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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.h>
82 #include <linux/reciprocal_div.h>
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 #define BOND_LINK_MON_INTERV 0
91 #define BOND_LINK_ARP_INTERV 0
93 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
94 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
95 static int num_peer_notif = 1;
96 static int miimon = BOND_LINK_MON_INTERV;
99 static int use_carrier = 1;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static int min_links;
105 static char *ad_select;
106 static char *xmit_hash_policy;
107 static int arp_interval = BOND_LINK_ARP_INTERV;
108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
109 static char *arp_validate;
110 static char *arp_all_targets;
111 static char *fail_over_mac;
112 static int all_slaves_active;
113 static struct bond_params bonding_defaults;
114 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
115 static int packets_per_slave = 1;
116 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
118 module_param(max_bonds, int, 0);
119 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
120 module_param(tx_queues, int, 0);
121 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
122 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
124 "failover event (alias of num_unsol_na)");
125 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
126 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
127 "failover event (alias of num_grat_arp)");
128 module_param(miimon, int, 0);
129 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
130 module_param(updelay, int, 0);
131 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
132 module_param(downdelay, int, 0);
133 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
135 module_param(use_carrier, int, 0);
136 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
137 "0 for off, 1 for on (default)");
138 module_param(mode, charp, 0);
139 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
140 "1 for active-backup, 2 for balance-xor, "
141 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
142 "6 for balance-alb");
143 module_param(primary, charp, 0);
144 MODULE_PARM_DESC(primary, "Primary network device to use");
145 module_param(primary_reselect, charp, 0);
146 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
148 "0 for always (default), "
149 "1 for only if speed of primary is "
151 "2 for only on active slave "
153 module_param(lacp_rate, charp, 0);
154 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
155 "0 for slow, 1 for fast");
156 module_param(ad_select, charp, 0);
157 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
158 "0 for stable (default), 1 for bandwidth, "
160 module_param(min_links, int, 0);
161 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
163 module_param(xmit_hash_policy, charp, 0);
164 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
165 "0 for layer 2 (default), 1 for layer 3+4, "
166 "2 for layer 2+3, 3 for encap layer 2+3, "
167 "4 for encap layer 3+4");
168 module_param(arp_interval, int, 0);
169 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
170 module_param_array(arp_ip_target, charp, NULL, 0);
171 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
172 module_param(arp_validate, charp, 0);
173 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
174 "0 for none (default), 1 for active, "
175 "2 for backup, 3 for all");
176 module_param(arp_all_targets, charp, 0);
177 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
178 module_param(fail_over_mac, charp, 0);
179 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
180 "the same MAC; 0 for none (default), "
181 "1 for active, 2 for follow");
182 module_param(all_slaves_active, int, 0);
183 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
184 "by setting active flag for all slaves; "
185 "0 for never (default), 1 for always.");
186 module_param(resend_igmp, int, 0);
187 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
189 module_param(packets_per_slave, int, 0);
190 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
191 "mode; 0 for a random slave, 1 packet per "
192 "slave (default), >1 packets per slave.");
193 module_param(lp_interval, uint, 0);
194 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
195 "the bonding driver sends learning packets to "
196 "each slaves peer switch. The default is 1.");
198 /*----------------------------- Global variables ----------------------------*/
200 #ifdef CONFIG_NET_POLL_CONTROLLER
201 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
204 int bond_net_id __read_mostly;
206 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
207 static int arp_ip_count;
208 static int bond_mode = BOND_MODE_ROUNDROBIN;
209 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
210 static int lacp_fast;
212 const struct bond_parm_tbl bond_lacp_tbl[] = {
213 { "slow", AD_LACP_SLOW},
214 { "fast", AD_LACP_FAST},
218 const struct bond_parm_tbl bond_mode_tbl[] = {
219 { "balance-rr", BOND_MODE_ROUNDROBIN},
220 { "active-backup", BOND_MODE_ACTIVEBACKUP},
221 { "balance-xor", BOND_MODE_XOR},
222 { "broadcast", BOND_MODE_BROADCAST},
223 { "802.3ad", BOND_MODE_8023AD},
224 { "balance-tlb", BOND_MODE_TLB},
225 { "balance-alb", BOND_MODE_ALB},
229 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
230 { "layer2", BOND_XMIT_POLICY_LAYER2},
231 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
232 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
233 { "encap2+3", BOND_XMIT_POLICY_ENCAP23},
234 { "encap3+4", BOND_XMIT_POLICY_ENCAP34},
238 const struct bond_parm_tbl arp_all_targets_tbl[] = {
239 { "any", BOND_ARP_TARGETS_ANY},
240 { "all", BOND_ARP_TARGETS_ALL},
244 const struct bond_parm_tbl arp_validate_tbl[] = {
245 { "none", BOND_ARP_VALIDATE_NONE},
246 { "active", BOND_ARP_VALIDATE_ACTIVE},
247 { "backup", BOND_ARP_VALIDATE_BACKUP},
248 { "all", BOND_ARP_VALIDATE_ALL},
252 const struct bond_parm_tbl fail_over_mac_tbl[] = {
253 { "none", BOND_FOM_NONE},
254 { "active", BOND_FOM_ACTIVE},
255 { "follow", BOND_FOM_FOLLOW},
259 const struct bond_parm_tbl pri_reselect_tbl[] = {
260 { "always", BOND_PRI_RESELECT_ALWAYS},
261 { "better", BOND_PRI_RESELECT_BETTER},
262 { "failure", BOND_PRI_RESELECT_FAILURE},
266 struct bond_parm_tbl ad_select_tbl[] = {
267 { "stable", BOND_AD_STABLE},
268 { "bandwidth", BOND_AD_BANDWIDTH},
269 { "count", BOND_AD_COUNT},
273 /*-------------------------- Forward declarations ---------------------------*/
275 static int bond_init(struct net_device *bond_dev);
276 static void bond_uninit(struct net_device *bond_dev);
278 /*---------------------------- General routines -----------------------------*/
280 const char *bond_mode_name(int mode)
282 static const char *names[] = {
283 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
284 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
285 [BOND_MODE_XOR] = "load balancing (xor)",
286 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
287 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
288 [BOND_MODE_TLB] = "transmit load balancing",
289 [BOND_MODE_ALB] = "adaptive load balancing",
292 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
298 /*---------------------------------- VLAN -----------------------------------*/
301 * bond_dev_queue_xmit - Prepare skb for xmit.
303 * @bond: bond device that got this skb for tx.
304 * @skb: hw accel VLAN tagged skb to transmit
305 * @slave_dev: slave that is supposed to xmit this skbuff
307 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
308 struct net_device *slave_dev)
310 skb->dev = slave_dev;
312 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
313 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
314 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
316 if (unlikely(netpoll_tx_running(bond->dev)))
317 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
323 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
324 * We don't protect the slave list iteration with a lock because:
325 * a. This operation is performed in IOCTL context,
326 * b. The operation is protected by the RTNL semaphore in the 8021q code,
327 * c. Holding a lock with BH disabled while directly calling a base driver
328 * entry point is generally a BAD idea.
330 * The design of synchronization/protection for this operation in the 8021q
331 * module is good for one or more VLAN devices over a single physical device
332 * and cannot be extended for a teaming solution like bonding, so there is a
333 * potential race condition here where a net device from the vlan group might
334 * be referenced (either by a base driver or the 8021q code) while it is being
335 * removed from the system. However, it turns out we're not making matters
336 * worse, and if it works for regular VLAN usage it will work here too.
340 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
341 * @bond_dev: bonding net device that got called
342 * @vid: vlan id being added
344 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
345 __be16 proto, u16 vid)
347 struct bonding *bond = netdev_priv(bond_dev);
348 struct slave *slave, *rollback_slave;
349 struct list_head *iter;
352 bond_for_each_slave(bond, slave, iter) {
353 res = vlan_vid_add(slave->dev, proto, vid);
361 /* unwind to the slave that failed */
362 bond_for_each_slave(bond, rollback_slave, iter) {
363 if (rollback_slave == slave)
366 vlan_vid_del(rollback_slave->dev, proto, vid);
373 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
374 * @bond_dev: bonding net device that got called
375 * @vid: vlan id being removed
377 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
378 __be16 proto, u16 vid)
380 struct bonding *bond = netdev_priv(bond_dev);
381 struct list_head *iter;
384 bond_for_each_slave(bond, slave, iter)
385 vlan_vid_del(slave->dev, proto, vid);
387 if (bond_is_lb(bond))
388 bond_alb_clear_vlan(bond, vid);
393 /*------------------------------- Link status -------------------------------*/
396 * Set the carrier state for the master according to the state of its
397 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
398 * do special 802.3ad magic.
400 * Returns zero if carrier state does not change, nonzero if it does.
402 static int bond_set_carrier(struct bonding *bond)
404 struct list_head *iter;
407 if (!bond_has_slaves(bond))
410 if (bond->params.mode == BOND_MODE_8023AD)
411 return bond_3ad_set_carrier(bond);
413 bond_for_each_slave(bond, slave, iter) {
414 if (slave->link == BOND_LINK_UP) {
415 if (!netif_carrier_ok(bond->dev)) {
416 netif_carrier_on(bond->dev);
424 if (netif_carrier_ok(bond->dev)) {
425 netif_carrier_off(bond->dev);
432 * Get link speed and duplex from the slave's base driver
433 * using ethtool. If for some reason the call fails or the
434 * values are invalid, set speed and duplex to -1,
437 static void bond_update_speed_duplex(struct slave *slave)
439 struct net_device *slave_dev = slave->dev;
440 struct ethtool_cmd ecmd;
444 slave->speed = SPEED_UNKNOWN;
445 slave->duplex = DUPLEX_UNKNOWN;
447 res = __ethtool_get_settings(slave_dev, &ecmd);
451 slave_speed = ethtool_cmd_speed(&ecmd);
452 if (slave_speed == 0 || slave_speed == ((__u32) -1))
455 switch (ecmd.duplex) {
463 slave->speed = slave_speed;
464 slave->duplex = ecmd.duplex;
470 * if <dev> supports MII link status reporting, check its link status.
472 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
473 * depending upon the setting of the use_carrier parameter.
475 * Return either BMSR_LSTATUS, meaning that the link is up (or we
476 * can't tell and just pretend it is), or 0, meaning that the link is
479 * If reporting is non-zero, instead of faking link up, return -1 if
480 * both ETHTOOL and MII ioctls fail (meaning the device does not
481 * support them). If use_carrier is set, return whatever it says.
482 * It'd be nice if there was a good way to tell if a driver supports
483 * netif_carrier, but there really isn't.
485 static int bond_check_dev_link(struct bonding *bond,
486 struct net_device *slave_dev, int reporting)
488 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
489 int (*ioctl)(struct net_device *, struct ifreq *, int);
491 struct mii_ioctl_data *mii;
493 if (!reporting && !netif_running(slave_dev))
496 if (bond->params.use_carrier)
497 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
499 /* Try to get link status using Ethtool first. */
500 if (slave_dev->ethtool_ops->get_link)
501 return slave_dev->ethtool_ops->get_link(slave_dev) ?
504 /* Ethtool can't be used, fallback to MII ioctls. */
505 ioctl = slave_ops->ndo_do_ioctl;
507 /* TODO: set pointer to correct ioctl on a per team member */
508 /* bases to make this more efficient. that is, once */
509 /* we determine the correct ioctl, we will always */
510 /* call it and not the others for that team */
514 * We cannot assume that SIOCGMIIPHY will also read a
515 * register; not all network drivers (e.g., e100)
519 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
520 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
522 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
523 mii->reg_num = MII_BMSR;
524 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
525 return mii->val_out & BMSR_LSTATUS;
530 * If reporting, report that either there's no dev->do_ioctl,
531 * or both SIOCGMIIREG and get_link failed (meaning that we
532 * cannot report link status). If not reporting, pretend
535 return reporting ? -1 : BMSR_LSTATUS;
538 /*----------------------------- Multicast list ------------------------------*/
541 * Push the promiscuity flag down to appropriate slaves
543 static int bond_set_promiscuity(struct bonding *bond, int inc)
545 struct list_head *iter;
548 if (USES_PRIMARY(bond->params.mode)) {
549 /* write lock already acquired */
550 if (bond->curr_active_slave) {
551 err = dev_set_promiscuity(bond->curr_active_slave->dev,
557 bond_for_each_slave(bond, slave, iter) {
558 err = dev_set_promiscuity(slave->dev, inc);
567 * Push the allmulti flag down to all slaves
569 static int bond_set_allmulti(struct bonding *bond, int inc)
571 struct list_head *iter;
574 if (USES_PRIMARY(bond->params.mode)) {
575 /* write lock already acquired */
576 if (bond->curr_active_slave) {
577 err = dev_set_allmulti(bond->curr_active_slave->dev,
583 bond_for_each_slave(bond, slave, iter) {
584 err = dev_set_allmulti(slave->dev, inc);
593 * Retrieve the list of registered multicast addresses for the bonding
594 * device and retransmit an IGMP JOIN request to the current active
597 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
599 struct bonding *bond = container_of(work, struct bonding,
602 if (!rtnl_trylock()) {
603 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
606 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
608 if (bond->igmp_retrans > 1) {
609 bond->igmp_retrans--;
610 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
615 /* Flush bond's hardware addresses from slave
617 static void bond_hw_addr_flush(struct net_device *bond_dev,
618 struct net_device *slave_dev)
620 struct bonding *bond = netdev_priv(bond_dev);
622 dev_uc_unsync(slave_dev, bond_dev);
623 dev_mc_unsync(slave_dev, bond_dev);
625 if (bond->params.mode == BOND_MODE_8023AD) {
626 /* del lacpdu mc addr from mc list */
627 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
629 dev_mc_del(slave_dev, lacpdu_multicast);
633 /*--------------------------- Active slave change ---------------------------*/
635 /* Update the hardware address list and promisc/allmulti for the new and
636 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
637 * slaves up date at all times; only the USES_PRIMARY modes need to call
638 * this function to swap these settings during a failover.
