!curr || curr->link != BOND_LINK_UP)
723 if (bond->force_primary) {
724 bond->force_primary = false;
728 if (!curr || curr->link != BOND_LINK_UP)
731 /* At this point, prim and curr are both up */
732 switch (bond->params.primary_reselect) {
733 case BOND_PRI_RESELECT_ALWAYS:
735 case BOND_PRI_RESELECT_BETTER:
736 if (prim->speed < curr->speed)
738 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
741 case BOND_PRI_RESELECT_FAILURE:
744 netdev_err(bond->dev, "impossible primary_reselect %d\n",
745 bond->params.primary_reselect);
751 * bond_find_best_slave - select the best available slave to be the active one
752 * @bond: our bonding struct
754 static struct slave *bond_find_best_slave(struct bonding *bond)
756 struct slave *slave, *bestslave = NULL;
757 struct list_head *iter;
758 int mintime = bond->params.updelay;
760 slave = bond_choose_primary_or_current(bond);
764 bond_for_each_slave(bond, slave, iter) {
765 if (slave->link == BOND_LINK_UP)
767 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
768 slave->delay < mintime) {
769 mintime = slave->delay;
777 static bool bond_should_notify_peers(struct bonding *bond)
782 slave = rcu_dereference(bond->curr_active_slave);
785 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
786 slave ? slave->dev->name : "NULL");
788 if (!slave || !bond->send_peer_notif ||
789 !netif_carrier_ok(bond->dev) ||
790 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
797 * change_active_interface - change the active slave into the specified one
798 * @bond: our bonding struct
799 * @new: the new slave to make the active one
801 * Set the new slave to the bond's settings and unset them on the old
803 * Setting include flags, mc-list, promiscuity, allmulti, etc.
805 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
806 * because it is apparently the best available slave we have, even though its
807 * updelay hasn't timed out yet.
809 * Caller must hold RTNL.
811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
813 struct slave *old_active;
817 old_active = rtnl_dereference(bond->curr_active_slave);
819 if (old_active == new_active)
823 new_active->last_link_up = jiffies;
825 if (new_active->link == BOND_LINK_BACK) {
826 if (bond_uses_primary(bond)) {
827 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
828 new_active->dev->name,
829 (bond->params.updelay - new_active->delay) * bond->params.miimon);
832 new_active->delay = 0;
833 bond_set_slave_link_state(new_active, BOND_LINK_UP,
834 BOND_SLAVE_NOTIFY_NOW);
836 if (BOND_MODE(bond) == BOND_MODE_8023AD)
837 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
839 if (bond_is_lb(bond))
840 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
842 if (bond_uses_primary(bond)) {
843 netdev_info(bond->dev, "making interface %s the new active one\n",
844 new_active->dev->name);
849 if (bond_uses_primary(bond))
850 bond_hw_addr_swap(bond, new_active, old_active);
852 if (bond_is_lb(bond)) {
853 bond_alb_handle_active_change(bond, new_active);
855 bond_set_slave_inactive_flags(old_active,
856 BOND_SLAVE_NOTIFY_NOW);
858 bond_set_slave_active_flags(new_active,
859 BOND_SLAVE_NOTIFY_NOW);
861 rcu_assign_pointer(bond->curr_active_slave, new_active);
864 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
866 bond_set_slave_inactive_flags(old_active,
867 BOND_SLAVE_NOTIFY_NOW);
870 bool should_notify_peers = false;
872 bond_set_slave_active_flags(new_active,
873 BOND_SLAVE_NOTIFY_NOW);
875 if (bond->params.fail_over_mac)
876 bond_do_fail_over_mac(bond, new_active,
879 if (netif_running(bond->dev)) {
880 bond->send_peer_notif =
881 bond->params.num_peer_notif;
882 should_notify_peers =
883 bond_should_notify_peers(bond);
886 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
887 if (should_notify_peers)
888 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
893 /* resend IGMP joins since active slave has changed or
894 * all were sent on curr_active_slave.
895 * resend only if bond is brought up with the affected
896 * bonding modes and the retransmission is enabled
898 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
899 ((bond_uses_primary(bond) && new_active) ||
900 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
901 bond->igmp_retrans = bond->params.resend_igmp;
902 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
907 * bond_select_active_slave - select a new active slave, if needed
908 * @bond: our bonding struct
910 * This functions should be called when one of the following occurs:
911 * - The old curr_active_slave has been released or lost its link.
912 * - The primary_slave has got its link back.
913 * - A slave has got its link back and there's no old curr_active_slave.
915 * Caller must hold RTNL.
917 void bond_select_active_slave(struct bonding *bond)
919 struct slave *best_slave;
924 best_slave = bond_find_best_slave(bond);
925 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
926 bond_change_active_slave(bond, best_slave);
927 rv = bond_set_carrier(bond);
931 if (netif_carrier_ok(bond->dev))
932 netdev_info(bond->dev, "first active interface up!\n");
934 netdev_info(bond->dev, "now running without any active interface!\n");
938 #ifdef CONFIG_NET_POLL_CONTROLLER
939 static inline int slave_enable_netpoll(struct slave *slave)
944 np = kzalloc(sizeof(*np), GFP_KERNEL);
949 err = __netpoll_setup(np, slave->dev);
958 static inline void slave_disable_netpoll(struct slave *slave)
960 struct netpoll *np = slave->np;
966 __netpoll_free_async(np);
969 static void bond_poll_controller(struct net_device *bond_dev)
971 struct bonding *bond = netdev_priv(bond_dev);
972 struct slave *slave = NULL;
973 struct list_head *iter;
974 struct ad_info ad_info;
975 struct netpoll_info *ni;
976 const struct net_device_ops *ops;
978 if (BOND_MODE(bond) == BOND_MODE_8023AD)
979 if (bond_3ad_get_active_agg_info(bond, &ad_info))
982 bond_for_each_slave_rcu(bond, slave, iter) {
983 ops = slave->dev->netdev_ops;
984 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
987 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
988 struct aggregator *agg =
989 SLAVE_AD_INFO(slave)->port.aggregator;
992 agg->aggregator_identifier != ad_info.aggregator_id)
996 ni = rcu_dereference_bh(slave->dev->npinfo);
997 if (down_trylock(&ni->dev_lock))
999 ops->ndo_poll_controller(slave->dev);
1004 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1006 struct bonding *bond = netdev_priv(bond_dev);
1007 struct list_head *iter;
1008 struct slave *slave;
1010 bond_for_each_slave(bond, slave, iter)
1011 if (bond_slave_is_up(slave))
1012 slave_disable_netpoll(slave);
1015 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1017 struct bonding *bond = netdev_priv(dev);
1018 struct list_head *iter;
1019 struct slave *slave;
1022 bond_for_each_slave(bond, slave, iter) {
1023 err = slave_enable_netpoll(slave);
1025 bond_netpoll_cleanup(dev);
1032 static inline int slave_enable_netpoll(struct slave *slave)
1036 static inline void slave_disable_netpoll(struct slave *slave)
1039 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1044 /*---------------------------------- IOCTL ----------------------------------*/
1046 static netdev_features_t bond_fix_features(struct net_device *dev,
1047 netdev_features_t features)
1049 struct bonding *bond = netdev_priv(dev);
1050 struct list_head *iter;
1051 netdev_features_t mask;
1052 struct slave *slave;
1056 features &= ~NETIF_F_ONE_FOR_ALL;
1057 features |= NETIF_F_ALL_FOR_ALL;
1059 bond_for_each_slave(bond, slave, iter) {
1060 features = netdev_increment_features(features,
1061 slave->dev->features,
1064 features = netdev_add_tso_features(features, mask);
1069 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1070 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1071 NETIF_F_HIGHDMA | NETIF_F_LRO)
1073 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1074 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1076 static void bond_compute_features(struct bonding *bond)
1078 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1079 IFF_XMIT_DST_RELEASE_PERM;
1080 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1081 netdev_features_t enc_features = BOND_ENC_FEATURES;
1082 struct net_device *bond_dev = bond->dev;
1083 struct list_head *iter;
1084 struct slave *slave;
1085 unsigned short max_hard_header_len = ETH_HLEN;
1086 unsigned int gso_max_size = GSO_MAX_SIZE;
1087 u16 gso_max_segs = GSO_MAX_SEGS;
1089 if (!bond_has_slaves(bond))
1091 vlan_features &= NETIF_F_ALL_FOR_ALL;
1093 bond_for_each_slave(bond, slave, iter) {
1094 vlan_features = netdev_increment_features(vlan_features,
1095 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1097 enc_features = netdev_increment_features(enc_features,
1098 slave->dev->hw_enc_features,
1100 dst_release_flag &= slave->dev->priv_flags;
1101 if (slave->dev->hard_header_len > max_hard_header_len)
1102 max_hard_header_len = slave->dev->hard_header_len;
1104 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1105 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1107 bond_dev->hard_header_len = max_hard_header_len;
1110 bond_dev->vlan_features = vlan_features;
1111 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1112 NETIF_F_HW_VLAN_CTAG_TX |
1113 NETIF_F_HW_VLAN_STAG_TX;
1114 bond_dev->gso_max_segs = gso_max_segs;
1115 netif_set_gso_max_size(bond_dev, gso_max_size);
1117 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1118 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1119 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1120 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1122 netdev_change_features(bond_dev);
1125 static void bond_setup_by_slave(struct net_device *bond_dev,
1126 struct net_device *slave_dev)
1128 bond_dev->header_ops = slave_dev->header_ops;
1130 bond_dev->type = slave_dev->type;
1131 bond_dev->hard_header_len = slave_dev->hard_header_len;
1132 bond_dev->needed_headroom = slave_dev->needed_headroom;
1133 bond_dev->addr_len = slave_dev->addr_len;
1135 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1136 slave_dev->addr_len);
1139 /* On bonding slaves other than the currently active slave, suppress
1140 * duplicates except for alb non-mcast/bcast.
1142 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1143 struct slave *slave,
1144 struct bonding *bond)
1146 if (bond_is_slave_inactive(slave)) {
1147 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1148 skb->pkt_type != PACKET_BROADCAST &&
1149 skb->pkt_type != PACKET_MULTICAST)
1156 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1158 struct sk_buff *skb = *pskb;
1159 struct slave *slave;
1160 struct bonding *bond;
1161 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1163 int ret = RX_HANDLER_ANOTHER;
1165 skb = skb_share_check(skb, GFP_ATOMIC);
1167 return RX_HANDLER_CONSUMED;
1171 slave = bond_slave_get_rcu(skb->dev);
1174 recv_probe = ACCESS_ONCE(bond->recv_probe);
1176 ret = recv_probe(skb, bond, slave);
1177 if (ret == RX_HANDLER_CONSUMED) {
1184 * For packets determined by bond_should_deliver_exact_match() call to
1185 * be suppressed we want to make an exception for link-local packets.
1186 * This is necessary for e.g. LLDP daemons to be able to monitor
1187 * inactive slave links without being forced to bind to them
1190 * At the same time, packets that are passed to the bonding master
1191 * (including link-local ones) can have their originating interface
1192 * determined via PACKET_ORIGDEV socket option.
