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_keys.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
89 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
90 static int num_peer_notif = 1;
94 static int use_carrier = 1;
97 static char *primary_reselect;
98 static char *lacp_rate;
100 static char *ad_select;
101 static char *xmit_hash_policy;
102 static int arp_interval;
103 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
104 static char *arp_validate;
105 static char *arp_all_targets;
106 static char *fail_over_mac;
107 static int all_slaves_active;
108 static struct bond_params bonding_defaults;
109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
110 static int packets_per_slave = 1;
111 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
113 module_param(max_bonds, int, 0);
114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
115 module_param(tx_queues, int, 0);
116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
117 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
118 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
119 "failover event (alias of num_unsol_na)");
120 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
122 "failover event (alias of num_grat_arp)");
123 module_param(miimon, int, 0);
124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
125 module_param(updelay, int, 0);
126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
127 module_param(downdelay, int, 0);
128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
130 module_param(use_carrier, int, 0);
131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132 "0 for off, 1 for on (default)");
133 module_param(mode, charp, 0);
134 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
135 "1 for active-backup, 2 for balance-xor, "
136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137 "6 for balance-alb");
138 module_param(primary, charp, 0);
139 MODULE_PARM_DESC(primary, "Primary network device to use");
140 module_param(primary_reselect, charp, 0);
141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
143 "0 for always (default), "
144 "1 for only if speed of primary is "
146 "2 for only on active slave "
148 module_param(lacp_rate, charp, 0);
149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
150 "0 for slow, 1 for fast");
151 module_param(ad_select, charp, 0);
152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
153 "0 for stable (default), 1 for bandwidth, "
155 module_param(min_links, int, 0);
156 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
158 module_param(xmit_hash_policy, charp, 0);
159 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
160 "0 for layer 2 (default), 1 for layer 3+4, "
161 "2 for layer 2+3, 3 for encap layer 2+3, "
162 "4 for encap layer 3+4");
163 module_param(arp_interval, int, 0);
164 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
165 module_param_array(arp_ip_target, charp, NULL, 0);
166 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
167 module_param(arp_validate, charp, 0);
168 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
169 "0 for none (default), 1 for active, "
170 "2 for backup, 3 for all");
171 module_param(arp_all_targets, charp, 0);
172 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
173 module_param(fail_over_mac, charp, 0);
174 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
175 "the same MAC; 0 for none (default), "
176 "1 for active, 2 for follow");
177 module_param(all_slaves_active, int, 0);
178 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
179 "by setting active flag for all slaves; "
180 "0 for never (default), 1 for always.");
181 module_param(resend_igmp, int, 0);
182 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
184 module_param(packets_per_slave, int, 0);
185 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
186 "mode; 0 for a random slave, 1 packet per "
187 "slave (default), >1 packets per slave.");
188 module_param(lp_interval, uint, 0);
189 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
190 "the bonding driver sends learning packets to "
191 "each slaves peer switch. The default is 1.");
193 /*----------------------------- Global variables ----------------------------*/
195 #ifdef CONFIG_NET_POLL_CONTROLLER
196 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
199 int bond_net_id __read_mostly;
201 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
202 static int arp_ip_count;
203 static int bond_mode = BOND_MODE_ROUNDROBIN;
204 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
205 static int lacp_fast;
207 /*-------------------------- Forward declarations ---------------------------*/
209 static int bond_init(struct net_device *bond_dev);
210 static void bond_uninit(struct net_device *bond_dev);
211 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
212 struct rtnl_link_stats64 *stats);
213 static void bond_slave_arr_handler(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 static 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,
371 static void bond_update_speed_duplex(struct slave *slave)
373 struct net_device *slave_dev = slave->dev;
374 struct ethtool_cmd ecmd;
378 slave->speed = SPEED_UNKNOWN;
379 slave->duplex = DUPLEX_UNKNOWN;
381 res = __ethtool_get_settings(slave_dev, &ecmd);
385 slave_speed = ethtool_cmd_speed(&ecmd);
386 if (slave_speed == 0 || slave_speed == ((__u32) -1))
389 switch (ecmd.duplex) {
397 slave->speed = slave_speed;
398 slave->duplex = ecmd.duplex;
403 const char *bond_slave_link_status(s8 link)
419 /* if <dev> supports MII link status reporting, check its link status.
421 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
422 * depending upon the setting of the use_carrier parameter.
424 * Return either BMSR_LSTATUS, meaning that the link is up (or we
425 * can't tell and just pretend it is), or 0, meaning that the link is
428 * If reporting is non-zero, instead of faking link up, return -1 if
429 * both ETHTOOL and MII ioctls fail (meaning the device does not
430 * support them). If use_carrier is set, return whatever it says.
431 * It'd be nice if there was a good way to tell if a driver supports
432 * netif_carrier, but there really isn't.
434 static int bond_check_dev_link(struct bonding *bond,
435 struct net_device *slave_dev, int reporting)
437 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
438 int (*ioctl)(struct net_device *, struct ifreq *, int);
440 struct mii_ioctl_data *mii;
442 if (!reporting && !netif_running(slave_dev))
445 if (bond->params.use_carrier)
446 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
448 /* Try to get link status using Ethtool first. */
449 if (slave_dev->ethtool_ops->get_link)
450 return slave_dev->ethtool_ops->get_link(slave_dev) ?
453 /* Ethtool can't be used, fallback to MII ioctls. */
454 ioctl = slave_ops->ndo_do_ioctl;
456 /* TODO: set pointer to correct ioctl on a per team member
457 * bases to make this more efficient. that is, once
458 * we determine the correct ioctl, we will always
459 * call it and not the others for that team
463 /* We cannot assume that SIOCGMIIPHY will also read a
464 * register; not all network drivers (e.g., e100)
468 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
469 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
471 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
472 mii->reg_num = MII_BMSR;
473 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
474 return mii->val_out & BMSR_LSTATUS;
478 /* If reporting, report that either there's no dev->do_ioctl,
479 * or both SIOCGMIIREG and get_link failed (meaning that we
480 * cannot report link status). If not reporting, pretend
483 return reporting ? -1 : BMSR_LSTATUS;
486 /*----------------------------- Multicast list ------------------------------*/
488 /* Push the promiscuity flag down to appropriate slaves */
489 static int bond_set_promiscuity(struct bonding *bond, int inc)
491 struct list_head *iter;
494 if (bond_uses_primary(bond)) {
495 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
498 err = dev_set_promiscuity(curr_active->dev, inc);
502 bond_for_each_slave(bond, slave, iter) {
503 err = dev_set_promiscuity(slave->dev, inc);
511 /* Push the allmulti flag down to all slaves */
512 static int bond_set_allmulti(struct bonding *bond, int inc)
514 struct list_head *iter;
517 if (bond_uses_primary(bond)) {
518 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
521 err = dev_set_allmulti(curr_active->dev, inc);
525 bond_for_each_slave(bond, slave, iter) {
526 err = dev_set_allmulti(slave->dev, inc);
534 /* Retrieve the list of registered multicast addresses for the bonding
535 * device and retransmit an IGMP JOIN request to the current active
538 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
540 struct bonding *bond = container_of(work, struct bonding,
543 if (!rtnl_trylock()) {
544 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
547 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
549 if (bond->igmp_retrans > 1) {
550 bond->igmp_retrans--;
551 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
556 /* Flush bond's hardware addresses from slave */
557 static void bond_hw_addr_flush(struct net_device *bond_dev,
558 struct net_device *slave_dev)
560 struct bonding *bond = netdev_priv(bond_dev);
562 dev_uc_unsync(slave_dev, bond_dev);
563 dev_mc_unsync(slave_dev, bond_dev);
565 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
566 /* del lacpdu mc addr from mc list */
567 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
569 dev_mc_del(slave_dev, lacpdu_multicast);
573 /*--------------------------- Active slave change ---------------------------*/
575 /* Update the hardware address list and promisc/allmulti for the new and
576 * old active slaves (if any). Modes that are not using primary keep all
577 * slaves up date at all times; only the modes that use primary need to call
578 * this function to swap these settings during a failover.
580 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
581 struct slave *old_active)
584 if (bond->dev->flags & IFF_PROMISC)
585 dev_set_promiscuity(old_active->dev, -1);
587 if (bond->dev->flags & IFF_ALLMULTI)
588 dev_set_allmulti(old_active->dev, -1);
590 bond_hw_addr_flush(bond->dev, old_active->dev);
594 /* FIXME: Signal errors upstream. */
595 if (bond->dev->flags & IFF_PROMISC)
596 dev_set_promiscuity(new_active->dev, 1);
598 if (bond->dev->flags & IFF_ALLMULTI)
599 dev_set_allmulti(new_active->dev, 1);
601 netif_addr_lock_bh(bond->dev);
602 dev_uc_sync(new_active->dev, bond->dev);
603 dev_mc_sync(new_active->dev, bond->dev);
604 netif_addr_unlock_bh(bond->dev);
609 * bond_set_dev_addr - clone slave's address to bond
610 * @bond_dev: bond net device
611 * @slave_dev: slave net device
613 * Should be called with RTNL held.
615 static void bond_set_dev_addr(struct net_device *bond_dev,
616 struct net_device *slave_dev)
618 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
619 bond_dev, slave_dev, slave_dev->addr_len);
620 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
621 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
622 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
625 /* bond_do_fail_over_mac
627 * Perform special MAC address swapping for fail_over_mac settings
631 static void bond_do_fail_over_mac(struct bonding *bond,
632 struct slave *new_active,
633 struct slave *old_active)
635 u8 tmp_mac[ETH_ALEN];
636 struct sockaddr saddr;
639 switch (bond->params.fail_over_mac) {
640 case BOND_FOM_ACTIVE:
642 bond_set_dev_addr(bond->dev, new_active->dev);
644 case BOND_FOM_FOLLOW:
645 /* if new_active && old_active, swap them
646 * if just old_active, do nothing (going to no active slave)
647 * if just new_active, set new_active to bond's MAC
653 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
654 ether_addr_copy(saddr.sa_data,
655 old_active->dev->dev_addr);
656 saddr.sa_family = new_active->dev->type;
658 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
659 saddr.sa_family = bond->dev->type;
662 rv = dev_set_mac_address(new_active->dev, &saddr);
664 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
665 -rv, new_active->dev->name);
672 ether_addr_copy(saddr.sa_data, tmp_mac);
673 saddr.sa_family = old_active->dev->type;
675 rv = dev_set_mac_address(old_active->dev, &saddr);
677 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
678 -rv, new_active->dev->name);
682 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
683 bond->params.fail_over_mac);
689 static bool bond_should_change_active(struct bonding *bond)
691 struct slave *prim = rtnl_dereference(bond->primary_slave);
692 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
694 if (!prim || !curr || curr->link != BOND_LINK_UP)
696 if (bond->force_primary) {
697 bond->force_primary = false;
700 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
701 (prim->speed < curr->speed ||
702 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
704 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
710 * find_best_interface - select the best available slave to be the active one
711 * @bond: our bonding struct
713 static struct slave *bond_find_best_slave(struct bonding *bond)
715 struct slave *slave, *bestslave = NULL, *primary;
716 struct list_head *iter;
717 int mintime = bond->params.updelay;
719 primary = rtnl_dereference(bond->primary_slave);
720 if (primary && primary->link == BOND_LINK_UP &&
721 bond_should_change_active(bond))
724 bond_for_each_slave(bond, slave, iter) {
725 if (slave->link == BOND_LINK_UP)
727 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
728 slave->delay < mintime) {
729 mintime = slave->delay;
737 static bool bond_should_notify_peers(struct bonding *bond)
742 slave = rcu_dereference(bond->curr_active_slave);
745 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
746 slave ? slave->dev->name : "NULL");
748 if (!slave || !bond->send_peer_notif ||
749 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
756 * change_active_interface - change the active slave into the specified one
757 * @bond: our bonding struct
758 * @new: the new slave to make the active one
760 * Set the new slave to the bond's settings and unset them on the old
762 * Setting include flags, mc-list, promiscuity, allmulti, etc.
764 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
765 * because it is apparently the best available slave we have, even though its
766 * updelay hasn't timed out yet.
768 * Caller must hold RTNL.