640 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
641 struct slave *old_active)
646 if (bond->dev->flags & IFF_PROMISC)
647 dev_set_promiscuity(old_active->dev, -1);
649 if (bond->dev->flags & IFF_ALLMULTI)
650 dev_set_allmulti(old_active->dev, -1);
652 bond_hw_addr_flush(bond->dev, old_active->dev);
656 /* FIXME: Signal errors upstream. */
657 if (bond->dev->flags & IFF_PROMISC)
658 dev_set_promiscuity(new_active->dev, 1);
660 if (bond->dev->flags & IFF_ALLMULTI)
661 dev_set_allmulti(new_active->dev, 1);
663 netif_addr_lock_bh(bond->dev);
664 dev_uc_sync(new_active->dev, bond->dev);
665 dev_mc_sync(new_active->dev, bond->dev);
666 netif_addr_unlock_bh(bond->dev);
671 * bond_set_dev_addr - clone slave's address to bond
672 * @bond_dev: bond net device
673 * @slave_dev: slave net device
675 * Should be called with RTNL held.
677 static void bond_set_dev_addr(struct net_device *bond_dev,
678 struct net_device *slave_dev)
680 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
681 bond_dev, slave_dev, slave_dev->addr_len);
682 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
683 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
684 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
688 * bond_do_fail_over_mac
690 * Perform special MAC address swapping for fail_over_mac settings
692 * Called with RTNL, curr_slave_lock for write_bh.
694 static void bond_do_fail_over_mac(struct bonding *bond,
695 struct slave *new_active,
696 struct slave *old_active)
697 __releases(&bond->curr_slave_lock)
698 __acquires(&bond->curr_slave_lock)
700 u8 tmp_mac[ETH_ALEN];
701 struct sockaddr saddr;
704 switch (bond->params.fail_over_mac) {
705 case BOND_FOM_ACTIVE:
707 write_unlock_bh(&bond->curr_slave_lock);
708 bond_set_dev_addr(bond->dev, new_active->dev);
709 write_lock_bh(&bond->curr_slave_lock);
712 case BOND_FOM_FOLLOW:
714 * if new_active && old_active, swap them
715 * if just old_active, do nothing (going to no active slave)
716 * if just new_active, set new_active to bond's MAC
721 write_unlock_bh(&bond->curr_slave_lock);
724 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
725 memcpy(saddr.sa_data, old_active->dev->dev_addr,
727 saddr.sa_family = new_active->dev->type;
729 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
730 saddr.sa_family = bond->dev->type;
733 rv = dev_set_mac_address(new_active->dev, &saddr);
735 pr_err("%s: Error %d setting MAC of slave %s\n",
736 bond->dev->name, -rv, new_active->dev->name);
743 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
744 saddr.sa_family = old_active->dev->type;
746 rv = dev_set_mac_address(old_active->dev, &saddr);
748 pr_err("%s: Error %d setting MAC of slave %s\n",
749 bond->dev->name, -rv, new_active->dev->name);
751 write_lock_bh(&bond->curr_slave_lock);
754 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
755 bond->dev->name, bond->params.fail_over_mac);
761 static bool bond_should_change_active(struct bonding *bond)
763 struct slave *prim = bond->primary_slave;
764 struct slave *curr = bond->curr_active_slave;
766 if (!prim || !curr || curr->link != BOND_LINK_UP)
768 if (bond->force_primary) {
769 bond->force_primary = false;
772 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
773 (prim->speed < curr->speed ||
774 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
776 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
782 * find_best_interface - select the best available slave to be the active one
783 * @bond: our bonding struct
785 static struct slave *bond_find_best_slave(struct bonding *bond)
787 struct slave *slave, *bestslave = NULL;
788 struct list_head *iter;
789 int mintime = bond->params.updelay;
791 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
792 bond_should_change_active(bond))
793 return bond->primary_slave;
795 bond_for_each_slave(bond, slave, iter) {
796 if (slave->link == BOND_LINK_UP)
798 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
799 slave->delay < mintime) {
800 mintime = slave->delay;
808 static bool bond_should_notify_peers(struct bonding *bond)
813 slave = rcu_dereference(bond->curr_active_slave);
816 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
817 bond->dev->name, slave ? slave->dev->name : "NULL");
819 if (!slave || !bond->send_peer_notif ||
820 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
827 * change_active_interface - change the active slave into the specified one
828 * @bond: our bonding struct
829 * @new: the new slave to make the active one
831 * Set the new slave to the bond's settings and unset them on the old
833 * Setting include flags, mc-list, promiscuity, allmulti, etc.
835 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
836 * because it is apparently the best available slave we have, even though its
837 * updelay hasn't timed out yet.
839 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
841 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
843 struct slave *old_active = bond->curr_active_slave;
845 if (old_active == new_active)
849 new_active->jiffies = jiffies;
851 if (new_active->link == BOND_LINK_BACK) {
852 if (USES_PRIMARY(bond->params.mode)) {
853 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
854 bond->dev->name, new_active->dev->name,
855 (bond->params.updelay - new_active->delay) * bond->params.miimon);
858 new_active->delay = 0;
859 new_active->link = BOND_LINK_UP;
861 if (bond->params.mode == BOND_MODE_8023AD)
862 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
864 if (bond_is_lb(bond))
865 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
867 if (USES_PRIMARY(bond->params.mode)) {
868 pr_info("%s: making interface %s the new active one.\n",
869 bond->dev->name, new_active->dev->name);
874 if (USES_PRIMARY(bond->params.mode))
875 bond_hw_addr_swap(bond, new_active, old_active);
877 if (bond_is_lb(bond)) {
878 bond_alb_handle_active_change(bond, new_active);
880 bond_set_slave_inactive_flags(old_active);
882 bond_set_slave_active_flags(new_active);
884 rcu_assign_pointer(bond->curr_active_slave, new_active);
887 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
889 bond_set_slave_inactive_flags(old_active);
892 bool should_notify_peers = false;
894 bond_set_slave_active_flags(new_active);
896 if (bond->params.fail_over_mac)
897 bond_do_fail_over_mac(bond, new_active,
900 if (netif_running(bond->dev)) {
901 bond->send_peer_notif =
902 bond->params.num_peer_notif;
903 should_notify_peers =
904 bond_should_notify_peers(bond);
907 write_unlock_bh(&bond->curr_slave_lock);
909 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
910 if (should_notify_peers)
911 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
914 write_lock_bh(&bond->curr_slave_lock);
918 /* resend IGMP joins since active slave has changed or
919 * all were sent on curr_active_slave.
920 * resend only if bond is brought up with the affected
921 * bonding modes and the retransmission is enabled */
922 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
923 ((USES_PRIMARY(bond->params.mode) && new_active) ||
924 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
925 bond->igmp_retrans = bond->params.resend_igmp;
926 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
931 * bond_select_active_slave - select a new active slave, if needed
932 * @bond: our bonding struct
934 * This functions should be called when one of the following occurs:
935 * - The old curr_active_slave has been released or lost its link.
936 * - The primary_slave has got its link back.
937 * - A slave has got its link back and there's no old curr_active_slave.
939 * Caller must hold curr_slave_lock for write_bh.
941 void bond_select_active_slave(struct bonding *bond)
943 struct slave *best_slave;
946 best_slave = bond_find_best_slave(bond);
947 if (best_slave != bond->curr_active_slave) {
948 bond_change_active_slave(bond, best_slave);
949 rv = bond_set_carrier(bond);
953 if (netif_carrier_ok(bond->dev)) {
954 pr_info("%s: first active interface up!\n",
957 pr_info("%s: now running without any active interface !\n",
963 #ifdef CONFIG_NET_POLL_CONTROLLER
964 static inline int slave_enable_netpoll(struct slave *slave)
969 np = kzalloc(sizeof(*np), GFP_ATOMIC);
974 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
983 static inline void slave_disable_netpoll(struct slave *slave)
985 struct netpoll *np = slave->np;
991 __netpoll_free_async(np);
993 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
995 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
997 if (!slave_dev->netdev_ops->ndo_poll_controller)
1002 static void bond_poll_controller(struct net_device *bond_dev)
1006 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1008 struct bonding *bond = netdev_priv(bond_dev);
1009 struct list_head *iter;
1010 struct slave *slave;
1012 bond_for_each_slave(bond, slave, iter)
1013 if (IS_UP(slave->dev))
1014 slave_disable_netpoll(slave);
1017 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1019 struct bonding *bond = netdev_priv(dev);
1020 struct list_head *iter;
1021 struct slave *slave;
1024 bond_for_each_slave(bond, slave, iter) {
1025 err = slave_enable_netpoll(slave);
1027 bond_netpoll_cleanup(dev);
1034 static inline int slave_enable_netpoll(struct slave *slave)
1038 static inline void slave_disable_netpoll(struct slave *slave)
1041 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1046 /*---------------------------------- IOCTL ----------------------------------*/
1048 static netdev_features_t bond_fix_features(struct net_device *dev,
1049 netdev_features_t features)
1051 struct bonding *bond = netdev_priv(dev);
1052 struct list_head *iter;
1053 netdev_features_t mask;
1054 struct slave *slave;
1056 if (!bond_has_slaves(bond)) {
1057 /* Disable adding VLANs to empty bond. But why? --mq */
1058 features |= NETIF_F_VLAN_CHALLENGED;
1063 features &= ~NETIF_F_ONE_FOR_ALL;
1064 features |= NETIF_F_ALL_FOR_ALL;
1066 bond_for_each_slave(bond, slave, iter) {
1067 features = netdev_increment_features(features,
1068 slave->dev->features,
1071 features = netdev_add_tso_features(features, mask);
1076 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1077 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1078 NETIF_F_HIGHDMA | NETIF_F_LRO)
1080 static void bond_compute_features(struct bonding *bond)
1082 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1083 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1084 struct net_device *bond_dev = bond->dev;
1085 struct list_head *iter;
1086 struct slave *slave;
1087 unsigned short max_hard_header_len = ETH_HLEN;
1088 unsigned int gso_max_size = GSO_MAX_SIZE;
1089 u16 gso_max_segs = GSO_MAX_SEGS;
1091 if (!bond_has_slaves(bond))
1094 bond_for_each_slave(bond, slave, iter) {
1095 vlan_features = netdev_increment_features(vlan_features,
1096 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1098 dst_release_flag &= slave->dev->priv_flags;
1099 if (slave->dev->hard_header_len > max_hard_header_len)
1100 max_hard_header_len = slave->dev->hard_header_len;
1102 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1103 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1107 bond_dev->vlan_features = vlan_features;
1108 bond_dev->hard_header_len = max_hard_header_len;
1109 bond_dev->gso_max_segs = gso_max_segs;
1110 netif_set_gso_max_size(bond_dev, gso_max_size);
1112 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1113 bond_dev->priv_flags = flags | dst_release_flag;
1115 netdev_change_features(bond_dev);
1118 static void bond_setup_by_slave(struct net_device *bond_dev,
1119 struct net_device *slave_dev)
1121 bond_dev->header_ops = slave_dev->header_ops;
1123 bond_dev->type = slave_dev->type;
1124 bond_dev->hard_header_len = slave_dev->hard_header_len;
1125 bond_dev->addr_len = slave_dev->addr_len;
1127 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1128 slave_dev->addr_len);
1131 /* On bonding slaves other than the currently active slave, suppress
1132 * duplicates except for alb non-mcast/bcast.