1194 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1195 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1196 return RX_HANDLER_PASS;
1197 return RX_HANDLER_EXACT;
1200 skb->dev = bond->dev;
1202 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1203 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1204 skb->pkt_type == PACKET_HOST) {
1206 if (unlikely(skb_cow_head(skb,
1207 skb->data - skb_mac_header(skb)))) {
1209 return RX_HANDLER_CONSUMED;
1211 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1212 bond->dev->addr_len);
1218 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1220 switch (BOND_MODE(bond)) {
1221 case BOND_MODE_ROUNDROBIN:
1222 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1223 case BOND_MODE_ACTIVEBACKUP:
1224 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1225 case BOND_MODE_BROADCAST:
1226 return NETDEV_LAG_TX_TYPE_BROADCAST;
1228 case BOND_MODE_8023AD:
1229 return NETDEV_LAG_TX_TYPE_HASH;
1231 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1235 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1237 struct netdev_lag_upper_info lag_upper_info;
1240 lag_upper_info.tx_type = bond_lag_tx_type(bond);
1241 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1245 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1249 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1251 netdev_upper_dev_unlink(slave->dev, bond->dev);
1252 slave->dev->flags &= ~IFF_SLAVE;
1253 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1256 static void slave_kobj_release(struct kobject *kobj)
1258 struct slave *slave = to_slave(kobj);
1259 struct bonding *bond = bond_get_bond_by_slave(slave);
1261 cancel_delayed_work_sync(&slave->notify_work);
1262 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1263 kfree(SLAVE_AD_INFO(slave));
1268 static struct kobj_type slave_ktype = {
1269 .release = slave_kobj_release,
1271 .sysfs_ops = &slave_sysfs_ops,
1275 static int bond_kobj_init(struct slave *slave)
1279 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1280 &(slave->dev->dev.kobj), "bonding_slave");
1282 kobject_put(&slave->kobj);
1287 static struct slave *bond_alloc_slave(struct bonding *bond,
1288 struct net_device *slave_dev)
1290 struct slave *slave = NULL;
1292 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1297 slave->dev = slave_dev;
1298 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1300 if (bond_kobj_init(slave))
1303 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1304 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1306 if (!SLAVE_AD_INFO(slave)) {
1307 kobject_put(&slave->kobj);
1315 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1317 info->bond_mode = BOND_MODE(bond);
1318 info->miimon = bond->params.miimon;
1319 info->num_slaves = bond->slave_cnt;
1322 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1324 strcpy(info->slave_name, slave->dev->name);
1325 info->link = slave->link;
1326 info->state = bond_slave_state(slave);
1327 info->link_failure_count = slave->link_failure_count;
1330 static void bond_netdev_notify_work(struct work_struct *_work)
1332 struct slave *slave = container_of(_work, struct slave,
1335 if (rtnl_trylock()) {
1336 struct netdev_bonding_info binfo;
1338 bond_fill_ifslave(slave, &binfo.slave);
1339 bond_fill_ifbond(slave->bond, &binfo.master);
1340 netdev_bonding_info_change(slave->dev, &binfo);
1343 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1347 void bond_queue_slave_event(struct slave *slave)
1349 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1352 void bond_lower_state_changed(struct slave *slave)
1354 struct netdev_lag_lower_state_info info;
1356 info.link_up = slave->link == BOND_LINK_UP ||
1357 slave->link == BOND_LINK_FAIL;
1358 info.tx_enabled = bond_is_active_slave(slave);
1359 netdev_lower_state_changed(slave->dev, &info);
1362 /* enslave device <slave> to bond device <master> */
1363 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1365 struct bonding *bond = netdev_priv(bond_dev);
1366 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1367 struct slave *new_slave = NULL, *prev_slave;
1368 struct sockaddr_storage ss;
1372 if (!bond->params.use_carrier &&
1373 slave_dev->ethtool_ops->get_link == NULL &&
1374 slave_ops->ndo_do_ioctl == NULL) {
1375 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1379 /* already in-use? */
1380 if (netdev_is_rx_handler_busy(slave_dev)) {
1381 netdev_err(bond_dev,
1382 "Error: Device is in use and cannot be enslaved\n");
1386 if (bond_dev == slave_dev) {
1387 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1391 /* vlan challenged mutual exclusion */
1392 /* no need to lock since we're protected by rtnl_lock */
1393 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1394 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1396 if (vlan_uses_dev(bond_dev)) {
1397 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1398 slave_dev->name, bond_dev->name);
1401 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1402 slave_dev->name, slave_dev->name,
1406 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1410 /* Old ifenslave binaries are no longer supported. These can
1411 * be identified with moderate accuracy by the state of the slave:
1412 * the current ifenslave will set the interface down prior to
1413 * enslaving it; the old ifenslave will not.
1415 if (slave_dev->flags & IFF_UP) {
1416 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1421 /* set bonding device ether type by slave - bonding netdevices are
1422 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1423 * there is a need to override some of the type dependent attribs/funcs.
1425 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1426 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1428 if (!bond_has_slaves(bond)) {
1429 if (bond_dev->type != slave_dev->type) {
1430 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1431 bond_dev->type, slave_dev->type);
1433 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1435 res = notifier_to_errno(res);
1437 netdev_err(bond_dev, "refused to change device type\n");
1441 /* Flush unicast and multicast addresses */
1442 dev_uc_flush(bond_dev);
1443 dev_mc_flush(bond_dev);
1445 if (slave_dev->type != ARPHRD_ETHER)
1446 bond_setup_by_slave(bond_dev, slave_dev);
1448 ether_setup(bond_dev);
1449 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1452 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1455 } else if (bond_dev->type != slave_dev->type) {
1456 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1457 slave_dev->name, slave_dev->type, bond_dev->type);
1461 if (slave_dev->type == ARPHRD_INFINIBAND &&
1462 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1463 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1466 goto err_undo_flags;
1469 if (!slave_ops->ndo_set_mac_address ||
1470 slave_dev->type == ARPHRD_INFINIBAND) {
1471 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1472 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1473 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1474 if (!bond_has_slaves(bond)) {
1475 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1476 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1478 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");
1480 goto err_undo_flags;
1485 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1487 /* If this is the first slave, then we need to set the master's hardware
1488 * address to be the same as the slave's.
1490 if (!bond_has_slaves(bond) &&
1491 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1492 bond_set_dev_addr(bond->dev, slave_dev);
1494 new_slave = bond_alloc_slave(bond, slave_dev);
1497 goto err_undo_flags;
1500 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1501 * is set via sysfs or module option if desired.
1503 new_slave->queue_id = 0;
1505 /* Save slave's original mtu and then set it to match the bond */
1506 new_slave->original_mtu = slave_dev->mtu;
1507 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1509 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1513 /* Save slave's original ("permanent") mac address for modes
1514 * that need it, and for restoring it upon release, and then
1515 * set it to the master's address
1517 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1518 slave_dev->addr_len);
1520 if (!bond->params.fail_over_mac ||
1521 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1522 /* Set slave to master's mac address. The application already
1523 * set the master's mac address to that of the first slave
1525 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1526 ss.ss_family = slave_dev->type;
1527 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1529 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1530 goto err_restore_mtu;
1534 /* set slave flag before open to prevent IPv6 addrconf */
1535 slave_dev->flags |= IFF_SLAVE;
1537 /* open the slave since the application closed it */
1538 res = dev_open(slave_dev);
1540 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1541 goto err_restore_mac;
1544 slave_dev->priv_flags |= IFF_BONDING;
1545 /* initialize slave stats */
1546 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1548 if (bond_is_lb(bond)) {
1549 /* bond_alb_init_slave() must be called before all other stages since
1550 * it might fail and we do not want to have to undo everything
1552 res = bond_alb_init_slave(bond, new_slave);
1557 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1559 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1561 goto err_hwaddr_unsync;
1564 prev_slave = bond_last_slave(bond);
1566 new_slave->delay = 0;
1567 new_slave->link_failure_count = 0;
1569 if (bond_update_speed_duplex(new_slave) &&
1570 bond_needs_speed_duplex(bond))
1571 new_slave->link = BOND_LINK_DOWN;
1573 new_slave->last_rx = jiffies -
1574 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1575 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1576 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1578 if (bond->params.miimon && !bond->params.use_carrier) {
1579 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1581 if ((link_reporting == -1) && !bond->params.arp_interval) {
1582 /* miimon is set but a bonded network driver
1583 * does not support ETHTOOL/MII and
1584 * arp_interval is not set. Note: if
1585 * use_carrier is enabled, we will never go
1586 * here (because netif_carrier is always
1587 * supported); thus, we don't need to change
1588 * the messages for netif_carrier.
1590 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1592 } else if (link_reporting == -1) {
1593 /* unable get link status using mii/ethtool */
1594 netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1599 /* check for initial state */
1600 new_slave->link = BOND_LINK_NOCHANGE;
1601 if (bond->params.miimon) {
1602 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1603 if (bond->params.updelay) {
1604 bond_set_slave_link_state(new_slave,
1606 BOND_SLAVE_NOTIFY_NOW);
1607 new_slave->delay = bond->params.updelay;
1609 bond_set_slave_link_state(new_slave,
1611 BOND_SLAVE_NOTIFY_NOW);
1614 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1615 BOND_SLAVE_NOTIFY_NOW);
1617 } else if (bond->params.arp_interval) {
1618 bond_set_slave_link_state(new_slave,
1619 (netif_carrier_ok(slave_dev) ?
1620 BOND_LINK_UP : BOND_LINK_DOWN),
1621 BOND_SLAVE_NOTIFY_NOW);
1623 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1624 BOND_SLAVE_NOTIFY_NOW);
1627 if (new_slave->link != BOND_LINK_DOWN)
1628 new_slave->last_link_up = jiffies;
1629 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1630 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1631 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1633 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1634 /* if there is a primary slave, remember it */
1635 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1636 rcu_assign_pointer(bond->primary_slave, new_slave);
1637 bond->force_primary = true;
1641 switch (BOND_MODE(bond)) {
1642 case BOND_MODE_ACTIVEBACKUP:
1643 bond_set_slave_inactive_flags(new_slave,
1644 BOND_SLAVE_NOTIFY_NOW);
1646 case BOND_MODE_8023AD:
1647 /* in 802.3ad mode, the internal mechanism
1648 * will activate the slaves in the selected
1651 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1652 /* if this is the first slave */
1654 SLAVE_AD_INFO(new_slave)->id = 1;
1655 /* Initialize AD with the number of times that the AD timer is called in 1 second
1656 * can be called only after the mac address of the bond is set
1658 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1660 SLAVE_AD_INFO(new_slave)->id =
1661 SLAVE_AD_INFO(prev_slave)->id + 1;
1664 bond_3ad_bind_slave(new_slave);
1668 bond_set_active_slave(new_slave);
1669 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1672 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1674 /* always active in trunk mode */
1675 bond_set_active_slave(new_slave);
1677 /* In trunking mode there is little meaning to curr_active_slave
1678 * anyway (it holds no special properties of the bond device),
1679 * so we can change it without calling change_active_interface()
1681 if (!rcu_access_pointer(bond->curr_active_slave) &&
1682 new_slave->link == BOND_LINK_UP)
1683 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1686 } /* switch(bond_mode) */
1688 #ifdef CONFIG_NET_POLL_CONTROLLER
1689 if (bond->dev->npinfo) {
1690 if (slave_enable_netpoll(new_slave)) {
1691 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1698 if (!(bond_dev->features & NETIF_F_LRO))
1699 dev_disable_lro(slave_dev);
1701 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1704 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1708 res = bond_master_upper_dev_link(bond, new_slave);
1710 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1711 goto err_unregister;
1714 res = bond_sysfs_slave_add(new_slave);
1716 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1717 goto err_upper_unlink;
1720 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1722 /* If the mode uses primary, then the following is handled by
1723 * bond_change_active_slave().
1725 if (!bond_uses_primary(bond)) {
1726 /* set promiscuity level to new slave */
1727 if (bond_dev->flags & IFF_PROMISC) {
1728 res = dev_set_promiscuity(slave_dev, 1);
1733 /* set allmulti level to new slave */
1734 if (bond_dev->flags & IFF_ALLMULTI) {
1735 res = dev_set_allmulti(slave_dev, 1);
1737 if (bond_dev->flags & IFF_PROMISC)
1738 dev_set_promiscuity(slave_dev, -1);
1743 netif_addr_lock_bh(bond_dev);
1744 dev_mc_sync_multiple(slave_dev, bond_dev);
1745 dev_uc_sync_multiple(slave_dev, bond_dev);
1746 netif_addr_unlock_bh(bond_dev);
1748 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1749 /* add lacpdu mc addr to mc list */
1750 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1752 dev_mc_add(slave_dev, lacpdu_multicast);
1757 bond_compute_features(bond);
1758 bond_set_carrier(bond);
1760 if (bond_uses_primary(bond)) {
1762 bond_select_active_slave(bond);
1763 unblock_netpoll_tx();
1766 if (bond_mode_uses_xmit_hash(bond))
1767 bond_update_slave_arr(bond, NULL);
1770 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1772 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1773 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1775 /* enslave is successful */
1776 bond_queue_slave_event(new_slave);
1779 /* Undo stages on error */
1781 bond_sysfs_slave_del(new_slave);
1784 bond_upper_dev_unlink(bond, new_slave);
1787 netdev_rx_handler_unregister(slave_dev);
1790 vlan_vids_del_by_dev(slave_dev, bond_dev);
1791 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1792 RCU_INIT_POINTER(bond->primary_slave, NULL);
1793 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1795 bond_change_active_slave(bond, NULL);
1796 bond_select_active_slave(bond);
1797 unblock_netpoll_tx();
1799 /* either primary_slave or curr_active_slave might've changed */
1801 slave_disable_netpoll(new_slave);
1804 if (!bond_uses_primary(bond))
1805 bond_hw_addr_flush(bond_dev, slave_dev);
1808 if (!netif_is_bond_master(slave_dev))
1809 slave_dev->priv_flags &= ~IFF_BONDING;
1810 dev_close(slave_dev);
1813 slave_dev->flags &= ~IFF_SLAVE;
1814 if (!bond->params.fail_over_mac ||
1815 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1816 /* XXX TODO - fom follow mode needs to change master's
1817 * MAC if this slave's MAC is in use by the bond, or at
1818 * least print a warning.
