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;
117 module_param(max_bonds, int, 0);
118 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
119 module_param(tx_queues, int, 0);
120 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
121 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
122 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
123 "failover event (alias of num_unsol_na)");
124 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
125 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
126 "failover event (alias of num_grat_arp)");
127 module_param(miimon, int, 0);
128 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
129 module_param(updelay, int, 0);
130 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
131 module_param(downdelay, int, 0);
132 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
134 module_param(use_carrier, int, 0);
135 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
136 "0 for off, 1 for on (default)");
137 module_param(mode, charp, 0);
138 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
139 "1 for active-backup, 2 for balance-xor, "
140 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
141 "6 for balance-alb");
142 module_param(primary, charp, 0);
143 MODULE_PARM_DESC(primary, "Primary network device to use");
144 module_param(primary_reselect, charp, 0);
145 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
147 "0 for always (default), "
148 "1 for only if speed of primary is "
150 "2 for only on active slave "
152 module_param(lacp_rate, charp, 0);
153 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
154 "0 for slow, 1 for fast");
155 module_param(ad_select, charp, 0);
156 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
157 "0 for stable (default), 1 for bandwidth, "
159 module_param(min_links, int, 0);
160 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
162 module_param(xmit_hash_policy, charp, 0);
163 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
164 "0 for layer 2 (default), 1 for layer 3+4, "
165 "2 for layer 2+3, 3 for encap layer 2+3, "
166 "4 for encap layer 3+4");
167 module_param(arp_interval, int, 0);
168 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
169 module_param_array(arp_ip_target, charp, NULL, 0);
170 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
171 module_param(arp_validate, charp, 0);
172 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
173 "0 for none (default), 1 for active, "
174 "2 for backup, 3 for all");
175 module_param(arp_all_targets, charp, 0);
176 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
177 module_param(fail_over_mac, charp, 0);
178 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
179 "the same MAC; 0 for none (default), "
180 "1 for active, 2 for follow");
181 module_param(all_slaves_active, int, 0);
182 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
183 "by setting active flag for all slaves; "
184 "0 for never (default), 1 for always.");
185 module_param(resend_igmp, int, 0);
186 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
188 module_param(packets_per_slave, int, 0);
189 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
190 "mode; 0 for a random slave, 1 packet per "
191 "slave (default), >1 packets per slave.");
193 /*----------------------------- Global variables ----------------------------*/
195 #ifdef CONFIG_NET_POLL_CONTROLLER
196 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
199 int bond_net_id __read_mostly;
201 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
202 static int arp_ip_count;
203 static int bond_mode = BOND_MODE_ROUNDROBIN;
204 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
205 static int lacp_fast;
207 const struct bond_parm_tbl bond_lacp_tbl[] = {
208 { "slow", AD_LACP_SLOW},
209 { "fast", AD_LACP_FAST},
213 const struct bond_parm_tbl bond_mode_tbl[] = {
214 { "balance-rr", BOND_MODE_ROUNDROBIN},
215 { "active-backup", BOND_MODE_ACTIVEBACKUP},
216 { "balance-xor", BOND_MODE_XOR},
217 { "broadcast", BOND_MODE_BROADCAST},
218 { "802.3ad", BOND_MODE_8023AD},
219 { "balance-tlb", BOND_MODE_TLB},
220 { "balance-alb", BOND_MODE_ALB},
224 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
225 { "layer2", BOND_XMIT_POLICY_LAYER2},
226 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
227 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
228 { "encap2+3", BOND_XMIT_POLICY_ENCAP23},
229 { "encap3+4", BOND_XMIT_POLICY_ENCAP34},
233 const struct bond_parm_tbl arp_all_targets_tbl[] = {
234 { "any", BOND_ARP_TARGETS_ANY},
235 { "all", BOND_ARP_TARGETS_ALL},
239 const struct bond_parm_tbl arp_validate_tbl[] = {
240 { "none", BOND_ARP_VALIDATE_NONE},
241 { "active", BOND_ARP_VALIDATE_ACTIVE},
242 { "backup", BOND_ARP_VALIDATE_BACKUP},
243 { "all", BOND_ARP_VALIDATE_ALL},
247 const struct bond_parm_tbl fail_over_mac_tbl[] = {
248 { "none", BOND_FOM_NONE},
249 { "active", BOND_FOM_ACTIVE},
250 { "follow", BOND_FOM_FOLLOW},
254 const struct bond_parm_tbl pri_reselect_tbl[] = {
255 { "always", BOND_PRI_RESELECT_ALWAYS},
256 { "better", BOND_PRI_RESELECT_BETTER},
257 { "failure", BOND_PRI_RESELECT_FAILURE},
261 struct bond_parm_tbl ad_select_tbl[] = {
262 { "stable", BOND_AD_STABLE},
263 { "bandwidth", BOND_AD_BANDWIDTH},
264 { "count", BOND_AD_COUNT},
268 /*-------------------------- Forward declarations ---------------------------*/
270 static int bond_init(struct net_device *bond_dev);
271 static void bond_uninit(struct net_device *bond_dev);
273 /*---------------------------- General routines -----------------------------*/
275 const char *bond_mode_name(int mode)
277 static const char *names[] = {
278 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
279 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
280 [BOND_MODE_XOR] = "load balancing (xor)",
281 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
282 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
283 [BOND_MODE_TLB] = "transmit load balancing",
284 [BOND_MODE_ALB] = "adaptive load balancing",
287 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
293 /*---------------------------------- VLAN -----------------------------------*/
296 * bond_dev_queue_xmit - Prepare skb for xmit.
298 * @bond: bond device that got this skb for tx.
299 * @skb: hw accel VLAN tagged skb to transmit
300 * @slave_dev: slave that is supposed to xmit this skbuff
302 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
303 struct net_device *slave_dev)
305 skb->dev = slave_dev;
307 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
308 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
309 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
311 if (unlikely(netpoll_tx_running(bond->dev)))
312 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
320 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
321 * We don't protect the slave list iteration with a lock because:
322 * a. This operation is performed in IOCTL context,
323 * b. The operation is protected by the RTNL semaphore in the 8021q code,
324 * c. Holding a lock with BH disabled while directly calling a base driver
325 * entry point is generally a BAD idea.
327 * The design of synchronization/protection for this operation in the 8021q
328 * module is good for one or more VLAN devices over a single physical device
329 * and cannot be extended for a teaming solution like bonding, so there is a
330 * potential race condition here where a net device from the vlan group might
331 * be referenced (either by a base driver or the 8021q code) while it is being
332 * removed from the system. However, it turns out we're not making matters
333 * worse, and if it works for regular VLAN usage it will work here too.
337 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
338 * @bond_dev: bonding net device that got called
339 * @vid: vlan id being added
341 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
342 __be16 proto, u16 vid)
344 struct bonding *bond = netdev_priv(bond_dev);
345 struct slave *slave, *rollback_slave;
346 struct list_head *iter;
349 bond_for_each_slave(bond, slave, iter) {
350 res = vlan_vid_add(slave->dev, proto, vid);
358 /* unwind to the slave that failed */
359 bond_for_each_slave(bond, rollback_slave, iter) {
360 if (rollback_slave == slave)
363 vlan_vid_del(rollback_slave->dev, proto, vid);
370 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
371 * @bond_dev: bonding net device that got called
372 * @vid: vlan id being removed
374 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
375 __be16 proto, u16 vid)
377 struct bonding *bond = netdev_priv(bond_dev);
378 struct list_head *iter;
381 bond_for_each_slave(bond, slave, iter)
382 vlan_vid_del(slave->dev, proto, vid);
384 if (bond_is_lb(bond))
385 bond_alb_clear_vlan(bond, vid);
390 /*------------------------------- Link status -------------------------------*/
393 * Set the carrier state for the master according to the state of its
394 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
395 * do special 802.3ad magic.
397 * Returns zero if carrier state does not change, nonzero if it does.