1134 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1135 struct slave *slave,
1136 struct bonding *bond)
1138 if (bond_is_slave_inactive(slave)) {
1139 if (bond->params.mode == BOND_MODE_ALB &&
1140 skb->pkt_type != PACKET_BROADCAST &&
1141 skb->pkt_type != PACKET_MULTICAST)
1148 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1150 struct sk_buff *skb = *pskb;
1151 struct slave *slave;
1152 struct bonding *bond;
1153 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1155 int ret = RX_HANDLER_ANOTHER;
1157 skb = skb_share_check(skb, GFP_ATOMIC);
1159 return RX_HANDLER_CONSUMED;
1163 slave = bond_slave_get_rcu(skb->dev);
1166 if (bond->params.arp_interval)
1167 slave->dev->last_rx = jiffies;
1169 recv_probe = ACCESS_ONCE(bond->recv_probe);
1171 ret = recv_probe(skb, bond, slave);
1172 if (ret == RX_HANDLER_CONSUMED) {
1178 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1179 return RX_HANDLER_EXACT;
1182 skb->dev = bond->dev;
1184 if (bond->params.mode == BOND_MODE_ALB &&
1185 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1186 skb->pkt_type == PACKET_HOST) {
1188 if (unlikely(skb_cow_head(skb,
1189 skb->data - skb_mac_header(skb)))) {
1191 return RX_HANDLER_CONSUMED;
1193 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1199 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1200 struct net_device *slave_dev,
1201 struct slave *slave)
1205 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1208 slave_dev->flags |= IFF_SLAVE;
1209 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1213 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1214 struct net_device *slave_dev)
1216 netdev_upper_dev_unlink(slave_dev, bond_dev);
1217 slave_dev->flags &= ~IFF_SLAVE;
1218 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1221 /* enslave device <slave> to bond device <master> */
1222 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1224 struct bonding *bond = netdev_priv(bond_dev);
1225 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1226 struct slave *new_slave = NULL, *prev_slave;
1227 struct sockaddr addr;
1231 if (!bond->params.use_carrier &&
1232 slave_dev->ethtool_ops->get_link == NULL &&
1233 slave_ops->ndo_do_ioctl == NULL) {
1234 pr_warning("%s: Warning: no link monitoring support for %s\n",
1235 bond_dev->name, slave_dev->name);
1238 /* already enslaved */
1239 if (slave_dev->flags & IFF_SLAVE) {
1240 pr_debug("Error, Device was already enslaved\n");
1244 /* vlan challenged mutual exclusion */
1245 /* no need to lock since we're protected by rtnl_lock */
1246 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1247 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1248 if (vlan_uses_dev(bond_dev)) {
1249 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1250 bond_dev->name, slave_dev->name, bond_dev->name);
1253 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1254 bond_dev->name, slave_dev->name,
1255 slave_dev->name, bond_dev->name);
1258 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1262 * Old ifenslave binaries are no longer supported. These can
1263 * be identified with moderate accuracy by the state of the slave:
1264 * the current ifenslave will set the interface down prior to
1265 * enslaving it; the old ifenslave will not.
1267 if ((slave_dev->flags & IFF_UP)) {
1268 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1271 goto err_undo_flags;
1274 /* set bonding device ether type by slave - bonding netdevices are
1275 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1276 * there is a need to override some of the type dependent attribs/funcs.
1278 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1279 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1281 if (!bond_has_slaves(bond)) {
1282 if (bond_dev->type != slave_dev->type) {
1283 pr_debug("%s: change device type from %d to %d\n",
1285 bond_dev->type, slave_dev->type);
1287 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1289 res = notifier_to_errno(res);
1291 pr_err("%s: refused to change device type\n",
1294 goto err_undo_flags;
1297 /* Flush unicast and multicast addresses */
1298 dev_uc_flush(bond_dev);
1299 dev_mc_flush(bond_dev);
1301 if (slave_dev->type != ARPHRD_ETHER)
1302 bond_setup_by_slave(bond_dev, slave_dev);
1304 ether_setup(bond_dev);
1305 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1308 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1311 } else if (bond_dev->type != slave_dev->type) {
1312 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1314 slave_dev->type, bond_dev->type);
1316 goto err_undo_flags;
1319 if (slave_ops->ndo_set_mac_address == NULL) {
1320 if (!bond_has_slaves(bond)) {
1321 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1323 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1324 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1325 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1328 goto err_undo_flags;
1332 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1334 /* If this is the first slave, then we need to set the master's hardware
1335 * address to be the same as the slave's. */
1336 if (!bond_has_slaves(bond) &&
1337 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1338 bond_set_dev_addr(bond->dev, slave_dev);
1340 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1343 goto err_undo_flags;
1346 * Set the new_slave's queue_id to be zero. Queue ID mapping
1347 * is set via sysfs or module option if desired.
1349 new_slave->queue_id = 0;
1351 /* Save slave's original mtu and then set it to match the bond */
1352 new_slave->original_mtu = slave_dev->mtu;
1353 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1355 pr_debug("Error %d calling dev_set_mtu\n", res);
1360 * Save slave's original ("permanent") mac address for modes
1361 * that need it, and for restoring it upon release, and then
1362 * set it to the master's address
1364 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1366 if (!bond->params.fail_over_mac) {
1368 * Set slave to master's mac address. The application already
1369 * set the master's mac address to that of the first slave
1371 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1372 addr.sa_family = slave_dev->type;
1373 res = dev_set_mac_address(slave_dev, &addr);
1375 pr_debug("Error %d calling set_mac_address\n", res);
1376 goto err_restore_mtu;
1380 /* open the slave since the application closed it */
1381 res = dev_open(slave_dev);
1383 pr_debug("Opening slave %s failed\n", slave_dev->name);
1384 goto err_restore_mac;
1387 new_slave->bond = bond;
1388 new_slave->dev = slave_dev;
1389 slave_dev->priv_flags |= IFF_BONDING;
1391 if (bond_is_lb(bond)) {
1392 /* bond_alb_init_slave() must be called before all other stages since
1393 * it might fail and we do not want to have to undo everything
1395 res = bond_alb_init_slave(bond, new_slave);
1400 /* If the mode USES_PRIMARY, then the following is handled by
1401 * bond_change_active_slave().
1403 if (!USES_PRIMARY(bond->params.mode)) {
1404 /* set promiscuity level to new slave */
1405 if (bond_dev->flags & IFF_PROMISC) {
1406 res = dev_set_promiscuity(slave_dev, 1);
1411 /* set allmulti level to new slave */
1412 if (bond_dev->flags & IFF_ALLMULTI) {
1413 res = dev_set_allmulti(slave_dev, 1);
1418 netif_addr_lock_bh(bond_dev);
1420 dev_mc_sync_multiple(slave_dev, bond_dev);
1421 dev_uc_sync_multiple(slave_dev, bond_dev);
1423 netif_addr_unlock_bh(bond_dev);
1426 if (bond->params.mode == BOND_MODE_8023AD) {
1427 /* add lacpdu mc addr to mc list */
1428 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1430 dev_mc_add(slave_dev, lacpdu_multicast);
1433 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1435 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1436 bond_dev->name, slave_dev->name);
1440 prev_slave = bond_last_slave(bond);
1442 new_slave->delay = 0;
1443 new_slave->link_failure_count = 0;
1445 bond_update_speed_duplex(new_slave);
1447 new_slave->last_arp_rx = jiffies -
1448 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1449 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1450 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1452 if (bond->params.miimon && !bond->params.use_carrier) {
1453 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1455 if ((link_reporting == -1) && !bond->params.arp_interval) {
1457 * miimon is set but a bonded network driver
1458 * does not support ETHTOOL/MII and
1459 * arp_interval is not set. Note: if
1460 * use_carrier is enabled, we will never go
1461 * here (because netif_carrier is always
1462 * supported); thus, we don't need to change
1463 * the messages for netif_carrier.
1465 pr_warning("%s: Warning: 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",
1466 bond_dev->name, slave_dev->name);
1467 } else if (link_reporting == -1) {
1468 /* unable get link status using mii/ethtool */
1469 pr_warning("%s: Warning: 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",
1470 bond_dev->name, slave_dev->name);
1474 /* check for initial state */
1475 if (bond->params.miimon) {
1476 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1477 if (bond->params.updelay) {
1478 new_slave->link = BOND_LINK_BACK;
1479 new_slave->delay = bond->params.updelay;
1481 new_slave->link = BOND_LINK_UP;
1484 new_slave->link = BOND_LINK_DOWN;
1486 } else if (bond->params.arp_interval) {
1487 new_slave->link = (netif_carrier_ok(slave_dev) ?
1488 BOND_LINK_UP : BOND_LINK_DOWN);
1490 new_slave->link = BOND_LINK_UP;
1493 if (new_slave->link != BOND_LINK_DOWN)
1494 new_slave->jiffies = jiffies;
1495 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1496 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1497 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1499 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1500 /* if there is a primary slave, remember it */
1501 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1502 bond->primary_slave = new_slave;
1503 bond->force_primary = true;
1507 switch (bond->params.mode) {
1508 case BOND_MODE_ACTIVEBACKUP:
1509 bond_set_slave_inactive_flags(new_slave);
1511 case BOND_MODE_8023AD:
1512 /* in 802.3ad mode, the internal mechanism
1513 * will activate the slaves in the selected
1516 bond_set_slave_inactive_flags(new_slave);
1517 /* if this is the first slave */
1519 SLAVE_AD_INFO(new_slave).id = 1;
1520 /* Initialize AD with the number of times that the AD timer is called in 1 second
1521 * can be called only after the mac address of the bond is set
1523 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1525 SLAVE_AD_INFO(new_slave).id =
1526 SLAVE_AD_INFO(prev_slave).id + 1;
1529 bond_3ad_bind_slave(new_slave);
1533 bond_set_active_slave(new_slave);
1534 bond_set_slave_inactive_flags(new_slave);
1537 pr_debug("This slave is always active in trunk mode\n");
1539 /* always active in trunk mode */
1540 bond_set_active_slave(new_slave);
1542 /* In trunking mode there is little meaning to curr_active_slave
1543 * anyway (it holds no special properties of the bond device),
1544 * so we can change it without calling change_active_interface()
1546 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1547 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1550 } /* switch(bond_mode) */
1552 #ifdef CONFIG_NET_POLL_CONTROLLER
1553 slave_dev->npinfo = bond->dev->npinfo;
1554 if (slave_dev->npinfo) {
1555 if (slave_enable_netpoll(new_slave)) {
1556 read_unlock(&bond->lock);
1557 pr_info("Error, %s: master_dev is using netpoll, "
1558 "but new slave device does not support netpoll.\n",
1566 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1569 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1573 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1575 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1576 goto err_unregister;
1580 bond_compute_features(bond);
1581 bond_set_carrier(bond);
1583 if (USES_PRIMARY(bond->params.mode)) {
1584 write_lock_bh(&bond->curr_slave_lock);
1585 bond_select_active_slave(bond);
1586 write_unlock_bh(&bond->curr_slave_lock);
1589 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1590 bond_dev->name, slave_dev->name,
1591 bond_is_active_slave(new_slave) ? "n active" : " backup",
1592 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1594 /* enslave is successful */
1597 /* Undo stages on error */
1599 netdev_rx_handler_unregister(slave_dev);
1602 if (!USES_PRIMARY(bond->params.mode))
1603 bond_hw_addr_flush(bond_dev, slave_dev);
1605 vlan_vids_del_by_dev(slave_dev, bond_dev);
1606 if (bond->primary_slave == new_slave)
1607 bond->primary_slave = NULL;
1608 if (bond->curr_active_slave == new_slave) {
1609 write_lock_bh(&bond->curr_slave_lock);
1610 bond_change_active_slave(bond, NULL);
1611 bond_select_active_slave(bond);
1612 write_unlock_bh(&bond->curr_slave_lock);
1614 slave_disable_netpoll(new_slave);
1617 slave_dev->priv_flags &= ~IFF_BONDING;
1618 dev_close(slave_dev);
1621 if (!bond->params.fail_over_mac) {
1622 /* XXX TODO - fom follow mode needs to change master's
1623 * MAC if this slave's MAC is in use by the bond, or at
1624 * least print a warning.
1626 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1627 addr.sa_family = slave_dev->type;
1628 dev_set_mac_address(slave_dev, &addr);
1632 dev_set_mtu(slave_dev, new_slave->original_mtu);
1638 /* Enslave of first slave has failed and we need to fix master's mac */
1639 if (!bond_has_slaves(bond) &&
1640 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1641 eth_hw_addr_random(bond_dev);
1647 * Try to release the slave device <slave> from the bond device <master>
1648 * It is legal to access curr_active_slave without a lock because all the function
1649 * is write-locked. If "all" is true it means that the function is being called
1650 * while destroying a bond interface and all slaves are being released.
1652 * The rules for slave state should be:
1653 * for Active/Backup:
1654 * Active stays on all backups go down
1655 * for Bonded connections:
1656 * The first up interface should be left on and all others downed.
1658 static int __bond_release_one(struct net_device *bond_dev,
1659 struct net_device *slave_dev,
1662 struct bonding *bond = netdev_priv(bond_dev);
1663 struct slave *slave, *oldcurrent;
1664 struct sockaddr addr;
1665 int old_flags = bond_dev->flags;
1666 netdev_features_t old_features = bond_dev->features;
1668 /* slave is not a slave or master is not master of this slave */
1669 if (!(slave_dev->flags & IFF_SLAVE) ||
1670 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1671 pr_err("%s: Error: cannot release %s.\n",
1672 bond_dev->name, slave_dev->name);
1678 slave = bond_get_slave_by_dev(bond, slave_dev);
1680 /* not a slave of this bond */
1681 pr_info("%s: %s not enslaved\n",
1682 bond_dev->name, slave_dev->name);
1683 unblock_netpoll_tx();
1687 /* release the slave from its bond */
1690 bond_upper_dev_unlink(bond_dev, slave_dev);
1691 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1692 * for this slave anymore.