770 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
772 struct slave *old_active;
776 old_active = rtnl_dereference(bond->curr_active_slave);
778 if (old_active == new_active)
782 new_active->last_link_up = jiffies;
784 if (new_active->link == BOND_LINK_BACK) {
785 if (bond_uses_primary(bond)) {
786 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
787 new_active->dev->name,
788 (bond->params.updelay - new_active->delay) * bond->params.miimon);
791 new_active->delay = 0;
792 new_active->link = BOND_LINK_UP;
794 if (BOND_MODE(bond) == BOND_MODE_8023AD)
795 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
797 if (bond_is_lb(bond))
798 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
800 if (bond_uses_primary(bond)) {
801 netdev_info(bond->dev, "making interface %s the new active one\n",
802 new_active->dev->name);
807 if (bond_uses_primary(bond))
808 bond_hw_addr_swap(bond, new_active, old_active);
810 if (bond_is_lb(bond)) {
811 bond_alb_handle_active_change(bond, new_active);
813 bond_set_slave_inactive_flags(old_active,
814 BOND_SLAVE_NOTIFY_NOW);
816 bond_set_slave_active_flags(new_active,
817 BOND_SLAVE_NOTIFY_NOW);
819 rcu_assign_pointer(bond->curr_active_slave, new_active);
822 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
824 bond_set_slave_inactive_flags(old_active,
825 BOND_SLAVE_NOTIFY_NOW);
828 bool should_notify_peers = false;
830 bond_set_slave_active_flags(new_active,
831 BOND_SLAVE_NOTIFY_NOW);
833 if (bond->params.fail_over_mac)
834 bond_do_fail_over_mac(bond, new_active,
837 if (netif_running(bond->dev)) {
838 bond->send_peer_notif =
839 bond->params.num_peer_notif;
840 should_notify_peers =
841 bond_should_notify_peers(bond);
844 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
845 if (should_notify_peers)
846 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
851 /* resend IGMP joins since active slave has changed or
852 * all were sent on curr_active_slave.
853 * resend only if bond is brought up with the affected
854 * bonding modes and the retransmission is enabled
856 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
857 ((bond_uses_primary(bond) && new_active) ||
858 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
859 bond->igmp_retrans = bond->params.resend_igmp;
860 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
865 * bond_select_active_slave - select a new active slave, if needed
866 * @bond: our bonding struct
868 * This functions should be called when one of the following occurs:
869 * - The old curr_active_slave has been released or lost its link.
870 * - The primary_slave has got its link back.
871 * - A slave has got its link back and there's no old curr_active_slave.
873 * Caller must hold RTNL.
875 void bond_select_active_slave(struct bonding *bond)
877 struct slave *best_slave;
882 best_slave = bond_find_best_slave(bond);
883 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
884 bond_change_active_slave(bond, best_slave);
885 rv = bond_set_carrier(bond);
889 if (netif_carrier_ok(bond->dev)) {
890 netdev_info(bond->dev, "first active interface up!\n");
892 netdev_info(bond->dev, "now running without any active interface!\n");
897 #ifdef CONFIG_NET_POLL_CONTROLLER
898 static inline int slave_enable_netpoll(struct slave *slave)
903 np = kzalloc(sizeof(*np), GFP_KERNEL);
908 err = __netpoll_setup(np, slave->dev);
917 static inline void slave_disable_netpoll(struct slave *slave)
919 struct netpoll *np = slave->np;
925 __netpoll_free_async(np);
928 static void bond_poll_controller(struct net_device *bond_dev)
932 static void bond_netpoll_cleanup(struct net_device *bond_dev)
934 struct bonding *bond = netdev_priv(bond_dev);
935 struct list_head *iter;
938 bond_for_each_slave(bond, slave, iter)
939 if (bond_slave_is_up(slave))
940 slave_disable_netpoll(slave);
943 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
945 struct bonding *bond = netdev_priv(dev);
946 struct list_head *iter;
950 bond_for_each_slave(bond, slave, iter) {
951 err = slave_enable_netpoll(slave);
953 bond_netpoll_cleanup(dev);
960 static inline int slave_enable_netpoll(struct slave *slave)
964 static inline void slave_disable_netpoll(struct slave *slave)
967 static void bond_netpoll_cleanup(struct net_device *bond_dev)
972 /*---------------------------------- IOCTL ----------------------------------*/
974 static netdev_features_t bond_fix_features(struct net_device *dev,
975 netdev_features_t features)
977 struct bonding *bond = netdev_priv(dev);
978 struct list_head *iter;
979 netdev_features_t mask;
983 features &= ~NETIF_F_ONE_FOR_ALL;
984 features |= NETIF_F_ALL_FOR_ALL;
986 bond_for_each_slave(bond, slave, iter) {
987 features = netdev_increment_features(features,
988 slave->dev->features,
991 features = netdev_add_tso_features(features, mask);
996 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
997 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
998 NETIF_F_HIGHDMA | NETIF_F_LRO)
1000 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1001 NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL)
1003 static void bond_compute_features(struct bonding *bond)
1005 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1006 IFF_XMIT_DST_RELEASE_PERM;
1007 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1008 netdev_features_t enc_features = BOND_ENC_FEATURES;
1009 struct net_device *bond_dev = bond->dev;
1010 struct list_head *iter;
1011 struct slave *slave;
1012 unsigned short max_hard_header_len = ETH_HLEN;
1013 unsigned int gso_max_size = GSO_MAX_SIZE;
1014 u16 gso_max_segs = GSO_MAX_SEGS;
1016 if (!bond_has_slaves(bond))
1018 vlan_features &= NETIF_F_ALL_FOR_ALL;
1020 bond_for_each_slave(bond, slave, iter) {
1021 vlan_features = netdev_increment_features(vlan_features,
1022 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1024 enc_features = netdev_increment_features(enc_features,
1025 slave->dev->hw_enc_features,
1027 dst_release_flag &= slave->dev->priv_flags;
1028 if (slave->dev->hard_header_len > max_hard_header_len)
1029 max_hard_header_len = slave->dev->hard_header_len;
1031 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1032 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1036 bond_dev->vlan_features = vlan_features;
1037 bond_dev->hw_enc_features = enc_features;
1038 bond_dev->hard_header_len = max_hard_header_len;
1039 bond_dev->gso_max_segs = gso_max_segs;
1040 netif_set_gso_max_size(bond_dev, gso_max_size);
1042 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1043 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1044 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1045 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1047 netdev_change_features(bond_dev);
1050 static void bond_setup_by_slave(struct net_device *bond_dev,
1051 struct net_device *slave_dev)
1053 bond_dev->header_ops = slave_dev->header_ops;
1055 bond_dev->type = slave_dev->type;
1056 bond_dev->hard_header_len = slave_dev->hard_header_len;
1057 bond_dev->addr_len = slave_dev->addr_len;
1059 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1060 slave_dev->addr_len);
1063 /* On bonding slaves other than the currently active slave, suppress
1064 * duplicates except for alb non-mcast/bcast.
1066 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1067 struct slave *slave,
1068 struct bonding *bond)
1070 if (bond_is_slave_inactive(slave)) {
1071 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1072 skb->pkt_type != PACKET_BROADCAST &&
1073 skb->pkt_type != PACKET_MULTICAST)
1080 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1082 struct sk_buff *skb = *pskb;
1083 struct slave *slave;
1084 struct bonding *bond;
1085 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1087 int ret = RX_HANDLER_ANOTHER;
1089 skb = skb_share_check(skb, GFP_ATOMIC);
1091 return RX_HANDLER_CONSUMED;
1095 slave = bond_slave_get_rcu(skb->dev);
1098 recv_probe = ACCESS_ONCE(bond->recv_probe);
1100 ret = recv_probe(skb, bond, slave);
1101 if (ret == RX_HANDLER_CONSUMED) {
1107 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1108 return RX_HANDLER_EXACT;
1111 skb->dev = bond->dev;
1113 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1114 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1115 skb->pkt_type == PACKET_HOST) {
1117 if (unlikely(skb_cow_head(skb,
1118 skb->data - skb_mac_header(skb)))) {
1120 return RX_HANDLER_CONSUMED;
1122 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1128 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1129 struct net_device *slave_dev,
1130 struct slave *slave)
1134 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1137 slave_dev->flags |= IFF_SLAVE;
1138 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1142 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1143 struct net_device *slave_dev)
1145 netdev_upper_dev_unlink(slave_dev, bond_dev);
1146 slave_dev->flags &= ~IFF_SLAVE;
1147 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1150 static struct slave *bond_alloc_slave(struct bonding *bond)
1152 struct slave *slave = NULL;
1154 slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1158 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1159 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1161 if (!SLAVE_AD_INFO(slave)) {
1169 static void bond_free_slave(struct slave *slave)
1171 struct bonding *bond = bond_get_bond_by_slave(slave);
1173 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1174 kfree(SLAVE_AD_INFO(slave));
1179 /* enslave device <slave> to bond device <master> */
1180 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1182 struct bonding *bond = netdev_priv(bond_dev);
1183 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1184 struct slave *new_slave = NULL, *prev_slave;
1185 struct sockaddr addr;
1189 if (!bond->params.use_carrier &&
1190 slave_dev->ethtool_ops->get_link == NULL &&
1191 slave_ops->ndo_do_ioctl == NULL) {
1192 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1196 /* already enslaved */
1197 if (slave_dev->flags & IFF_SLAVE) {
1198 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1202 if (bond_dev == slave_dev) {
1203 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1207 /* vlan challenged mutual exclusion */
1208 /* no need to lock since we're protected by rtnl_lock */
1209 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1210 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1212 if (vlan_uses_dev(bond_dev)) {
1213 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1214 slave_dev->name, bond_dev->name);
1217 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1218 slave_dev->name, slave_dev->name,
1222 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1226 /* Old ifenslave binaries are no longer supported. These can
1227 * be identified with moderate accuracy by the state of the slave:
1228 * the current ifenslave will set the interface down prior to
1229 * enslaving it; the old ifenslave will not.
1231 if ((slave_dev->flags & IFF_UP)) {
1232 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1235 goto err_undo_flags;
1238 /* set bonding device ether type by slave - bonding netdevices are
1239 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1240 * there is a need to override some of the type dependent attribs/funcs.
1242 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1243 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1245 if (!bond_has_slaves(bond)) {
1246 if (bond_dev->type != slave_dev->type) {
1247 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1248 bond_dev->type, slave_dev->type);
1250 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1252 res = notifier_to_errno(res);
1254 netdev_err(bond_dev, "refused to change device type\n");
1256 goto err_undo_flags;
1259 /* Flush unicast and multicast addresses */
1260 dev_uc_flush(bond_dev);
1261 dev_mc_flush(bond_dev);
1263 if (slave_dev->type != ARPHRD_ETHER)
1264 bond_setup_by_slave(bond_dev, slave_dev);
1266 ether_setup(bond_dev);
1267 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1270 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1273 } else if (bond_dev->type != slave_dev->type) {
1274 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1275 slave_dev->name, slave_dev->type, bond_dev->type);
1277 goto err_undo_flags;
1280 if (slave_ops->ndo_set_mac_address == NULL) {
1281 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1282 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1283 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1284 if (!bond_has_slaves(bond)) {
1285 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1286 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1288 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");
1290 goto err_undo_flags;
1295 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1297 /* If this is the first slave, then we need to set the master's hardware
1298 * address to be the same as the slave's.
1300 if (!bond_has_slaves(bond) &&
1301 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1302 bond_set_dev_addr(bond->dev, slave_dev);
1304 new_slave = bond_alloc_slave(bond);
1307 goto err_undo_flags;
1310 new_slave->bond = bond;
1311 new_slave->dev = slave_dev;
1312 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1313 * is set via sysfs or module option if desired.
1315 new_slave->queue_id = 0;
1317 /* Save slave's original mtu and then set it to match the bond */
1318 new_slave->original_mtu = slave_dev->mtu;
1319 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1321 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1325 /* Save slave's original ("permanent") mac address for modes
1326 * that need it, and for restoring it upon release, and then
1327 * set it to the master's address
1329 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1331 if (!bond->params.fail_over_mac ||
1332 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1333 /* Set slave to master's mac address. The application already
1334 * set the master's mac address to that of the first slave
1336 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1337 addr.sa_family = slave_dev->type;
1338 res = dev_set_mac_address(slave_dev, &addr);
1340 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1341 goto err_restore_mtu;
1345 /* open the slave since the application closed it */
1346 res = dev_open(slave_dev);
1348 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1349 goto err_restore_mac;
1352 slave_dev->priv_flags |= IFF_BONDING;
1353 /* initialize slave stats */
1354 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1356 if (bond_is_lb(bond)) {
1357 /* bond_alb_init_slave() must be called before all other stages since
1358 * it might fail and we do not want to have to undo everything
1360 res = bond_alb_init_slave(bond, new_slave);
1365 /* If the mode uses primary, then the following is handled by
1366 * bond_change_active_slave().