1820 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1821 new_slave->dev->addr_len);
1822 ss.ss_family = slave_dev->type;
1823 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1827 dev_set_mtu(slave_dev, new_slave->original_mtu);
1830 kobject_put(&new_slave->kobj);
1833 /* Enslave of first slave has failed and we need to fix master's mac */
1834 if (!bond_has_slaves(bond)) {
1835 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1836 slave_dev->dev_addr))
1837 eth_hw_addr_random(bond_dev);
1838 if (bond_dev->type != ARPHRD_ETHER) {
1839 dev_close(bond_dev);
1840 ether_setup(bond_dev);
1841 bond_dev->flags |= IFF_MASTER;
1842 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1849 /* Try to release the slave device <slave> from the bond device <master>
1850 * It is legal to access curr_active_slave without a lock because all the function
1851 * is RTNL-locked. If "all" is true it means that the function is being called
1852 * while destroying a bond interface and all slaves are being released.
1854 * The rules for slave state should be:
1855 * for Active/Backup:
1856 * Active stays on all backups go down
1857 * for Bonded connections:
1858 * The first up interface should be left on and all others downed.
1860 static int __bond_release_one(struct net_device *bond_dev,
1861 struct net_device *slave_dev,
1862 bool all, bool unregister)
1864 struct bonding *bond = netdev_priv(bond_dev);
1865 struct slave *slave, *oldcurrent;
1866 struct sockaddr_storage ss;
1867 int old_flags = bond_dev->flags;
1868 netdev_features_t old_features = bond_dev->features;
1870 /* slave is not a slave or master is not master of this slave */
1871 if (!(slave_dev->flags & IFF_SLAVE) ||
1872 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1873 netdev_dbg(bond_dev, "cannot release %s\n",
1880 slave = bond_get_slave_by_dev(bond, slave_dev);
1882 /* not a slave of this bond */
1883 netdev_info(bond_dev, "%s not enslaved\n",
1885 unblock_netpoll_tx();
1889 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1891 bond_sysfs_slave_del(slave);
1893 /* recompute stats just before removing the slave */
1894 bond_get_stats(bond->dev, &bond->bond_stats);
1896 bond_upper_dev_unlink(bond, slave);
1897 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1898 * for this slave anymore.
1900 netdev_rx_handler_unregister(slave_dev);
1902 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1903 bond_3ad_unbind_slave(slave);
1905 if (bond_mode_uses_xmit_hash(bond))
1906 bond_update_slave_arr(bond, slave);
1908 netdev_info(bond_dev, "Releasing %s interface %s\n",
1909 bond_is_active_slave(slave) ? "active" : "backup",
1912 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1914 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1916 if (!all && (!bond->params.fail_over_mac ||
1917 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1918 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1919 bond_has_slaves(bond))
1920 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",
1921 slave_dev->name, slave->perm_hwaddr,
1922 bond_dev->name, slave_dev->name);
1925 if (rtnl_dereference(bond->primary_slave) == slave)
1926 RCU_INIT_POINTER(bond->primary_slave, NULL);
1928 if (oldcurrent == slave)
1929 bond_change_active_slave(bond, NULL);
1931 if (bond_is_lb(bond)) {
1932 /* Must be called only after the slave has been
1933 * detached from the list and the curr_active_slave
1934 * has been cleared (if our_slave == old_current),
1935 * but before a new active slave is selected.
1937 bond_alb_deinit_slave(bond, slave);
1941 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1942 } else if (oldcurrent == slave) {
1943 /* Note that we hold RTNL over this sequence, so there
1944 * is no concern that another slave add/remove event
1947 bond_select_active_slave(bond);
1950 if (!bond_has_slaves(bond)) {
1951 bond_set_carrier(bond);
1952 eth_hw_addr_random(bond_dev);
1953 bond->nest_level = SINGLE_DEPTH_NESTING;
1955 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1958 unblock_netpoll_tx();
1962 if (!bond_has_slaves(bond)) {
1963 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1964 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1967 bond_compute_features(bond);
1968 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1969 (old_features & NETIF_F_VLAN_CHALLENGED))
1970 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1971 slave_dev->name, bond_dev->name);
1973 vlan_vids_del_by_dev(slave_dev, bond_dev);
1975 /* If the mode uses primary, then this case was handled above by
1976 * bond_change_active_slave(..., NULL)
1978 if (!bond_uses_primary(bond)) {
1979 /* unset promiscuity level from slave
1980 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1981 * of the IFF_PROMISC flag in the bond_dev, but we need the
1982 * value of that flag before that change, as that was the value
1983 * when this slave was attached, so we cache at the start of the
1984 * function and use it here. Same goes for ALLMULTI below
1986 if (old_flags & IFF_PROMISC)
1987 dev_set_promiscuity(slave_dev, -1);
1989 /* unset allmulti level from slave */
1990 if (old_flags & IFF_ALLMULTI)
1991 dev_set_allmulti(slave_dev, -1);
1993 bond_hw_addr_flush(bond_dev, slave_dev);
1996 slave_disable_netpoll(slave);
1998 /* close slave before restoring its mac address */
1999 dev_close(slave_dev);
2001 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2002 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2003 /* restore original ("permanent") mac address */
2004 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2005 slave->dev->addr_len);
2006 ss.ss_family = slave_dev->type;
2007 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
2011 __dev_set_mtu(slave_dev, slave->original_mtu);
2013 dev_set_mtu(slave_dev, slave->original_mtu);
2015 if (!netif_is_bond_master(slave_dev))
2016 slave_dev->priv_flags &= ~IFF_BONDING;
2018 kobject_put(&slave->kobj);
2023 /* A wrapper used because of ndo_del_link */
2024 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2026 return __bond_release_one(bond_dev, slave_dev, false, false);
2029 /* First release a slave and then destroy the bond if no more slaves are left.
2030 * Must be under rtnl_lock when this function is called.
2032 static int bond_release_and_destroy(struct net_device *bond_dev,
2033 struct net_device *slave_dev)
2035 struct bonding *bond = netdev_priv(bond_dev);
2038 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2039 if (ret == 0 && !bond_has_slaves(bond) &&
2040 bond_dev->reg_state != NETREG_UNREGISTERING) {
2041 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2042 netdev_info(bond_dev, "Destroying bond %s\n",
2044 bond_remove_proc_entry(bond);
2045 unregister_netdevice(bond_dev);
2050 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2052 struct bonding *bond = netdev_priv(bond_dev);
2053 bond_fill_ifbond(bond, info);
2056 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2058 struct bonding *bond = netdev_priv(bond_dev);
2059 struct list_head *iter;
2060 int i = 0, res = -ENODEV;
2061 struct slave *slave;
2063 bond_for_each_slave(bond, slave, iter) {
2064 if (i++ == (int)info->slave_id) {
2066 bond_fill_ifslave(slave, info);
2074 /*-------------------------------- Monitoring -------------------------------*/
2076 /* called with rcu_read_lock() */
2077 static int bond_miimon_inspect(struct bonding *bond)
2079 int link_state, commit = 0;
2080 struct list_head *iter;
2081 struct slave *slave;
2082 bool ignore_updelay;
2084 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2086 bond_for_each_slave_rcu(bond, slave, iter) {
2087 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2089 link_state = bond_check_dev_link(bond, slave->dev, 0);
2091 switch (slave->link) {
2096 bond_propose_link_state(slave, BOND_LINK_FAIL);
2098 slave->delay = bond->params.downdelay;
2100 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2102 BOND_MODE_ACTIVEBACKUP) ?
2103 (bond_is_active_slave(slave) ?
2104 "active " : "backup ") : "",
2106 bond->params.downdelay * bond->params.miimon);
2109 case BOND_LINK_FAIL:
2111 /* recovered before downdelay expired */
2112 bond_propose_link_state(slave, BOND_LINK_UP);
2113 slave->last_link_up = jiffies;
2114 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2115 (bond->params.downdelay - slave->delay) *
2116 bond->params.miimon,
2122 if (slave->delay <= 0) {
2123 bond_propose_link_state(slave, BOND_LINK_DOWN);
2131 case BOND_LINK_DOWN:
2135 bond_propose_link_state(slave, BOND_LINK_BACK);
2137 slave->delay = bond->params.updelay;
2140 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2142 ignore_updelay ? 0 :
2143 bond->params.updelay *
2144 bond->params.miimon);
2147 case BOND_LINK_BACK:
2149 bond_propose_link_state(slave, BOND_LINK_DOWN);
2150 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2151 (bond->params.updelay - slave->delay) *
2152 bond->params.miimon,
2161 if (slave->delay <= 0) {
2162 bond_propose_link_state(slave, BOND_LINK_UP);
2164 ignore_updelay = false;
2176 static void bond_miimon_commit(struct bonding *bond)
2178 struct list_head *iter;
2179 struct slave *slave, *primary;
2181 bond_for_each_slave(bond, slave, iter) {
2182 switch (slave->link_new_state) {
2183 case BOND_LINK_NOCHANGE:
2184 /* For 802.3ad mode, check current slave speed and
2185 * duplex again in case its port was disabled after
2186 * invalid speed/duplex reporting but recovered before
2187 * link monitoring could make a decision on the actual
2190 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2191 slave->link == BOND_LINK_UP)
2192 bond_3ad_adapter_speed_duplex_changed(slave);
2196 if (bond_update_speed_duplex(slave) &&
2197 bond_needs_speed_duplex(bond)) {
2198 slave->link = BOND_LINK_DOWN;
2199 if (net_ratelimit())
2200 netdev_warn(bond->dev,
2201 "failed to get link speed/duplex for %s\n",
2205 bond_set_slave_link_state(slave, BOND_LINK_UP,
2206 BOND_SLAVE_NOTIFY_NOW);
2207 slave->last_link_up = jiffies;
2209 primary = rtnl_dereference(bond->primary_slave);
2210 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2211 /* prevent it from being the active one */
2212 bond_set_backup_slave(slave);
2213 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2214 /* make it immediately active */
2215 bond_set_active_slave(slave);
2218 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2220 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2221 slave->duplex ? "full" : "half");
2223 /* notify ad that the link status has changed */
2224 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2225 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2227 if (bond_is_lb(bond))
2228 bond_alb_handle_link_change(bond, slave,
2231 if (BOND_MODE(bond) == BOND_MODE_XOR)
2232 bond_update_slave_arr(bond, NULL);
2234 if (!bond->curr_active_slave || slave == primary)
2239 case BOND_LINK_DOWN:
2240 if (slave->link_failure_count < UINT_MAX)
2241 slave->link_failure_count++;
2243 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2244 BOND_SLAVE_NOTIFY_NOW);
2246 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2247 BOND_MODE(bond) == BOND_MODE_8023AD)
2248 bond_set_slave_inactive_flags(slave,
2249 BOND_SLAVE_NOTIFY_NOW);
2251 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2254 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2255 bond_3ad_handle_link_change(slave,
2258 if (bond_is_lb(bond))
2259 bond_alb_handle_link_change(bond, slave,
2262 if (BOND_MODE(bond) == BOND_MODE_XOR)
2263 bond_update_slave_arr(bond, NULL);
2265 if (slave == rcu_access_pointer(bond->curr_active_slave))
2271 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2272 slave->link_new_state, slave->dev->name);
2273 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2280 bond_select_active_slave(bond);
2281 unblock_netpoll_tx();
2284 bond_set_carrier(bond);
2289 * Really a wrapper that splits the mii monitor into two phases: an
2290 * inspection, then (if inspection indicates something needs to be done)
2291 * an acquisition of appropriate locks followed by a commit phase to
2292 * implement whatever link state changes are indicated.