1694 netdev_rx_handler_unregister(slave_dev);
1695 write_lock_bh(&bond->lock);
1697 /* Inform AD package of unbinding of slave. */
1698 if (bond->params.mode == BOND_MODE_8023AD)
1699 bond_3ad_unbind_slave(slave);
1701 write_unlock_bh(&bond->lock);
1703 pr_info("%s: releasing %s interface %s\n",
1705 bond_is_active_slave(slave) ? "active" : "backup",
1708 oldcurrent = bond->curr_active_slave;
1710 bond->current_arp_slave = NULL;
1712 if (!all && !bond->params.fail_over_mac) {
1713 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1714 bond_has_slaves(bond))
1715 pr_warn("%s: Warning: 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",
1716 bond_dev->name, slave_dev->name,
1718 bond_dev->name, slave_dev->name);
1721 if (bond->primary_slave == slave)
1722 bond->primary_slave = NULL;
1724 if (oldcurrent == slave) {
1725 write_lock_bh(&bond->curr_slave_lock);
1726 bond_change_active_slave(bond, NULL);
1727 write_unlock_bh(&bond->curr_slave_lock);
1730 if (bond_is_lb(bond)) {
1731 /* Must be called only after the slave has been
1732 * detached from the list and the curr_active_slave
1733 * has been cleared (if our_slave == old_current),
1734 * but before a new active slave is selected.
1736 bond_alb_deinit_slave(bond, slave);
1740 rcu_assign_pointer(bond->curr_active_slave, NULL);
1741 } else if (oldcurrent == slave) {
1743 * Note that we hold RTNL over this sequence, so there
1744 * is no concern that another slave add/remove event
1747 write_lock_bh(&bond->curr_slave_lock);
1749 bond_select_active_slave(bond);
1751 write_unlock_bh(&bond->curr_slave_lock);
1754 if (!bond_has_slaves(bond)) {
1755 bond_set_carrier(bond);
1756 eth_hw_addr_random(bond_dev);
1758 if (vlan_uses_dev(bond_dev)) {
1759 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1760 bond_dev->name, bond_dev->name);
1761 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1766 unblock_netpoll_tx();
1769 if (!bond_has_slaves(bond)) {
1770 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1771 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1774 bond_compute_features(bond);
1775 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1776 (old_features & NETIF_F_VLAN_CHALLENGED))
1777 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1778 bond_dev->name, slave_dev->name, bond_dev->name);
1780 /* must do this from outside any spinlocks */
1781 vlan_vids_del_by_dev(slave_dev, bond_dev);
1783 /* If the mode USES_PRIMARY, then this cases was handled above by
1784 * bond_change_active_slave(..., NULL)
1786 if (!USES_PRIMARY(bond->params.mode)) {
1787 /* unset promiscuity level from slave
1788 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1789 * of the IFF_PROMISC flag in the bond_dev, but we need the
1790 * value of that flag before that change, as that was the value
1791 * when this slave was attached, so we cache at the start of the
1792 * function and use it here. Same goes for ALLMULTI below
1794 if (old_flags & IFF_PROMISC)
1795 dev_set_promiscuity(slave_dev, -1);
1797 /* unset allmulti level from slave */
1798 if (old_flags & IFF_ALLMULTI)
1799 dev_set_allmulti(slave_dev, -1);
1801 bond_hw_addr_flush(bond_dev, slave_dev);
1804 slave_disable_netpoll(slave);
1806 /* close slave before restoring its mac address */
1807 dev_close(slave_dev);
1809 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1810 /* restore original ("permanent") mac address */
1811 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1812 addr.sa_family = slave_dev->type;
1813 dev_set_mac_address(slave_dev, &addr);
1816 dev_set_mtu(slave_dev, slave->original_mtu);
1818 slave_dev->priv_flags &= ~IFF_BONDING;
1822 return 0; /* deletion OK */
1825 /* A wrapper used because of ndo_del_link */
1826 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1828 return __bond_release_one(bond_dev, slave_dev, false);
1832 * First release a slave and then destroy the bond if no more slaves are left.
1833 * Must be under rtnl_lock when this function is called.
1835 static int bond_release_and_destroy(struct net_device *bond_dev,
1836 struct net_device *slave_dev)
1838 struct bonding *bond = netdev_priv(bond_dev);
1841 ret = bond_release(bond_dev, slave_dev);
1842 if (ret == 0 && !bond_has_slaves(bond)) {
1843 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1844 pr_info("%s: destroying bond %s.\n",
1845 bond_dev->name, bond_dev->name);
1846 unregister_netdevice(bond_dev);
1851 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1853 struct bonding *bond = netdev_priv(bond_dev);
1855 info->bond_mode = bond->params.mode;
1856 info->miimon = bond->params.miimon;
1858 read_lock(&bond->lock);
1859 info->num_slaves = bond->slave_cnt;
1860 read_unlock(&bond->lock);
1865 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1867 struct bonding *bond = netdev_priv(bond_dev);
1868 struct list_head *iter;
1869 int i = 0, res = -ENODEV;
1870 struct slave *slave;
1872 read_lock(&bond->lock);
1873 bond_for_each_slave(bond, slave, iter) {
1874 if (i++ == (int)info->slave_id) {
1876 strcpy(info->slave_name, slave->dev->name);
1877 info->link = slave->link;
1878 info->state = bond_slave_state(slave);
1879 info->link_failure_count = slave->link_failure_count;
1883 read_unlock(&bond->lock);
1888 /*-------------------------------- Monitoring -------------------------------*/
1891 static int bond_miimon_inspect(struct bonding *bond)
1893 int link_state, commit = 0;
1894 struct list_head *iter;
1895 struct slave *slave;
1896 bool ignore_updelay;
1898 ignore_updelay = !bond->curr_active_slave ? true : false;
1900 bond_for_each_slave_rcu(bond, slave, iter) {
1901 slave->new_link = BOND_LINK_NOCHANGE;
1903 link_state = bond_check_dev_link(bond, slave->dev, 0);
1905 switch (slave->link) {
1910 slave->link = BOND_LINK_FAIL;
1911 slave->delay = bond->params.downdelay;
1913 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1915 (bond->params.mode ==
1916 BOND_MODE_ACTIVEBACKUP) ?
1917 (bond_is_active_slave(slave) ?
1918 "active " : "backup ") : "",
1920 bond->params.downdelay * bond->params.miimon);
1923 case BOND_LINK_FAIL:
1926 * recovered before downdelay expired
1928 slave->link = BOND_LINK_UP;
1929 slave->jiffies = jiffies;
1930 pr_info("%s: link status up again after %d ms for interface %s.\n",
1932 (bond->params.downdelay - slave->delay) *
1933 bond->params.miimon,
1938 if (slave->delay <= 0) {
1939 slave->new_link = BOND_LINK_DOWN;
1947 case BOND_LINK_DOWN:
1951 slave->link = BOND_LINK_BACK;
1952 slave->delay = bond->params.updelay;
1955 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1956 bond->dev->name, slave->dev->name,
1957 ignore_updelay ? 0 :
1958 bond->params.updelay *
1959 bond->params.miimon);
1962 case BOND_LINK_BACK:
1964 slave->link = BOND_LINK_DOWN;
1965 pr_info("%s: link status down again after %d ms for interface %s.\n",
1967 (bond->params.updelay - slave->delay) *
1968 bond->params.miimon,
1977 if (slave->delay <= 0) {
1978 slave->new_link = BOND_LINK_UP;
1980 ignore_updelay = false;
1992 static void bond_miimon_commit(struct bonding *bond)
1994 struct list_head *iter;
1995 struct slave *slave;
1997 bond_for_each_slave(bond, slave, iter) {
1998 switch (slave->new_link) {
1999 case BOND_LINK_NOCHANGE:
2003 slave->link = BOND_LINK_UP;
2004 slave->jiffies = jiffies;
2006 if (bond->params.mode == BOND_MODE_8023AD) {
2007 /* prevent it from being the active one */
2008 bond_set_backup_slave(slave);
2009 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2010 /* make it immediately active */
2011 bond_set_active_slave(slave);
2012 } else if (slave != bond->primary_slave) {
2013 /* prevent it from being the active one */
2014 bond_set_backup_slave(slave);
2017 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2018 bond->dev->name, slave->dev->name,
2019 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2020 slave->duplex ? "full" : "half");
2022 /* notify ad that the link status has changed */
2023 if (bond->params.mode == BOND_MODE_8023AD)
2024 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2026 if (bond_is_lb(bond))
2027 bond_alb_handle_link_change(bond, slave,
2030 if (!bond->curr_active_slave ||
2031 (slave == bond->primary_slave))
2036 case BOND_LINK_DOWN:
2037 if (slave->link_failure_count < UINT_MAX)
2038 slave->link_failure_count++;
2040 slave->link = BOND_LINK_DOWN;
2042 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2043 bond->params.mode == BOND_MODE_8023AD)
2044 bond_set_slave_inactive_flags(slave);
2046 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2047 bond->dev->name, slave->dev->name);
2049 if (bond->params.mode == BOND_MODE_8023AD)
2050 bond_3ad_handle_link_change(slave,
2053 if (bond_is_lb(bond))
2054 bond_alb_handle_link_change(bond, slave,
2057 if (slave == bond->curr_active_slave)
2063 pr_err("%s: invalid new link %d on slave %s\n",
2064 bond->dev->name, slave->new_link,
2066 slave->new_link = BOND_LINK_NOCHANGE;
2074 write_lock_bh(&bond->curr_slave_lock);
2075 bond_select_active_slave(bond);
2076 write_unlock_bh(&bond->curr_slave_lock);
2077 unblock_netpoll_tx();
2080 bond_set_carrier(bond);
2086 * Really a wrapper that splits the mii monitor into two phases: an
2087 * inspection, then (if inspection indicates something needs to be done)
2088 * an acquisition of appropriate locks followed by a commit phase to
2089 * implement whatever link state changes are indicated.
2091 static void bond_mii_monitor(struct work_struct *work)
2093 struct bonding *bond = container_of(work, struct bonding,
2095 bool should_notify_peers = false;
2096 unsigned long delay;
2098 delay = msecs_to_jiffies(bond->params.miimon);
2100 if (!bond_has_slaves(bond))
2105 should_notify_peers = bond_should_notify_peers(bond);
2107 if (bond_miimon_inspect(bond)) {
2110 /* Race avoidance with bond_close cancel of workqueue */
2111 if (!rtnl_trylock()) {
2113 should_notify_peers = false;
2117 bond_miimon_commit(bond);
2119 rtnl_unlock(); /* might sleep, hold no other locks */
2124 if (bond->params.miimon)
2125 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2127 if (should_notify_peers) {
2128 if (!rtnl_trylock())
2130 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2135 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2137 struct net_device *upper;
2138 struct list_head *iter;
2141 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2145 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2146 if (ip == bond_confirm_addr(upper, 0, ip)) {
2157 * We go to the (large) trouble of VLAN tagging ARP frames because
2158 * switches in VLAN mode (especially if ports are configured as
2159 * "native" to a VLAN) might not pass non-tagged frames.
2161 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2163 struct sk_buff *skb;
2165 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2166 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2168 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2169 NULL, slave_dev->dev_addr, NULL);
2172 pr_err("ARP packet allocation failed\n");
2176 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2178 pr_err("failed to insert VLAN tag\n");
2186 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2188 struct net_device *upper, *vlan_upper;
2189 struct list_head *iter, *vlan_iter;
2191 __be32 *targets = bond->params.arp_targets, addr;
2194 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2195 pr_debug("basa: target %pI4\n", &targets[i]);
2197 /* Find out through which dev should the packet go */
2198 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2201 pr_debug("%s: no route to arp_ip_target %pI4\n",
2202 bond->dev->name, &targets[i]);
2208 /* bond device itself */
2209 if (rt->dst.dev == bond->dev)
2213 /* first we search only for vlan devices. for every vlan
2214 * found we verify its upper dev list, searching for the
2215 * rt->dst.dev. If found we save the tag of the vlan and
2216 * proceed to send the packet.