1368 if (!bond_uses_primary(bond)) {
1369 /* set promiscuity level to new slave */
1370 if (bond_dev->flags & IFF_PROMISC) {
1371 res = dev_set_promiscuity(slave_dev, 1);
1376 /* set allmulti level to new slave */
1377 if (bond_dev->flags & IFF_ALLMULTI) {
1378 res = dev_set_allmulti(slave_dev, 1);
1383 netif_addr_lock_bh(bond_dev);
1385 dev_mc_sync_multiple(slave_dev, bond_dev);
1386 dev_uc_sync_multiple(slave_dev, bond_dev);
1388 netif_addr_unlock_bh(bond_dev);
1391 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1392 /* add lacpdu mc addr to mc list */
1393 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1395 dev_mc_add(slave_dev, lacpdu_multicast);
1398 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1400 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1405 prev_slave = bond_last_slave(bond);
1407 new_slave->delay = 0;
1408 new_slave->link_failure_count = 0;
1410 bond_update_speed_duplex(new_slave);
1412 new_slave->last_rx = jiffies -
1413 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1414 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1415 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1417 if (bond->params.miimon && !bond->params.use_carrier) {
1418 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1420 if ((link_reporting == -1) && !bond->params.arp_interval) {
1421 /* miimon is set but a bonded network driver
1422 * does not support ETHTOOL/MII and
1423 * arp_interval is not set. Note: if
1424 * use_carrier is enabled, we will never go
1425 * here (because netif_carrier is always
1426 * supported); thus, we don't need to change
1427 * the messages for netif_carrier.
1429 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",
1431 } else if (link_reporting == -1) {
1432 /* unable get link status using mii/ethtool */
1433 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",
1438 /* check for initial state */
1439 if (bond->params.miimon) {
1440 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1441 if (bond->params.updelay) {
1442 new_slave->link = BOND_LINK_BACK;
1443 new_slave->delay = bond->params.updelay;
1445 new_slave->link = BOND_LINK_UP;
1448 new_slave->link = BOND_LINK_DOWN;
1450 } else if (bond->params.arp_interval) {
1451 new_slave->link = (netif_carrier_ok(slave_dev) ?
1452 BOND_LINK_UP : BOND_LINK_DOWN);
1454 new_slave->link = BOND_LINK_UP;
1457 if (new_slave->link != BOND_LINK_DOWN)
1458 new_slave->last_link_up = jiffies;
1459 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1460 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1461 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1463 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1464 /* if there is a primary slave, remember it */
1465 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1466 rcu_assign_pointer(bond->primary_slave, new_slave);
1467 bond->force_primary = true;
1471 switch (BOND_MODE(bond)) {
1472 case BOND_MODE_ACTIVEBACKUP:
1473 bond_set_slave_inactive_flags(new_slave,
1474 BOND_SLAVE_NOTIFY_NOW);
1476 case BOND_MODE_8023AD:
1477 /* in 802.3ad mode, the internal mechanism
1478 * will activate the slaves in the selected
1481 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1482 /* if this is the first slave */
1484 SLAVE_AD_INFO(new_slave)->id = 1;
1485 /* Initialize AD with the number of times that the AD timer is called in 1 second
1486 * can be called only after the mac address of the bond is set
1488 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1490 SLAVE_AD_INFO(new_slave)->id =
1491 SLAVE_AD_INFO(prev_slave)->id + 1;
1494 bond_3ad_bind_slave(new_slave);
1498 bond_set_active_slave(new_slave);
1499 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1502 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1504 /* always active in trunk mode */
1505 bond_set_active_slave(new_slave);
1507 /* In trunking mode there is little meaning to curr_active_slave
1508 * anyway (it holds no special properties of the bond device),
1509 * so we can change it without calling change_active_interface()
1511 if (!rcu_access_pointer(bond->curr_active_slave) &&
1512 new_slave->link == BOND_LINK_UP)
1513 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1516 } /* switch(bond_mode) */
1518 #ifdef CONFIG_NET_POLL_CONTROLLER
1519 slave_dev->npinfo = bond->dev->npinfo;
1520 if (slave_dev->npinfo) {
1521 if (slave_enable_netpoll(new_slave)) {
1522 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1529 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1532 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1536 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1538 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1539 goto err_unregister;
1542 res = bond_sysfs_slave_add(new_slave);
1544 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1545 goto err_upper_unlink;
1549 bond_compute_features(bond);
1550 bond_set_carrier(bond);
1552 if (bond_uses_primary(bond)) {
1554 bond_select_active_slave(bond);
1555 unblock_netpoll_tx();
1558 if (bond_mode_uses_xmit_hash(bond))
1559 bond_update_slave_arr(bond, NULL);
1561 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1563 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1564 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1566 /* enslave is successful */
1569 /* Undo stages on error */
1571 bond_upper_dev_unlink(bond_dev, slave_dev);
1574 netdev_rx_handler_unregister(slave_dev);
1577 if (!bond_uses_primary(bond))
1578 bond_hw_addr_flush(bond_dev, slave_dev);
1580 vlan_vids_del_by_dev(slave_dev, bond_dev);
1581 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1582 RCU_INIT_POINTER(bond->primary_slave, NULL);
1583 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1585 bond_change_active_slave(bond, NULL);
1586 bond_select_active_slave(bond);
1587 unblock_netpoll_tx();
1589 /* either primary_slave or curr_active_slave might've changed */
1591 slave_disable_netpoll(new_slave);
1594 slave_dev->priv_flags &= ~IFF_BONDING;
1595 dev_close(slave_dev);
1598 if (!bond->params.fail_over_mac ||
1599 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1600 /* XXX TODO - fom follow mode needs to change master's
1601 * MAC if this slave's MAC is in use by the bond, or at
1602 * least print a warning.
1604 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1605 addr.sa_family = slave_dev->type;
1606 dev_set_mac_address(slave_dev, &addr);
1610 dev_set_mtu(slave_dev, new_slave->original_mtu);
1613 bond_free_slave(new_slave);
1616 /* Enslave of first slave has failed and we need to fix master's mac */
1617 if (!bond_has_slaves(bond) &&
1618 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1619 eth_hw_addr_random(bond_dev);
1624 /* Try to release the slave device <slave> from the bond device <master>
1625 * It is legal to access curr_active_slave without a lock because all the function
1626 * is RTNL-locked. If "all" is true it means that the function is being called
1627 * while destroying a bond interface and all slaves are being released.
1629 * The rules for slave state should be:
1630 * for Active/Backup:
1631 * Active stays on all backups go down
1632 * for Bonded connections:
1633 * The first up interface should be left on and all others downed.
1635 static int __bond_release_one(struct net_device *bond_dev,
1636 struct net_device *slave_dev,
1639 struct bonding *bond = netdev_priv(bond_dev);
1640 struct slave *slave, *oldcurrent;
1641 struct sockaddr addr;
1642 int old_flags = bond_dev->flags;
1643 netdev_features_t old_features = bond_dev->features;
1645 /* slave is not a slave or master is not master of this slave */
1646 if (!(slave_dev->flags & IFF_SLAVE) ||
1647 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1648 netdev_err(bond_dev, "cannot release %s\n",
1655 slave = bond_get_slave_by_dev(bond, slave_dev);
1657 /* not a slave of this bond */
1658 netdev_info(bond_dev, "%s not enslaved\n",
1660 unblock_netpoll_tx();
1664 bond_sysfs_slave_del(slave);
1666 /* recompute stats just before removing the slave */
1667 bond_get_stats(bond->dev, &bond->bond_stats);
1669 bond_upper_dev_unlink(bond_dev, slave_dev);
1670 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1671 * for this slave anymore.
1673 netdev_rx_handler_unregister(slave_dev);
1675 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1676 bond_3ad_unbind_slave(slave);
1678 if (bond_mode_uses_xmit_hash(bond))
1679 bond_update_slave_arr(bond, slave);
1681 netdev_info(bond_dev, "Releasing %s interface %s\n",
1682 bond_is_active_slave(slave) ? "active" : "backup",
1685 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1687 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1689 if (!all && (!bond->params.fail_over_mac ||
1690 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1691 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1692 bond_has_slaves(bond))
1693 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",
1694 slave_dev->name, slave->perm_hwaddr,
1695 bond_dev->name, slave_dev->name);
1698 if (rtnl_dereference(bond->primary_slave) == slave)
1699 RCU_INIT_POINTER(bond->primary_slave, NULL);
1701 if (oldcurrent == slave)
1702 bond_change_active_slave(bond, NULL);
1704 if (bond_is_lb(bond)) {
1705 /* Must be called only after the slave has been
1706 * detached from the list and the curr_active_slave
1707 * has been cleared (if our_slave == old_current),
1708 * but before a new active slave is selected.
1710 bond_alb_deinit_slave(bond, slave);
1714 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1715 } else if (oldcurrent == slave) {
1716 /* Note that we hold RTNL over this sequence, so there
1717 * is no concern that another slave add/remove event
1720 bond_select_active_slave(bond);
1723 if (!bond_has_slaves(bond)) {
1724 bond_set_carrier(bond);
1725 eth_hw_addr_random(bond_dev);
1728 unblock_netpoll_tx();
1732 if (!bond_has_slaves(bond)) {
1733 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1734 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1737 bond_compute_features(bond);
1738 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1739 (old_features & NETIF_F_VLAN_CHALLENGED))
1740 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1741 slave_dev->name, bond_dev->name);
1743 vlan_vids_del_by_dev(slave_dev, bond_dev);
1745 /* If the mode uses primary, then this case was handled above by
1746 * bond_change_active_slave(..., NULL)
1748 if (!bond_uses_primary(bond)) {
1749 /* unset promiscuity level from slave
1750 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1751 * of the IFF_PROMISC flag in the bond_dev, but we need the
1752 * value of that flag before that change, as that was the value
1753 * when this slave was attached, so we cache at the start of the
1754 * function and use it here. Same goes for ALLMULTI below
1756 if (old_flags & IFF_PROMISC)
1757 dev_set_promiscuity(slave_dev, -1);
1759 /* unset allmulti level from slave */
1760 if (old_flags & IFF_ALLMULTI)
1761 dev_set_allmulti(slave_dev, -1);
1763 bond_hw_addr_flush(bond_dev, slave_dev);
1766 slave_disable_netpoll(slave);
1768 /* close slave before restoring its mac address */
1769 dev_close(slave_dev);
1771 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1772 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1773 /* restore original ("permanent") mac address */
1774 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1775 addr.sa_family = slave_dev->type;
1776 dev_set_mac_address(slave_dev, &addr);
1779 dev_set_mtu(slave_dev, slave->original_mtu);
1781 slave_dev->priv_flags &= ~IFF_BONDING;
1783 bond_free_slave(slave);
1788 /* A wrapper used because of ndo_del_link */
1789 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1791 return __bond_release_one(bond_dev, slave_dev, false);
1794 /* First release a slave and then destroy the bond if no more slaves are left.
1795 * Must be under rtnl_lock when this function is called.
1797 static int bond_release_and_destroy(struct net_device *bond_dev,
1798 struct net_device *slave_dev)
1800 struct bonding *bond = netdev_priv(bond_dev);
1803 ret = bond_release(bond_dev, slave_dev);
1804 if (ret == 0 && !bond_has_slaves(bond)) {
1805 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1806 netdev_info(bond_dev, "Destroying bond %s\n",
1808 unregister_netdevice(bond_dev);
1813 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1815 struct bonding *bond = netdev_priv(bond_dev);
1817 info->bond_mode = BOND_MODE(bond);
1818 info->miimon = bond->params.miimon;
1819 info->num_slaves = bond->slave_cnt;
1824 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1826 struct bonding *bond = netdev_priv(bond_dev);
1827 struct list_head *iter;
1828 int i = 0, res = -ENODEV;
1829 struct slave *slave;
1831 bond_for_each_slave(bond, slave, iter) {
1832 if (i++ == (int)info->slave_id) {
1834 strcpy(info->slave_name, slave->dev->name);
1835 info->link = slave->link;
1836 info->state = bond_slave_state(slave);
1837 info->link_failure_count = slave->link_failure_count;
1845 /*-------------------------------- Monitoring -------------------------------*/
1847 /* called with rcu_read_lock() */
1848 static int bond_miimon_inspect(struct bonding *bond)
1850 int link_state, commit = 0;
1851 struct list_head *iter;
1852 struct slave *slave;
1853 bool ignore_updelay;
1855 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1857 bond_for_each_slave_rcu(bond, slave, iter) {
1858 slave->new_link = BOND_LINK_NOCHANGE;
1860 link_state = bond_check_dev_link(bond, slave->dev, 0);
1862 switch (slave->link) {
1867 slave->link = BOND_LINK_FAIL;
1868 slave->delay = bond->params.downdelay;
1870 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1872 BOND_MODE_ACTIVEBACKUP) ?