399 static int bond_set_carrier(struct bonding *bond)
401 struct list_head *iter;
404 if (!bond_has_slaves(bond))
407 if (bond->params.mode == BOND_MODE_8023AD)
408 return bond_3ad_set_carrier(bond);
410 bond_for_each_slave(bond, slave, iter) {
411 if (slave->link == BOND_LINK_UP) {
412 if (!netif_carrier_ok(bond->dev)) {
413 netif_carrier_on(bond->dev);
421 if (netif_carrier_ok(bond->dev)) {
422 netif_carrier_off(bond->dev);
429 * Get link speed and duplex from the slave's base driver
430 * using ethtool. If for some reason the call fails or the
431 * values are invalid, set speed and duplex to -1,
434 static void bond_update_speed_duplex(struct slave *slave)
436 struct net_device *slave_dev = slave->dev;
437 struct ethtool_cmd ecmd;
441 slave->speed = SPEED_UNKNOWN;
442 slave->duplex = DUPLEX_UNKNOWN;
444 res = __ethtool_get_settings(slave_dev, &ecmd);
448 slave_speed = ethtool_cmd_speed(&ecmd);
449 if (slave_speed == 0 || slave_speed == ((__u32) -1))
452 switch (ecmd.duplex) {
460 slave->speed = slave_speed;
461 slave->duplex = ecmd.duplex;
467 * if <dev> supports MII link status reporting, check its link status.
469 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
470 * depending upon the setting of the use_carrier parameter.
472 * Return either BMSR_LSTATUS, meaning that the link is up (or we
473 * can't tell and just pretend it is), or 0, meaning that the link is
476 * If reporting is non-zero, instead of faking link up, return -1 if
477 * both ETHTOOL and MII ioctls fail (meaning the device does not
478 * support them). If use_carrier is set, return whatever it says.
479 * It'd be nice if there was a good way to tell if a driver supports
480 * netif_carrier, but there really isn't.
482 static int bond_check_dev_link(struct bonding *bond,
483 struct net_device *slave_dev, int reporting)
485 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
486 int (*ioctl)(struct net_device *, struct ifreq *, int);
488 struct mii_ioctl_data *mii;
490 if (!reporting && !netif_running(slave_dev))
493 if (bond->params.use_carrier)
494 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
496 /* Try to get link status using Ethtool first. */
497 if (slave_dev->ethtool_ops->get_link)
498 return slave_dev->ethtool_ops->get_link(slave_dev) ?
501 /* Ethtool can't be used, fallback to MII ioctls. */
502 ioctl = slave_ops->ndo_do_ioctl;
504 /* TODO: set pointer to correct ioctl on a per team member */
505 /* bases to make this more efficient. that is, once */
506 /* we determine the correct ioctl, we will always */
507 /* call it and not the others for that team */
511 * We cannot assume that SIOCGMIIPHY will also read a
512 * register; not all network drivers (e.g., e100)
516 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
517 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
519 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
520 mii->reg_num = MII_BMSR;
521 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
522 return mii->val_out & BMSR_LSTATUS;
527 * If reporting, report that either there's no dev->do_ioctl,
528 * or both SIOCGMIIREG and get_link failed (meaning that we
529 * cannot report link status). If not reporting, pretend
532 return reporting ? -1 : BMSR_LSTATUS;
535 /*----------------------------- Multicast list ------------------------------*/
538 * Push the promiscuity flag down to appropriate slaves
540 static int bond_set_promiscuity(struct bonding *bond, int inc)
542 struct list_head *iter;
545 if (USES_PRIMARY(bond->params.mode)) {
546 /* write lock already acquired */
547 if (bond->curr_active_slave) {
548 err = dev_set_promiscuity(bond->curr_active_slave->dev,
554 bond_for_each_slave(bond, slave, iter) {
555 err = dev_set_promiscuity(slave->dev, inc);
564 * Push the allmulti flag down to all slaves
566 static int bond_set_allmulti(struct bonding *bond, int inc)
568 struct list_head *iter;
571 if (USES_PRIMARY(bond->params.mode)) {
572 /* write lock already acquired */
573 if (bond->curr_active_slave) {
574 err = dev_set_allmulti(bond->curr_active_slave->dev,
580 bond_for_each_slave(bond, slave, iter) {
581 err = dev_set_allmulti(slave->dev, inc);
590 * Retrieve the list of registered multicast addresses for the bonding
591 * device and retransmit an IGMP JOIN request to the current active
594 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
596 struct bonding *bond = container_of(work, struct bonding,
599 if (!rtnl_trylock()) {
600 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
603 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
605 if (bond->igmp_retrans > 1) {
606 bond->igmp_retrans--;
607 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
612 /* Flush bond's hardware addresses from slave
614 static void bond_hw_addr_flush(struct net_device *bond_dev,
615 struct net_device *slave_dev)
617 struct bonding *bond = netdev_priv(bond_dev);
619 dev_uc_unsync(slave_dev, bond_dev);
620 dev_mc_unsync(slave_dev, bond_dev);
622 if (bond->params.mode == BOND_MODE_8023AD) {
623 /* del lacpdu mc addr from mc list */
624 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
626 dev_mc_del(slave_dev, lacpdu_multicast);
630 /*--------------------------- Active slave change ---------------------------*/
632 /* Update the hardware address list and promisc/allmulti for the new and
633 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
634 * slaves up date at all times; only the USES_PRIMARY modes need to call
635 * this function to swap these settings during a failover.
637 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
638 struct slave *old_active)
643 if (bond->dev->flags & IFF_PROMISC)
644 dev_set_promiscuity(old_active->dev, -1);
646 if (bond->dev->flags & IFF_ALLMULTI)
647 dev_set_allmulti(old_active->dev, -1);
649 bond_hw_addr_flush(bond->dev, old_active->dev);
653 /* FIXME: Signal errors upstream. */
654 if (bond->dev->flags & IFF_PROMISC)
655 dev_set_promiscuity(new_active->dev, 1);
657 if (bond->dev->flags & IFF_ALLMULTI)
658 dev_set_allmulti(new_active->dev, 1);
660 netif_addr_lock_bh(bond->dev);
661 dev_uc_sync(new_active->dev, bond->dev);
662 dev_mc_sync(new_active->dev, bond->dev);
663 netif_addr_unlock_bh(bond->dev);
668 * bond_set_dev_addr - clone slave's address to bond
669 * @bond_dev: bond net device
670 * @slave_dev: slave net device
672 * Should be called with RTNL held.
674 static void bond_set_dev_addr(struct net_device *bond_dev,
675 struct net_device *slave_dev)
677 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
678 bond_dev, slave_dev, slave_dev->addr_len);
679 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
680 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
681 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
685 * bond_do_fail_over_mac
687 * Perform special MAC address swapping for fail_over_mac settings
689 * Called with RTNL, curr_slave_lock for write_bh.