2220 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2222 if (!is_vlan_dev(vlan_upper))
2224 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2226 if (upper == rt->dst.dev) {
2227 vlan_id = vlan_dev_vlan_id(vlan_upper);
2234 /* if the device we're looking for is not on top of any of
2235 * our upper vlans, then just search for any dev that
2236 * matches, and in case it's a vlan - save the id
2238 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2239 if (upper == rt->dst.dev) {
2240 /* if it's a vlan - get its VID */
2241 if (is_vlan_dev(upper))
2242 vlan_id = vlan_dev_vlan_id(upper);
2250 /* Not our device - skip */
2251 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2252 bond->dev->name, &targets[i],
2253 rt->dst.dev ? rt->dst.dev->name : "NULL");
2259 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2261 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2266 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2270 if (!sip || !bond_has_this_ip(bond, tip)) {
2271 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2275 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2277 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2280 slave->last_arp_rx = jiffies;
2281 slave->target_last_arp_rx[i] = jiffies;
2284 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2285 struct slave *slave)
2287 struct arphdr *arp = (struct arphdr *)skb->data;
2288 unsigned char *arp_ptr;
2292 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2293 return RX_HANDLER_ANOTHER;
2295 read_lock(&bond->lock);
2297 if (!slave_do_arp_validate(bond, slave))
2300 alen = arp_hdr_len(bond->dev);
2302 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2303 bond->dev->name, skb->dev->name);
2305 if (alen > skb_headlen(skb)) {
2306 arp = kmalloc(alen, GFP_ATOMIC);
2309 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2313 if (arp->ar_hln != bond->dev->addr_len ||
2314 skb->pkt_type == PACKET_OTHERHOST ||
2315 skb->pkt_type == PACKET_LOOPBACK ||
2316 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2317 arp->ar_pro != htons(ETH_P_IP) ||
2321 arp_ptr = (unsigned char *)(arp + 1);
2322 arp_ptr += bond->dev->addr_len;
2323 memcpy(&sip, arp_ptr, 4);
2324 arp_ptr += 4 + bond->dev->addr_len;
2325 memcpy(&tip, arp_ptr, 4);
2327 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2328 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2329 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2333 * Backup slaves won't see the ARP reply, but do come through
2334 * here for each ARP probe (so we swap the sip/tip to validate
2335 * the probe). In a "redundant switch, common router" type of
2336 * configuration, the ARP probe will (hopefully) travel from
2337 * the active, through one switch, the router, then the other
2338 * switch before reaching the backup.
2340 * We 'trust' the arp requests if there is an active slave and
2341 * it received valid arp reply(s) after it became active. This
2342 * is done to avoid endless looping when we can't reach the
2343 * arp_ip_target and fool ourselves with our own arp requests.
2345 if (bond_is_active_slave(slave))
2346 bond_validate_arp(bond, slave, sip, tip);
2347 else if (bond->curr_active_slave &&
2348 time_after(slave_last_rx(bond, bond->curr_active_slave),
2349 bond->curr_active_slave->jiffies))
2350 bond_validate_arp(bond, slave, tip, sip);
2353 read_unlock(&bond->lock);
2354 if (arp != (struct arphdr *)skb->data)
2356 return RX_HANDLER_ANOTHER;
2359 /* function to verify if we're in the arp_interval timeslice, returns true if
2360 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2361 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2363 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2366 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2368 return time_in_range(jiffies,
2369 last_act - delta_in_ticks,
2370 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2374 * this function is called regularly to monitor each slave's link
2375 * ensuring that traffic is being sent and received when arp monitoring
2376 * is used in load-balancing mode. if the adapter has been dormant, then an
2377 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2378 * arp monitoring in active backup mode.
2380 static void bond_loadbalance_arp_mon(struct work_struct *work)
2382 struct bonding *bond = container_of(work, struct bonding,
2384 struct slave *slave, *oldcurrent;
2385 struct list_head *iter;
2386 int do_failover = 0;
2388 if (!bond_has_slaves(bond))
2393 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2394 /* see if any of the previous devices are up now (i.e. they have
2395 * xmt and rcv traffic). the curr_active_slave does not come into
2396 * the picture unless it is null. also, slave->jiffies is not needed
2397 * here because we send an arp on each slave and give a slave as
2398 * long as it needs to get the tx/rx within the delta.
2399 * TODO: what about up/down delay in arp mode? it wasn't here before
2402 bond_for_each_slave_rcu(bond, slave, iter) {
2403 unsigned long trans_start = dev_trans_start(slave->dev);
2405 if (slave->link != BOND_LINK_UP) {
2406 if (bond_time_in_interval(bond, trans_start, 1) &&
2407 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2409 slave->link = BOND_LINK_UP;
2410 bond_set_active_slave(slave);
2412 /* primary_slave has no meaning in round-robin
2413 * mode. the window of a slave being up and
2414 * curr_active_slave being null after enslaving
2418 pr_info("%s: link status definitely up for interface %s, ",
2423 pr_info("%s: interface %s is now up\n",
2429 /* slave->link == BOND_LINK_UP */
2431 /* not all switches will respond to an arp request
2432 * when the source ip is 0, so don't take the link down
2433 * if we don't know our ip yet
2435 if (!bond_time_in_interval(bond, trans_start, 2) ||
2436 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2438 slave->link = BOND_LINK_DOWN;
2439 bond_set_backup_slave(slave);
2441 if (slave->link_failure_count < UINT_MAX)
2442 slave->link_failure_count++;
2444 pr_info("%s: 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 (IS_UP(slave->dev))
2461 bond_arp_send_all(bond, slave);
2467 /* the bond_select_active_slave must hold RTNL
2468 * and curr_slave_lock for write.
2470 if (!rtnl_trylock())
2473 write_lock_bh(&bond->curr_slave_lock);
2475 bond_select_active_slave(bond);
2477 write_unlock_bh(&bond->curr_slave_lock);
2478 unblock_netpoll_tx();
2483 if (bond->params.arp_interval)
2484 queue_delayed_work(bond->wq, &bond->arp_work,
2485 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 hold.
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;
2516 * Give slaves 2*delta after being enslaved or made
2517 * active. This avoids bouncing, as the last receive
2518 * times need a full ARP monitor cycle to be updated.
2520 if (bond_time_in_interval(bond, slave->jiffies, 2))
2524 * Backup slave is down if:
2525 * - No current_arp_slave AND
2526 * - more than 3*delta since last receive AND
2527 * - the bond has an IP address
2529 * Note: a non-null current_arp_slave indicates
2530 * the curr_active_slave went down and we are
2531 * searching for a new one; under this condition
2532 * we only take the curr_active_slave down - this
2533 * gives each slave a chance to tx/rx traffic
2534 * before being taken out
2536 if (!bond_is_active_slave(slave) &&
2537 !bond->current_arp_slave &&
2538 !bond_time_in_interval(bond, last_rx, 3)) {
2539 slave->new_link = BOND_LINK_DOWN;
2544 * Active slave is down if:
2545 * - more than 2*delta since transmitting OR
2546 * - (more than 2*delta since receive AND
2547 * the bond has an IP address)
2549 trans_start = dev_trans_start(slave->dev);
2550 if (bond_is_active_slave(slave) &&
2551 (!bond_time_in_interval(bond, trans_start, 2) ||
2552 !bond_time_in_interval(bond, last_rx, 2))) {
2553 slave->new_link = BOND_LINK_DOWN;
2562 * Called to commit link state changes noted by inspection step of
2563 * active-backup mode ARP monitor.
2565 * Called with RTNL hold.
2567 static void bond_ab_arp_commit(struct bonding *bond)
2569 unsigned long trans_start;
2570 struct list_head *iter;
2571 struct slave *slave;
2573 bond_for_each_slave(bond, slave, iter) {
2574 switch (slave->new_link) {
2575 case BOND_LINK_NOCHANGE:
2579 trans_start = dev_trans_start(slave->dev);
2580 if (bond->curr_active_slave != slave ||
2581 (!bond->curr_active_slave &&
2582 bond_time_in_interval(bond, trans_start, 1))) {
2583 slave->link = BOND_LINK_UP;
2584 if (bond->current_arp_slave) {
2585 bond_set_slave_inactive_flags(
2586 bond->current_arp_slave);
2587 bond->current_arp_slave = NULL;
2590 pr_info("%s: link status definitely up for interface %s.\n",
2591 bond->dev->name, slave->dev->name);
2593 if (!bond->curr_active_slave ||
2594 (slave == 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);
2608 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2609 bond->dev->name, slave->dev->name);
2611 if (slave == bond->curr_active_slave) {
2612 bond->current_arp_slave = NULL;
2619 pr_err("%s: impossible: new_link %d on slave %s\n",
2620 bond->dev->name, slave->new_link,
2628 write_lock_bh(&bond->curr_slave_lock);
2629 bond_select_active_slave(bond);
2630 write_unlock_bh(&bond->curr_slave_lock);
2631 unblock_netpoll_tx();
2634 bond_set_carrier(bond);
2638 * Send ARP probes for active-backup mode ARP monitor.
2640 * Called with rcu_read_lock hold.
2642 static void bond_ab_arp_probe(struct bonding *bond)
2644 struct slave *slave, *before = NULL, *new_slave = NULL,
2645 *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2646 struct list_head *iter;
2649 read_lock(&bond->curr_slave_lock);
2651 if (curr_arp_slave && bond->curr_active_slave)
2652 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2653 curr_arp_slave->dev->name,
2654 bond->curr_active_slave->dev->name);
2656 if (bond->curr_active_slave) {
2657 bond_arp_send_all(bond, bond->curr_active_slave);
2658 read_unlock(&bond->curr_slave_lock);
2662 read_unlock(&bond->curr_slave_lock);
2664 /* if we don't have a curr_active_slave, search for the next available
2665 * backup slave from the current_arp_slave and make it the candidate
2666 * for becoming the curr_active_slave
2669 if (!curr_arp_slave) {
2670 curr_arp_slave = bond_first_slave_rcu(bond);
2671 if (!curr_arp_slave)
2675 bond_set_slave_inactive_flags(curr_arp_slave);
2677 bond_for_each_slave_rcu(bond, slave, iter) {
2678 if (!found && !before && IS_UP(slave->dev))
2681 if (found && !new_slave && IS_UP(slave->dev))
2683 /* if the link state is up at this point, we
2684 * mark it down - this can happen if we have
2685 * simultaneous link failures and
2686 * reselect_active_interface doesn't make this
2687 * one the current slave so it is still marked
2688 * up when it is actually down
2690 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2691 slave->link = BOND_LINK_DOWN;
2692 if (slave->link_failure_count < UINT_MAX)
2693 slave->link_failure_count++;
2695 bond_set_slave_inactive_flags(slave);
2697 pr_info("%s: backup interface %s is now down.\n",
2698 bond->dev->name, slave->dev->name);
2700 if (slave == curr_arp_slave)
2704 if (!new_slave && before)
2710 new_slave->link = BOND_LINK_BACK;
2711 bond_set_slave_active_flags(new_slave);
2712 bond_arp_send_all(bond, new_slave);
2713 new_slave->jiffies = jiffies;
2714 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2717 static void bond_activebackup_arp_mon(struct work_struct *work)
2719 struct bonding *bond = container_of(work, struct bonding,
2721 bool should_notify_peers = false;
2724 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2726 if (!bond_has_slaves(bond))
2731 should_notify_peers = bond_should_notify_peers(bond);
2733 if (bond_ab_arp_inspect(bond)) {
2736 /* Race avoidance with bond_close flush of workqueue */
2737 if (!rtnl_trylock()) {
2739 should_notify_peers = false;
2743 bond_ab_arp_commit(bond);
2749 bond_ab_arp_probe(bond);
2753 if (bond->params.arp_interval)
2754 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2756 if (should_notify_peers) {
2757 if (!rtnl_trylock())
2759 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2764 /*-------------------------- netdev event handling --------------------------*/
2767 * Change device name
2769 static int bond_event_changename(struct bonding *bond)
2771 bond_remove_proc_entry(bond);
2772 bond_create_proc_entry(bond);
2774 bond_debug_reregister(bond);
2779 static int bond_master_netdev_event(unsigned long event,
2780 struct net_device *bond_dev)
2782 struct bonding *event_bond = netdev_priv(bond_dev);
2785 case NETDEV_CHANGENAME:
2786 return bond_event_changename(event_bond);
2787 case NETDEV_UNREGISTER:
2788 bond_remove_proc_entry(event_bond);
2790 case NETDEV_REGISTER:
2791 bond_create_proc_entry(event_bond);
2793 case NETDEV_NOTIFY_PEERS:
2794 if (event_bond->send_peer_notif)
2795 event_bond->send_peer_notif--;
2804 static int bond_slave_netdev_event(unsigned long event,
2805 struct net_device *slave_dev)
2807 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2808 struct bonding *bond;
2809 struct net_device *bond_dev;
2813 /* A netdev event can be generated while enslaving a device
2814 * before netdev_rx_handler_register is called in which case
2815 * slave will be NULL
2819 bond_dev = slave->bond->dev;
2823 case NETDEV_UNREGISTER:
2824 if (bond_dev->type != ARPHRD_ETHER)
2825 bond_release_and_destroy(bond_dev, slave_dev);
2827 bond_release(bond_dev, slave_dev);
2831 old_speed = slave->speed;
2832 old_duplex = slave->duplex;
2834 bond_update_speed_duplex(slave);
2836 if (bond->params.mode == BOND_MODE_8023AD) {
2837 if (old_speed != slave->speed)
2838 bond_3ad_adapter_speed_changed(slave);
2839 if (old_duplex != slave->duplex)
2840 bond_3ad_adapter_duplex_changed(slave);
2845 * ... Or is it this?
2848 case NETDEV_CHANGEMTU:
2850 * TODO: Should slaves be allowed to
2851 * independently alter their MTU? For
2852 * an active-backup bond, slaves need
2853 * not be the same type of device, so
2854 * MTUs may vary. For other modes,
2855 * slaves arguably should have the
2856 * same MTUs. To do this, we'd need to
2857 * take over the slave's change_mtu
2858 * function for the duration of their
2862 case NETDEV_CHANGENAME:
2864 * TODO: handle changing the primary's name
2867 case NETDEV_FEAT_CHANGE:
2868 bond_compute_features(bond);
2870 case NETDEV_RESEND_IGMP:
2871 /* Propagate to master device */
2872 call_netdevice_notifiers(event, slave->bond->dev);
2882 * bond_netdev_event: handle netdev notifier chain events.