1873 (bond_is_active_slave(slave) ?
1874 "active " : "backup ") : "",
1876 bond->params.downdelay * bond->params.miimon);
1879 case BOND_LINK_FAIL:
1881 /* recovered before downdelay expired */
1882 slave->link = BOND_LINK_UP;
1883 slave->last_link_up = jiffies;
1884 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1885 (bond->params.downdelay - slave->delay) *
1886 bond->params.miimon,
1891 if (slave->delay <= 0) {
1892 slave->new_link = BOND_LINK_DOWN;
1900 case BOND_LINK_DOWN:
1904 slave->link = BOND_LINK_BACK;
1905 slave->delay = bond->params.updelay;
1908 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1910 ignore_updelay ? 0 :
1911 bond->params.updelay *
1912 bond->params.miimon);
1915 case BOND_LINK_BACK:
1917 slave->link = BOND_LINK_DOWN;
1918 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
1919 (bond->params.updelay - slave->delay) *
1920 bond->params.miimon,
1929 if (slave->delay <= 0) {
1930 slave->new_link = BOND_LINK_UP;
1932 ignore_updelay = false;
1944 static void bond_miimon_commit(struct bonding *bond)
1946 struct list_head *iter;
1947 struct slave *slave, *primary;
1949 bond_for_each_slave(bond, slave, iter) {
1950 switch (slave->new_link) {
1951 case BOND_LINK_NOCHANGE:
1955 slave->link = BOND_LINK_UP;
1956 slave->last_link_up = jiffies;
1958 primary = rtnl_dereference(bond->primary_slave);
1959 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1960 /* prevent it from being the active one */
1961 bond_set_backup_slave(slave);
1962 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1963 /* make it immediately active */
1964 bond_set_active_slave(slave);
1965 } else if (slave != primary) {
1966 /* prevent it from being the active one */
1967 bond_set_backup_slave(slave);
1970 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
1972 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1973 slave->duplex ? "full" : "half");
1975 /* notify ad that the link status has changed */
1976 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1977 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
1979 if (bond_is_lb(bond))
1980 bond_alb_handle_link_change(bond, slave,
1983 if (BOND_MODE(bond) == BOND_MODE_XOR)
1984 bond_update_slave_arr(bond, NULL);
1986 if (!bond->curr_active_slave || slave == primary)
1991 case BOND_LINK_DOWN:
1992 if (slave->link_failure_count < UINT_MAX)
1993 slave->link_failure_count++;
1995 slave->link = BOND_LINK_DOWN;
1997 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
1998 BOND_MODE(bond) == BOND_MODE_8023AD)
1999 bond_set_slave_inactive_flags(slave,
2000 BOND_SLAVE_NOTIFY_NOW);
2002 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2005 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2006 bond_3ad_handle_link_change(slave,
2009 if (bond_is_lb(bond))
2010 bond_alb_handle_link_change(bond, slave,
2013 if (BOND_MODE(bond) == BOND_MODE_XOR)
2014 bond_update_slave_arr(bond, NULL);
2016 if (slave == rcu_access_pointer(bond->curr_active_slave))
2022 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2023 slave->new_link, slave->dev->name);
2024 slave->new_link = BOND_LINK_NOCHANGE;
2031 bond_select_active_slave(bond);
2032 unblock_netpoll_tx();
2035 bond_set_carrier(bond);
2040 * Really a wrapper that splits the mii monitor into two phases: an
2041 * inspection, then (if inspection indicates something needs to be done)
2042 * an acquisition of appropriate locks followed by a commit phase to
2043 * implement whatever link state changes are indicated.
2045 static void bond_mii_monitor(struct work_struct *work)
2047 struct bonding *bond = container_of(work, struct bonding,
2049 bool should_notify_peers = false;
2050 unsigned long delay;
2052 delay = msecs_to_jiffies(bond->params.miimon);
2054 if (!bond_has_slaves(bond))
2059 should_notify_peers = bond_should_notify_peers(bond);
2061 if (bond_miimon_inspect(bond)) {
2064 /* Race avoidance with bond_close cancel of workqueue */
2065 if (!rtnl_trylock()) {
2067 should_notify_peers = false;
2071 bond_miimon_commit(bond);
2073 rtnl_unlock(); /* might sleep, hold no other locks */
2078 if (bond->params.miimon)
2079 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2081 if (should_notify_peers) {
2082 if (!rtnl_trylock())
2084 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2089 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2091 struct net_device *upper;
2092 struct list_head *iter;
2095 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2099 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2100 if (ip == bond_confirm_addr(upper, 0, ip)) {
2110 /* We go to the (large) trouble of VLAN tagging ARP frames because
2111 * switches in VLAN mode (especially if ports are configured as
2112 * "native" to a VLAN) might not pass non-tagged frames.
2114 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2115 __be32 dest_ip, __be32 src_ip,
2116 struct bond_vlan_tag *tags)
2118 struct sk_buff *skb;
2119 struct bond_vlan_tag *outer_tag = tags;
2121 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2122 arp_op, slave_dev->name, &dest_ip, &src_ip);
2124 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2125 NULL, slave_dev->dev_addr, NULL);
2128 net_err_ratelimited("ARP packet allocation failed\n");
2132 if (!tags || tags->vlan_proto == VLAN_N_VID)
2137 /* Go through all the tags backwards and add them to the packet */
2138 while (tags->vlan_proto != VLAN_N_VID) {
2139 if (!tags->vlan_id) {
2144 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2145 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2146 skb = __vlan_put_tag(skb, tags->vlan_proto,
2149 net_err_ratelimited("failed to insert inner VLAN tag\n");
2155 /* Set the outer tag */
2156 if (outer_tag->vlan_id) {
2157 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2158 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2159 skb = vlan_put_tag(skb, outer_tag->vlan_proto,
2160 outer_tag->vlan_id);
2162 net_err_ratelimited("failed to insert outer VLAN tag\n");
2171 /* Validate the device path between the @start_dev and the @end_dev.
2172 * The path is valid if the @end_dev is reachable through device
2174 * When the path is validated, collect any vlan information in the
2177 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2178 struct net_device *end_dev,
2181 struct bond_vlan_tag *tags;
2182 struct net_device *upper;
2183 struct list_head *iter;
2185 if (start_dev == end_dev) {
2186 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2188 return ERR_PTR(-ENOMEM);
2189 tags[level].vlan_proto = VLAN_N_VID;
2193 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2194 tags = bond_verify_device_path(upper, end_dev, level + 1);
2195 if (IS_ERR_OR_NULL(tags)) {
2200 if (is_vlan_dev(upper)) {
2201 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2202 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2211 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2214 struct bond_vlan_tag *tags;
2215 __be32 *targets = bond->params.arp_targets, addr;
2218 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2219 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2222 /* Find out through which dev should the packet go */
2223 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2226 /* there's no route to target - try to send arp
2227 * probe to generate any traffic (arp_validate=0)
2229 if (bond->params.arp_validate)
2230 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2233 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2238 /* bond device itself */
2239 if (rt->dst.dev == bond->dev)
2243 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2246 if (!IS_ERR_OR_NULL(tags))
2249 /* Not our device - skip */
2250 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2251 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2257 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2259 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2265 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2269 if (!sip || !bond_has_this_ip(bond, tip)) {
2270 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2275 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2277 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2281 slave->last_rx = jiffies;
2282 slave->target_last_arp_rx[i] = jiffies;
2285 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2286 struct slave *slave)
2288 struct arphdr *arp = (struct arphdr *)skb->data;
2289 struct slave *curr_active_slave;
2290 unsigned char *arp_ptr;
2292 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2294 if (!slave_do_arp_validate(bond, slave)) {
2295 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2296 !slave_do_arp_validate_only(bond))
2297 slave->last_rx = jiffies;
2298 return RX_HANDLER_ANOTHER;
2299 } else if (!is_arp) {
2300 return RX_HANDLER_ANOTHER;
2303 alen = arp_hdr_len(bond->dev);
2305 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2308 if (alen > skb_headlen(skb)) {
2309 arp = kmalloc(alen, GFP_ATOMIC);
2312 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2316 if (arp->ar_hln != bond->dev->addr_len ||
2317 skb->pkt_type == PACKET_OTHERHOST ||
2318 skb->pkt_type == PACKET_LOOPBACK ||
2319 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2320 arp->ar_pro != htons(ETH_P_IP) ||
2324 arp_ptr = (unsigned char *)(arp + 1);
2325 arp_ptr += bond->dev->addr_len;
2326 memcpy(&sip, arp_ptr, 4);
2327 arp_ptr += 4 + bond->dev->addr_len;
2328 memcpy(&tip, arp_ptr, 4);
2330 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2331 slave->dev->name, bond_slave_state(slave),
2332 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2335 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2337 /* Backup slaves won't see the ARP reply, but do come through
2338 * here for each ARP probe (so we swap the sip/tip to validate
2339 * the probe). In a "redundant switch, common router" type of
2340 * configuration, the ARP probe will (hopefully) travel from
2341 * the active, through one switch, the router, then the other
2342 * switch before reaching the backup.
2344 * We 'trust' the arp requests if there is an active slave and
2345 * it received valid arp reply(s) after it became active. This
2346 * is done to avoid endless looping when we can't reach the
2347 * arp_ip_target and fool ourselves with our own arp requests.
2350 if (bond_is_active_slave(slave))
2351 bond_validate_arp(bond, slave, sip, tip);
2352 else if (curr_active_slave &&
2353 time_after(slave_last_rx(bond, curr_active_slave),
2354 curr_active_slave->last_link_up))
2355 bond_validate_arp(bond, slave, tip, sip);
2358 if (arp != (struct arphdr *)skb->data)
2360 return RX_HANDLER_ANOTHER;
2363 /* function to verify if we're in the arp_interval timeslice, returns true if
2364 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2365 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2367 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2370 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2372 return time_in_range(jiffies,
2373 last_act - delta_in_ticks,
2374 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2377 /* This function is called regularly to monitor each slave's link
2378 * ensuring that traffic is being sent and received when arp monitoring
2379 * is used in load-balancing mode. if the adapter has been dormant, then an
2380 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2381 * arp monitoring in active backup mode.
2383 static void bond_loadbalance_arp_mon(struct work_struct *work)
2385 struct bonding *bond = container_of(work, struct bonding,
2387 struct slave *slave, *oldcurrent;
2388 struct list_head *iter;
2389 int do_failover = 0, slave_state_changed = 0;
2391 if (!bond_has_slaves(bond))
2396 oldcurrent = rcu_dereference(bond->curr_active_slave);
2397 /* see if any of the previous devices are up now (i.e. they have
2398 * xmt and rcv traffic). the curr_active_slave does not come into
2399 * the picture unless it is null. also, slave->last_link_up is not
2400 * needed here because we send an arp on each slave and give a slave
2401 * as long as it needs to get the tx/rx within the delta.
2402 * TODO: what about up/down delay in arp mode? it wasn't here before
2405 bond_for_each_slave_rcu(bond, slave, iter) {
2406 unsigned long trans_start = dev_trans_start(slave->dev);
2408 if (slave->link != BOND_LINK_UP) {
2409 if (bond_time_in_interval(bond, trans_start, 1) &&
2410 bond_time_in_interval(bond, slave->last_rx, 1)) {
2412 slave->link = BOND_LINK_UP;
2413 slave_state_changed = 1;
2415 /* primary_slave has no meaning in round-robin
2416 * mode. the window of a slave being up and
2417 * curr_active_slave being null after enslaving
2421 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2425 netdev_info(bond->dev, "interface %s is now up\n",
2430 /* slave->link == BOND_LINK_UP */
2432 /* not all switches will respond to an arp request
2433 * when the source ip is 0, so don't take the link down
2434 * if we don't know our ip yet
2436 if (!bond_time_in_interval(bond, trans_start, 2) ||
2437 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2439 slave->link = BOND_LINK_DOWN;
2440 slave_state_changed = 1;
2442 if (slave->link_failure_count < UINT_MAX)
2443 slave->link_failure_count++;
2445 netdev_info(bond->dev, "interface %s is now down\n",
2448 if (slave == oldcurrent)
2453 /* note: if switch is in round-robin mode, all links
2454 * must tx arp to ensure all links rx an arp - otherwise
2455 * links may oscillate or not come up at all; if switch is
2456 * in something like xor mode, there is nothing we can
2457 * do - all replies will be rx'ed on same link causing slaves
2458 * to be unstable during low/no traffic periods
2460 if (bond_slave_is_up(slave))
2461 bond_arp_send_all(bond, slave);
2466 if (do_failover || slave_state_changed) {
2467 if (!rtnl_trylock())
2470 if (slave_state_changed) {
2471 bond_slave_state_change(bond);
2472 if (BOND_MODE(bond) == BOND_MODE_XOR)
2473 bond_update_slave_arr(bond, NULL);
2477 bond_select_active_slave(bond);
2478 unblock_netpoll_tx();
2484 if (bond->params.arp_interval)
2485 queue_delayed_work(bond->wq, &bond->arp_work,
2486 msecs_to_jiffies(bond->params.arp_interval));
2489 /* Called to inspect slaves for active-backup mode ARP monitor link state
2490 * changes. Sets new_link in slaves to specify what action should take
2491 * place for the slave. Returns 0 if no changes are found, >0 if changes
2492 * to link states must be committed.