691 static void bond_do_fail_over_mac(struct bonding *bond,
692 struct slave *new_active,
693 struct slave *old_active)
694 __releases(&bond->curr_slave_lock)
695 __acquires(&bond->curr_slave_lock)
697 u8 tmp_mac[ETH_ALEN];
698 struct sockaddr saddr;
701 switch (bond->params.fail_over_mac) {
702 case BOND_FOM_ACTIVE:
704 write_unlock_bh(&bond->curr_slave_lock);
705 bond_set_dev_addr(bond->dev, new_active->dev);
706 write_lock_bh(&bond->curr_slave_lock);
709 case BOND_FOM_FOLLOW:
711 * if new_active && old_active, swap them
712 * if just old_active, do nothing (going to no active slave)
713 * if just new_active, set new_active to bond's MAC
718 write_unlock_bh(&bond->curr_slave_lock);
721 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
722 memcpy(saddr.sa_data, old_active->dev->dev_addr,
724 saddr.sa_family = new_active->dev->type;
726 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
727 saddr.sa_family = bond->dev->type;
730 rv = dev_set_mac_address(new_active->dev, &saddr);
732 pr_err("%s: Error %d setting MAC of slave %s\n",
733 bond->dev->name, -rv, new_active->dev->name);
740 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
741 saddr.sa_family = old_active->dev->type;
743 rv = dev_set_mac_address(old_active->dev, &saddr);
745 pr_err("%s: Error %d setting MAC of slave %s\n",
746 bond->dev->name, -rv, new_active->dev->name);
748 write_lock_bh(&bond->curr_slave_lock);
751 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
752 bond->dev->name, bond->params.fail_over_mac);
758 static bool bond_should_change_active(struct bonding *bond)
760 struct slave *prim = bond->primary_slave;
761 struct slave *curr = bond->curr_active_slave;
763 if (!prim || !curr || curr->link != BOND_LINK_UP)
765 if (bond->force_primary) {
766 bond->force_primary = false;
769 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
770 (prim->speed < curr->speed ||
771 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
773 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
779 * find_best_interface - select the best available slave to be the active one
780 * @bond: our bonding struct
782 static struct slave *bond_find_best_slave(struct bonding *bond)
784 struct slave *slave, *bestslave = NULL;
785 struct list_head *iter;
786 int mintime = bond->params.updelay;
788 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
789 bond_should_change_active(bond))
790 return bond->primary_slave;
792 bond_for_each_slave(bond, slave, iter) {
793 if (slave->link == BOND_LINK_UP)
795 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
796 slave->delay < mintime) {
797 mintime = slave->delay;
805 static bool bond_should_notify_peers(struct bonding *bond)
810 slave = rcu_dereference(bond->curr_active_slave);
813 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
814 bond->dev->name, slave ? slave->dev->name : "NULL");
816 if (!slave || !bond->send_peer_notif ||
817 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
824 * change_active_interface - change the active slave into the specified one
825 * @bond: our bonding struct
826 * @new: the new slave to make the active one
828 * Set the new slave to the bond's settings and unset them on the old
830 * Setting include flags, mc-list, promiscuity, allmulti, etc.
832 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
833 * because it is apparently the best available slave we have, even though its
834 * updelay hasn't timed out yet.
836 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
838 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
840 struct slave *old_active = bond->curr_active_slave;
842 if (old_active == new_active)
846 new_active->jiffies = jiffies;
848 if (new_active->link == BOND_LINK_BACK) {
849 if (USES_PRIMARY(bond->params.mode)) {
850 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
851 bond->dev->name, new_active->dev->name,
852 (bond->params.updelay - new_active->delay) * bond->params.miimon);
855 new_active->delay = 0;
856 new_active->link = BOND_LINK_UP;
858 if (bond->params.mode == BOND_MODE_8023AD)
859 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
861 if (bond_is_lb(bond))
862 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
864 if (USES_PRIMARY(bond->params.mode)) {
865 pr_info("%s: making interface %s the new active one.\n",
866 bond->dev->name, new_active->dev->name);
871 if (USES_PRIMARY(bond->params.mode))
872 bond_hw_addr_swap(bond, new_active, old_active);
874 if (bond_is_lb(bond)) {
875 bond_alb_handle_active_change(bond, new_active);
877 bond_set_slave_inactive_flags(old_active);
879 bond_set_slave_active_flags(new_active);
881 rcu_assign_pointer(bond->curr_active_slave, new_active);
884 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
886 bond_set_slave_inactive_flags(old_active);
889 bool should_notify_peers = false;
891 bond_set_slave_active_flags(new_active);
893 if (bond->params.fail_over_mac)
894 bond_do_fail_over_mac(bond, new_active,
897 if (netif_running(bond->dev)) {
898 bond->send_peer_notif =
899 bond->params.num_peer_notif;
900 should_notify_peers =
901 bond_should_notify_peers(bond);
904 write_unlock_bh(&bond->curr_slave_lock);
906 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
907 if (should_notify_peers)
908 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
911 write_lock_bh(&bond->curr_slave_lock);
915 /* resend IGMP joins since active slave has changed or
916 * all were sent on curr_active_slave.
917 * resend only if bond is brought up with the affected
918 * bonding modes and the retransmission is enabled */
919 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
920 ((USES_PRIMARY(bond->params.mode) && new_active) ||
921 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
922 bond->igmp_retrans = bond->params.resend_igmp;
923 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
928 * bond_select_active_slave - select a new active slave, if needed
929 * @bond: our bonding struct
931 * This functions should be called when one of the following occurs:
932 * - The old curr_active_slave has been released or lost its link.
933 * - The primary_slave has got its link back.
934 * - A slave has got its link back and there's no old curr_active_slave.
936 * Caller must hold curr_slave_lock for write_bh.
938 void bond_select_active_slave(struct bonding *bond)
940 struct slave *best_slave;
943 best_slave = bond_find_best_slave(bond);
944 if (best_slave != bond->curr_active_slave) {
945 bond_change_active_slave(bond, best_slave);
946 rv = bond_set_carrier(bond);
950 if (netif_carrier_ok(bond->dev)) {
951 pr_info("%s: first active interface up!\n",
954 pr_info("%s: now running without any active interface !\n",
960 #ifdef CONFIG_NET_POLL_CONTROLLER
961 static inline int slave_enable_netpoll(struct slave *slave)
966 np = kzalloc(sizeof(*np), GFP_ATOMIC);
971 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
980 static inline void slave_disable_netpoll(struct slave *slave)
982 struct netpoll *np = slave->np;
988 __netpoll_free_async(np);
990 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
992 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
994 if (!slave_dev->netdev_ops->ndo_poll_controller)
999 static void bond_poll_controller(struct net_device *bond_dev)
1003 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1005 struct bonding *bond = netdev_priv(bond_dev);
1006 struct list_head *iter;
1007 struct slave *slave;
1009 bond_for_each_slave(bond, slave, iter)
1010 if (IS_UP(slave->dev))
1011 slave_disable_netpoll(slave);
1014 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1016 struct bonding *bond = netdev_priv(dev);
1017 struct list_head *iter;
1018 struct slave *slave;
1021 bond_for_each_slave(bond, slave, iter) {
1022 err = slave_enable_netpoll(slave);
1024 bond_netpoll_cleanup(dev);
1031 static inline int slave_enable_netpoll(struct slave *slave)
1035 static inline void slave_disable_netpoll(struct slave *slave)
1038 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1043 /*---------------------------------- IOCTL ----------------------------------*/
1045 static netdev_features_t bond_fix_features(struct net_device *dev,
1046 netdev_features_t features)
1048 struct bonding *bond = netdev_priv(dev);
1049 struct list_head *iter;
1050 netdev_features_t mask;
1051 struct slave *slave;
1053 if (!bond_has_slaves(bond)) {
1054 /* Disable adding VLANs to empty bond. But why? --mq */
1055 features |= NETIF_F_VLAN_CHALLENGED;
1060 features &= ~NETIF_F_ONE_FOR_ALL;
1061 features |= NETIF_F_ALL_FOR_ALL;
1063 bond_for_each_slave(bond, slave, iter) {
1064 features = netdev_increment_features(features,
1065 slave->dev->features,
1068 features = netdev_add_tso_features(features, mask);
1073 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1074 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1075 NETIF_F_HIGHDMA | NETIF_F_LRO)
1077 static void bond_compute_features(struct bonding *bond)
1079 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1080 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1081 struct net_device *bond_dev = bond->dev;
1082 struct list_head *iter;
1083 struct slave *slave;
1084 unsigned short max_hard_header_len = ETH_HLEN;
1085 unsigned int gso_max_size = GSO_MAX_SIZE;
1086 u16 gso_max_segs = GSO_MAX_SEGS;
1088 if (!bond_has_slaves(bond))
1091 bond_for_each_slave(bond, slave, iter) {
1092 vlan_features = netdev_increment_features(vlan_features,
1093 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1095 dst_release_flag &= slave->dev->priv_flags;
1096 if (slave->dev->hard_header_len > max_hard_header_len)
1097 max_hard_header_len = slave->dev->hard_header_len;
1099 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1100 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1104 bond_dev->vlan_features = vlan_features;
1105 bond_dev->hard_header_len = max_hard_header_len;
1106 bond_dev->gso_max_segs = gso_max_segs;
1107 netif_set_gso_max_size(bond_dev, gso_max_size);
1109 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1110 bond_dev->priv_flags = flags | dst_release_flag;
1112 netdev_change_features(bond_dev);
1115 static void bond_setup_by_slave(struct net_device *bond_dev,
1116 struct net_device *slave_dev)
1118 bond_dev->header_ops = slave_dev->header_ops;
1120 bond_dev->type = slave_dev->type;
1121 bond_dev->hard_header_len = slave_dev->hard_header_len;
1122 bond_dev->addr_len = slave_dev->addr_len;
1124 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1125 slave_dev->addr_len);
1128 /* On bonding slaves other than the currently active slave, suppress
1129 * duplicates except for alb non-mcast/bcast.