2294 static void bond_mii_monitor(struct work_struct *work)
2296 struct bonding *bond = container_of(work, struct bonding,
2298 bool should_notify_peers = false;
2299 unsigned long delay;
2300 struct slave *slave;
2301 struct list_head *iter;
2303 delay = msecs_to_jiffies(bond->params.miimon);
2305 if (!bond_has_slaves(bond))
2310 should_notify_peers = bond_should_notify_peers(bond);
2312 if (bond_miimon_inspect(bond)) {
2315 /* Race avoidance with bond_close cancel of workqueue */
2316 if (!rtnl_trylock()) {
2318 should_notify_peers = false;
2322 bond_for_each_slave(bond, slave, iter) {
2323 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2325 bond_miimon_commit(bond);
2327 rtnl_unlock(); /* might sleep, hold no other locks */
2332 if (bond->params.miimon)
2333 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2335 if (should_notify_peers) {
2336 if (!rtnl_trylock())
2338 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2343 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2345 __be32 ip = *((__be32 *)data);
2347 return ip == bond_confirm_addr(upper, 0, ip);
2350 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2354 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2358 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2365 /* We go to the (large) trouble of VLAN tagging ARP frames because
2366 * switches in VLAN mode (especially if ports are configured as
2367 * "native" to a VLAN) might not pass non-tagged frames.
2369 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2370 __be32 dest_ip, __be32 src_ip,
2371 struct bond_vlan_tag *tags)
2373 struct sk_buff *skb;
2374 struct bond_vlan_tag *outer_tag = tags;
2376 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2377 arp_op, slave_dev->name, &dest_ip, &src_ip);
2379 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2380 NULL, slave_dev->dev_addr, NULL);
2383 net_err_ratelimited("ARP packet allocation failed\n");
2387 if (!tags || tags->vlan_proto == VLAN_N_VID)
2392 /* Go through all the tags backwards and add them to the packet */
2393 while (tags->vlan_proto != VLAN_N_VID) {
2394 if (!tags->vlan_id) {
2399 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2400 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2401 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2404 net_err_ratelimited("failed to insert inner VLAN tag\n");
2410 /* Set the outer tag */
2411 if (outer_tag->vlan_id) {
2412 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2413 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2414 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2415 outer_tag->vlan_id);
2422 /* Validate the device path between the @start_dev and the @end_dev.
2423 * The path is valid if the @end_dev is reachable through device
2425 * When the path is validated, collect any vlan information in the
2428 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2429 struct net_device *end_dev,
2432 struct bond_vlan_tag *tags;
2433 struct net_device *upper;
2434 struct list_head *iter;
2436 if (start_dev == end_dev) {
2437 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2439 return ERR_PTR(-ENOMEM);
2440 tags[level].vlan_proto = VLAN_N_VID;
2444 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2445 tags = bond_verify_device_path(upper, end_dev, level + 1);
2446 if (IS_ERR_OR_NULL(tags)) {
2451 if (is_vlan_dev(upper)) {
2452 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2453 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2462 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2465 struct bond_vlan_tag *tags;
2466 __be32 *targets = bond->params.arp_targets, addr;
2469 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2470 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2473 /* Find out through which dev should the packet go */
2474 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2477 /* there's no route to target - try to send arp
2478 * probe to generate any traffic (arp_validate=0)
2480 if (bond->params.arp_validate)
2481 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2484 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2489 /* bond device itself */
2490 if (rt->dst.dev == bond->dev)
2494 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2497 if (!IS_ERR_OR_NULL(tags))
2500 /* Not our device - skip */
2501 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2502 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2508 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2510 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2516 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2520 if (!sip || !bond_has_this_ip(bond, tip)) {
2521 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2526 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2528 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2532 slave->last_rx = jiffies;
2533 slave->target_last_arp_rx[i] = jiffies;
2536 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2537 struct slave *slave)
2539 struct arphdr *arp = (struct arphdr *)skb->data;
2540 struct slave *curr_active_slave, *curr_arp_slave;
2541 unsigned char *arp_ptr;
2543 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2545 if (!slave_do_arp_validate(bond, slave)) {
2546 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2547 !slave_do_arp_validate_only(bond))
2548 slave->last_rx = jiffies;
2549 return RX_HANDLER_ANOTHER;
2550 } else if (!is_arp) {
2551 return RX_HANDLER_ANOTHER;
2554 alen = arp_hdr_len(bond->dev);
2556 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2559 if (alen > skb_headlen(skb)) {
2560 arp = kmalloc(alen, GFP_ATOMIC);
2563 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2567 if (arp->ar_hln != bond->dev->addr_len ||
2568 skb->pkt_type == PACKET_OTHERHOST ||
2569 skb->pkt_type == PACKET_LOOPBACK ||
2570 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2571 arp->ar_pro != htons(ETH_P_IP) ||
2575 arp_ptr = (unsigned char *)(arp + 1);
2576 arp_ptr += bond->dev->addr_len;
2577 memcpy(&sip, arp_ptr, 4);
2578 arp_ptr += 4 + bond->dev->addr_len;
2579 memcpy(&tip, arp_ptr, 4);
2581 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2582 slave->dev->name, bond_slave_state(slave),
2583 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2586 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2587 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2589 /* We 'trust' the received ARP enough to validate it if:
2591 * (a) the slave receiving the ARP is active (which includes the
2592 * current ARP slave, if any), or
2594 * (b) the receiving slave isn't active, but there is a currently
2595 * active slave and it received valid arp reply(s) after it became
2596 * the currently active slave, or
2598 * (c) there is an ARP slave that sent an ARP during the prior ARP
2599 * interval, and we receive an ARP reply on any slave. We accept
2600 * these because switch FDB update delays may deliver the ARP
2601 * reply to a slave other than the sender of the ARP request.
2603 * Note: for (b), backup slaves are receiving the broadcast ARP
2604 * request, not a reply. This request passes from the sending
2605 * slave through the L2 switch(es) to the receiving slave. Since
2606 * this is checking the request, sip/tip are swapped for
2609 * This is done to avoid endless looping when we can't reach the
2610 * arp_ip_target and fool ourselves with our own arp requests.
2612 if (bond_is_active_slave(slave))
2613 bond_validate_arp(bond, slave, sip, tip);
2614 else if (curr_active_slave &&
2615 time_after(slave_last_rx(bond, curr_active_slave),
2616 curr_active_slave->last_link_up))
2617 bond_validate_arp(bond, slave, tip, sip);
2618 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2619 bond_time_in_interval(bond,
2620 dev_trans_start(curr_arp_slave->dev), 1))
2621 bond_validate_arp(bond, slave, sip, tip);
2624 if (arp != (struct arphdr *)skb->data)
2626 return RX_HANDLER_ANOTHER;
2629 /* function to verify if we're in the arp_interval timeslice, returns true if
2630 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2631 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2633 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2636 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2638 return time_in_range(jiffies,
2639 last_act - delta_in_ticks,
2640 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2643 /* This function is called regularly to monitor each slave's link
2644 * ensuring that traffic is being sent and received when arp monitoring
2645 * is used in load-balancing mode. if the adapter has been dormant, then an
2646 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2647 * arp monitoring in active backup mode.
2649 static void bond_loadbalance_arp_mon(struct bonding *bond)
2651 struct slave *slave, *oldcurrent;
2652 struct list_head *iter;
2653 int do_failover = 0, slave_state_changed = 0;
2655 if (!bond_has_slaves(bond))
2660 oldcurrent = rcu_dereference(bond->curr_active_slave);
2661 /* see if any of the previous devices are up now (i.e. they have
2662 * xmt and rcv traffic). the curr_active_slave does not come into
2663 * the picture unless it is null. also, slave->last_link_up is not
2664 * needed here because we send an arp on each slave and give a slave
2665 * as long as it needs to get the tx/rx within the delta.
2666 * TODO: what about up/down delay in arp mode? it wasn't here before
2669 bond_for_each_slave_rcu(bond, slave, iter) {
2670 unsigned long trans_start = dev_trans_start(slave->dev);
2672 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2674 if (slave->link != BOND_LINK_UP) {
2675 if (bond_time_in_interval(bond, trans_start, 1) &&
2676 bond_time_in_interval(bond, slave->last_rx, 1)) {
2678 bond_propose_link_state(slave, BOND_LINK_UP);
2679 slave_state_changed = 1;
2681 /* primary_slave has no meaning in round-robin
2682 * mode. the window of a slave being up and
2683 * curr_active_slave being null after enslaving
2687 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2691 netdev_info(bond->dev, "interface %s is now up\n",
2696 /* slave->link == BOND_LINK_UP */
2698 /* not all switches will respond to an arp request
2699 * when the source ip is 0, so don't take the link down
2700 * if we don't know our ip yet
2702 if (!bond_time_in_interval(bond, trans_start, 2) ||
2703 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2705 bond_propose_link_state(slave, BOND_LINK_DOWN);
2706 slave_state_changed = 1;
2708 if (slave->link_failure_count < UINT_MAX)
2709 slave->link_failure_count++;
2711 netdev_info(bond->dev, "interface %s is now down\n",
2714 if (slave == oldcurrent)
2719 /* note: if switch is in round-robin mode, all links
2720 * must tx arp to ensure all links rx an arp - otherwise
2721 * links may oscillate or not come up at all; if switch is
2722 * in something like xor mode, there is nothing we can
2723 * do - all replies will be rx'ed on same link causing slaves
2724 * to be unstable during low/no traffic periods
2726 if (bond_slave_is_up(slave))
2727 bond_arp_send_all(bond, slave);
2732 if (do_failover || slave_state_changed) {
2733 if (!rtnl_trylock())
2736 bond_for_each_slave(bond, slave, iter) {
2737 if (slave->link_new_state != BOND_LINK_NOCHANGE)
2738 slave->link = slave->link_new_state;
2741 if (slave_state_changed) {
2742 bond_slave_state_change(bond);
2743 if (BOND_MODE(bond) == BOND_MODE_XOR)
2744 bond_update_slave_arr(bond, NULL);
2748 bond_select_active_slave(bond);
2749 unblock_netpoll_tx();
2755 if (bond->params.arp_interval)
2756 queue_delayed_work(bond->wq, &bond->arp_work,
2757 msecs_to_jiffies(bond->params.arp_interval));
2760 /* Called to inspect slaves for active-backup mode ARP monitor link state
2761 * changes. Sets proposed link state in slaves to specify what action
2762 * should take place for the slave. Returns 0 if no changes are found, >0
2763 * if changes to link states must be committed.
2765 * Called with rcu_read_lock held.
2767 static int bond_ab_arp_inspect(struct bonding *bond)
2769 unsigned long trans_start, last_rx;
2770 struct list_head *iter;
2771 struct slave *slave;
2774 bond_for_each_slave_rcu(bond, slave, iter) {
2775 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2776 last_rx = slave_last_rx(bond, slave);
2778 if (slave->link != BOND_LINK_UP) {
2779 if (bond_time_in_interval(bond, last_rx, 1)) {
2780 bond_propose_link_state(slave, BOND_LINK_UP);
2782 } else if (slave->link == BOND_LINK_BACK) {
2783 bond_propose_link_state(slave, BOND_LINK_FAIL);
2789 /* Give slaves 2*delta after being enslaved or made
2790 * active. This avoids bouncing, as the last receive
2791 * times need a full ARP monitor cycle to be updated.
2793 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2796 /* Backup slave is down if:
2797 * - No current_arp_slave AND
2798 * - more than 3*delta since last receive AND
2799 * - the bond has an IP address
2801 * Note: a non-null current_arp_slave indicates
2802 * the curr_active_slave went down and we are
2803 * searching for a new one; under this condition
2804 * we only take the curr_active_slave down - this
2805 * gives each slave a chance to tx/rx traffic
2806 * before being taken out
2808 if (!bond_is_active_slave(slave) &&
2809 !rcu_access_pointer(bond->current_arp_slave) &&
2810 !bond_time_in_interval(bond, last_rx, 3)) {
2811 bond_propose_link_state(slave, BOND_LINK_DOWN);
2815 /* Active slave is down if:
2816 * - more than 2*delta since transmitting OR
2817 * - (more than 2*delta since receive AND
2818 * the bond has an IP address)
2820 trans_start = dev_trans_start(slave->dev);
2821 if (bond_is_active_slave(slave) &&
2822 (!bond_time_in_interval(bond, trans_start, 2) ||
2823 !bond_time_in_interval(bond, last_rx, 2))) {
2824 bond_propose_link_state(slave, BOND_LINK_DOWN);
2832 /* Called to commit link state changes noted by inspection step of
2833 * active-backup mode ARP monitor.
2835 * Called with RTNL hold.