2494 * Called with rcu_read_lock held.
2496 static int bond_ab_arp_inspect(struct bonding *bond)
2498 unsigned long trans_start, last_rx;
2499 struct list_head *iter;
2500 struct slave *slave;
2503 bond_for_each_slave_rcu(bond, slave, iter) {
2504 slave->new_link = BOND_LINK_NOCHANGE;
2505 last_rx = slave_last_rx(bond, slave);
2507 if (slave->link != BOND_LINK_UP) {
2508 if (bond_time_in_interval(bond, last_rx, 1)) {
2509 slave->new_link = BOND_LINK_UP;
2515 /* Give slaves 2*delta after being enslaved or made
2516 * active. This avoids bouncing, as the last receive
2517 * times need a full ARP monitor cycle to be updated.
2519 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2522 /* Backup slave is down if:
2523 * - No current_arp_slave AND
2524 * - more than 3*delta since last receive AND
2525 * - the bond has an IP address
2527 * Note: a non-null current_arp_slave indicates
2528 * the curr_active_slave went down and we are
2529 * searching for a new one; under this condition
2530 * we only take the curr_active_slave down - this
2531 * gives each slave a chance to tx/rx traffic
2532 * before being taken out
2534 if (!bond_is_active_slave(slave) &&
2535 !rcu_access_pointer(bond->current_arp_slave) &&
2536 !bond_time_in_interval(bond, last_rx, 3)) {
2537 slave->new_link = BOND_LINK_DOWN;
2541 /* Active slave is down if:
2542 * - more than 2*delta since transmitting OR
2543 * - (more than 2*delta since receive AND
2544 * the bond has an IP address)
2546 trans_start = dev_trans_start(slave->dev);
2547 if (bond_is_active_slave(slave) &&
2548 (!bond_time_in_interval(bond, trans_start, 2) ||
2549 !bond_time_in_interval(bond, last_rx, 2))) {
2550 slave->new_link = BOND_LINK_DOWN;
2558 /* Called to commit link state changes noted by inspection step of
2559 * active-backup mode ARP monitor.
2561 * Called with RTNL hold.
2563 static void bond_ab_arp_commit(struct bonding *bond)
2565 unsigned long trans_start;
2566 struct list_head *iter;
2567 struct slave *slave;
2569 bond_for_each_slave(bond, slave, iter) {
2570 switch (slave->new_link) {
2571 case BOND_LINK_NOCHANGE:
2575 trans_start = dev_trans_start(slave->dev);
2576 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2577 (!rtnl_dereference(bond->curr_active_slave) &&
2578 bond_time_in_interval(bond, trans_start, 1))) {
2579 struct slave *current_arp_slave;
2581 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2582 slave->link = BOND_LINK_UP;
2583 if (current_arp_slave) {
2584 bond_set_slave_inactive_flags(
2586 BOND_SLAVE_NOTIFY_NOW);
2587 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2590 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2593 if (!rtnl_dereference(bond->curr_active_slave) ||
2594 slave == rtnl_dereference(bond->primary_slave))
2601 case BOND_LINK_DOWN:
2602 if (slave->link_failure_count < UINT_MAX)
2603 slave->link_failure_count++;
2605 slave->link = BOND_LINK_DOWN;
2606 bond_set_slave_inactive_flags(slave,
2607 BOND_SLAVE_NOTIFY_NOW);
2609 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2612 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2613 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2620 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2621 slave->new_link, slave->dev->name);
2627 bond_select_active_slave(bond);
2628 unblock_netpoll_tx();
2631 bond_set_carrier(bond);
2634 /* Send ARP probes for active-backup mode ARP monitor.
2636 * Called with rcu_read_lock held.
2638 static bool bond_ab_arp_probe(struct bonding *bond)
2640 struct slave *slave, *before = NULL, *new_slave = NULL,
2641 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2642 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2643 struct list_head *iter;
2645 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2647 if (curr_arp_slave && curr_active_slave)
2648 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2649 curr_arp_slave->dev->name,
2650 curr_active_slave->dev->name);
2652 if (curr_active_slave) {
2653 bond_arp_send_all(bond, curr_active_slave);
2654 return should_notify_rtnl;
2657 /* if we don't have a curr_active_slave, search for the next available
2658 * backup slave from the current_arp_slave and make it the candidate
2659 * for becoming the curr_active_slave
2662 if (!curr_arp_slave) {
2663 curr_arp_slave = bond_first_slave_rcu(bond);
2664 if (!curr_arp_slave)
2665 return should_notify_rtnl;
2668 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2670 bond_for_each_slave_rcu(bond, slave, iter) {
2671 if (!found && !before && bond_slave_is_up(slave))
2674 if (found && !new_slave && bond_slave_is_up(slave))
2676 /* if the link state is up at this point, we
2677 * mark it down - this can happen if we have
2678 * simultaneous link failures and
2679 * reselect_active_interface doesn't make this
2680 * one the current slave so it is still marked
2681 * up when it is actually down
2683 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2684 slave->link = BOND_LINK_DOWN;
2685 if (slave->link_failure_count < UINT_MAX)
2686 slave->link_failure_count++;
2688 bond_set_slave_inactive_flags(slave,
2689 BOND_SLAVE_NOTIFY_LATER);
2691 netdev_info(bond->dev, "backup interface %s is now down\n",
2694 if (slave == curr_arp_slave)
2698 if (!new_slave && before)
2704 new_slave->link = BOND_LINK_BACK;
2705 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2706 bond_arp_send_all(bond, new_slave);
2707 new_slave->last_link_up = jiffies;
2708 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2711 bond_for_each_slave_rcu(bond, slave, iter) {
2712 if (slave->should_notify) {
2713 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2717 return should_notify_rtnl;
2720 static void bond_activebackup_arp_mon(struct work_struct *work)
2722 struct bonding *bond = container_of(work, struct bonding,
2724 bool should_notify_peers = false;
2725 bool should_notify_rtnl = false;
2728 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2730 if (!bond_has_slaves(bond))
2735 should_notify_peers = bond_should_notify_peers(bond);
2737 if (bond_ab_arp_inspect(bond)) {
2740 /* Race avoidance with bond_close flush of workqueue */
2741 if (!rtnl_trylock()) {
2743 should_notify_peers = false;
2747 bond_ab_arp_commit(bond);
2753 should_notify_rtnl = bond_ab_arp_probe(bond);
2757 if (bond->params.arp_interval)
2758 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2760 if (should_notify_peers || should_notify_rtnl) {
2761 if (!rtnl_trylock())
2764 if (should_notify_peers)
2765 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2767 if (should_notify_rtnl)
2768 bond_slave_state_notify(bond);
2774 /*-------------------------- netdev event handling --------------------------*/
2776 /* Change device name */
2777 static int bond_event_changename(struct bonding *bond)
2779 bond_remove_proc_entry(bond);
2780 bond_create_proc_entry(bond);
2782 bond_debug_reregister(bond);
2787 static int bond_master_netdev_event(unsigned long event,
2788 struct net_device *bond_dev)
2790 struct bonding *event_bond = netdev_priv(bond_dev);
2793 case NETDEV_CHANGENAME:
2794 return bond_event_changename(event_bond);
2795 case NETDEV_UNREGISTER:
2796 bond_remove_proc_entry(event_bond);
2798 case NETDEV_REGISTER:
2799 bond_create_proc_entry(event_bond);
2801 case NETDEV_NOTIFY_PEERS:
2802 if (event_bond->send_peer_notif)
2803 event_bond->send_peer_notif--;
2812 static int bond_slave_netdev_event(unsigned long event,
2813 struct net_device *slave_dev)
2815 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2816 struct bonding *bond;
2817 struct net_device *bond_dev;
2821 /* A netdev event can be generated while enslaving a device
2822 * before netdev_rx_handler_register is called in which case
2823 * slave will be NULL
2827 bond_dev = slave->bond->dev;
2829 primary = rtnl_dereference(bond->primary_slave);
2832 case NETDEV_UNREGISTER:
2833 if (bond_dev->type != ARPHRD_ETHER)
2834 bond_release_and_destroy(bond_dev, slave_dev);
2836 bond_release(bond_dev, slave_dev);
2840 old_speed = slave->speed;
2841 old_duplex = slave->duplex;
2843 bond_update_speed_duplex(slave);
2845 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2846 if (old_speed != slave->speed)
2847 bond_3ad_adapter_speed_changed(slave);
2848 if (old_duplex != slave->duplex)
2849 bond_3ad_adapter_duplex_changed(slave);
2851 /* Refresh slave-array if applicable!
2852 * If the setup does not use miimon or arpmon (mode-specific!),
2853 * then these events will not cause the slave-array to be
2854 * refreshed. This will cause xmit to use a slave that is not
2855 * usable. Avoid such situation by refeshing the array at these
2856 * events. If these (miimon/arpmon) parameters are configured
2857 * then array gets refreshed twice and that should be fine!
2859 if (bond_mode_uses_xmit_hash(bond))
2860 bond_update_slave_arr(bond, NULL);
2863 if (bond_mode_uses_xmit_hash(bond))
2864 bond_update_slave_arr(bond, NULL);
2866 case NETDEV_CHANGEMTU:
2867 /* TODO: Should slaves be allowed to
2868 * independently alter their MTU? For
2869 * an active-backup bond, slaves need
2870 * not be the same type of device, so
2871 * MTUs may vary. For other modes,
2872 * slaves arguably should have the
2873 * same MTUs. To do this, we'd need to
2874 * take over the slave's change_mtu
2875 * function for the duration of their
2879 case NETDEV_CHANGENAME:
2880 /* we don't care if we don't have primary set */
2881 if (!bond_uses_primary(bond) ||
2882 !bond->params.primary[0])
2885 if (slave == primary) {
2886 /* slave's name changed - he's no longer primary */
2887 RCU_INIT_POINTER(bond->primary_slave, NULL);
2888 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2889 /* we have a new primary slave */
2890 rcu_assign_pointer(bond->primary_slave, slave);
2891 } else { /* we didn't change primary - exit */
2895 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2896 primary ? slave_dev->name : "none");
2899 bond_select_active_slave(bond);
2900 unblock_netpoll_tx();
2902 case NETDEV_FEAT_CHANGE:
2903 bond_compute_features(bond);
2905 case NETDEV_RESEND_IGMP:
2906 /* Propagate to master device */
2907 call_netdevice_notifiers(event, slave->bond->dev);
2916 /* bond_netdev_event: handle netdev notifier chain events.