1131 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1132 struct slave *slave,
1133 struct bonding *bond)
1135 if (bond_is_slave_inactive(slave)) {
1136 if (bond->params.mode == BOND_MODE_ALB &&
1137 skb->pkt_type != PACKET_BROADCAST &&
1138 skb->pkt_type != PACKET_MULTICAST)
1145 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1147 struct sk_buff *skb = *pskb;
1148 struct slave *slave;
1149 struct bonding *bond;
1150 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1152 int ret = RX_HANDLER_ANOTHER;
1154 skb = skb_share_check(skb, GFP_ATOMIC);
1156 return RX_HANDLER_CONSUMED;
1160 slave = bond_slave_get_rcu(skb->dev);
1163 if (bond->params.arp_interval)
1164 slave->dev->last_rx = jiffies;
1166 recv_probe = ACCESS_ONCE(bond->recv_probe);
1168 ret = recv_probe(skb, bond, slave);
1169 if (ret == RX_HANDLER_CONSUMED) {
1175 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1176 return RX_HANDLER_EXACT;
1179 skb->dev = bond->dev;
1181 if (bond->params.mode == BOND_MODE_ALB &&
1182 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1183 skb->pkt_type == PACKET_HOST) {
1185 if (unlikely(skb_cow_head(skb,
1186 skb->data - skb_mac_header(skb)))) {
1188 return RX_HANDLER_CONSUMED;
1190 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1196 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1197 struct net_device *slave_dev,
1198 struct slave *slave)
1202 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1205 slave_dev->flags |= IFF_SLAVE;
1206 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1210 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1211 struct net_device *slave_dev)
1213 netdev_upper_dev_unlink(slave_dev, bond_dev);
1214 slave_dev->flags &= ~IFF_SLAVE;
1215 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1218 /* enslave device <slave> to bond device <master> */
1219 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1221 struct bonding *bond = netdev_priv(bond_dev);
1222 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1223 struct slave *new_slave = NULL, *prev_slave;
1224 struct sockaddr addr;
1228 if (!bond->params.use_carrier &&
1229 slave_dev->ethtool_ops->get_link == NULL &&
1230 slave_ops->ndo_do_ioctl == NULL) {
1231 pr_warning("%s: Warning: no link monitoring support for %s\n",
1232 bond_dev->name, slave_dev->name);
1235 /* already enslaved */
1236 if (slave_dev->flags & IFF_SLAVE) {
1237 pr_debug("Error, Device was already enslaved\n");
1241 /* vlan challenged mutual exclusion */
1242 /* no need to lock since we're protected by rtnl_lock */
1243 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1244 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1245 if (vlan_uses_dev(bond_dev)) {
1246 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1247 bond_dev->name, slave_dev->name, bond_dev->name);
1250 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1251 bond_dev->name, slave_dev->name,
1252 slave_dev->name, bond_dev->name);
1255 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1259 * Old ifenslave binaries are no longer supported. These can
1260 * be identified with moderate accuracy by the state of the slave:
1261 * the current ifenslave will set the interface down prior to
1262 * enslaving it; the old ifenslave will not.
1264 if ((slave_dev->flags & IFF_UP)) {
1265 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1268 goto err_undo_flags;
1271 /* set bonding device ether type by slave - bonding netdevices are
1272 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1273 * there is a need to override some of the type dependent attribs/funcs.
1275 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1276 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1278 if (!bond_has_slaves(bond)) {
1279 if (bond_dev->type != slave_dev->type) {
1280 pr_debug("%s: change device type from %d to %d\n",
1282 bond_dev->type, slave_dev->type);
1284 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1286 res = notifier_to_errno(res);
1288 pr_err("%s: refused to change device type\n",
1291 goto err_undo_flags;
1294 /* Flush unicast and multicast addresses */
1295 dev_uc_flush(bond_dev);
1296 dev_mc_flush(bond_dev);
1298 if (slave_dev->type != ARPHRD_ETHER)
1299 bond_setup_by_slave(bond_dev, slave_dev);
1301 ether_setup(bond_dev);
1302 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1305 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1308 } else if (bond_dev->type != slave_dev->type) {
1309 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1311 slave_dev->type, bond_dev->type);
1313 goto err_undo_flags;
1316 if (slave_ops->ndo_set_mac_address == NULL) {
1317 if (!bond_has_slaves(bond)) {
1318 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1320 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1321 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1322 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",
1325 goto err_undo_flags;
1329 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1331 /* If this is the first slave, then we need to set the master's hardware
1332 * address to be the same as the slave's. */
1333 if (!bond_has_slaves(bond) &&
1334 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1335 bond_set_dev_addr(bond->dev, slave_dev);
1337 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1340 goto err_undo_flags;
1343 * Set the new_slave's queue_id to be zero. Queue ID mapping
1344 * is set via sysfs or module option if desired.
1346 new_slave->queue_id = 0;
1348 /* Save slave's original mtu and then set it to match the bond */
1349 new_slave->original_mtu = slave_dev->mtu;
1350 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1352 pr_debug("Error %d calling dev_set_mtu\n", res);
1357 * Save slave's original ("permanent") mac address for modes
1358 * that need it, and for restoring it upon release, and then
1359 * set it to the master's address
1361 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1363 if (!bond->params.fail_over_mac) {
1365 * Set slave to master's mac address. The application already
1366 * set the master's mac address to that of the first slave
1368 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1369 addr.sa_family = slave_dev->type;
1370 res = dev_set_mac_address(slave_dev, &addr);
1372 pr_debug("Error %d calling set_mac_address\n", res);
1373 goto err_restore_mtu;
1377 /* open the slave since the application closed it */
1378 res = dev_open(slave_dev);
1380 pr_debug("Opening slave %s failed\n", slave_dev->name);
1381 goto err_restore_mac;
1384 new_slave->bond = bond;
1385 new_slave->dev = slave_dev;
1386 slave_dev->priv_flags |= IFF_BONDING;
1388 if (bond_is_lb(bond)) {
1389 /* bond_alb_init_slave() must be called before all other stages since
1390 * it might fail and we do not want to have to undo everything
1392 res = bond_alb_init_slave(bond, new_slave);
1397 /* If the mode USES_PRIMARY, then the following is handled by
1398 * bond_change_active_slave().
1400 if (!USES_PRIMARY(bond->params.mode)) {
1401 /* set promiscuity level to new slave */
1402 if (bond_dev->flags & IFF_PROMISC) {
1403 res = dev_set_promiscuity(slave_dev, 1);
1408 /* set allmulti level to new slave */
1409 if (bond_dev->flags & IFF_ALLMULTI) {
1410 res = dev_set_allmulti(slave_dev, 1);
1415 netif_addr_lock_bh(bond_dev);
1417 dev_mc_sync_multiple(slave_dev, bond_dev);
1418 dev_uc_sync_multiple(slave_dev, bond_dev);
1420 netif_addr_unlock_bh(bond_dev);
1423 if (bond->params.mode == BOND_MODE_8023AD) {
1424 /* add lacpdu mc addr to mc list */
1425 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1427 dev_mc_add(slave_dev, lacpdu_multicast);
1430 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1432 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1433 bond_dev->name, slave_dev->name);
1437 prev_slave = bond_last_slave(bond);
1439 new_slave->delay = 0;
1440 new_slave->link_failure_count = 0;
1442 bond_update_speed_duplex(new_slave);
1444 new_slave->last_arp_rx = jiffies -
1445 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1446 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1447 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1449 if (bond->params.miimon && !bond->params.use_carrier) {
1450 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1452 if ((link_reporting == -1) && !bond->params.arp_interval) {
1454 * miimon is set but a bonded network driver
1455 * does not support ETHTOOL/MII and
1456 * arp_interval is not set. Note: if
1457 * use_carrier is enabled, we will never go
1458 * here (because netif_carrier is always
1459 * supported); thus, we don't need to change
1460 * the messages for netif_carrier.