2884 * This function receives events for the netdev chain. The caller (an
2885 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2886 * locks for us to safely manipulate the slave devices (RTNL lock,
2889 static int bond_netdev_event(struct notifier_block *this,
2890 unsigned long event, void *ptr)
2892 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2894 pr_debug("event_dev: %s, event: %lx\n",
2895 event_dev ? event_dev->name : "None",
2898 if (!(event_dev->priv_flags & IFF_BONDING))
2901 if (event_dev->flags & IFF_MASTER) {
2902 pr_debug("IFF_MASTER\n");
2903 return bond_master_netdev_event(event, event_dev);
2906 if (event_dev->flags & IFF_SLAVE) {
2907 pr_debug("IFF_SLAVE\n");
2908 return bond_slave_netdev_event(event, event_dev);
2914 static struct notifier_block bond_netdev_notifier = {
2915 .notifier_call = bond_netdev_event,
2918 /*---------------------------- Hashing Policies -----------------------------*/
2920 /* L2 hash helper */
2921 static inline u32 bond_eth_hash(struct sk_buff *skb)
2923 struct ethhdr *data = (struct ethhdr *)skb->data;
2925 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2926 return data->h_dest[5] ^ data->h_source[5];
2931 /* Extract the appropriate headers based on bond's xmit policy */
2932 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2933 struct flow_keys *fk)
2935 const struct ipv6hdr *iph6;
2936 const struct iphdr *iph;
2937 int noff, proto = -1;
2939 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2940 return skb_flow_dissect(skb, fk);
2943 noff = skb_network_offset(skb);
2944 if (skb->protocol == htons(ETH_P_IP)) {
2945 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2948 fk->src = iph->saddr;
2949 fk->dst = iph->daddr;
2950 noff += iph->ihl << 2;
2951 if (!ip_is_fragment(iph))
2952 proto = iph->protocol;
2953 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2954 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2956 iph6 = ipv6_hdr(skb);
2957 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2958 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2959 noff += sizeof(*iph6);
2960 proto = iph6->nexthdr;
2964 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2965 fk->ports = skb_flow_get_ports(skb, noff, proto);
2971 * bond_xmit_hash - generate a hash value based on the xmit policy
2972 * @bond: bonding device
2973 * @skb: buffer to use for headers
2974 * @count: modulo value
2976 * This function will extract the necessary headers from the skb buffer and use
2977 * them to generate a hash based on the xmit_policy set in the bonding device
2978 * which will be reduced modulo count before returning.
2980 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
2982 struct flow_keys flow;
2985 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2986 !bond_flow_dissect(bond, skb, &flow))
2987 return bond_eth_hash(skb) % count;
2989 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
2990 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
2991 hash = bond_eth_hash(skb);
2993 hash = (__force u32)flow.ports;
2994 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
2995 hash ^= (hash >> 16);
2996 hash ^= (hash >> 8);
2998 return hash % count;
3001 /*-------------------------- Device entry points ----------------------------*/
3003 static void bond_work_init_all(struct bonding *bond)
3005 INIT_DELAYED_WORK(&bond->mcast_work,
3006 bond_resend_igmp_join_requests_delayed);
3007 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3008 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3009 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3010 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3012 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3013 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3016 static void bond_work_cancel_all(struct bonding *bond)
3018 cancel_delayed_work_sync(&bond->mii_work);
3019 cancel_delayed_work_sync(&bond->arp_work);
3020 cancel_delayed_work_sync(&bond->alb_work);
3021 cancel_delayed_work_sync(&bond->ad_work);
3022 cancel_delayed_work_sync(&bond->mcast_work);
3025 static int bond_open(struct net_device *bond_dev)
3027 struct bonding *bond = netdev_priv(bond_dev);
3028 struct list_head *iter;
3029 struct slave *slave;
3031 /* reset slave->backup and slave->inactive */
3032 read_lock(&bond->lock);
3033 if (bond_has_slaves(bond)) {
3034 read_lock(&bond->curr_slave_lock);
3035 bond_for_each_slave(bond, slave, iter) {
3036 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3037 && (slave != bond->curr_active_slave)) {
3038 bond_set_slave_inactive_flags(slave);
3040 bond_set_slave_active_flags(slave);
3043 read_unlock(&bond->curr_slave_lock);
3045 read_unlock(&bond->lock);
3047 bond_work_init_all(bond);
3049 if (bond_is_lb(bond)) {
3050 /* bond_alb_initialize must be called before the timer
3053 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3055 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3058 if (bond->params.miimon) /* link check interval, in milliseconds. */
3059 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3061 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3062 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3063 if (bond->params.arp_validate)
3064 bond->recv_probe = bond_arp_rcv;
3067 if (bond->params.mode == BOND_MODE_8023AD) {
3068 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3069 /* register to receive LACPDUs */
3070 bond->recv_probe = bond_3ad_lacpdu_recv;
3071 bond_3ad_initiate_agg_selection(bond, 1);
3077 static int bond_close(struct net_device *bond_dev)
3079 struct bonding *bond = netdev_priv(bond_dev);
3081 bond_work_cancel_all(bond);
3082 bond->send_peer_notif = 0;
3083 if (bond_is_lb(bond))
3084 bond_alb_deinitialize(bond);
3085 bond->recv_probe = NULL;
3090 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3091 struct rtnl_link_stats64 *stats)
3093 struct bonding *bond = netdev_priv(bond_dev);
3094 struct rtnl_link_stats64 temp;
3095 struct list_head *iter;
3096 struct slave *slave;
3098 memset(stats, 0, sizeof(*stats));
3100 read_lock_bh(&bond->lock);
3101 bond_for_each_slave(bond, slave, iter) {
3102 const struct rtnl_link_stats64 *sstats =
3103 dev_get_stats(slave->dev, &temp);
3105 stats->rx_packets += sstats->rx_packets;
3106 stats->rx_bytes += sstats->rx_bytes;
3107 stats->rx_errors += sstats->rx_errors;
3108 stats->rx_dropped += sstats->rx_dropped;
3110 stats->tx_packets += sstats->tx_packets;
3111 stats->tx_bytes += sstats->tx_bytes;
3112 stats->tx_errors += sstats->tx_errors;
3113 stats->tx_dropped += sstats->tx_dropped;
3115 stats->multicast += sstats->multicast;
3116 stats->collisions += sstats->collisions;
3118 stats->rx_length_errors += sstats->rx_length_errors;
3119 stats->rx_over_errors += sstats->rx_over_errors;
3120 stats->rx_crc_errors += sstats->rx_crc_errors;
3121 stats->rx_frame_errors += sstats->rx_frame_errors;
3122 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3123 stats->rx_missed_errors += sstats->rx_missed_errors;
3125 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3126 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3127 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3128 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3129 stats->tx_window_errors += sstats->tx_window_errors;
3131 read_unlock_bh(&bond->lock);
3136 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3138 struct bonding *bond = netdev_priv(bond_dev);
3139 struct net_device *slave_dev = NULL;
3140 struct ifbond k_binfo;
3141 struct ifbond __user *u_binfo = NULL;
3142 struct ifslave k_sinfo;
3143 struct ifslave __user *u_sinfo = NULL;
3144 struct mii_ioctl_data *mii = NULL;
3148 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3160 * We do this again just in case we were called by SIOCGMIIREG
3161 * instead of SIOCGMIIPHY.
3168 if (mii->reg_num == 1) {
3170 read_lock(&bond->lock);
3171 read_lock(&bond->curr_slave_lock);
3172 if (netif_carrier_ok(bond->dev))
3173 mii->val_out = BMSR_LSTATUS;
3175 read_unlock(&bond->curr_slave_lock);
3176 read_unlock(&bond->lock);
3180 case BOND_INFO_QUERY_OLD:
3181 case SIOCBONDINFOQUERY:
3182 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3184 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3187 res = bond_info_query(bond_dev, &k_binfo);
3189 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3193 case BOND_SLAVE_INFO_QUERY_OLD:
3194 case SIOCBONDSLAVEINFOQUERY:
3195 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3197 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3200 res = bond_slave_info_query(bond_dev, &k_sinfo);
3202 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3211 net = dev_net(bond_dev);
3213 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3216 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3218 pr_debug("slave_dev=%p:\n", slave_dev);
3223 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3225 case BOND_ENSLAVE_OLD:
3226 case SIOCBONDENSLAVE:
3227 res = bond_enslave(bond_dev, slave_dev);
3229 case BOND_RELEASE_OLD:
3230 case SIOCBONDRELEASE:
3231 res = bond_release(bond_dev, slave_dev);
3233 case BOND_SETHWADDR_OLD:
3234 case SIOCBONDSETHWADDR:
3235 bond_set_dev_addr(bond_dev, slave_dev);
3238 case BOND_CHANGE_ACTIVE_OLD:
3239 case SIOCBONDCHANGEACTIVE:
3240 res = bond_option_active_slave_set(bond, slave_dev);
3252 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3254 struct bonding *bond = netdev_priv(bond_dev);
3256 if (change & IFF_PROMISC)
3257 bond_set_promiscuity(bond,
3258 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3260 if (change & IFF_ALLMULTI)
3261 bond_set_allmulti(bond,
3262 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3265 static void bond_set_rx_mode(struct net_device *bond_dev)
3267 struct bonding *bond = netdev_priv(bond_dev);
3268 struct list_head *iter;
3269 struct slave *slave;
3273 if (USES_PRIMARY(bond->params.mode)) {
3274 slave = rcu_dereference(bond->curr_active_slave);
3276 dev_uc_sync(slave->dev, bond_dev);
3277 dev_mc_sync(slave->dev, bond_dev);
3280 bond_for_each_slave_rcu(bond, slave, iter) {
3281 dev_uc_sync_multiple(slave->dev, bond_dev);
3282 dev_mc_sync_multiple(slave->dev, bond_dev);
3288 static int bond_neigh_init(struct neighbour *n)
3290 struct bonding *bond = netdev_priv(n->dev);
3291 const struct net_device_ops *slave_ops;
3292 struct neigh_parms parms;
3293 struct slave *slave;
3296 slave = bond_first_slave(bond);
3299 slave_ops = slave->dev->netdev_ops;
3300 if (!slave_ops->ndo_neigh_setup)
3303 parms.neigh_setup = NULL;
3304 parms.neigh_cleanup = NULL;
3305 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3310 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3311 * after the last slave has been detached. Assumes that all slaves
3312 * utilize the same neigh_cleanup (true at this writing as only user
3315 n->parms->neigh_cleanup = parms.neigh_cleanup;
3317 if (!parms.neigh_setup)
3320 return parms.neigh_setup(n);
3324 * The bonding ndo_neigh_setup is called at init time beofre any
3325 * slave exists. So we must declare proxy setup function which will
3326 * be used at run time to resolve the actual slave neigh param setup.
3328 * It's also called by master devices (such as vlans) to setup their
3329 * underlying devices. In that case - do nothing, we're already set up from
3332 static int bond_neigh_setup(struct net_device *dev,
3333 struct neigh_parms *parms)
3335 /* modify only our neigh_parms */
3336 if (parms->dev == dev)
3337 parms->neigh_setup = bond_neigh_init;
3343 * Change the MTU of all of a master's slaves to match the master
3345 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3347 struct bonding *bond = netdev_priv(bond_dev);
3348 struct slave *slave, *rollback_slave;
3349 struct list_head *iter;
3352 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3353 (bond_dev ? bond_dev->name : "None"), new_mtu);
3355 /* Can't hold bond->lock with bh disabled here since
3356 * some base drivers panic. On the other hand we can't
3357 * hold bond->lock without bh disabled because we'll
3358 * deadlock. The only solution is to rely on the fact
3359 * that we're under rtnl_lock here, and the slaves
3360 * list won't change. This doesn't solve the problem
3361 * of setting the slave's MTU while it is
3362 * transmitting, but the assumption is that the base
3363 * driver can handle that.
3365 * TODO: figure out a way to safely iterate the slaves
3366 * list, but without holding a lock around the actual
3367 * call to the base driver.
3370 bond_for_each_slave(bond, slave, iter) {
3371 pr_debug("s %p c_m %p\n",
3373 slave->dev->netdev_ops->ndo_change_mtu);
3375 res = dev_set_mtu(slave->dev, new_mtu);
3378 /* If we failed to set the slave's mtu to the new value
3379 * we must abort the operation even in ACTIVE_BACKUP
3380 * mode, because if we allow the backup slaves to have
3381 * different mtu values than the active slave we'll
3382 * need to change their mtu when doing a failover. That
3383 * means changing their mtu from timer context, which
3384 * is probably not a good idea.
3386 pr_debug("err %d %s\n", res, slave->dev->name);
3391 bond_dev->mtu = new_mtu;
3396 /* unwind from head to the slave that failed */
3397 bond_for_each_slave(bond, rollback_slave, iter) {
3400 if (rollback_slave == slave)
3403 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3405 pr_debug("unwind err %d dev %s\n",
3406 tmp_res, rollback_slave->dev->name);
3416 * Note that many devices must be down to change the HW address, and
3417 * downing the master releases all slaves. We can make bonds full of
3418 * bonding devices to test this, however.