2918 * This function receives events for the netdev chain. The caller (an
2919 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2920 * locks for us to safely manipulate the slave devices (RTNL lock,
2923 static int bond_netdev_event(struct notifier_block *this,
2924 unsigned long event, void *ptr)
2926 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2928 netdev_dbg(event_dev, "event: %lx\n", event);
2930 if (!(event_dev->priv_flags & IFF_BONDING))
2933 if (event_dev->flags & IFF_MASTER) {
2934 netdev_dbg(event_dev, "IFF_MASTER\n");
2935 return bond_master_netdev_event(event, event_dev);
2938 if (event_dev->flags & IFF_SLAVE) {
2939 netdev_dbg(event_dev, "IFF_SLAVE\n");
2940 return bond_slave_netdev_event(event, event_dev);
2946 static struct notifier_block bond_netdev_notifier = {
2947 .notifier_call = bond_netdev_event,
2950 /*---------------------------- Hashing Policies -----------------------------*/
2952 /* L2 hash helper */
2953 static inline u32 bond_eth_hash(struct sk_buff *skb)
2955 struct ethhdr *ep, hdr_tmp;
2957 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
2959 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
2963 /* Extract the appropriate headers based on bond's xmit policy */
2964 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2965 struct flow_keys *fk)
2967 const struct ipv6hdr *iph6;
2968 const struct iphdr *iph;
2969 int noff, proto = -1;
2971 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2972 return skb_flow_dissect(skb, fk);
2975 noff = skb_network_offset(skb);
2976 if (skb->protocol == htons(ETH_P_IP)) {
2977 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
2980 fk->src = iph->saddr;
2981 fk->dst = iph->daddr;
2982 noff += iph->ihl << 2;
2983 if (!ip_is_fragment(iph))
2984 proto = iph->protocol;
2985 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2986 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
2988 iph6 = ipv6_hdr(skb);
2989 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2990 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2991 noff += sizeof(*iph6);
2992 proto = iph6->nexthdr;
2996 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2997 fk->ports = skb_flow_get_ports(skb, noff, proto);
3003 * bond_xmit_hash - generate a hash value based on the xmit policy
3004 * @bond: bonding device
3005 * @skb: buffer to use for headers
3007 * This function will extract the necessary headers from the skb buffer and use
3008 * them to generate a hash based on the xmit_policy set in the bonding device
3010 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3012 struct flow_keys flow;
3015 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3016 !bond_flow_dissect(bond, skb, &flow))
3017 return bond_eth_hash(skb);
3019 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3020 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3021 hash = bond_eth_hash(skb);
3023 hash = (__force u32)flow.ports;
3024 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3025 hash ^= (hash >> 16);
3026 hash ^= (hash >> 8);
3031 /*-------------------------- Device entry points ----------------------------*/
3033 static void bond_work_init_all(struct bonding *bond)
3035 INIT_DELAYED_WORK(&bond->mcast_work,
3036 bond_resend_igmp_join_requests_delayed);
3037 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3038 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3039 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3040 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3042 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3043 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3044 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3047 static void bond_work_cancel_all(struct bonding *bond)
3049 cancel_delayed_work_sync(&bond->mii_work);
3050 cancel_delayed_work_sync(&bond->arp_work);
3051 cancel_delayed_work_sync(&bond->alb_work);
3052 cancel_delayed_work_sync(&bond->ad_work);
3053 cancel_delayed_work_sync(&bond->mcast_work);
3054 cancel_delayed_work_sync(&bond->slave_arr_work);
3057 static int bond_open(struct net_device *bond_dev)
3059 struct bonding *bond = netdev_priv(bond_dev);
3060 struct list_head *iter;
3061 struct slave *slave;
3063 /* reset slave->backup and slave->inactive */
3064 if (bond_has_slaves(bond)) {
3065 bond_for_each_slave(bond, slave, iter) {
3066 if (bond_uses_primary(bond) &&
3067 slave != rcu_access_pointer(bond->curr_active_slave)) {
3068 bond_set_slave_inactive_flags(slave,
3069 BOND_SLAVE_NOTIFY_NOW);
3071 bond_set_slave_active_flags(slave,
3072 BOND_SLAVE_NOTIFY_NOW);
3077 bond_work_init_all(bond);
3079 if (bond_is_lb(bond)) {
3080 /* bond_alb_initialize must be called before the timer
3083 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3085 if (bond->params.tlb_dynamic_lb)
3086 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3089 if (bond->params.miimon) /* link check interval, in milliseconds. */
3090 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3092 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3093 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3094 bond->recv_probe = bond_arp_rcv;
3097 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3098 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3099 /* register to receive LACPDUs */
3100 bond->recv_probe = bond_3ad_lacpdu_recv;
3101 bond_3ad_initiate_agg_selection(bond, 1);
3104 if (bond_mode_uses_xmit_hash(bond))
3105 bond_update_slave_arr(bond, NULL);
3110 static int bond_close(struct net_device *bond_dev)
3112 struct bonding *bond = netdev_priv(bond_dev);
3114 bond_work_cancel_all(bond);
3115 bond->send_peer_notif = 0;
3116 if (bond_is_lb(bond))
3117 bond_alb_deinitialize(bond);
3118 bond->recv_probe = NULL;
3123 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3124 struct rtnl_link_stats64 *stats)
3126 struct bonding *bond = netdev_priv(bond_dev);
3127 struct rtnl_link_stats64 temp;
3128 struct list_head *iter;
3129 struct slave *slave;
3131 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3133 bond_for_each_slave(bond, slave, iter) {
3134 const struct rtnl_link_stats64 *sstats =
3135 dev_get_stats(slave->dev, &temp);
3136 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3138 stats->rx_packets += sstats->rx_packets - pstats->rx_packets;
3139 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3140 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3141 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3143 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3144 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3145 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3146 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3148 stats->multicast += sstats->multicast - pstats->multicast;
3149 stats->collisions += sstats->collisions - pstats->collisions;
3151 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3152 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3153 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3154 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3155 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3156 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3158 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3159 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3160 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3161 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3162 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3164 /* save off the slave stats for the next run */
3165 memcpy(pstats, sstats, sizeof(*sstats));
3167 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3172 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3174 struct bonding *bond = netdev_priv(bond_dev);
3175 struct net_device *slave_dev = NULL;
3176 struct ifbond k_binfo;
3177 struct ifbond __user *u_binfo = NULL;
3178 struct ifslave k_sinfo;
3179 struct ifslave __user *u_sinfo = NULL;
3180 struct mii_ioctl_data *mii = NULL;
3181 struct bond_opt_value newval;
3185 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3196 /* We do this again just in case we were called by SIOCGMIIREG
3197 * instead of SIOCGMIIPHY.
3203 if (mii->reg_num == 1) {
3205 if (netif_carrier_ok(bond->dev))
3206 mii->val_out = BMSR_LSTATUS;
3210 case BOND_INFO_QUERY_OLD:
3211 case SIOCBONDINFOQUERY:
3212 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3214 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3217 res = bond_info_query(bond_dev, &k_binfo);
3219 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3223 case BOND_SLAVE_INFO_QUERY_OLD:
3224 case SIOCBONDSLAVEINFOQUERY:
3225 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3227 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3230 res = bond_slave_info_query(bond_dev, &k_sinfo);
3232 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3240 net = dev_net(bond_dev);
3242 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3245 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3247 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3252 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3254 case BOND_ENSLAVE_OLD:
3255 case SIOCBONDENSLAVE:
3256 res = bond_enslave(bond_dev, slave_dev);
3258 case BOND_RELEASE_OLD:
3259 case SIOCBONDRELEASE:
3260 res = bond_release(bond_dev, slave_dev);
3262 case BOND_SETHWADDR_OLD:
3263 case SIOCBONDSETHWADDR:
3264 bond_set_dev_addr(bond_dev, slave_dev);
3267 case BOND_CHANGE_ACTIVE_OLD:
3268 case SIOCBONDCHANGEACTIVE:
3269 bond_opt_initstr(&newval, slave_dev->name);
3270 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3279 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3281 struct bonding *bond = netdev_priv(bond_dev);
3283 if (change & IFF_PROMISC)
3284 bond_set_promiscuity(bond,
3285 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3287 if (change & IFF_ALLMULTI)
3288 bond_set_allmulti(bond,
3289 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3292 static void bond_set_rx_mode(struct net_device *bond_dev)
3294 struct bonding *bond = netdev_priv(bond_dev);
3295 struct list_head *iter;
3296 struct slave *slave;
3299 if (bond_uses_primary(bond)) {
3300 slave = rcu_dereference(bond->curr_active_slave);
3302 dev_uc_sync(slave->dev, bond_dev);
3303 dev_mc_sync(slave->dev, bond_dev);
3306 bond_for_each_slave_rcu(bond, slave, iter) {
3307 dev_uc_sync_multiple(slave->dev, bond_dev);
3308 dev_mc_sync_multiple(slave->dev, bond_dev);
3314 static int bond_neigh_init(struct neighbour *n)
3316 struct bonding *bond = netdev_priv(n->dev);
3317 const struct net_device_ops *slave_ops;
3318 struct neigh_parms parms;
3319 struct slave *slave;
3322 slave = bond_first_slave(bond);
3325 slave_ops = slave->dev->netdev_ops;
3326 if (!slave_ops->ndo_neigh_setup)
3329 parms.neigh_setup = NULL;
3330 parms.neigh_cleanup = NULL;
3331 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3335 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3336 * after the last slave has been detached. Assumes that all slaves
3337 * utilize the same neigh_cleanup (true at this writing as only user
3340 n->parms->neigh_cleanup = parms.neigh_cleanup;
3342 if (!parms.neigh_setup)
3345 return parms.neigh_setup(n);
3348 /* The bonding ndo_neigh_setup is called at init time beofre any
3349 * slave exists. So we must declare proxy setup function which will
3350 * be used at run time to resolve the actual slave neigh param setup.
3352 * It's also called by master devices (such as vlans) to setup their
3353 * underlying devices. In that case - do nothing, we're already set up from
3356 static int bond_neigh_setup(struct net_device *dev,
3357 struct neigh_parms *parms)
3359 /* modify only our neigh_parms */
3360 if (parms->dev == dev)
3361 parms->neigh_setup = bond_neigh_init;
3366 /* Change the MTU of all of a master's slaves to match the master */
3367 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3369 struct bonding *bond = netdev_priv(bond_dev);
3370 struct slave *slave, *rollback_slave;
3371 struct list_head *iter;
3374 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3376 bond_for_each_slave(bond, slave, iter) {
3377 netdev_dbg(bond_dev, "s %p c_m %p\n",
3378 slave, slave->dev->netdev_ops->ndo_change_mtu);
3380 res = dev_set_mtu(slave->dev, new_mtu);
3383 /* If we failed to set the slave's mtu to the new value
3384 * we must abort the operation even in ACTIVE_BACKUP
3385 * mode, because if we allow the backup slaves to have
3386 * different mtu values than the active slave we'll
3387 * need to change their mtu when doing a failover. That
3388 * means changing their mtu from timer context, which
3389 * is probably not a good idea.
3391 netdev_dbg(bond_dev, "err %d %s\n", res,
3397 bond_dev->mtu = new_mtu;
3402 /* unwind from head to the slave that failed */
3403 bond_for_each_slave(bond, rollback_slave, iter) {
3406 if (rollback_slave == slave)
3409 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3411 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3412 tmp_res, rollback_slave->dev->name);
3419 /* Change HW address
3421 * Note that many devices must be down to change the HW address, and
3422 * downing the master releases all slaves. We can make bonds full of
3423 * bonding devices to test this, however.
3425 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3427 struct bonding *bond = netdev_priv(bond_dev);
3428 struct slave *slave, *rollback_slave;
3429 struct sockaddr *sa = addr, tmp_sa;
3430 struct list_head *iter;
3433 if (BOND_MODE(bond) == BOND_MODE_ALB)
3434 return bond_alb_set_mac_address(bond_dev, addr);
3437 netdev_dbg(bond_dev, "bond=%p\n", bond);
3439 /* If fail_over_mac is enabled, do nothing and return success.
3440 * Returning an error causes ifenslave to fail.
3442 if (bond->params.fail_over_mac &&
3443 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3446 if (!is_valid_ether_addr(sa->sa_data))
3447 return -EADDRNOTAVAIL;
3449 bond_for_each_slave(bond, slave, iter) {
3450 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3451 res = dev_set_mac_address(slave->dev, addr);
3453 /* TODO: consider downing the slave
3455 * User should expect communications
3456 * breakage anyway until ARP finish
3459 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3465 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3469 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3470 tmp_sa.sa_family = bond_dev->type;
3472 /* unwind from head to the slave that failed */
3473 bond_for_each_slave(bond, rollback_slave, iter) {
3476 if (rollback_slave == slave)
3479 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3481 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3482 tmp_res, rollback_slave->dev->name);
3490 * bond_xmit_slave_id - transmit skb through slave with slave_id
3491 * @bond: bonding device that is transmitting
3492 * @skb: buffer to transmit
3493 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3495 * This function tries to transmit through slave with slave_id but in case
3496 * it fails, it tries to find the first available slave for transmission.
3497 * The skb is consumed in all cases, thus the function is void.
3499 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3501 struct list_head *iter;
3502 struct slave *slave;
3505 /* Here we start from the slave with slave_id */
3506 bond_for_each_slave_rcu(bond, slave, iter) {
3508 if (bond_slave_can_tx(slave)) {
3509 bond_dev_queue_xmit(bond, skb, slave->dev);
3515 /* Here we start from the first slave up to slave_id */
3517 bond_for_each_slave_rcu(bond, slave, iter) {
3520 if (bond_slave_can_tx(slave)) {
3521 bond_dev_queue_xmit(bond, skb, slave->dev);
3525 /* no slave that can tx has been found */
3526 dev_kfree_skb_any(skb);
3530 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3531 * @bond: bonding device to use
3533 * Based on the value of the bonding device's packets_per_slave parameter
3534 * this function generates a slave id, which is usually used as the next
3535 * slave to transmit through.