1462 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",
1463 bond_dev->name, slave_dev->name);
1464 } else if (link_reporting == -1) {
1465 /* unable get link status using mii/ethtool */
1466 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",
1467 bond_dev->name, slave_dev->name);
1471 /* check for initial state */
1472 if (bond->params.miimon) {
1473 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1474 if (bond->params.updelay) {
1475 new_slave->link = BOND_LINK_BACK;
1476 new_slave->delay = bond->params.updelay;
1478 new_slave->link = BOND_LINK_UP;
1481 new_slave->link = BOND_LINK_DOWN;
1483 } else if (bond->params.arp_interval) {
1484 new_slave->link = (netif_carrier_ok(slave_dev) ?
1485 BOND_LINK_UP : BOND_LINK_DOWN);
1487 new_slave->link = BOND_LINK_UP;
1490 if (new_slave->link != BOND_LINK_DOWN)
1491 new_slave->jiffies = jiffies;
1492 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1493 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1494 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1496 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1497 /* if there is a primary slave, remember it */
1498 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1499 bond->primary_slave = new_slave;
1500 bond->force_primary = true;
1504 switch (bond->params.mode) {
1505 case BOND_MODE_ACTIVEBACKUP:
1506 bond_set_slave_inactive_flags(new_slave);
1508 case BOND_MODE_8023AD:
1509 /* in 802.3ad mode, the internal mechanism
1510 * will activate the slaves in the selected
1513 bond_set_slave_inactive_flags(new_slave);
1514 /* if this is the first slave */
1516 SLAVE_AD_INFO(new_slave).id = 1;
1517 /* Initialize AD with the number of times that the AD timer is called in 1 second
1518 * can be called only after the mac address of the bond is set
1520 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1522 SLAVE_AD_INFO(new_slave).id =
1523 SLAVE_AD_INFO(prev_slave).id + 1;
1526 bond_3ad_bind_slave(new_slave);
1530 bond_set_active_slave(new_slave);
1531 bond_set_slave_inactive_flags(new_slave);
1534 pr_debug("This slave is always active in trunk mode\n");
1536 /* always active in trunk mode */
1537 bond_set_active_slave(new_slave);
1539 /* In trunking mode there is little meaning to curr_active_slave
1540 * anyway (it holds no special properties of the bond device),
1541 * so we can change it without calling change_active_interface()
1543 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1544 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1547 } /* switch(bond_mode) */
1549 #ifdef CONFIG_NET_POLL_CONTROLLER
1550 slave_dev->npinfo = bond->dev->npinfo;
1551 if (slave_dev->npinfo) {
1552 if (slave_enable_netpoll(new_slave)) {
1553 read_unlock(&bond->lock);
1554 pr_info("Error, %s: master_dev is using netpoll, "
1555 "but new slave device does not support netpoll.\n",
1563 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1566 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1570 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1572 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1573 goto err_unregister;
1577 bond_compute_features(bond);
1578 bond_set_carrier(bond);
1580 if (USES_PRIMARY(bond->params.mode)) {
1581 write_lock_bh(&bond->curr_slave_lock);
1582 bond_select_active_slave(bond);
1583 write_unlock_bh(&bond->curr_slave_lock);
1586 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1587 bond_dev->name, slave_dev->name,
1588 bond_is_active_slave(new_slave) ? "n active" : " backup",
1589 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1591 /* enslave is successful */
1594 /* Undo stages on error */
1596 netdev_rx_handler_unregister(slave_dev);
1599 if (!USES_PRIMARY(bond->params.mode))
1600 bond_hw_addr_flush(bond_dev, slave_dev);
1602 vlan_vids_del_by_dev(slave_dev, bond_dev);
1603 if (bond->primary_slave == new_slave)
1604 bond->primary_slave = NULL;
1605 if (bond->curr_active_slave == new_slave) {
1606 write_lock_bh(&bond->curr_slave_lock);
1607 bond_change_active_slave(bond, NULL);
1608 bond_select_active_slave(bond);
1609 write_unlock_bh(&bond->curr_slave_lock);
1611 slave_disable_netpoll(new_slave);
1614 slave_dev->priv_flags &= ~IFF_BONDING;
1615 dev_close(slave_dev);
1618 if (!bond->params.fail_over_mac) {
1619 /* XXX TODO - fom follow mode needs to change master's
1620 * MAC if this slave's MAC is in use by the bond, or at
1621 * least print a warning.
1623 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1624 addr.sa_family = slave_dev->type;
1625 dev_set_mac_address(slave_dev, &addr);
1629 dev_set_mtu(slave_dev, new_slave->original_mtu);
1635 /* Enslave of first slave has failed and we need to fix master's mac */
1636 if (!bond_has_slaves(bond) &&
1637 ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1638 eth_hw_addr_random(bond_dev);
1644 * Try to release the slave device <slave> from the bond device <master>
1645 * It is legal to access curr_active_slave without a lock because all the function
1646 * is write-locked. If "all" is true it means that the function is being called
1647 * while destroying a bond interface and all slaves are being released.
1649 * The rules for slave state should be:
1650 * for Active/Backup:
1651 * Active stays on all backups go down
1652 * for Bonded connections:
1653 * The first up interface should be left on and all others downed.
1655 static int __bond_release_one(struct net_device *bond_dev,
1656 struct net_device *slave_dev,
1659 struct bonding *bond = netdev_priv(bond_dev);
1660 struct slave *slave, *oldcurrent;
1661 struct sockaddr addr;
1662 int old_flags = bond_dev->flags;
1663 netdev_features_t old_features = bond_dev->features;
1665 /* slave is not a slave or master is not master of this slave */
1666 if (!(slave_dev->flags & IFF_SLAVE) ||
1667 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1668 pr_err("%s: Error: cannot release %s.\n",
1669 bond_dev->name, slave_dev->name);
1675 slave = bond_get_slave_by_dev(bond, slave_dev);
1677 /* not a slave of this bond */
1678 pr_info("%s: %s not enslaved\n",
1679 bond_dev->name, slave_dev->name);
1680 unblock_netpoll_tx();
1684 /* release the slave from its bond */
1687 bond_upper_dev_unlink(bond_dev, slave_dev);
1688 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1689 * for this slave anymore.
1691 netdev_rx_handler_unregister(slave_dev);
1692 write_lock_bh(&bond->lock);
1694 /* Inform AD package of unbinding of slave. */
1695 if (bond->params.mode == BOND_MODE_8023AD)
1696 bond_3ad_unbind_slave(slave);
1698 write_unlock_bh(&bond->lock);
1700 pr_info("%s: releasing %s interface %s\n",
1702 bond_is_active_slave(slave) ? "active" : "backup",
1705 oldcurrent = bond->curr_active_slave;
1707 bond->current_arp_slave = NULL;
1709 if (!all && !bond->params.fail_over_mac) {
1710 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1711 bond_has_slaves(bond))
1712 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",
1713 bond_dev->name, slave_dev->name,
1715 bond_dev->name, slave_dev->name);
1718 if (bond->primary_slave == slave)
1719 bond->primary_slave = NULL;
1721 if (oldcurrent == slave) {
1722 write_lock_bh(&bond->curr_slave_lock);
1723 bond_change_active_slave(bond, NULL);
1724 write_unlock_bh(&bond->curr_slave_lock);
1727 if (bond_is_lb(bond)) {
1728 /* Must be called only after the slave has been
1729 * detached from the list and the curr_active_slave
1730 * has been cleared (if our_slave == old_current),
1731 * but before a new active slave is selected.