2837 static void bond_ab_arp_commit(struct bonding *bond)
2839 unsigned long trans_start;
2840 struct list_head *iter;
2841 struct slave *slave;
2843 bond_for_each_slave(bond, slave, iter) {
2844 switch (slave->link_new_state) {
2845 case BOND_LINK_NOCHANGE:
2849 trans_start = dev_trans_start(slave->dev);
2850 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2851 (!rtnl_dereference(bond->curr_active_slave) &&
2852 bond_time_in_interval(bond, trans_start, 1))) {
2853 struct slave *current_arp_slave;
2855 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2856 bond_set_slave_link_state(slave, BOND_LINK_UP,
2857 BOND_SLAVE_NOTIFY_NOW);
2858 if (current_arp_slave) {
2859 bond_set_slave_inactive_flags(
2861 BOND_SLAVE_NOTIFY_NOW);
2862 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2865 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2868 if (!rtnl_dereference(bond->curr_active_slave) ||
2869 slave == rtnl_dereference(bond->primary_slave))
2876 case BOND_LINK_DOWN:
2877 if (slave->link_failure_count < UINT_MAX)
2878 slave->link_failure_count++;
2880 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2881 BOND_SLAVE_NOTIFY_NOW);
2882 bond_set_slave_inactive_flags(slave,
2883 BOND_SLAVE_NOTIFY_NOW);
2885 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2888 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2889 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2895 case BOND_LINK_FAIL:
2896 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2897 BOND_SLAVE_NOTIFY_NOW);
2898 bond_set_slave_inactive_flags(slave,
2899 BOND_SLAVE_NOTIFY_NOW);
2901 /* A slave has just been enslaved and has become
2902 * the current active slave.
2904 if (rtnl_dereference(bond->curr_active_slave))
2905 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2909 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2910 slave->link_new_state, slave->dev->name);
2916 bond_select_active_slave(bond);
2917 unblock_netpoll_tx();
2920 bond_set_carrier(bond);
2923 /* Send ARP probes for active-backup mode ARP monitor.
2925 * Called with rcu_read_lock held.
2927 static bool bond_ab_arp_probe(struct bonding *bond)
2929 struct slave *slave, *before = NULL, *new_slave = NULL,
2930 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2931 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2932 struct list_head *iter;
2934 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2936 if (curr_arp_slave && curr_active_slave)
2937 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2938 curr_arp_slave->dev->name,
2939 curr_active_slave->dev->name);
2941 if (curr_active_slave) {
2942 bond_arp_send_all(bond, curr_active_slave);
2943 return should_notify_rtnl;
2946 /* if we don't have a curr_active_slave, search for the next available
2947 * backup slave from the current_arp_slave and make it the candidate
2948 * for becoming the curr_active_slave
2951 if (!curr_arp_slave) {
2952 curr_arp_slave = bond_first_slave_rcu(bond);
2953 if (!curr_arp_slave)
2954 return should_notify_rtnl;
2957 bond_for_each_slave_rcu(bond, slave, iter) {
2958 if (!found && !before && bond_slave_is_up(slave))
2961 if (found && !new_slave && bond_slave_is_up(slave))
2963 /* if the link state is up at this point, we
2964 * mark it down - this can happen if we have
2965 * simultaneous link failures and
2966 * reselect_active_interface doesn't make this
2967 * one the current slave so it is still marked
2968 * up when it is actually down
2970 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2971 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2972 BOND_SLAVE_NOTIFY_LATER);
2973 if (slave->link_failure_count < UINT_MAX)
2974 slave->link_failure_count++;
2976 bond_set_slave_inactive_flags(slave,
2977 BOND_SLAVE_NOTIFY_LATER);
2979 netdev_info(bond->dev, "backup interface %s is now down\n",
2982 if (slave == curr_arp_slave)
2986 if (!new_slave && before)
2992 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2993 BOND_SLAVE_NOTIFY_LATER);
2994 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2995 bond_arp_send_all(bond, new_slave);
2996 new_slave->last_link_up = jiffies;
2997 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3000 bond_for_each_slave_rcu(bond, slave, iter) {
3001 if (slave->should_notify || slave->should_notify_link) {
3002 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3006 return should_notify_rtnl;
3009 static void bond_activebackup_arp_mon(struct bonding *bond)
3011 bool should_notify_peers = false;
3012 bool should_notify_rtnl = false;
3015 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3017 if (!bond_has_slaves(bond))
3022 should_notify_peers = bond_should_notify_peers(bond);
3024 if (bond_ab_arp_inspect(bond)) {
3027 /* Race avoidance with bond_close flush of workqueue */
3028 if (!rtnl_trylock()) {
3030 should_notify_peers = false;
3034 bond_ab_arp_commit(bond);
3040 should_notify_rtnl = bond_ab_arp_probe(bond);
3044 if (bond->params.arp_interval)
3045 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3047 if (should_notify_peers || should_notify_rtnl) {
3048 if (!rtnl_trylock())
3051 if (should_notify_peers)
3052 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3054 if (should_notify_rtnl) {
3055 bond_slave_state_notify(bond);
3056 bond_slave_link_notify(bond);
3063 static void bond_arp_monitor(struct work_struct *work)
3065 struct bonding *bond = container_of(work, struct bonding,
3068 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3069 bond_activebackup_arp_mon(bond);
3071 bond_loadbalance_arp_mon(bond);
3074 /*-------------------------- netdev event handling --------------------------*/
3076 /* Change device name */
3077 static int bond_event_changename(struct bonding *bond)
3079 bond_remove_proc_entry(bond);
3080 bond_create_proc_entry(bond);
3082 bond_debug_reregister(bond);
3087 static int bond_master_netdev_event(unsigned long event,
3088 struct net_device *bond_dev)
3090 struct bonding *event_bond = netdev_priv(bond_dev);
3093 case NETDEV_CHANGENAME:
3094 return bond_event_changename(event_bond);
3095 case NETDEV_UNREGISTER:
3096 bond_remove_proc_entry(event_bond);
3098 case NETDEV_REGISTER:
3099 bond_create_proc_entry(event_bond);
3101 case NETDEV_NOTIFY_PEERS:
3102 if (event_bond->send_peer_notif)
3103 event_bond->send_peer_notif--;
3112 static int bond_slave_netdev_event(unsigned long event,
3113 struct net_device *slave_dev)
3115 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3116 struct bonding *bond;
3117 struct net_device *bond_dev;
3119 /* A netdev event can be generated while enslaving a device
3120 * before netdev_rx_handler_register is called in which case
3121 * slave will be NULL
3125 bond_dev = slave->bond->dev;
3127 primary = rtnl_dereference(bond->primary_slave);
3130 case NETDEV_UNREGISTER:
3131 if (bond_dev->type != ARPHRD_ETHER)
3132 bond_release_and_destroy(bond_dev, slave_dev);
3134 __bond_release_one(bond_dev, slave_dev, false, true);
3138 bond_update_speed_duplex(slave);
3139 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3140 bond_3ad_adapter_speed_duplex_changed(slave);
3143 /* Refresh slave-array if applicable!
3144 * If the setup does not use miimon or arpmon (mode-specific!),
3145 * then these events will not cause the slave-array to be
3146 * refreshed. This will cause xmit to use a slave that is not
3147 * usable. Avoid such situation by refeshing the array at these
3148 * events. If these (miimon/arpmon) parameters are configured
3149 * then array gets refreshed twice and that should be fine!
3151 if (bond_mode_uses_xmit_hash(bond))
3152 bond_update_slave_arr(bond, NULL);
3154 case NETDEV_CHANGEMTU:
3155 /* TODO: Should slaves be allowed to
3156 * independently alter their MTU? For
3157 * an active-backup bond, slaves need
3158 * not be the same type of device, so
3159 * MTUs may vary. For other modes,
3160 * slaves arguably should have the
3161 * same MTUs. To do this, we'd need to
3162 * take over the slave's change_mtu
3163 * function for the duration of their
3167 case NETDEV_CHANGENAME:
3168 /* we don't care if we don't have primary set */
3169 if (!bond_uses_primary(bond) ||
3170 !bond->params.primary[0])
3173 if (slave == primary) {
3174 /* slave's name changed - he's no longer primary */
3175 RCU_INIT_POINTER(bond->primary_slave, NULL);
3176 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3177 /* we have a new primary slave */
3178 rcu_assign_pointer(bond->primary_slave, slave);
3179 } else { /* we didn't change primary - exit */
3183 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3184 primary ? slave_dev->name : "none");
3187 bond_select_active_slave(bond);
3188 unblock_netpoll_tx();
3190 case NETDEV_FEAT_CHANGE:
3191 bond_compute_features(bond);
3193 case NETDEV_RESEND_IGMP:
3194 /* Propagate to master device */
3195 call_netdevice_notifiers(event, slave->bond->dev);
3204 /* bond_netdev_event: handle netdev notifier chain events.
3206 * This function receives events for the netdev chain. The caller (an
3207 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3208 * locks for us to safely manipulate the slave devices (RTNL lock,
3211 static int bond_netdev_event(struct notifier_block *this,
3212 unsigned long event, void *ptr)
3214 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3216 netdev_dbg(event_dev, "event: %lx\n", event);
3218 if (!(event_dev->priv_flags & IFF_BONDING))
3221 if (event_dev->flags & IFF_MASTER) {
3224 netdev_dbg(event_dev, "IFF_MASTER\n");
3225 ret = bond_master_netdev_event(event, event_dev);
3226 if (ret != NOTIFY_DONE)
3230 if (event_dev->flags & IFF_SLAVE) {
3231 netdev_dbg(event_dev, "IFF_SLAVE\n");
3232 return bond_slave_netdev_event(event, event_dev);
3238 static struct notifier_block bond_netdev_notifier = {
3239 .notifier_call = bond_netdev_event,
3242 /*---------------------------- Hashing Policies -----------------------------*/
3244 /* L2 hash helper */
3245 static inline u32 bond_eth_hash(struct sk_buff *skb)
3247 struct ethhdr *ep, hdr_tmp;
3249 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3251 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3255 /* Extract the appropriate headers based on bond's xmit policy */
3256 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3257 struct flow_keys *fk)
3259 const struct ipv6hdr *iph6;
3260 const struct iphdr *iph;
3261 int noff, proto = -1;
3263 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3264 return skb_flow_dissect_flow_keys(skb, fk, 0);
3266 fk->ports.ports = 0;
3267 noff = skb_network_offset(skb);
3268 if (skb->protocol == htons(ETH_P_IP)) {
3269 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3272 iph_to_flow_copy_v4addrs(fk, iph);
3273 noff += iph->ihl << 2;
3274 if (!ip_is_fragment(iph))
3275 proto = iph->protocol;
3276 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3277 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3279 iph6 = ipv6_hdr(skb);
3280 iph_to_flow_copy_v6addrs(fk, iph6);
3281 noff += sizeof(*iph6);
3282 proto = iph6->nexthdr;
3286 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3287 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3293 * bond_xmit_hash - generate a hash value based on the xmit policy
3294 * @bond: bonding device
3295 * @skb: buffer to use for headers
3297 * This function will extract the necessary headers from the skb buffer and use
3298 * them to generate a hash based on the xmit_policy set in the bonding device
3300 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3302 struct flow_keys flow;
3305 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3309 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3310 !bond_flow_dissect(bond, skb, &flow))
3311 return bond_eth_hash(skb);
3313 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3314 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3315 hash = bond_eth_hash(skb);
3317 hash = (__force u32)flow.ports.ports;
3318 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3319 (__force u32)flow_get_u32_src(&flow);
3320 hash ^= (hash >> 16);
3321 hash ^= (hash >> 8);
3326 /*-------------------------- Device entry points ----------------------------*/
3328 void bond_work_init_all(struct bonding *bond)
3330 INIT_DELAYED_WORK(&bond->mcast_work,
3331 bond_resend_igmp_join_requests_delayed);
3332 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3333 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3334 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3335 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3336 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3339 static void bond_work_cancel_all(struct bonding *bond)
3341 cancel_delayed_work_sync(&bond->mii_work);
3342 cancel_delayed_work_sync(&bond->arp_work);
3343 cancel_delayed_work_sync(&bond->alb_work);
3344 cancel_delayed_work_sync(&bond->ad_work);
3345 cancel_delayed_work_sync(&bond->mcast_work);
3346 cancel_delayed_work_sync(&bond->slave_arr_work);
3349 static int bond_open(struct net_device *bond_dev)
3351 struct bonding *bond = netdev_priv(bond_dev);
3352 struct list_head *iter;
3353 struct slave *slave;
3355 /* reset slave->backup and slave->inactive */
3356 if (bond_has_slaves(bond)) {
3357 bond_for_each_slave(bond, slave, iter) {
3358 if (bond_uses_primary(bond) &&
3359 slave != rcu_access_pointer(bond->curr_active_slave)) {
3360 bond_set_slave_inactive_flags(slave,
3361 BOND_SLAVE_NOTIFY_NOW);
3362 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3363 bond_set_slave_active_flags(slave,
3364 BOND_SLAVE_NOTIFY_NOW);
3369 if (bond_is_lb(bond)) {
3370 /* bond_alb_initialize must be called before the timer
3373 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3375 if (bond->params.tlb_dynamic_lb)
3376 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3379 if (bond->params.miimon) /* link check interval, in milliseconds. */
3380 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3382 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3383 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3384 bond->recv_probe = bond_arp_rcv;
3387 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3388 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3389 /* register to receive LACPDUs */
3390 bond->recv_probe = bond_3ad_lacpdu_recv;
3391 bond_3ad_initiate_agg_selection(bond, 1);
3394 if (bond_mode_uses_xmit_hash(bond))
3395 bond_update_slave_arr(bond, NULL);
3400 static int bond_close(struct net_device *bond_dev)
3402 struct bonding *bond = netdev_priv(bond_dev);
3404 bond_work_cancel_all(bond);
3405 bond->send_peer_notif = 0;
3406 if (bond_is_lb(bond))
3407 bond_alb_deinitialize(bond);
3408 bond->recv_probe = NULL;
3413 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3414 * that some drivers can provide 32bit values only.