3537 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3540 struct reciprocal_value reciprocal_packets_per_slave;
3541 int packets_per_slave = bond->params.packets_per_slave;
3543 switch (packets_per_slave) {
3545 slave_id = prandom_u32();
3548 slave_id = bond->rr_tx_counter;
3551 reciprocal_packets_per_slave =
3552 bond->params.reciprocal_packets_per_slave;
3553 slave_id = reciprocal_divide(bond->rr_tx_counter,
3554 reciprocal_packets_per_slave);
3557 bond->rr_tx_counter++;
3562 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3564 struct bonding *bond = netdev_priv(bond_dev);
3565 struct iphdr *iph = ip_hdr(skb);
3566 struct slave *slave;
3569 /* Start with the curr_active_slave that joined the bond as the
3570 * default for sending IGMP traffic. For failover purposes one
3571 * needs to maintain some consistency for the interface that will
3572 * send the join/membership reports. The curr_active_slave found
3573 * will send all of this type of traffic.
3575 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3576 slave = rcu_dereference(bond->curr_active_slave);
3578 bond_dev_queue_xmit(bond, skb, slave->dev);
3580 bond_xmit_slave_id(bond, skb, 0);
3582 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3584 if (likely(slave_cnt)) {
3585 slave_id = bond_rr_gen_slave_id(bond);
3586 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3588 dev_kfree_skb_any(skb);
3592 return NETDEV_TX_OK;
3595 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3596 * the bond has a usable interface.
3598 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3600 struct bonding *bond = netdev_priv(bond_dev);
3601 struct slave *slave;
3603 slave = rcu_dereference(bond->curr_active_slave);
3605 bond_dev_queue_xmit(bond, skb, slave->dev);
3607 dev_kfree_skb_any(skb);
3609 return NETDEV_TX_OK;
3612 /* Use this to update slave_array when (a) it's not appropriate to update
3613 * slave_array right away (note that update_slave_array() may sleep)
3614 * and / or (b) RTNL is not held.
3616 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3618 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3621 /* Slave array work handler. Holds only RTNL */
3622 static void bond_slave_arr_handler(struct work_struct *work)
3624 struct bonding *bond = container_of(work, struct bonding,
3625 slave_arr_work.work);
3628 if (!rtnl_trylock())
3631 ret = bond_update_slave_arr(bond, NULL);
3634 pr_warn_ratelimited("Failed to update slave array from WT\n");
3640 bond_slave_arr_work_rearm(bond, 1);
3643 /* Build the usable slaves array in control path for modes that use xmit-hash
3644 * to determine the slave interface -
3645 * (a) BOND_MODE_8023AD
3647 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3649 * The caller is expected to hold RTNL only and NO other lock!
3651 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3653 struct slave *slave;
3654 struct list_head *iter;
3655 struct bond_up_slave *new_arr, *old_arr;
3660 #ifdef CONFIG_LOCKDEP
3661 WARN_ON(lockdep_is_held(&bond->mode_lock));
3664 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3668 pr_err("Failed to build slave-array.\n");
3671 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3672 struct ad_info ad_info;
3674 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3675 pr_debug("bond_3ad_get_active_agg_info failed\n");
3676 kfree_rcu(new_arr, rcu);
3677 /* No active aggragator means it's not safe to use
3678 * the previous array.
3680 old_arr = rtnl_dereference(bond->slave_arr);
3682 RCU_INIT_POINTER(bond->slave_arr, NULL);
3683 kfree_rcu(old_arr, rcu);
3687 slaves_in_agg = ad_info.ports;
3688 agg_id = ad_info.aggregator_id;
3690 bond_for_each_slave(bond, slave, iter) {
3691 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3692 struct aggregator *agg;
3694 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3695 if (!agg || agg->aggregator_identifier != agg_id)
3698 if (!bond_slave_can_tx(slave))
3700 if (skipslave == slave)
3702 new_arr->arr[new_arr->count++] = slave;
3705 old_arr = rtnl_dereference(bond->slave_arr);
3706 rcu_assign_pointer(bond->slave_arr, new_arr);
3708 kfree_rcu(old_arr, rcu);
3710 if (ret != 0 && skipslave) {
3713 /* Rare situation where caller has asked to skip a specific
3714 * slave but allocation failed (most likely!). BTW this is
3715 * only possible when the call is initiated from
3716 * __bond_release_one(). In this situation; overwrite the
3717 * skipslave entry in the array with the last entry from the
3718 * array to avoid a situation where the xmit path may choose
3719 * this to-be-skipped slave to send a packet out.
3721 old_arr = rtnl_dereference(bond->slave_arr);
3722 for (idx = 0; idx < old_arr->count; idx++) {
3723 if (skipslave == old_arr->arr[idx]) {
3725 old_arr->arr[old_arr->count-1];
3734 /* Use this Xmit function for 3AD as well as XOR modes. The current
3735 * usable slave array is formed in the control path. The xmit function
3736 * just calculates hash and sends the packet out.
3738 int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3740 struct bonding *bond = netdev_priv(dev);
3741 struct slave *slave;
3742 struct bond_up_slave *slaves;
3745 slaves = rcu_dereference(bond->slave_arr);
3746 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3747 if (likely(count)) {
3748 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3749 bond_dev_queue_xmit(bond, skb, slave->dev);
3751 dev_kfree_skb_any(skb);
3752 atomic_long_inc(&dev->tx_dropped);
3755 return NETDEV_TX_OK;
3758 /* in broadcast mode, we send everything to all usable interfaces. */
3759 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3761 struct bonding *bond = netdev_priv(bond_dev);
3762 struct slave *slave = NULL;
3763 struct list_head *iter;
3765 bond_for_each_slave_rcu(bond, slave, iter) {
3766 if (bond_is_last_slave(bond, slave))
3768 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3769 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3772 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3773 bond_dev->name, __func__);
3776 bond_dev_queue_xmit(bond, skb2, slave->dev);
3779 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3780 bond_dev_queue_xmit(bond, skb, slave->dev);
3782 dev_kfree_skb_any(skb);
3784 return NETDEV_TX_OK;
3787 /*------------------------- Device initialization ---------------------------*/
3789 /* Lookup the slave that corresponds to a qid */
3790 static inline int bond_slave_override(struct bonding *bond,
3791 struct sk_buff *skb)
3793 struct slave *slave = NULL;
3794 struct list_head *iter;
3796 if (!skb->queue_mapping)
3799 /* Find out if any slaves have the same mapping as this skb. */
3800 bond_for_each_slave_rcu(bond, slave, iter) {
3801 if (slave->queue_id == skb->queue_mapping) {
3802 if (bond_slave_can_tx(slave)) {
3803 bond_dev_queue_xmit(bond, skb, slave->dev);
3806 /* If the slave isn't UP, use default transmit policy. */
3815 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3816 void *accel_priv, select_queue_fallback_t fallback)
3818 /* This helper function exists to help dev_pick_tx get the correct
3819 * destination queue. Using a helper function skips a call to
3820 * skb_tx_hash and will put the skbs in the queue we expect on their
3821 * way down to the bonding driver.
3823 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3825 /* Save the original txq to restore before passing to the driver */
3826 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3828 if (unlikely(txq >= dev->real_num_tx_queues)) {
3830 txq -= dev->real_num_tx_queues;
3831 } while (txq >= dev->real_num_tx_queues);
3836 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3838 struct bonding *bond = netdev_priv(dev);
3840 if (bond_should_override_tx_queue(bond) &&
3841 !bond_slave_override(bond, skb))
3842 return NETDEV_TX_OK;
3844 switch (BOND_MODE(bond)) {
3845 case BOND_MODE_ROUNDROBIN:
3846 return bond_xmit_roundrobin(skb, dev);
3847 case BOND_MODE_ACTIVEBACKUP:
3848 return bond_xmit_activebackup(skb, dev);
3849 case BOND_MODE_8023AD:
3851 return bond_3ad_xor_xmit(skb, dev);
3852 case BOND_MODE_BROADCAST:
3853 return bond_xmit_broadcast(skb, dev);
3855 return bond_alb_xmit(skb, dev);
3857 return bond_tlb_xmit(skb, dev);
3859 /* Should never happen, mode already checked */
3860 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3862 dev_kfree_skb_any(skb);
3863 return NETDEV_TX_OK;
3867 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3869 struct bonding *bond = netdev_priv(dev);
3870 netdev_tx_t ret = NETDEV_TX_OK;
3872 /* If we risk deadlock from transmitting this in the
3873 * netpoll path, tell netpoll to queue the frame for later tx
3875 if (unlikely(is_netpoll_tx_blocked(dev)))
3876 return NETDEV_TX_BUSY;
3879 if (bond_has_slaves(bond))
3880 ret = __bond_start_xmit(skb, dev);
3882 dev_kfree_skb_any(skb);
3888 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3889 struct ethtool_cmd *ecmd)
3891 struct bonding *bond = netdev_priv(bond_dev);
3892 unsigned long speed = 0;
3893 struct list_head *iter;
3894 struct slave *slave;
3896 ecmd->duplex = DUPLEX_UNKNOWN;
3897 ecmd->port = PORT_OTHER;
3899 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3900 * do not need to check mode. Though link speed might not represent
3901 * the true receive or transmit bandwidth (not all modes are symmetric)
3902 * this is an accurate maximum.
3904 bond_for_each_slave(bond, slave, iter) {
3905 if (bond_slave_can_tx(slave)) {
3906 if (slave->speed != SPEED_UNKNOWN)
3907 speed += slave->speed;
3908 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3909 slave->duplex != DUPLEX_UNKNOWN)
3910 ecmd->duplex = slave->duplex;
3913 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3918 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3919 struct ethtool_drvinfo *drvinfo)
3921 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3922 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3923 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3927 static const struct ethtool_ops bond_ethtool_ops = {
3928 .get_drvinfo = bond_ethtool_get_drvinfo,
3929 .get_settings = bond_ethtool_get_settings,
3930 .get_link = ethtool_op_get_link,
3933 static const struct net_device_ops bond_netdev_ops = {
3934 .ndo_init = bond_init,
3935 .ndo_uninit = bond_uninit,
3936 .ndo_open = bond_open,
3937 .ndo_stop = bond_close,
3938 .ndo_start_xmit = bond_start_xmit,
3939 .ndo_select_queue = bond_select_queue,
3940 .ndo_get_stats64 = bond_get_stats,
3941 .ndo_do_ioctl = bond_do_ioctl,
3942 .ndo_change_rx_flags = bond_change_rx_flags,
3943 .ndo_set_rx_mode = bond_set_rx_mode,
3944 .ndo_change_mtu = bond_change_mtu,
3945 .ndo_set_mac_address = bond_set_mac_address,
3946 .ndo_neigh_setup = bond_neigh_setup,
3947 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3948 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3949 #ifdef CONFIG_NET_POLL_CONTROLLER
3950 .ndo_netpoll_setup = bond_netpoll_setup,
3951 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3952 .ndo_poll_controller = bond_poll_controller,
3954 .ndo_add_slave = bond_enslave,
3955 .ndo_del_slave = bond_release,
3956 .ndo_fix_features = bond_fix_features,
3959 static const struct device_type bond_type = {
3963 static void bond_destructor(struct net_device *bond_dev)
3965 struct bonding *bond = netdev_priv(bond_dev);
3967 destroy_workqueue(bond->wq);
3968 free_netdev(bond_dev);
3971 void bond_setup(struct net_device *bond_dev)
3973 struct bonding *bond = netdev_priv(bond_dev);
3975 spin_lock_init(&bond->mode_lock);
3976 bond->params = bonding_defaults;
3978 /* Initialize pointers */
3979 bond->dev = bond_dev;
3981 /* Initialize the device entry points */
3982 ether_setup(bond_dev);
3983 bond_dev->netdev_ops = &bond_netdev_ops;
3984 bond_dev->ethtool_ops = &bond_ethtool_ops;
3986 bond_dev->destructor = bond_destructor;
3988 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3990 /* Initialize the device options */
3991 bond_dev->tx_queue_len = 0;
3992 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3993 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
3994 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3996 /* don't acquire bond device's netif_tx_lock when transmitting */
3997 bond_dev->features |= NETIF_F_LLTX;
3999 /* By default, we declare the bond to be fully
4000 * VLAN hardware accelerated capable. Special
4001 * care is taken in the various xmit functions
4002 * when there are slaves that are not hw accel
4006 /* Don't allow bond devices to change network namespaces. */
4007 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4009 bond_dev->hw_features = BOND_VLAN_FEATURES |
4010 NETIF_F_HW_VLAN_CTAG_TX |
4011 NETIF_F_HW_VLAN_CTAG_RX |
4012 NETIF_F_HW_VLAN_CTAG_FILTER;
4014 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4015 bond_dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
4016 bond_dev->features |= bond_dev->hw_features;
4019 /* Destroy a bonding device.