1733 bond_alb_deinit_slave(bond, slave);
1737 rcu_assign_pointer(bond->curr_active_slave, NULL);
1738 } else if (oldcurrent == slave) {
1740 * Note that we hold RTNL over this sequence, so there
1741 * is no concern that another slave add/remove event
1744 write_lock_bh(&bond->curr_slave_lock);
1746 bond_select_active_slave(bond);
1748 write_unlock_bh(&bond->curr_slave_lock);
1751 if (!bond_has_slaves(bond)) {
1752 bond_set_carrier(bond);
1753 eth_hw_addr_random(bond_dev);
1755 if (vlan_uses_dev(bond_dev)) {
1756 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1757 bond_dev->name, bond_dev->name);
1758 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1763 unblock_netpoll_tx();
1766 if (!bond_has_slaves(bond)) {
1767 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1768 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1771 bond_compute_features(bond);
1772 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1773 (old_features & NETIF_F_VLAN_CHALLENGED))
1774 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1775 bond_dev->name, slave_dev->name, bond_dev->name);
1777 /* must do this from outside any spinlocks */
1778 vlan_vids_del_by_dev(slave_dev, bond_dev);
1780 /* If the mode USES_PRIMARY, then this cases was handled above by
1781 * bond_change_active_slave(..., NULL)
1783 if (!USES_PRIMARY(bond->params.mode)) {
1784 /* unset promiscuity level from slave
1785 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1786 * of the IFF_PROMISC flag in the bond_dev, but we need the
1787 * value of that flag before that change, as that was the value
1788 * when this slave was attached, so we cache at the start of the
1789 * function and use it here. Same goes for ALLMULTI below
1791 if (old_flags & IFF_PROMISC)
1792 dev_set_promiscuity(slave_dev, -1);
1794 /* unset allmulti level from slave */
1795 if (old_flags & IFF_ALLMULTI)
1796 dev_set_allmulti(slave_dev, -1);
1798 bond_hw_addr_flush(bond_dev, slave_dev);
1801 slave_disable_netpoll(slave);
1803 /* close slave before restoring its mac address */
1804 dev_close(slave_dev);
1806 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1807 /* restore original ("permanent") mac address */
1808 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1809 addr.sa_family = slave_dev->type;
1810 dev_set_mac_address(slave_dev, &addr);
1813 dev_set_mtu(slave_dev, slave->original_mtu);
1815 slave_dev->priv_flags &= ~IFF_BONDING;
1819 return 0; /* deletion OK */
1822 /* A wrapper used because of ndo_del_link */
1823 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1825 return __bond_release_one(bond_dev, slave_dev, false);
1829 * First release a slave and then destroy the bond if no more slaves are left.
1830 * Must be under rtnl_lock when this function is called.
1832 static int bond_release_and_destroy(struct net_device *bond_dev,
1833 struct net_device *slave_dev)
1835 struct bonding *bond = netdev_priv(bond_dev);
1838 ret = bond_release(bond_dev, slave_dev);
1839 if (ret == 0 && !bond_has_slaves(bond)) {
1840 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1841 pr_info("%s: destroying bond %s.\n",
1842 bond_dev->name, bond_dev->name);
1843 unregister_netdevice(bond_dev);
1848 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1850 struct bonding *bond = netdev_priv(bond_dev);
1852 info->bond_mode = bond->params.mode;
1853 info->miimon = bond->params.miimon;
1855 read_lock(&bond->lock);
1856 info->num_slaves = bond->slave_cnt;
1857 read_unlock(&bond->lock);
1862 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1864 struct bonding *bond = netdev_priv(bond_dev);
1865 struct list_head *iter;
1866 int i = 0, res = -ENODEV;
1867 struct slave *slave;
1869 read_lock(&bond->lock);
1870 bond_for_each_slave(bond, slave, iter) {
1871 if (i++ == (int)info->slave_id) {
1873 strcpy(info->slave_name, slave->dev->name);
1874 info->link = slave->link;
1875 info->state = bond_slave_state(slave);
1876 info->link_failure_count = slave->link_failure_count;
1880 read_unlock(&bond->lock);
1885 /*-------------------------------- Monitoring -------------------------------*/
1888 static int bond_miimon_inspect(struct bonding *bond)
1890 int link_state, commit = 0;
1891 struct list_head *iter;
1892 struct slave *slave;
1893 bool ignore_updelay;
1895 ignore_updelay = !bond->curr_active_slave ? true : false;
1897 bond_for_each_slave_rcu(bond, slave, iter) {
1898 slave->new_link = BOND_LINK_NOCHANGE;
1900 link_state = bond_check_dev_link(bond, slave->dev, 0);
1902 switch (slave->link) {
1907 slave->link = BOND_LINK_FAIL;
1908 slave->delay = bond->params.downdelay;
1910 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1912 (bond->params.mode ==
1913 BOND_MODE_ACTIVEBACKUP) ?
1914 (bond_is_active_slave(slave) ?
1915 "active " : "backup ") : "",
1917 bond->params.downdelay * bond->params.miimon);
1920 case BOND_LINK_FAIL:
1923 * recovered before downdelay expired
1925 slave->link = BOND_LINK_UP;
1926 slave->jiffies = jiffies;
1927 pr_info("%s: link status up again after %d ms for interface %s.\n",
1929 (bond->params.downdelay - slave->delay) *
1930 bond->params.miimon,
1935 if (slave->delay <= 0) {
1936 slave->new_link = BOND_LINK_DOWN;
1944 case BOND_LINK_DOWN:
1948 slave->link = BOND_LINK_BACK;
1949 slave->delay = bond->params.updelay;
1952 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1953 bond->dev->name, slave->dev->name,
1954 ignore_updelay ? 0 :
1955 bond->params.updelay *
1956 bond->params.miimon);
1959 case BOND_LINK_BACK:
1961 slave->link = BOND_LINK_DOWN;
1962 pr_info("%s: link status down again after %d ms for interface %s.\n",
1964 (bond->params.updelay - slave->delay) *
1965 bond->params.miimon,
1974 if (slave->delay <= 0) {
1975 slave->new_link = BOND_LINK_UP;
1977 ignore_updelay = false;
1989 static void bond_miimon_commit(struct bonding *bond)
1991 struct list_head *iter;
1992 struct slave *slave;
1994 bond_for_each_slave(bond, slave, iter) {
1995 switch (slave->new_link) {
1996 case BOND_LINK_NOCHANGE:
2000 slave->link = BOND_LINK_UP;
2001 slave->jiffies = jiffies;
2003 if (bond->params.mode == BOND_MODE_8023AD) {
2004 /* prevent it from being the active one */
2005 bond_set_backup_slave(slave);
2006 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2007 /* make it immediately active */
2008 bond_set_active_slave(slave);
2009 } else if (slave != bond->primary_slave) {
2010 /* prevent it from being the active one */
2011 bond_set_backup_slave(slave);
2014 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2015 bond->dev->name, slave->dev->name,
2016 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2017 slave->duplex ? "full" : "half");
2019 /* notify ad that the link status has changed */
2020 if (bond->params.mode == BOND_MODE_8023AD)
2021 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2023 if (bond_is_lb(bond))
2024 bond_alb_handle_link_change(bond, slave,
2027 if (!bond->curr_active_slave ||
2028 (slave == bond->primary_slave))
2033 case BOND_LINK_DOWN:
2034 if (slave->link_failure_count < UINT_MAX)
2035 slave->link_failure_count++;
2037 slave->link = BOND_LINK_DOWN;
2039 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2040 bond->params.mode == BOND_MODE_8023AD)
2041 bond_set_slave_inactive_flags(slave);
2043 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2044 bond->dev->name, slave->dev->name);
2046 if (bond->params.mode == BOND_MODE_8023AD)
2047 bond_3ad_handle_link_change(slave,
2050 if (bond_is_lb(bond))
2051 bond_alb_handle_link_change(bond, slave,
2054 if (slave == bond->curr_active_slave)
2060 pr_err("%s: invalid new link %d on slave %s\n",
2061 bond->dev->name, slave->new_link,
2063 slave->new_link = BOND_LINK_NOCHANGE;
2071 write_lock_bh(&bond->curr_slave_lock);
2072 bond_select_active_slave(bond);
2073 write_unlock_bh(&bond->curr_slave_lock);
2074 unblock_netpoll_tx();
2077 bond_set_carrier(bond);
2083 * Really a wrapper that splits the mii monitor into two phases: an
2084 * inspection, then (if inspection indicates something needs to be done)
2085 * an acquisition of appropriate locks followed by a commit phase to
2086 * implement whatever link state changes are indicated.