3416 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3417 const struct rtnl_link_stats64 *_new,
3418 const struct rtnl_link_stats64 *_old)
3420 const u64 *new = (const u64 *)_new;
3421 const u64 *old = (const u64 *)_old;
3422 u64 *res = (u64 *)_res;
3425 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3428 s64 delta = nv - ov;
3430 /* detects if this particular field is 32bit only */
3431 if (((nv | ov) >> 32) == 0)
3432 delta = (s64)(s32)((u32)nv - (u32)ov);
3434 /* filter anomalies, some drivers reset their stats
3435 * at down/up events.
3442 static int bond_get_nest_level(struct net_device *bond_dev)
3444 struct bonding *bond = netdev_priv(bond_dev);
3446 return bond->nest_level;
3449 static void bond_get_stats(struct net_device *bond_dev,
3450 struct rtnl_link_stats64 *stats)
3452 struct bonding *bond = netdev_priv(bond_dev);
3453 struct rtnl_link_stats64 temp;
3454 struct list_head *iter;
3455 struct slave *slave;
3457 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3458 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3461 bond_for_each_slave_rcu(bond, slave, iter) {
3462 const struct rtnl_link_stats64 *new =
3463 dev_get_stats(slave->dev, &temp);
3465 bond_fold_stats(stats, new, &slave->slave_stats);
3467 /* save off the slave stats for the next run */
3468 memcpy(&slave->slave_stats, new, sizeof(*new));
3472 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3473 spin_unlock(&bond->stats_lock);
3476 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3478 struct bonding *bond = netdev_priv(bond_dev);
3479 struct net_device *slave_dev = NULL;
3480 struct ifbond k_binfo;
3481 struct ifbond __user *u_binfo = NULL;
3482 struct ifslave k_sinfo;
3483 struct ifslave __user *u_sinfo = NULL;
3484 struct mii_ioctl_data *mii = NULL;
3485 struct bond_opt_value newval;
3489 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3500 /* We do this again just in case we were called by SIOCGMIIREG
3501 * instead of SIOCGMIIPHY.
3507 if (mii->reg_num == 1) {
3509 if (netif_carrier_ok(bond->dev))
3510 mii->val_out = BMSR_LSTATUS;
3514 case BOND_INFO_QUERY_OLD:
3515 case SIOCBONDINFOQUERY:
3516 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3518 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3521 bond_info_query(bond_dev, &k_binfo);
3522 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3526 case BOND_SLAVE_INFO_QUERY_OLD:
3527 case SIOCBONDSLAVEINFOQUERY:
3528 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3530 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3533 res = bond_slave_info_query(bond_dev, &k_sinfo);
3535 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3543 net = dev_net(bond_dev);
3545 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3548 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3550 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3555 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3557 case BOND_ENSLAVE_OLD:
3558 case SIOCBONDENSLAVE:
3559 res = bond_enslave(bond_dev, slave_dev);
3561 case BOND_RELEASE_OLD:
3562 case SIOCBONDRELEASE:
3563 res = bond_release(bond_dev, slave_dev);
3565 case BOND_SETHWADDR_OLD:
3566 case SIOCBONDSETHWADDR:
3567 bond_set_dev_addr(bond_dev, slave_dev);
3570 case BOND_CHANGE_ACTIVE_OLD:
3571 case SIOCBONDCHANGEACTIVE:
3572 bond_opt_initstr(&newval, slave_dev->name);
3573 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3583 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3585 struct bonding *bond = netdev_priv(bond_dev);
3587 if (change & IFF_PROMISC)
3588 bond_set_promiscuity(bond,
3589 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3591 if (change & IFF_ALLMULTI)
3592 bond_set_allmulti(bond,
3593 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3596 static void bond_set_rx_mode(struct net_device *bond_dev)
3598 struct bonding *bond = netdev_priv(bond_dev);
3599 struct list_head *iter;
3600 struct slave *slave;
3603 if (bond_uses_primary(bond)) {
3604 slave = rcu_dereference(bond->curr_active_slave);
3606 dev_uc_sync(slave->dev, bond_dev);
3607 dev_mc_sync(slave->dev, bond_dev);
3610 bond_for_each_slave_rcu(bond, slave, iter) {
3611 dev_uc_sync_multiple(slave->dev, bond_dev);
3612 dev_mc_sync_multiple(slave->dev, bond_dev);
3618 static int bond_neigh_init(struct neighbour *n)
3620 struct bonding *bond = netdev_priv(n->dev);
3621 const struct net_device_ops *slave_ops;
3622 struct neigh_parms parms;
3623 struct slave *slave;
3626 slave = bond_first_slave(bond);
3629 slave_ops = slave->dev->netdev_ops;
3630 if (!slave_ops->ndo_neigh_setup)
3633 parms.neigh_setup = NULL;
3634 parms.neigh_cleanup = NULL;
3635 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3639 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3640 * after the last slave has been detached. Assumes that all slaves
3641 * utilize the same neigh_cleanup (true at this writing as only user
3644 n->parms->neigh_cleanup = parms.neigh_cleanup;
3646 if (!parms.neigh_setup)
3649 return parms.neigh_setup(n);
3652 /* The bonding ndo_neigh_setup is called at init time beofre any
3653 * slave exists. So we must declare proxy setup function which will
3654 * be used at run time to resolve the actual slave neigh param setup.
3656 * It's also called by master devices (such as vlans) to setup their
3657 * underlying devices. In that case - do nothing, we're already set up from
3660 static int bond_neigh_setup(struct net_device *dev,
3661 struct neigh_parms *parms)
3663 /* modify only our neigh_parms */
3664 if (parms->dev == dev)
3665 parms->neigh_setup = bond_neigh_init;
3670 /* Change the MTU of all of a master's slaves to match the master */
3671 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3673 struct bonding *bond = netdev_priv(bond_dev);
3674 struct slave *slave, *rollback_slave;
3675 struct list_head *iter;
3678 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3680 bond_for_each_slave(bond, slave, iter) {
3681 netdev_dbg(bond_dev, "s %p c_m %p\n",
3682 slave, slave->dev->netdev_ops->ndo_change_mtu);
3684 res = dev_set_mtu(slave->dev, new_mtu);
3687 /* If we failed to set the slave's mtu to the new value
3688 * we must abort the operation even in ACTIVE_BACKUP
3689 * mode, because if we allow the backup slaves to have
3690 * different mtu values than the active slave we'll
3691 * need to change their mtu when doing a failover. That
3692 * means changing their mtu from timer context, which
3693 * is probably not a good idea.
3695 netdev_dbg(bond_dev, "err %d %s\n", res,
3701 bond_dev->mtu = new_mtu;
3706 /* unwind from head to the slave that failed */
3707 bond_for_each_slave(bond, rollback_slave, iter) {
3710 if (rollback_slave == slave)
3713 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3715 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3716 tmp_res, rollback_slave->dev->name);
3723 /* Change HW address
3725 * Note that many devices must be down to change the HW address, and
3726 * downing the master releases all slaves. We can make bonds full of
3727 * bonding devices to test this, however.
3729 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3731 struct bonding *bond = netdev_priv(bond_dev);
3732 struct slave *slave, *rollback_slave;
3733 struct sockaddr_storage *ss = addr, tmp_ss;
3734 struct list_head *iter;
3737 if (BOND_MODE(bond) == BOND_MODE_ALB)
3738 return bond_alb_set_mac_address(bond_dev, addr);
3741 netdev_dbg(bond_dev, "bond=%p\n", bond);
3743 /* If fail_over_mac is enabled, do nothing and return success.
3744 * Returning an error causes ifenslave to fail.
3746 if (bond->params.fail_over_mac &&
3747 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3750 if (!is_valid_ether_addr(ss->__data))
3751 return -EADDRNOTAVAIL;
3753 bond_for_each_slave(bond, slave, iter) {
3754 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3755 res = dev_set_mac_address(slave->dev, addr);
3757 /* TODO: consider downing the slave
3759 * User should expect communications
3760 * breakage anyway until ARP finish
3763 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3769 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3773 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3774 tmp_ss.ss_family = bond_dev->type;
3776 /* unwind from head to the slave that failed */
3777 bond_for_each_slave(bond, rollback_slave, iter) {
3780 if (rollback_slave == slave)
3783 tmp_res = dev_set_mac_address(rollback_slave->dev,
3784 (struct sockaddr *)&tmp_ss);
3786 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3787 tmp_res, rollback_slave->dev->name);
3795 * bond_xmit_slave_id - transmit skb through slave with slave_id
3796 * @bond: bonding device that is transmitting
3797 * @skb: buffer to transmit
3798 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3800 * This function tries to transmit through slave with slave_id but in case
3801 * it fails, it tries to find the first available slave for transmission.
3802 * The skb is consumed in all cases, thus the function is void.
3804 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3806 struct list_head *iter;
3807 struct slave *slave;
3810 /* Here we start from the slave with slave_id */
3811 bond_for_each_slave_rcu(bond, slave, iter) {
3813 if (bond_slave_can_tx(slave)) {
3814 bond_dev_queue_xmit(bond, skb, slave->dev);
3820 /* Here we start from the first slave up to slave_id */
3822 bond_for_each_slave_rcu(bond, slave, iter) {
3825 if (bond_slave_can_tx(slave)) {
3826 bond_dev_queue_xmit(bond, skb, slave->dev);
3830 /* no slave that can tx has been found */
3831 bond_tx_drop(bond->dev, skb);
3835 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3836 * @bond: bonding device to use
3838 * Based on the value of the bonding device's packets_per_slave parameter
3839 * this function generates a slave id, which is usually used as the next
3840 * slave to transmit through.
3842 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3845 struct reciprocal_value reciprocal_packets_per_slave;
3846 int packets_per_slave = bond->params.packets_per_slave;
3848 switch (packets_per_slave) {
3850 slave_id = prandom_u32();
3853 slave_id = bond->rr_tx_counter;
3856 reciprocal_packets_per_slave =
3857 bond->params.reciprocal_packets_per_slave;
3858 slave_id = reciprocal_divide(bond->rr_tx_counter,
3859 reciprocal_packets_per_slave);
3862 bond->rr_tx_counter++;
3867 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3869 struct bonding *bond = netdev_priv(bond_dev);
3870 struct slave *slave;
3874 /* Start with the curr_active_slave that joined the bond as the
3875 * default for sending IGMP traffic. For failover purposes one
3876 * needs to maintain some consistency for the interface that will
3877 * send the join/membership reports. The curr_active_slave found
3878 * will send all of this type of traffic.
3880 if (skb->protocol == htons(ETH_P_IP)) {
3881 int noff = skb_network_offset(skb);
3884 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3888 if (iph->protocol == IPPROTO_IGMP) {
3889 slave = rcu_dereference(bond->curr_active_slave);
3891 bond_dev_queue_xmit(bond, skb, slave->dev);
3893 bond_xmit_slave_id(bond, skb, 0);
3894 return NETDEV_TX_OK;
3899 slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3900 if (likely(slave_cnt)) {
3901 slave_id = bond_rr_gen_slave_id(bond);
3902 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3904 bond_tx_drop(bond_dev, skb);
3906 return NETDEV_TX_OK;
3909 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3910 * the bond has a usable interface.