4020 * Must be under rtnl_lock when this function is called.
4022 static void bond_uninit(struct net_device *bond_dev)
4024 struct bonding *bond = netdev_priv(bond_dev);
4025 struct list_head *iter;
4026 struct slave *slave;
4027 struct bond_up_slave *arr;
4029 bond_netpoll_cleanup(bond_dev);
4031 /* Release the bonded slaves */
4032 bond_for_each_slave(bond, slave, iter)
4033 __bond_release_one(bond_dev, slave->dev, true);
4034 netdev_info(bond_dev, "Released all slaves\n");
4036 arr = rtnl_dereference(bond->slave_arr);
4038 RCU_INIT_POINTER(bond->slave_arr, NULL);
4039 kfree_rcu(arr, rcu);
4042 list_del(&bond->bond_list);
4044 bond_debug_unregister(bond);
4047 /*------------------------- Module initialization ---------------------------*/
4049 static int bond_check_params(struct bond_params *params)
4051 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4052 struct bond_opt_value newval;
4053 const struct bond_opt_value *valptr;
4054 int arp_all_targets_value;
4056 /* Convert string parameters. */
4058 bond_opt_initstr(&newval, mode);
4059 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4061 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4064 bond_mode = valptr->value;
4067 if (xmit_hash_policy) {
4068 if ((bond_mode != BOND_MODE_XOR) &&
4069 (bond_mode != BOND_MODE_8023AD) &&
4070 (bond_mode != BOND_MODE_TLB)) {
4071 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4072 bond_mode_name(bond_mode));
4074 bond_opt_initstr(&newval, xmit_hash_policy);
4075 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4078 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4082 xmit_hashtype = valptr->value;
4087 if (bond_mode != BOND_MODE_8023AD) {
4088 pr_info("lacp_rate param is irrelevant in mode %s\n",
4089 bond_mode_name(bond_mode));
4091 bond_opt_initstr(&newval, lacp_rate);
4092 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4095 pr_err("Error: Invalid lacp rate \"%s\"\n",
4099 lacp_fast = valptr->value;
4104 bond_opt_initstr(&newval, ad_select);
4105 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4108 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4111 params->ad_select = valptr->value;
4112 if (bond_mode != BOND_MODE_8023AD)
4113 pr_warn("ad_select param only affects 802.3ad mode\n");
4115 params->ad_select = BOND_AD_STABLE;
4118 if (max_bonds < 0) {
4119 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4120 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4121 max_bonds = BOND_DEFAULT_MAX_BONDS;
4125 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4131 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4136 if (downdelay < 0) {
4137 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4138 downdelay, INT_MAX);
4142 if ((use_carrier != 0) && (use_carrier != 1)) {
4143 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4148 if (num_peer_notif < 0 || num_peer_notif > 255) {
4149 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4154 /* reset values for 802.3ad/TLB/ALB */
4155 if (!bond_mode_uses_arp(bond_mode)) {
4157 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");
4158 pr_warn("Forcing miimon to 100msec\n");
4159 miimon = BOND_DEFAULT_MIIMON;
4163 if (tx_queues < 1 || tx_queues > 255) {
4164 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4165 tx_queues, BOND_DEFAULT_TX_QUEUES);
4166 tx_queues = BOND_DEFAULT_TX_QUEUES;
4169 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4170 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4172 all_slaves_active = 0;
4175 if (resend_igmp < 0 || resend_igmp > 255) {
4176 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4177 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4178 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4181 bond_opt_initval(&newval, packets_per_slave);
4182 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4183 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4184 packets_per_slave, USHRT_MAX);
4185 packets_per_slave = 1;
4188 if (bond_mode == BOND_MODE_ALB) {
4189 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",
4194 if (updelay || downdelay) {
4195 /* just warn the user the up/down delay will have
4196 * no effect since miimon is zero...
4198 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",
4199 updelay, downdelay);
4202 /* don't allow arp monitoring */
4204 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4205 miimon, arp_interval);
4209 if ((updelay % miimon) != 0) {
4210 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4211 updelay, miimon, (updelay / miimon) * miimon);
4216 if ((downdelay % miimon) != 0) {
4217 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4219 (downdelay / miimon) * miimon);
4222 downdelay /= miimon;
4225 if (arp_interval < 0) {
4226 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4227 arp_interval, INT_MAX);
4231 for (arp_ip_count = 0, i = 0;
4232 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4235 /* not a complete check, but good enough to catch mistakes */
4236 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4237 !bond_is_ip_target_ok(ip)) {
4238 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4242 if (bond_get_targets_ip(arp_target, ip) == -1)
4243 arp_target[arp_ip_count++] = ip;
4245 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4250 if (arp_interval && !arp_ip_count) {
4251 /* don't allow arping if no arp_ip_target given... */
4252 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4258 if (!arp_interval) {
4259 pr_err("arp_validate requires arp_interval\n");
4263 bond_opt_initstr(&newval, arp_validate);
4264 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4267 pr_err("Error: invalid arp_validate \"%s\"\n",
4271 arp_validate_value = valptr->value;
4273 arp_validate_value = 0;
4276 arp_all_targets_value = 0;
4277 if (arp_all_targets) {
4278 bond_opt_initstr(&newval, arp_all_targets);
4279 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4282 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4284 arp_all_targets_value = 0;
4286 arp_all_targets_value = valptr->value;
4291 pr_info("MII link monitoring set to %d ms\n", miimon);
4292 } else if (arp_interval) {
4293 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4294 arp_validate_value);
4295 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4296 arp_interval, valptr->string, arp_ip_count);
4298 for (i = 0; i < arp_ip_count; i++)
4299 pr_cont(" %s", arp_ip_target[i]);
4303 } else if (max_bonds) {
4304 /* miimon and arp_interval not set, we need one so things
4305 * work as expected, see bonding.txt for details
4307 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");
4310 if (primary && !bond_mode_uses_primary(bond_mode)) {
4311 /* currently, using a primary only makes sense
4312 * in active backup, TLB or ALB modes
4314 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4315 primary, bond_mode_name(bond_mode));
4319 if (primary && primary_reselect) {
4320 bond_opt_initstr(&newval, primary_reselect);
4321 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4324 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4328 primary_reselect_value = valptr->value;
4330 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4333 if (fail_over_mac) {
4334 bond_opt_initstr(&newval, fail_over_mac);
4335 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4338 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4342 fail_over_mac_value = valptr->value;
4343 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4344 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4346 fail_over_mac_value = BOND_FOM_NONE;
4349 if (lp_interval == 0) {
4350 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4351 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4352 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4355 /* fill params struct with the proper values */
4356 params->mode = bond_mode;
4357 params->xmit_policy = xmit_hashtype;
4358 params->miimon = miimon;
4359 params->num_peer_notif = num_peer_notif;
4360 params->arp_interval = arp_interval;
4361 params->arp_validate = arp_validate_value;
4362 params->arp_all_targets = arp_all_targets_value;
4363 params->updelay = updelay;
4364 params->downdelay = downdelay;
4365 params->use_carrier = use_carrier;
4366 params->lacp_fast = lacp_fast;
4367 params->primary[0] = 0;
4368 params->primary_reselect = primary_reselect_value;
4369 params->fail_over_mac = fail_over_mac_value;
4370 params->tx_queues = tx_queues;
4371 params->all_slaves_active = all_slaves_active;
4372 params->resend_igmp = resend_igmp;
4373 params->min_links = min_links;
4374 params->lp_interval = lp_interval;
4375 params->packets_per_slave = packets_per_slave;
4376 params->tlb_dynamic_lb = 1; /* Default value */
4377 if (packets_per_slave > 0) {
4378 params->reciprocal_packets_per_slave =
4379 reciprocal_value(packets_per_slave);
4381 /* reciprocal_packets_per_slave is unused if
4382 * packets_per_slave is 0 or 1, just initialize it
4384 params->reciprocal_packets_per_slave =
4385 (struct reciprocal_value) { 0 };
4389 strncpy(params->primary, primary, IFNAMSIZ);
4390 params->primary[IFNAMSIZ - 1] = 0;
4393 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4398 static struct lock_class_key bonding_netdev_xmit_lock_key;
4399 static struct lock_class_key bonding_netdev_addr_lock_key;
4400 static struct lock_class_key bonding_tx_busylock_key;
4402 static void bond_set_lockdep_class_one(struct net_device *dev,
4403 struct netdev_queue *txq,
4406 lockdep_set_class(&txq->_xmit_lock,
4407 &bonding_netdev_xmit_lock_key);
4410 static void bond_set_lockdep_class(struct net_device *dev)
4412 lockdep_set_class(&dev->addr_list_lock,
4413 &bonding_netdev_addr_lock_key);
4414 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4415 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4418 /* Called from registration process */
4419 static int bond_init(struct net_device *bond_dev)
4421 struct bonding *bond = netdev_priv(bond_dev);
4422 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4424 netdev_dbg(bond_dev, "Begin bond_init\n");
4426 bond->wq = create_singlethread_workqueue(bond_dev->name);
4430 bond_set_lockdep_class(bond_dev);
4432 list_add_tail(&bond->bond_list, &bn->dev_list);
4434 bond_prepare_sysfs_group(bond);
4436 bond_debug_register(bond);
4438 /* Ensure valid dev_addr */
4439 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4440 bond_dev->addr_assign_type == NET_ADDR_PERM)
4441 eth_hw_addr_random(bond_dev);
4446 unsigned int bond_get_num_tx_queues(void)
4451 /* Create a new bond based on the specified name and bonding parameters.
4452 * If name is NULL, obtain a suitable "bond%d" name for us.
4453 * Caller must NOT hold rtnl_lock; we need to release it here before we
4454 * set up our sysfs entries.
4456 int bond_create(struct net *net, const char *name)
4458 struct net_device *bond_dev;
4463 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4464 name ? name : "bond%d", NET_NAME_UNKNOWN,
4465 bond_setup, tx_queues);
4467 pr_err("%s: eek! can't alloc netdev!\n", name);
4472 dev_net_set(bond_dev, net);
4473 bond_dev->rtnl_link_ops = &bond_link_ops;
4475 res = register_netdevice(bond_dev);
4477 netif_carrier_off(bond_dev);
4481 bond_destructor(bond_dev);
4485 static int __net_init bond_net_init(struct net *net)
4487 struct bond_net *bn = net_generic(net, bond_net_id);
4490 INIT_LIST_HEAD(&bn->dev_list);
4492 bond_create_proc_dir(bn);
4493 bond_create_sysfs(bn);
4498 static void __net_exit bond_net_exit(struct net *net)
4500 struct bond_net *bn = net_generic(net, bond_net_id);
4501 struct bonding *bond, *tmp_bond;
4504 bond_destroy_sysfs(bn);
4506 /* Kill off any bonds created after unregistering bond rtnl ops */
4508 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4509 unregister_netdevice_queue(bond->dev, &list);
4510 unregister_netdevice_many(&list);
4513 bond_destroy_proc_dir(bn);
4516 static struct pernet_operations bond_net_ops = {
4517 .init = bond_net_init,
4518 .exit = bond_net_exit,
4520 .size = sizeof(struct bond_net),
4523 static int __init bonding_init(void)
4528 pr_info("%s", bond_version);
4530 res = bond_check_params(&bonding_defaults);
4534 res = register_pernet_subsys(&bond_net_ops);
4538 res = bond_netlink_init();
4542 bond_create_debugfs();
4544 for (i = 0; i < max_bonds; i++) {
4545 res = bond_create(&init_net, NULL);
4550 register_netdevice_notifier(&bond_netdev_notifier);
4554 bond_destroy_debugfs();
4555 bond_netlink_fini();
4557 unregister_pernet_subsys(&bond_net_ops);
4562 static void __exit bonding_exit(void)
4564 unregister_netdevice_notifier(&bond_netdev_notifier);
4566 bond_destroy_debugfs();
4568 bond_netlink_fini();
4569 unregister_pernet_subsys(&bond_net_ops);
4571 #ifdef CONFIG_NET_POLL_CONTROLLER
4572 /* Make sure we don't have an imbalance on our netpoll blocking */
4573 WARN_ON(atomic_read(&netpoll_block_tx));
4577 module_init(bonding_init);
4578 module_exit(bonding_exit);
4579 MODULE_LICENSE("GPL");
4580 MODULE_VERSION(DRV_VERSION);
4581 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4582 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");