2088 void bond_mii_monitor(struct work_struct *work)
2090 struct bonding *bond = container_of(work, struct bonding,
2092 bool should_notify_peers = false;
2093 unsigned long delay;
2095 delay = msecs_to_jiffies(bond->params.miimon);
2097 if (!bond_has_slaves(bond))
2102 should_notify_peers = bond_should_notify_peers(bond);
2104 if (bond_miimon_inspect(bond)) {
2107 /* Race avoidance with bond_close cancel of workqueue */
2108 if (!rtnl_trylock()) {
2110 should_notify_peers = false;
2114 bond_miimon_commit(bond);
2116 rtnl_unlock(); /* might sleep, hold no other locks */
2121 if (bond->params.miimon)
2122 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2124 if (should_notify_peers) {
2125 if (!rtnl_trylock())
2127 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2132 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2134 struct net_device *upper;
2135 struct list_head *iter;
2138 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2142 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2143 if (ip == bond_confirm_addr(upper, 0, ip)) {
2154 * We go to the (large) trouble of VLAN tagging ARP frames because
2155 * switches in VLAN mode (especially if ports are configured as
2156 * "native" to a VLAN) might not pass non-tagged frames.
2158 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2160 struct sk_buff *skb;
2162 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2163 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2165 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2166 NULL, slave_dev->dev_addr, NULL);
2169 pr_err("ARP packet allocation failed\n");
2173 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2175 pr_err("failed to insert VLAN tag\n");
2183 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2185 struct net_device *upper, *vlan_upper;
2186 struct list_head *iter, *vlan_iter;
2188 __be32 *targets = bond->params.arp_targets, addr;
2191 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2192 pr_debug("basa: target %pI4\n", &targets[i]);
2194 /* Find out through which dev should the packet go */
2195 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2198 pr_debug("%s: no route to arp_ip_target %pI4\n",
2199 bond->dev->name, &targets[i]);
2205 /* bond device itself */
2206 if (rt->dst.dev == bond->dev)
2210 /* first we search only for vlan devices. for every vlan
2211 * found we verify its upper dev list, searching for the
2212 * rt->dst.dev. If found we save the tag of the vlan and
2213 * proceed to send the packet.
2217 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2219 if (!is_vlan_dev(vlan_upper))
2221 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2223 if (upper == rt->dst.dev) {
2224 vlan_id = vlan_dev_vlan_id(vlan_upper);
2231 /* if the device we're looking for is not on top of any of
2232 * our upper vlans, then just search for any dev that
2233 * matches, and in case it's a vlan - save the id
2235 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2236 if (upper == rt->dst.dev) {
2237 /* if it's a vlan - get its VID */
2238 if (is_vlan_dev(upper))
2239 vlan_id = vlan_dev_vlan_id(upper);
2247 /* Not our device - skip */
2248 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2249 bond->dev->name, &targets[i],
2250 rt->dst.dev ? rt->dst.dev->name : "NULL");
2256 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2258 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2263 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2267 if (!sip || !bond_has_this_ip(bond, tip)) {
2268 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2272 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2274 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2277 slave->last_arp_rx = jiffies;
2278 slave->target_last_arp_rx[i] = jiffies;
2281 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2282 struct slave *slave)
2284 struct arphdr *arp = (struct arphdr *)skb->data;
2285 unsigned char *arp_ptr;
2289 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2290 return RX_HANDLER_ANOTHER;
2292 read_lock(&bond->lock);
2294 if (!slave_do_arp_validate(bond, slave))
2297 alen = arp_hdr_len(bond->dev);
2299 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2300 bond->dev->name, skb->dev->name);
2302 if (alen > skb_headlen(skb)) {
2303 arp = kmalloc(alen, GFP_ATOMIC);
2306 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2310 if (arp->ar_hln != bond->dev->addr_len ||
2311 skb->pkt_type == PACKET_OTHERHOST ||
2312 skb->pkt_type == PACKET_LOOPBACK ||
2313 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2314 arp->ar_pro != htons(ETH_P_IP) ||
2318 arp_ptr = (unsigned char *)(arp + 1);
2319 arp_ptr += bond->dev->addr_len;
2320 memcpy(&sip, arp_ptr, 4);
2321 arp_ptr += 4 + bond->dev->addr_len;
2322 memcpy(&tip, arp_ptr, 4);
2324 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2325 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2326 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2330 * Backup slaves won't see the ARP reply, but do come through
2331 * here for each ARP probe (so we swap the sip/tip to validate
2332 * the probe). In a "redundant switch, common router" type of
2333 * configuration, the ARP probe will (hopefully) travel from
2334 * the active, through one switch, the router, then the other
2335 * switch before reaching the backup.
2337 * We 'trust' the arp requests if there is an active slave and
2338 * it received valid arp reply(s) after it became active. This
2339 * is done to avoid endless looping when we can't reach the
2340 * arp_ip_target and fool ourselves with our own arp requests.
2342 if (bond_is_active_slave(slave))
2343 bond_validate_arp(bond, slave, sip, tip);
2344 else if (bond->curr_active_slave &&
2345 time_after(slave_last_rx(bond, bond->curr_active_slave),
2346 bond->curr_active_slave->jiffies))
2347 bond_validate_arp(bond, slave, tip, sip);
2350 read_unlock(&bond->lock);
2351 if (arp != (struct arphdr *)skb->data)
2353 return RX_HANDLER_ANOTHER;
2356 /* function to verify if we're in the arp_interval timeslice, returns true if
2357 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2358 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2360 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2363 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2365 return time_in_range(jiffies,
2366 last_act - delta_in_ticks,
2367 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2371 * this function is called regularly to monitor each slave's link
2372 * ensuring that traffic is being sent and received when arp monitoring
2373 * is used in load-balancing mode. if the adapter has been dormant, then an
2374 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2375 * arp monitoring in active backup mode.
2377 void bond_loadbalance_arp_mon(struct work_struct *work)
2379 struct bonding *bond = container_of(work, struct bonding,
2381 struct slave *slave, *oldcurrent;
2382 struct list_head *iter;
2383 int do_failover = 0;
2385 if (!bond_has_slaves(bond))
2390 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2391 /* see if any of the previous devices are up now (i.e. they have
2392 * xmt and rcv traffic). the curr_active_slave does not come into
2393 * the picture unless it is null. also, slave->jiffies is not needed
2394 * here because we send an arp on each slave and give a slave as
2395 * long as it needs to get the tx/rx within the delta.
2396 * TODO: what about up/down delay in arp mode? it wasn't here before
2399 bond_for_each_slave_rcu(bond, slave, iter) {
2400 unsigned long trans_start = dev_trans_start(slave->dev);
2402 if (slave->link != BOND_LINK_UP) {
2403 if (bond_time_in_interval(bond, trans_start, 1) &&
2404 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2406 slave->link = BOND_LINK_UP;
2407 bond_set_active_slave(slave);
2409 /* primary_slave has no meaning in round-robin
2410 * mode. the window of a slave being up and
2411 * curr_active_slave being null after enslaving
2415 pr_info("%s: link status definitely up for interface %s, ",
2420 pr_info("%s: interface %s is now up\n",
2426 /* slave->link == BOND_LINK_UP */
2428 /* not all switches will respond to an arp request
2429 * when the source ip is 0, so don't take the link down
2430 * if we don't know our ip yet
2432 if (!bond_time_in_interval(bond, trans_start, 2) ||
2433 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2435 slave->link = BOND_LINK_DOWN;
2436 bond_set_backup_slave(slave);
2438 if (slave->link_failure_count < UINT_MAX)
2439 slave->link_failure_count++;
2441 pr_info("%s: interface %s is now down.\n",
2445 if (slave == oldcurrent)
2450 /* note: if switch is in round-robin mode, all links
2451 * must tx arp to ensure all links rx an arp - otherwise
2452 * links may oscillate or not come up at all; if switch is
2453 * in something like xor mode, there is nothing we can
2454 * do - all replies will be rx'ed on same link causing slaves
2455 * to be unstable during low/no traffic periods
2457 if (IS_UP(slave->dev))
2458 bond_arp_send_all(bond, slave);
2464 /* the bond_select_active_slave must hold RTNL
2465 * and curr_slave_lock for write.