3912 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3914 struct bonding *bond = netdev_priv(bond_dev);
3915 struct slave *slave;
3917 slave = rcu_dereference(bond->curr_active_slave);
3919 bond_dev_queue_xmit(bond, skb, slave->dev);
3921 bond_tx_drop(bond_dev, skb);
3923 return NETDEV_TX_OK;
3926 /* Use this to update slave_array when (a) it's not appropriate to update
3927 * slave_array right away (note that update_slave_array() may sleep)
3928 * and / or (b) RTNL is not held.
3930 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3932 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3935 /* Slave array work handler. Holds only RTNL */
3936 static void bond_slave_arr_handler(struct work_struct *work)
3938 struct bonding *bond = container_of(work, struct bonding,
3939 slave_arr_work.work);
3942 if (!rtnl_trylock())
3945 ret = bond_update_slave_arr(bond, NULL);
3948 pr_warn_ratelimited("Failed to update slave array from WT\n");
3954 bond_slave_arr_work_rearm(bond, 1);
3957 /* Build the usable slaves array in control path for modes that use xmit-hash
3958 * to determine the slave interface -
3959 * (a) BOND_MODE_8023AD
3961 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3963 * The caller is expected to hold RTNL only and NO other lock!
3965 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3967 struct slave *slave;
3968 struct list_head *iter;
3969 struct bond_up_slave *new_arr, *old_arr;
3973 #ifdef CONFIG_LOCKDEP
3974 WARN_ON(lockdep_is_held(&bond->mode_lock));
3977 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3981 pr_err("Failed to build slave-array.\n");
3984 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3985 struct ad_info ad_info;
3987 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3988 pr_debug("bond_3ad_get_active_agg_info failed\n");
3989 kfree_rcu(new_arr, rcu);
3990 /* No active aggragator means it's not safe to use
3991 * the previous array.
3993 old_arr = rtnl_dereference(bond->slave_arr);
3995 RCU_INIT_POINTER(bond->slave_arr, NULL);
3996 kfree_rcu(old_arr, rcu);
4000 agg_id = ad_info.aggregator_id;
4002 bond_for_each_slave(bond, slave, iter) {
4003 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4004 struct aggregator *agg;
4006 agg = SLAVE_AD_INFO(slave)->port.aggregator;
4007 if (!agg || agg->aggregator_identifier != agg_id)
4010 if (!bond_slave_can_tx(slave))
4012 if (skipslave == slave)
4014 new_arr->arr[new_arr->count++] = slave;
4017 old_arr = rtnl_dereference(bond->slave_arr);
4018 rcu_assign_pointer(bond->slave_arr, new_arr);
4020 kfree_rcu(old_arr, rcu);
4022 if (ret != 0 && skipslave) {
4025 /* Rare situation where caller has asked to skip a specific
4026 * slave but allocation failed (most likely!). BTW this is
4027 * only possible when the call is initiated from
4028 * __bond_release_one(). In this situation; overwrite the
4029 * skipslave entry in the array with the last entry from the
4030 * array to avoid a situation where the xmit path may choose
4031 * this to-be-skipped slave to send a packet out.
4033 old_arr = rtnl_dereference(bond->slave_arr);
4034 for (idx = 0; old_arr != NULL && idx < old_arr->count; idx++) {
4035 if (skipslave == old_arr->arr[idx]) {
4037 old_arr->arr[old_arr->count-1];
4046 /* Use this Xmit function for 3AD as well as XOR modes. The current
4047 * usable slave array is formed in the control path. The xmit function
4048 * just calculates hash and sends the packet out.
4050 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
4052 struct bonding *bond = netdev_priv(dev);
4053 struct slave *slave;
4054 struct bond_up_slave *slaves;
4057 slaves = rcu_dereference(bond->slave_arr);
4058 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
4059 if (likely(count)) {
4060 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4061 bond_dev_queue_xmit(bond, skb, slave->dev);
4063 bond_tx_drop(dev, skb);
4066 return NETDEV_TX_OK;
4069 /* in broadcast mode, we send everything to all usable interfaces. */
4070 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4072 struct bonding *bond = netdev_priv(bond_dev);
4073 struct slave *slave = NULL;
4074 struct list_head *iter;
4076 bond_for_each_slave_rcu(bond, slave, iter) {
4077 if (bond_is_last_slave(bond, slave))
4079 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4080 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4083 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4084 bond_dev->name, __func__);
4087 bond_dev_queue_xmit(bond, skb2, slave->dev);
4090 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4091 bond_dev_queue_xmit(bond, skb, slave->dev);
4093 bond_tx_drop(bond_dev, skb);
4095 return NETDEV_TX_OK;
4098 /*------------------------- Device initialization ---------------------------*/
4100 /* Lookup the slave that corresponds to a qid */
4101 static inline int bond_slave_override(struct bonding *bond,
4102 struct sk_buff *skb)
4104 struct slave *slave = NULL;
4105 struct list_head *iter;
4107 if (!skb->queue_mapping)
4110 /* Find out if any slaves have the same mapping as this skb. */
4111 bond_for_each_slave_rcu(bond, slave, iter) {
4112 if (slave->queue_id == skb->queue_mapping) {
4113 if (bond_slave_is_up(slave) &&
4114 slave->link == BOND_LINK_UP) {
4115 bond_dev_queue_xmit(bond, skb, slave->dev);
4118 /* If the slave isn't UP, use default transmit policy. */
4127 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4128 void *accel_priv, select_queue_fallback_t fallback)
4130 /* This helper function exists to help dev_pick_tx get the correct
4131 * destination queue. Using a helper function skips a call to
4132 * skb_tx_hash and will put the skbs in the queue we expect on their
4133 * way down to the bonding driver.
4135 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4137 /* Save the original txq to restore before passing to the driver */
4138 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4140 if (unlikely(txq >= dev->real_num_tx_queues)) {
4142 txq -= dev->real_num_tx_queues;
4143 } while (txq >= dev->real_num_tx_queues);
4148 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4150 struct bonding *bond = netdev_priv(dev);
4152 if (bond_should_override_tx_queue(bond) &&
4153 !bond_slave_override(bond, skb))
4154 return NETDEV_TX_OK;
4156 switch (BOND_MODE(bond)) {
4157 case BOND_MODE_ROUNDROBIN:
4158 return bond_xmit_roundrobin(skb, dev);
4159 case BOND_MODE_ACTIVEBACKUP:
4160 return bond_xmit_activebackup(skb, dev);
4161 case BOND_MODE_8023AD:
4163 return bond_3ad_xor_xmit(skb, dev);
4164 case BOND_MODE_BROADCAST:
4165 return bond_xmit_broadcast(skb, dev);
4167 return bond_alb_xmit(skb, dev);
4169 return bond_tlb_xmit(skb, dev);
4171 /* Should never happen, mode already checked */
4172 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4174 bond_tx_drop(dev, skb);
4175 return NETDEV_TX_OK;
4179 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4181 struct bonding *bond = netdev_priv(dev);
4182 netdev_tx_t ret = NETDEV_TX_OK;
4184 /* If we risk deadlock from transmitting this in the
4185 * netpoll path, tell netpoll to queue the frame for later tx
4187 if (unlikely(is_netpoll_tx_blocked(dev)))
4188 return NETDEV_TX_BUSY;
4191 if (bond_has_slaves(bond))
4192 ret = __bond_start_xmit(skb, dev);
4194 bond_tx_drop(dev, skb);
4200 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4202 if (speed == 0 || speed == SPEED_UNKNOWN)
4203 speed = slave->speed;
4205 speed = min(speed, slave->speed);
4210 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4211 struct ethtool_link_ksettings *cmd)
4213 struct bonding *bond = netdev_priv(bond_dev);
4214 struct list_head *iter;
4215 struct slave *slave;
4218 cmd->base.duplex = DUPLEX_UNKNOWN;
4219 cmd->base.port = PORT_OTHER;
4221 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4222 * do not need to check mode. Though link speed might not represent
4223 * the true receive or transmit bandwidth (not all modes are symmetric)
4224 * this is an accurate maximum.
4226 bond_for_each_slave(bond, slave, iter) {
4227 if (bond_slave_can_tx(slave)) {
4228 if (slave->speed != SPEED_UNKNOWN) {
4229 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4230 speed = bond_mode_bcast_speed(slave,
4233 speed += slave->speed;
4235 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4236 slave->duplex != DUPLEX_UNKNOWN)
4237 cmd->base.duplex = slave->duplex;
4240 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4245 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4246 struct ethtool_drvinfo *drvinfo)
4248 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4249 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4250 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4254 static const struct ethtool_ops bond_ethtool_ops = {
4255 .get_drvinfo = bond_ethtool_get_drvinfo,
4256 .get_link = ethtool_op_get_link,
4257 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4260 static const struct net_device_ops bond_netdev_ops = {
4261 .ndo_init = bond_init,
4262 .ndo_uninit = bond_uninit,
4263 .ndo_open = bond_open,
4264 .ndo_stop = bond_close,
4265 .ndo_start_xmit = bond_start_xmit,
4266 .ndo_select_queue = bond_select_queue,
4267 .ndo_get_stats64 = bond_get_stats,
4268 .ndo_do_ioctl = bond_do_ioctl,
4269 .ndo_change_rx_flags = bond_change_rx_flags,
4270 .ndo_set_rx_mode = bond_set_rx_mode,
4271 .ndo_change_mtu = bond_change_mtu,
4272 .ndo_set_mac_address = bond_set_mac_address,
4273 .ndo_neigh_setup = bond_neigh_setup,
4274 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4275 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4276 .ndo_get_lock_subclass = bond_get_nest_level,
4277 #ifdef CONFIG_NET_POLL_CONTROLLER
4278 .ndo_netpoll_setup = bond_netpoll_setup,
4279 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4280 .ndo_poll_controller = bond_poll_controller,
4282 .ndo_add_slave = bond_enslave,
4283 .ndo_del_slave = bond_release,
4284 .ndo_fix_features = bond_fix_features,
4285 .ndo_features_check = passthru_features_check,
4288 static const struct device_type bond_type = {
4292 static void bond_destructor(struct net_device *bond_dev)
4294 struct bonding *bond = netdev_priv(bond_dev);
4296 destroy_workqueue(bond->wq);
4299 void bond_setup(struct net_device *bond_dev)
4301 struct bonding *bond = netdev_priv(bond_dev);
4303 spin_lock_init(&bond->mode_lock);
4304 spin_lock_init(&bond->stats_lock);
4305 bond->params = bonding_defaults;
4307 /* Initialize pointers */
4308 bond->dev = bond_dev;
4310 /* Initialize the device entry points */
4311 ether_setup(bond_dev);
4312 bond_dev->max_mtu = ETH_MAX_MTU;
4313 bond_dev->netdev_ops = &bond_netdev_ops;
4314 bond_dev->ethtool_ops = &bond_ethtool_ops;
4316 bond_dev->needs_free_netdev = true;
4317 bond_dev->priv_destructor = bond_destructor;
4319 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4321 /* Initialize the device options */
4322 bond_dev->flags |= IFF_MASTER;
4323 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4324 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4326 /* don't acquire bond device's netif_tx_lock when transmitting */
4327 bond_dev->features |= NETIF_F_LLTX;
4329 /* By default, we declare the bond to be fully
4330 * VLAN hardware accelerated capable. Special
4331 * care is taken in the various xmit functions
4332 * when there are slaves that are not hw accel
4336 /* Don't allow bond devices to change network namespaces. */
4337 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4339 bond_dev->hw_features = BOND_VLAN_FEATURES |
4340 NETIF_F_HW_VLAN_CTAG_RX |
4341 NETIF_F_HW_VLAN_CTAG_FILTER;
4343 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4344 bond_dev->features |= bond_dev->hw_features;
4345 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX;
4348 /* Destroy a bonding device.
4349 * Must be under rtnl_lock when this function is called.