3420 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3422 struct bonding *bond = netdev_priv(bond_dev);
3423 struct slave *slave, *rollback_slave;
3424 struct sockaddr *sa = addr, tmp_sa;
3425 struct list_head *iter;
3428 if (bond->params.mode == BOND_MODE_ALB)
3429 return bond_alb_set_mac_address(bond_dev, addr);
3432 pr_debug("bond=%p, name=%s\n",
3433 bond, bond_dev ? bond_dev->name : "None");
3435 /* If fail_over_mac is enabled, do nothing and return success.
3436 * Returning an error causes ifenslave to fail.
3438 if (bond->params.fail_over_mac)
3441 if (!is_valid_ether_addr(sa->sa_data))
3442 return -EADDRNOTAVAIL;
3444 /* Can't hold bond->lock with bh disabled here since
3445 * some base drivers panic. On the other hand we can't
3446 * hold bond->lock without bh disabled because we'll
3447 * deadlock. The only solution is to rely on the fact
3448 * that we're under rtnl_lock here, and the slaves
3449 * list won't change. This doesn't solve the problem
3450 * of setting the slave's hw address while it is
3451 * transmitting, but the assumption is that the base
3452 * driver can handle that.
3454 * TODO: figure out a way to safely iterate the slaves
3455 * list, but without holding a lock around the actual
3456 * call to the base driver.
3459 bond_for_each_slave(bond, slave, iter) {
3460 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3461 pr_debug("slave %p %s\n", slave, slave->dev->name);
3463 if (slave_ops->ndo_set_mac_address == NULL) {
3465 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3469 res = dev_set_mac_address(slave->dev, addr);
3471 /* TODO: consider downing the slave
3473 * User should expect communications
3474 * breakage anyway until ARP finish
3477 pr_debug("err %d %s\n", res, slave->dev->name);
3483 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3487 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3488 tmp_sa.sa_family = bond_dev->type;
3490 /* unwind from head to the slave that failed */
3491 bond_for_each_slave(bond, rollback_slave, iter) {
3494 if (rollback_slave == slave)
3497 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3499 pr_debug("unwind err %d dev %s\n",
3500 tmp_res, rollback_slave->dev->name);
3508 * bond_xmit_slave_id - transmit skb through slave with slave_id
3509 * @bond: bonding device that is transmitting
3510 * @skb: buffer to transmit
3511 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3513 * This function tries to transmit through slave with slave_id but in case
3514 * it fails, it tries to find the first available slave for transmission.
3515 * The skb is consumed in all cases, thus the function is void.
3517 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3519 struct list_head *iter;
3520 struct slave *slave;
3523 /* Here we start from the slave with slave_id */
3524 bond_for_each_slave_rcu(bond, slave, iter) {
3526 if (slave_can_tx(slave)) {
3527 bond_dev_queue_xmit(bond, skb, slave->dev);
3533 /* Here we start from the first slave up to slave_id */
3535 bond_for_each_slave_rcu(bond, slave, iter) {
3538 if (slave_can_tx(slave)) {
3539 bond_dev_queue_xmit(bond, skb, slave->dev);
3543 /* no slave that can tx has been found */
3548 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3549 * @bond: bonding device to use
3551 * Based on the value of the bonding device's packets_per_slave parameter
3552 * this function generates a slave id, which is usually used as the next
3553 * slave to transmit through.
3555 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3557 int packets_per_slave = bond->params.packets_per_slave;
3560 switch (packets_per_slave) {
3562 slave_id = prandom_u32();
3565 slave_id = bond->rr_tx_counter;
3568 slave_id = reciprocal_divide(bond->rr_tx_counter,
3572 bond->rr_tx_counter++;
3577 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3579 struct bonding *bond = netdev_priv(bond_dev);
3580 struct iphdr *iph = ip_hdr(skb);
3581 struct slave *slave;
3584 /* Start with the curr_active_slave that joined the bond as the
3585 * default for sending IGMP traffic. For failover purposes one
3586 * needs to maintain some consistency for the interface that will
3587 * send the join/membership reports. The curr_active_slave found
3588 * will send all of this type of traffic.
3590 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3591 slave = rcu_dereference(bond->curr_active_slave);
3592 if (slave && slave_can_tx(slave))
3593 bond_dev_queue_xmit(bond, skb, slave->dev);
3595 bond_xmit_slave_id(bond, skb, 0);
3597 slave_id = bond_rr_gen_slave_id(bond);
3598 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3601 return NETDEV_TX_OK;
3605 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3606 * the bond has a usable interface.
3608 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3610 struct bonding *bond = netdev_priv(bond_dev);
3611 struct slave *slave;
3613 slave = rcu_dereference(bond->curr_active_slave);
3615 bond_dev_queue_xmit(bond, skb, slave->dev);
3619 return NETDEV_TX_OK;
3622 /* In bond_xmit_xor() , we determine the output device by using a pre-
3623 * determined xmit_hash_policy(), If the selected device is not enabled,
3624 * find the next active slave.
3626 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3628 struct bonding *bond = netdev_priv(bond_dev);
3630 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3632 return NETDEV_TX_OK;
3635 /* in broadcast mode, we send everything to all usable interfaces. */
3636 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3638 struct bonding *bond = netdev_priv(bond_dev);
3639 struct slave *slave = NULL;
3640 struct list_head *iter;
3642 bond_for_each_slave_rcu(bond, slave, iter) {
3643 if (bond_is_last_slave(bond, slave))
3645 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3646 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3649 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3653 /* bond_dev_queue_xmit always returns 0 */
3654 bond_dev_queue_xmit(bond, skb2, slave->dev);
3657 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3658 bond_dev_queue_xmit(bond, skb, slave->dev);
3662 return NETDEV_TX_OK;
3665 /*------------------------- Device initialization ---------------------------*/
3668 * Lookup the slave that corresponds to a qid
3670 static inline int bond_slave_override(struct bonding *bond,
3671 struct sk_buff *skb)
3673 struct slave *slave = NULL;
3674 struct list_head *iter;
3676 if (!skb->queue_mapping)
3679 /* Find out if any slaves have the same mapping as this skb. */
3680 bond_for_each_slave_rcu(bond, slave, iter) {
3681 if (slave->queue_id == skb->queue_mapping) {
3682 if (slave_can_tx(slave)) {
3683 bond_dev_queue_xmit(bond, skb, slave->dev);
3686 /* If the slave isn't UP, use default transmit policy. */
3695 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3698 * This helper function exists to help dev_pick_tx get the correct
3699 * destination queue. Using a helper function skips a call to
3700 * skb_tx_hash and will put the skbs in the queue we expect on their
3701 * way down to the bonding driver.
3703 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3706 * Save the original txq to restore before passing to the driver
3708 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3710 if (unlikely(txq >= dev->real_num_tx_queues)) {
3712 txq -= dev->real_num_tx_queues;
3713 } while (txq >= dev->real_num_tx_queues);
3718 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3720 struct bonding *bond = netdev_priv(dev);
3722 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3723 if (!bond_slave_override(bond, skb))
3724 return NETDEV_TX_OK;
3727 switch (bond->params.mode) {
3728 case BOND_MODE_ROUNDROBIN:
3729 return bond_xmit_roundrobin(skb, dev);
3730 case BOND_MODE_ACTIVEBACKUP:
3731 return bond_xmit_activebackup(skb, dev);
3733 return bond_xmit_xor(skb, dev);
3734 case BOND_MODE_BROADCAST:
3735 return bond_xmit_broadcast(skb, dev);
3736 case BOND_MODE_8023AD:
3737 return bond_3ad_xmit_xor(skb, dev);
3740 return bond_alb_xmit(skb, dev);
3742 /* Should never happen, mode already checked */
3743 pr_err("%s: Error: Unknown bonding mode %d\n",
3744 dev->name, bond->params.mode);
3747 return NETDEV_TX_OK;
3751 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3753 struct bonding *bond = netdev_priv(dev);
3754 netdev_tx_t ret = NETDEV_TX_OK;
3757 * If we risk deadlock from transmitting this in the
3758 * netpoll path, tell netpoll to queue the frame for later tx
3760 if (is_netpoll_tx_blocked(dev))
3761 return NETDEV_TX_BUSY;
3764 if (bond_has_slaves(bond))
3765 ret = __bond_start_xmit(skb, dev);
3773 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3774 struct ethtool_cmd *ecmd)
3776 struct bonding *bond = netdev_priv(bond_dev);
3777 unsigned long speed = 0;
3778 struct list_head *iter;
3779 struct slave *slave;
3781 ecmd->duplex = DUPLEX_UNKNOWN;
3782 ecmd->port = PORT_OTHER;
3784 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3785 * do not need to check mode. Though link speed might not represent
3786 * the true receive or transmit bandwidth (not all modes are symmetric)
3787 * this is an accurate maximum.
3789 read_lock(&bond->lock);
3790 bond_for_each_slave(bond, slave, iter) {
3791 if (SLAVE_IS_OK(slave)) {
3792 if (slave->speed != SPEED_UNKNOWN)
3793 speed += slave->speed;
3794 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3795 slave->duplex != DUPLEX_UNKNOWN)
3796 ecmd->duplex = slave->duplex;
3799 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3800 read_unlock(&bond->lock);
3805 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3806 struct ethtool_drvinfo *drvinfo)
3808 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3809 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3810 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3814 static const struct ethtool_ops bond_ethtool_ops = {
3815 .get_drvinfo = bond_ethtool_get_drvinfo,
3816 .get_settings = bond_ethtool_get_settings,
3817 .get_link = ethtool_op_get_link,
3820 static const struct net_device_ops bond_netdev_ops = {
3821 .ndo_init = bond_init,
3822 .ndo_uninit = bond_uninit,
3823 .ndo_open = bond_open,
3824 .ndo_stop = bond_close,
3825 .ndo_start_xmit = bond_start_xmit,
3826 .ndo_select_queue = bond_select_queue,
3827 .ndo_get_stats64 = bond_get_stats,
3828 .ndo_do_ioctl = bond_do_ioctl,
3829 .ndo_change_rx_flags = bond_change_rx_flags,
3830 .ndo_set_rx_mode = bond_set_rx_mode,
3831 .ndo_change_mtu = bond_change_mtu,
3832 .ndo_set_mac_address = bond_set_mac_address,
3833 .ndo_neigh_setup = bond_neigh_setup,
3834 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3835 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3836 #ifdef CONFIG_NET_POLL_CONTROLLER
3837 .ndo_netpoll_setup = bond_netpoll_setup,
3838 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3839 .ndo_poll_controller = bond_poll_controller,
3841 .ndo_add_slave = bond_enslave,
3842 .ndo_del_slave = bond_release,
3843 .ndo_fix_features = bond_fix_features,
3846 static const struct device_type bond_type = {
3850 static void bond_destructor(struct net_device *bond_dev)
3852 struct bonding *bond = netdev_priv(bond_dev);
3854 destroy_workqueue(bond->wq);
3855 free_netdev(bond_dev);
3858 void bond_setup(struct net_device *bond_dev)
3860 struct bonding *bond = netdev_priv(bond_dev);
3862 /* initialize rwlocks */
3863 rwlock_init(&bond->lock);
3864 rwlock_init(&bond->curr_slave_lock);
3865 bond->params = bonding_defaults;
3867 /* Initialize pointers */
3868 bond->dev = bond_dev;
3870 /* Initialize the device entry points */
3871 ether_setup(bond_dev);
3872 bond_dev->netdev_ops = &bond_netdev_ops;
3873 bond_dev->ethtool_ops = &bond_ethtool_ops;
3875 bond_dev->destructor = bond_destructor;
3877 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3879 /* Initialize the device options */
3880 bond_dev->tx_queue_len = 0;
3881 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3882 bond_dev->priv_flags |= IFF_BONDING;
3883 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3885 /* At first, we block adding VLANs. That's the only way to
3886 * prevent problems that occur when adding VLANs over an
3887 * empty bond. The block will be removed once non-challenged
3888 * slaves are enslaved.
3890 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3892 /* don't acquire bond device's netif_tx_lock when
3894 bond_dev->features |= NETIF_F_LLTX;
3896 /* By default, we declare the bond to be fully
3897 * VLAN hardware accelerated capable. Special
3898 * care is taken in the various xmit functions
3899 * when there are slaves that are not hw accel
3903 bond_dev->hw_features = BOND_VLAN_FEATURES |
3904 NETIF_F_HW_VLAN_CTAG_TX |
3905 NETIF_F_HW_VLAN_CTAG_RX |
3906 NETIF_F_HW_VLAN_CTAG_FILTER;
3908 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3909 bond_dev->features |= bond_dev->hw_features;
3913 * Destroy a bonding device.
3914 * Must be under rtnl_lock when this function is called.