4351 static void bond_uninit(struct net_device *bond_dev)
4353 struct bonding *bond = netdev_priv(bond_dev);
4354 struct list_head *iter;
4355 struct slave *slave;
4356 struct bond_up_slave *arr;
4358 bond_netpoll_cleanup(bond_dev);
4360 /* Release the bonded slaves */
4361 bond_for_each_slave(bond, slave, iter)
4362 __bond_release_one(bond_dev, slave->dev, true, true);
4363 netdev_info(bond_dev, "Released all slaves\n");
4365 arr = rtnl_dereference(bond->slave_arr);
4367 RCU_INIT_POINTER(bond->slave_arr, NULL);
4368 kfree_rcu(arr, rcu);
4371 list_del(&bond->bond_list);
4373 bond_debug_unregister(bond);
4376 /*------------------------- Module initialization ---------------------------*/
4378 static int bond_check_params(struct bond_params *params)
4380 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4381 struct bond_opt_value newval;
4382 const struct bond_opt_value *valptr;
4383 int arp_all_targets_value = 0;
4384 u16 ad_actor_sys_prio = 0;
4385 u16 ad_user_port_key = 0;
4386 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4388 int bond_mode = BOND_MODE_ROUNDROBIN;
4389 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4393 /* Convert string parameters. */
4395 bond_opt_initstr(&newval, mode);
4396 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4398 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4401 bond_mode = valptr->value;
4404 if (xmit_hash_policy) {
4405 if ((bond_mode != BOND_MODE_XOR) &&
4406 (bond_mode != BOND_MODE_8023AD) &&
4407 (bond_mode != BOND_MODE_TLB)) {
4408 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4409 bond_mode_name(bond_mode));
4411 bond_opt_initstr(&newval, xmit_hash_policy);
4412 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4415 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4419 xmit_hashtype = valptr->value;
4424 if (bond_mode != BOND_MODE_8023AD) {
4425 pr_info("lacp_rate param is irrelevant in mode %s\n",
4426 bond_mode_name(bond_mode));
4428 bond_opt_initstr(&newval, lacp_rate);
4429 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4432 pr_err("Error: Invalid lacp rate \"%s\"\n",
4436 lacp_fast = valptr->value;
4441 bond_opt_initstr(&newval, ad_select);
4442 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4445 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4448 params->ad_select = valptr->value;
4449 if (bond_mode != BOND_MODE_8023AD)
4450 pr_warn("ad_select param only affects 802.3ad mode\n");
4452 params->ad_select = BOND_AD_STABLE;
4455 if (max_bonds < 0) {
4456 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4457 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4458 max_bonds = BOND_DEFAULT_MAX_BONDS;
4462 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4468 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4473 if (downdelay < 0) {
4474 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4475 downdelay, INT_MAX);
4479 if ((use_carrier != 0) && (use_carrier != 1)) {
4480 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4485 if (num_peer_notif < 0 || num_peer_notif > 255) {
4486 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4491 /* reset values for 802.3ad/TLB/ALB */
4492 if (!bond_mode_uses_arp(bond_mode)) {
4494 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");
4495 pr_warn("Forcing miimon to 100msec\n");
4496 miimon = BOND_DEFAULT_MIIMON;
4500 if (tx_queues < 1 || tx_queues > 255) {
4501 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4502 tx_queues, BOND_DEFAULT_TX_QUEUES);
4503 tx_queues = BOND_DEFAULT_TX_QUEUES;
4506 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4507 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4509 all_slaves_active = 0;
4512 if (resend_igmp < 0 || resend_igmp > 255) {
4513 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4514 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4515 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4518 bond_opt_initval(&newval, packets_per_slave);
4519 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4520 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4521 packets_per_slave, USHRT_MAX);
4522 packets_per_slave = 1;
4525 if (bond_mode == BOND_MODE_ALB) {
4526 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",
4531 if (updelay || downdelay) {
4532 /* just warn the user the up/down delay will have
4533 * no effect since miimon is zero...
4535 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",
4536 updelay, downdelay);
4539 /* don't allow arp monitoring */
4541 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4542 miimon, arp_interval);
4546 if ((updelay % miimon) != 0) {
4547 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4548 updelay, miimon, (updelay / miimon) * miimon);
4553 if ((downdelay % miimon) != 0) {
4554 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4556 (downdelay / miimon) * miimon);
4559 downdelay /= miimon;
4562 if (arp_interval < 0) {
4563 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4564 arp_interval, INT_MAX);
4568 for (arp_ip_count = 0, i = 0;
4569 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4572 /* not a complete check, but good enough to catch mistakes */
4573 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4574 !bond_is_ip_target_ok(ip)) {
4575 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4579 if (bond_get_targets_ip(arp_target, ip) == -1)
4580 arp_target[arp_ip_count++] = ip;
4582 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4587 if (arp_interval && !arp_ip_count) {
4588 /* don't allow arping if no arp_ip_target given... */
4589 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4595 if (!arp_interval) {
4596 pr_err("arp_validate requires arp_interval\n");
4600 bond_opt_initstr(&newval, arp_validate);
4601 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4604 pr_err("Error: invalid arp_validate \"%s\"\n",
4608 arp_validate_value = valptr->value;
4610 arp_validate_value = 0;
4613 if (arp_all_targets) {
4614 bond_opt_initstr(&newval, arp_all_targets);
4615 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4618 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4620 arp_all_targets_value = 0;
4622 arp_all_targets_value = valptr->value;
4627 pr_info("MII link monitoring set to %d ms\n", miimon);
4628 } else if (arp_interval) {
4629 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4630 arp_validate_value);
4631 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4632 arp_interval, valptr->string, arp_ip_count);
4634 for (i = 0; i < arp_ip_count; i++)
4635 pr_cont(" %s", arp_ip_target[i]);
4639 } else if (max_bonds) {
4640 /* miimon and arp_interval not set, we need one so things
4641 * work as expected, see bonding.txt for details
4643 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");
4646 if (primary && !bond_mode_uses_primary(bond_mode)) {
4647 /* currently, using a primary only makes sense
4648 * in active backup, TLB or ALB modes
4650 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4651 primary, bond_mode_name(bond_mode));
4655 if (primary && primary_reselect) {
4656 bond_opt_initstr(&newval, primary_reselect);
4657 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4660 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4664 primary_reselect_value = valptr->value;
4666 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4669 if (fail_over_mac) {
4670 bond_opt_initstr(&newval, fail_over_mac);
4671 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4674 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4678 fail_over_mac_value = valptr->value;
4679 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4680 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4682 fail_over_mac_value = BOND_FOM_NONE;
4685 bond_opt_initstr(&newval, "default");
4686 valptr = bond_opt_parse(
4687 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4690 pr_err("Error: No ad_actor_sys_prio default value");
4693 ad_actor_sys_prio = valptr->value;
4695 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4698 pr_err("Error: No ad_user_port_key default value");
4701 ad_user_port_key = valptr->value;
4703 bond_opt_initstr(&newval, "default");
4704 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4706 pr_err("Error: No tlb_dynamic_lb default value");
4709 tlb_dynamic_lb = valptr->value;
4711 if (lp_interval == 0) {
4712 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4713 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4714 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4717 /* fill params struct with the proper values */
4718 params->mode = bond_mode;
4719 params->xmit_policy = xmit_hashtype;
4720 params->miimon = miimon;
4721 params->num_peer_notif = num_peer_notif;
4722 params->arp_interval = arp_interval;
4723 params->arp_validate = arp_validate_value;
4724 params->arp_all_targets = arp_all_targets_value;
4725 params->updelay = updelay;
4726 params->downdelay = downdelay;
4727 params->use_carrier = use_carrier;
4728 params->lacp_fast = lacp_fast;
4729 params->primary[0] = 0;
4730 params->primary_reselect = primary_reselect_value;
4731 params->fail_over_mac = fail_over_mac_value;
4732 params->tx_queues = tx_queues;
4733 params->all_slaves_active = all_slaves_active;
4734 params->resend_igmp = resend_igmp;
4735 params->min_links = min_links;
4736 params->lp_interval = lp_interval;
4737 params->packets_per_slave = packets_per_slave;
4738 params->tlb_dynamic_lb = tlb_dynamic_lb;
4739 params->ad_actor_sys_prio = ad_actor_sys_prio;
4740 eth_zero_addr(params->ad_actor_system);
4741 params->ad_user_port_key = ad_user_port_key;
4742 if (packets_per_slave > 0) {
4743 params->reciprocal_packets_per_slave =
4744 reciprocal_value(packets_per_slave);
4746 /* reciprocal_packets_per_slave is unused if
4747 * packets_per_slave is 0 or 1, just initialize it
4749 params->reciprocal_packets_per_slave =
4750 (struct reciprocal_value) { 0 };
4754 strncpy(params->primary, primary, IFNAMSIZ);
4755 params->primary[IFNAMSIZ - 1] = 0;
4758 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4763 /* Called from registration process */
4764 static int bond_init(struct net_device *bond_dev)
4766 struct bonding *bond = netdev_priv(bond_dev);
4767 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4769 netdev_dbg(bond_dev, "Begin bond_init\n");
4771 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4775 bond->nest_level = SINGLE_DEPTH_NESTING;
4776 netdev_lockdep_set_classes(bond_dev);
4778 list_add_tail(&bond->bond_list, &bn->dev_list);
4780 bond_prepare_sysfs_group(bond);
4782 bond_debug_register(bond);
4784 /* Ensure valid dev_addr */
4785 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4786 bond_dev->addr_assign_type == NET_ADDR_PERM)
4787 eth_hw_addr_random(bond_dev);
4792 unsigned int bond_get_num_tx_queues(void)
4797 /* Create a new bond based on the specified name and bonding parameters.
4798 * If name is NULL, obtain a suitable "bond%d" name for us.
4799 * Caller must NOT hold rtnl_lock; we need to release it here before we
4800 * set up our sysfs entries.
4802 int bond_create(struct net *net, const char *name)
4804 struct net_device *bond_dev;
4805 struct bonding *bond;
4806 struct alb_bond_info *bond_info;
4811 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4812 name ? name : "bond%d", NET_NAME_UNKNOWN,
4813 bond_setup, tx_queues);
4815 pr_err("%s: eek! can't alloc netdev!\n", name);
4821 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4822 * It is set to 0 by default which is wrong.
4824 bond = netdev_priv(bond_dev);
4825 bond_info = &(BOND_ALB_INFO(bond));
4826 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4828 dev_net_set(bond_dev, net);
4829 bond_dev->rtnl_link_ops = &bond_link_ops;
4831 res = register_netdevice(bond_dev);
4833 free_netdev(bond_dev);
4839 netif_carrier_off(bond_dev);
4841 bond_work_init_all(bond);
4847 static int __net_init bond_net_init(struct net *net)
4849 struct bond_net *bn = net_generic(net, bond_net_id);
4852 INIT_LIST_HEAD(&bn->dev_list);
4854 bond_create_proc_dir(bn);
4855 bond_create_sysfs(bn);
4860 static void __net_exit bond_net_exit(struct net *net)
4862 struct bond_net *bn = net_generic(net, bond_net_id);
4863 struct bonding *bond, *tmp_bond;
4866 bond_destroy_sysfs(bn);
4868 /* Kill off any bonds created after unregistering bond rtnl ops */
4870 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4871 unregister_netdevice_queue(bond->dev, &list);
4872 unregister_netdevice_many(&list);
4875 bond_destroy_proc_dir(bn);
4878 static struct pernet_operations bond_net_ops = {
4879 .init = bond_net_init,
4880 .exit = bond_net_exit,
4882 .size = sizeof(struct bond_net),
4885 static int __init bonding_init(void)
4890 pr_info("%s", bond_version);
4892 res = bond_check_params(&bonding_defaults);
4896 res = register_pernet_subsys(&bond_net_ops);
4900 res = bond_netlink_init();
4904 bond_create_debugfs();
4906 for (i = 0; i < max_bonds; i++) {
4907 res = bond_create(&init_net, NULL);
4912 register_netdevice_notifier(&bond_netdev_notifier);
4916 bond_destroy_debugfs();
4917 bond_netlink_fini();
4919 unregister_pernet_subsys(&bond_net_ops);
4924 static void __exit bonding_exit(void)
4926 unregister_netdevice_notifier(&bond_netdev_notifier);
4928 bond_destroy_debugfs();
4930 bond_netlink_fini();
4931 unregister_pernet_subsys(&bond_net_ops);
4933 #ifdef CONFIG_NET_POLL_CONTROLLER
4934 /* Make sure we don't have an imbalance on our netpoll blocking */
4935 WARN_ON(atomic_read(&netpoll_block_tx));
4939 module_init(bonding_init);
4940 module_exit(bonding_exit);
4941 MODULE_LICENSE("GPL");
4942 MODULE_VERSION(DRV_VERSION);
4943 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4944 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");