3916 static void bond_uninit(struct net_device *bond_dev)
3918 struct bonding *bond = netdev_priv(bond_dev);
3919 struct list_head *iter;
3920 struct slave *slave;
3922 bond_netpoll_cleanup(bond_dev);
3924 /* Release the bonded slaves */
3925 bond_for_each_slave(bond, slave, iter)
3926 __bond_release_one(bond_dev, slave->dev, true);
3927 pr_info("%s: released all slaves\n", bond_dev->name);
3929 list_del(&bond->bond_list);
3931 bond_debug_unregister(bond);
3934 /*------------------------- Module initialization ---------------------------*/
3936 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3940 for (i = 0; tbl[i].modename; i++)
3941 if (mode == tbl[i].mode)
3947 static int bond_parm_tbl_lookup_name(const char *modename,
3948 const struct bond_parm_tbl *tbl)
3952 for (i = 0; tbl[i].modename; i++)
3953 if (strcmp(modename, tbl[i].modename) == 0)
3960 * Convert string input module parms. Accept either the
3961 * number of the mode or its string name. A bit complicated because
3962 * some mode names are substrings of other names, and calls from sysfs
3963 * may have whitespace in the name (trailing newlines, for example).
3965 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3968 char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
3970 for (p = (char *)buf; *p; p++)
3971 if (!(isdigit(*p) || isspace(*p)))
3974 if (*p && sscanf(buf, "%20s", modestr) != 0)
3975 return bond_parm_tbl_lookup_name(modestr, tbl);
3976 else if (sscanf(buf, "%d", &modeint) != 0)
3977 return bond_parm_tbl_lookup(modeint, tbl);
3982 static int bond_check_params(struct bond_params *params)
3984 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3985 int arp_all_targets_value;
3988 * Convert string parameters.
3991 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
3992 if (bond_mode == -1) {
3993 pr_err("Error: Invalid bonding mode \"%s\"\n",
3994 mode == NULL ? "NULL" : mode);
3999 if (xmit_hash_policy) {
4000 if ((bond_mode != BOND_MODE_XOR) &&
4001 (bond_mode != BOND_MODE_8023AD)) {
4002 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4003 bond_mode_name(bond_mode));
4005 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4007 if (xmit_hashtype == -1) {
4008 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4009 xmit_hash_policy == NULL ? "NULL" :
4017 if (bond_mode != BOND_MODE_8023AD) {
4018 pr_info("lacp_rate param is irrelevant in mode %s\n",
4019 bond_mode_name(bond_mode));
4021 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4022 if (lacp_fast == -1) {
4023 pr_err("Error: Invalid lacp rate \"%s\"\n",
4024 lacp_rate == NULL ? "NULL" : lacp_rate);
4031 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4032 if (params->ad_select == -1) {
4033 pr_err("Error: Invalid ad_select \"%s\"\n",
4034 ad_select == NULL ? "NULL" : ad_select);
4038 if (bond_mode != BOND_MODE_8023AD) {
4039 pr_warning("ad_select param only affects 802.3ad mode\n");
4042 params->ad_select = BOND_AD_STABLE;
4045 if (max_bonds < 0) {
4046 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4047 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4048 max_bonds = BOND_DEFAULT_MAX_BONDS;
4052 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4053 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4054 miimon = BOND_LINK_MON_INTERV;
4058 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4063 if (downdelay < 0) {
4064 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4065 downdelay, INT_MAX);
4069 if ((use_carrier != 0) && (use_carrier != 1)) {
4070 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4075 if (num_peer_notif < 0 || num_peer_notif > 255) {
4076 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4081 /* reset values for 802.3ad/TLB/ALB */
4082 if (BOND_NO_USES_ARP(bond_mode)) {
4084 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4085 pr_warning("Forcing miimon to 100msec\n");
4086 miimon = BOND_DEFAULT_MIIMON;
4090 if (tx_queues < 1 || tx_queues > 255) {
4091 pr_warning("Warning: tx_queues (%d) should be between "
4092 "1 and 255, resetting to %d\n",
4093 tx_queues, BOND_DEFAULT_TX_QUEUES);
4094 tx_queues = BOND_DEFAULT_TX_QUEUES;
4097 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4098 pr_warning("Warning: all_slaves_active module parameter (%d), "
4099 "not of valid value (0/1), so it was set to "
4100 "0\n", all_slaves_active);
4101 all_slaves_active = 0;
4104 if (resend_igmp < 0 || resend_igmp > 255) {
4105 pr_warning("Warning: resend_igmp (%d) should be between "
4106 "0 and 255, resetting to %d\n",
4107 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4108 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4111 if (packets_per_slave < 0 || packets_per_slave > USHRT_MAX) {
4112 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4113 packets_per_slave, USHRT_MAX);
4114 packets_per_slave = 1;
4117 if (bond_mode == BOND_MODE_ALB) {
4118 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",
4123 if (updelay || downdelay) {
4124 /* just warn the user the up/down delay will have
4125 * no effect since miimon is zero...
4127 pr_warning("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",
4128 updelay, downdelay);
4131 /* don't allow arp monitoring */
4133 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4134 miimon, arp_interval);
4138 if ((updelay % miimon) != 0) {
4139 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4141 (updelay / miimon) * miimon);
4146 if ((downdelay % miimon) != 0) {
4147 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4149 (downdelay / miimon) * miimon);
4152 downdelay /= miimon;
4155 if (arp_interval < 0) {
4156 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4157 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4158 arp_interval = BOND_LINK_ARP_INTERV;
4161 for (arp_ip_count = 0, i = 0;
4162 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4163 /* not complete check, but should be good enough to
4166 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4167 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4168 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4172 if (bond_get_targets_ip(arp_target, ip) == -1)
4173 arp_target[arp_ip_count++] = ip;
4175 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4180 if (arp_interval && !arp_ip_count) {
4181 /* don't allow arping if no arp_ip_target given... */
4182 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4188 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4189 pr_err("arp_validate only supported in active-backup mode\n");
4192 if (!arp_interval) {
4193 pr_err("arp_validate requires arp_interval\n");
4197 arp_validate_value = bond_parse_parm(arp_validate,
4199 if (arp_validate_value == -1) {
4200 pr_err("Error: invalid arp_validate \"%s\"\n",
4201 arp_validate == NULL ? "NULL" : arp_validate);
4205 arp_validate_value = 0;
4207 arp_all_targets_value = 0;
4208 if (arp_all_targets) {
4209 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4210 arp_all_targets_tbl);
4212 if (arp_all_targets_value == -1) {
4213 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4215 arp_all_targets_value = 0;
4220 pr_info("MII link monitoring set to %d ms\n", miimon);
4221 } else if (arp_interval) {
4222 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4224 arp_validate_tbl[arp_validate_value].modename,
4227 for (i = 0; i < arp_ip_count; i++)
4228 pr_info(" %s", arp_ip_target[i]);
4232 } else if (max_bonds) {
4233 /* miimon and arp_interval not set, we need one so things
4234 * work as expected, see bonding.txt for details
4236 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");
4239 if (primary && !USES_PRIMARY(bond_mode)) {
4240 /* currently, using a primary only makes sense
4241 * in active backup, TLB or ALB modes
4243 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4244 primary, bond_mode_name(bond_mode));
4248 if (primary && primary_reselect) {
4249 primary_reselect_value = bond_parse_parm(primary_reselect,
4251 if (primary_reselect_value == -1) {
4252 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4254 NULL ? "NULL" : primary_reselect);
4258 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4261 if (fail_over_mac) {
4262 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4264 if (fail_over_mac_value == -1) {
4265 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4266 arp_validate == NULL ? "NULL" : arp_validate);
4270 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4271 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4273 fail_over_mac_value = BOND_FOM_NONE;
4276 if (lp_interval == 0) {
4277 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4278 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4279 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4282 /* fill params struct with the proper values */
4283 params->mode = bond_mode;
4284 params->xmit_policy = xmit_hashtype;
4285 params->miimon = miimon;
4286 params->num_peer_notif = num_peer_notif;
4287 params->arp_interval = arp_interval;
4288 params->arp_validate = arp_validate_value;
4289 params->arp_all_targets = arp_all_targets_value;
4290 params->updelay = updelay;
4291 params->downdelay = downdelay;
4292 params->use_carrier = use_carrier;
4293 params->lacp_fast = lacp_fast;
4294 params->primary[0] = 0;
4295 params->primary_reselect = primary_reselect_value;
4296 params->fail_over_mac = fail_over_mac_value;
4297 params->tx_queues = tx_queues;
4298 params->all_slaves_active = all_slaves_active;
4299 params->resend_igmp = resend_igmp;
4300 params->min_links = min_links;
4301 params->lp_interval = lp_interval;
4302 if (packets_per_slave > 1)
4303 params->packets_per_slave = reciprocal_value(packets_per_slave);
4305 params->packets_per_slave = packets_per_slave;
4307 strncpy(params->primary, primary, IFNAMSIZ);
4308 params->primary[IFNAMSIZ - 1] = 0;
4311 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4316 static struct lock_class_key bonding_netdev_xmit_lock_key;
4317 static struct lock_class_key bonding_netdev_addr_lock_key;
4318 static struct lock_class_key bonding_tx_busylock_key;
4320 static void bond_set_lockdep_class_one(struct net_device *dev,
4321 struct netdev_queue *txq,
4324 lockdep_set_class(&txq->_xmit_lock,
4325 &bonding_netdev_xmit_lock_key);
4328 static void bond_set_lockdep_class(struct net_device *dev)
4330 lockdep_set_class(&dev->addr_list_lock,
4331 &bonding_netdev_addr_lock_key);
4332 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4333 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4337 * Called from registration process
4339 static int bond_init(struct net_device *bond_dev)
4341 struct bonding *bond = netdev_priv(bond_dev);
4342 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4343 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4345 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4348 * Initialize locks that may be required during
4349 * en/deslave operations. All of the bond_open work
4350 * (of which this is part) should really be moved to
4351 * a phase prior to dev_open
4353 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4354 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4356 bond->wq = create_singlethread_workqueue(bond_dev->name);
4360 bond_set_lockdep_class(bond_dev);
4362 list_add_tail(&bond->bond_list, &bn->dev_list);
4364 bond_prepare_sysfs_group(bond);
4366 bond_debug_register(bond);
4368 /* Ensure valid dev_addr */
4369 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4370 bond_dev->addr_assign_type == NET_ADDR_PERM)
4371 eth_hw_addr_random(bond_dev);
4376 unsigned int bond_get_num_tx_queues(void)
4381 /* Create a new bond based on the specified name and bonding parameters.
4382 * If name is NULL, obtain a suitable "bond%d" name for us.
4383 * Caller must NOT hold rtnl_lock; we need to release it here before we
4384 * set up our sysfs entries.
4386 int bond_create(struct net *net, const char *name)
4388 struct net_device *bond_dev;
4393 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4394 name ? name : "bond%d",
4395 bond_setup, tx_queues);
4397 pr_err("%s: eek! can't alloc netdev!\n", name);
4402 dev_net_set(bond_dev, net);
4403 bond_dev->rtnl_link_ops = &bond_link_ops;
4405 res = register_netdevice(bond_dev);
4407 netif_carrier_off(bond_dev);
4411 bond_destructor(bond_dev);
4415 static int __net_init bond_net_init(struct net *net)
4417 struct bond_net *bn = net_generic(net, bond_net_id);
4420 INIT_LIST_HEAD(&bn->dev_list);
4422 bond_create_proc_dir(bn);
4423 bond_create_sysfs(bn);
4428 static void __net_exit bond_net_exit(struct net *net)
4430 struct bond_net *bn = net_generic(net, bond_net_id);
4431 struct bonding *bond, *tmp_bond;
4434 bond_destroy_sysfs(bn);
4435 bond_destroy_proc_dir(bn);
4437 /* Kill off any bonds created after unregistering bond rtnl ops */
4439 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4440 unregister_netdevice_queue(bond->dev, &list);
4441 unregister_netdevice_many(&list);
4445 static struct pernet_operations bond_net_ops = {
4446 .init = bond_net_init,
4447 .exit = bond_net_exit,
4449 .size = sizeof(struct bond_net),
4452 static int __init bonding_init(void)
4457 pr_info("%s", bond_version);
4459 res = bond_check_params(&bonding_defaults);
4463 res = register_pernet_subsys(&bond_net_ops);
4467 res = bond_netlink_init();
4471 bond_create_debugfs();
4473 for (i = 0; i < max_bonds; i++) {
4474 res = bond_create(&init_net, NULL);
4479 register_netdevice_notifier(&bond_netdev_notifier);
4483 bond_netlink_fini();
4485 unregister_pernet_subsys(&bond_net_ops);
4490 static void __exit bonding_exit(void)
4492 unregister_netdevice_notifier(&bond_netdev_notifier);
4494 bond_destroy_debugfs();
4496 bond_netlink_fini();
4497 unregister_pernet_subsys(&bond_net_ops);
4499 #ifdef CONFIG_NET_POLL_CONTROLLER
4501 * Make sure we don't have an imbalance on our netpoll blocking
4503 WARN_ON(atomic_read(&netpoll_block_tx));
4507 module_init(bonding_init);
4508 module_exit(bonding_exit);
4509 MODULE_LICENSE("GPL");
4510 MODULE_VERSION(DRV_VERSION);
4511 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4512 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");