2467 if (!rtnl_trylock())
2470 write_lock_bh(&bond->curr_slave_lock);
2472 bond_select_active_slave(bond);
2474 write_unlock_bh(&bond->curr_slave_lock);
2475 unblock_netpoll_tx();
2480 if (bond->params.arp_interval)
2481 queue_delayed_work(bond->wq, &bond->arp_work,
2482 msecs_to_jiffies(bond->params.arp_interval));
2486 * Called to inspect slaves for active-backup mode ARP monitor link state
2487 * changes. Sets new_link in slaves to specify what action should take
2488 * place for the slave. Returns 0 if no changes are found, >0 if changes
2489 * to link states must be committed.
2491 * Called with rcu_read_lock hold.
2493 static int bond_ab_arp_inspect(struct bonding *bond)
2495 unsigned long trans_start, last_rx;
2496 struct list_head *iter;
2497 struct slave *slave;
2500 bond_for_each_slave_rcu(bond, slave, iter) {
2501 slave->new_link = BOND_LINK_NOCHANGE;
2502 last_rx = slave_last_rx(bond, slave);
2504 if (slave->link != BOND_LINK_UP) {
2505 if (bond_time_in_interval(bond, last_rx, 1)) {
2506 slave->new_link = BOND_LINK_UP;
2513 * Give slaves 2*delta after being enslaved or made
2514 * active. This avoids bouncing, as the last receive
2515 * times need a full ARP monitor cycle to be updated.
2517 if (bond_time_in_interval(bond, slave->jiffies, 2))
2521 * Backup slave is down if:
2522 * - No current_arp_slave AND
2523 * - more than 3*delta since last receive AND
2524 * - the bond has an IP address
2526 * Note: a non-null current_arp_slave indicates
2527 * the curr_active_slave went down and we are
2528 * searching for a new one; under this condition
2529 * we only take the curr_active_slave down - this
2530 * gives each slave a chance to tx/rx traffic
2531 * before being taken out
2533 if (!bond_is_active_slave(slave) &&
2534 !bond->current_arp_slave &&
2535 !bond_time_in_interval(bond, last_rx, 3)) {
2536 slave->new_link = BOND_LINK_DOWN;
2541 * Active slave is down if:
2542 * - more than 2*delta since transmitting OR
2543 * - (more than 2*delta since receive AND
2544 * the bond has an IP address)
2546 trans_start = dev_trans_start(slave->dev);
2547 if (bond_is_active_slave(slave) &&
2548 (!bond_time_in_interval(bond, trans_start, 2) ||
2549 !bond_time_in_interval(bond, last_rx, 2))) {
2550 slave->new_link = BOND_LINK_DOWN;
2559 * Called to commit link state changes noted by inspection step of
2560 * active-backup mode ARP monitor.
2562 * Called with RTNL hold.
2564 static void bond_ab_arp_commit(struct bonding *bond)
2566 unsigned long trans_start;
2567 struct list_head *iter;
2568 struct slave *slave;
2570 bond_for_each_slave(bond, slave, iter) {
2571 switch (slave->new_link) {
2572 case BOND_LINK_NOCHANGE:
2576 trans_start = dev_trans_start(slave->dev);
2577 if (bond->curr_active_slave != slave ||
2578 (!bond->curr_active_slave &&
2579 bond_time_in_interval(bond, trans_start, 1))) {
2580 slave->link = BOND_LINK_UP;
2581 if (bond->current_arp_slave) {
2582 bond_set_slave_inactive_flags(
2583 bond->current_arp_slave);
2584 bond->current_arp_slave = NULL;
2587 pr_info("%s: link status definitely up for interface %s.\n",
2588 bond->dev->name, slave->dev->name);
2590 if (!bond->curr_active_slave ||
2591 (slave == bond->primary_slave))
2598 case BOND_LINK_DOWN:
2599 if (slave->link_failure_count < UINT_MAX)
2600 slave->link_failure_count++;
2602 slave->link = BOND_LINK_DOWN;
2603 bond_set_slave_inactive_flags(slave);
2605 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2606 bond->dev->name, slave->dev->name);
2608 if (slave == bond->curr_active_slave) {
2609 bond->current_arp_slave = NULL;
2616 pr_err("%s: impossible: new_link %d on slave %s\n",
2617 bond->dev->name, slave->new_link,
2625 write_lock_bh(&bond->curr_slave_lock);
2626 bond_select_active_slave(bond);
2627 write_unlock_bh(&bond->curr_slave_lock);
2628 unblock_netpoll_tx();
2631 bond_set_carrier(bond);
2635 * Send ARP probes for active-backup mode ARP monitor.
2637 * Called with rcu_read_lock hold.
2639 static void bond_ab_arp_probe(struct bonding *bond)
2641 struct slave *slave, *before = NULL, *new_slave = NULL,
2642 *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2643 struct list_head *iter;
2646 read_lock(&bond->curr_slave_lock);
2648 if (curr_arp_slave && bond->curr_active_slave)
2649 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2650 curr_arp_slave->dev->name,
2651 bond->curr_active_slave->dev->name);
2653 if (bond->curr_active_slave) {
2654 bond_arp_send_all(bond, bond->curr_active_slave);
2655 read_unlock(&bond->curr_slave_lock);
2659 read_unlock(&bond->curr_slave_lock);
2661 /* if we don't have a curr_active_slave, search for the next available
2662 * backup slave from the current_arp_slave and make it the candidate
2663 * for becoming the curr_active_slave
2666 if (!curr_arp_slave) {
2667 curr_arp_slave = bond_first_slave_rcu(bond);
2668 if (!curr_arp_slave)
2672 bond_set_slave_inactive_flags(curr_arp_slave);
2674 bond_for_each_slave_rcu(bond, slave, iter) {
2675 if (!found && !before && IS_UP(slave->dev))
2678 if (found && !new_slave && IS_UP(slave->dev))
2680 /* if the link state is up at this point, we
2681 * mark it down - this can happen if we have
2682 * simultaneous link failures and
2683 * reselect_active_interface doesn't make this
2684 * one the current slave so it is still marked
2685 * up when it is actually down
2687 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2688 slave->link = BOND_LINK_DOWN;
2689 if (slave->link_failure_count < UINT_MAX)
2690 slave->link_failure_count++;
2692 bond_set_slave_inactive_flags(slave);
2694 pr_info("%s: backup interface %s is now down.\n",
2695 bond->dev->name, slave->dev->name);
2697 if (slave == curr_arp_slave)
2701 if (!new_slave && before)
2707 new_slave->link = BOND_LINK_BACK;
2708 bond_set_slave_active_flags(new_slave);
2709 bond_arp_send_all(bond, new_slave);
2710 new_slave->jiffies = jiffies;
2711 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2714 void bond_activebackup_arp_mon(struct work_struct *work)
2716 struct bonding *bond = container_of(work, struct bonding,
2718 bool should_notify_peers = false;
2721 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2723 if (!bond_has_slaves(bond))
2728 should_notify_peers = bond_should_notify_peers(bond);
2730 if (bond_ab_arp_inspect(bond)) {
2733 /* Race avoidance with bond_close flush of workqueue */
2734 if (!rtnl_trylock()) {
2736 should_notify_peers = false;
2740 bond_ab_arp_commit(bond);
2746 bond_ab_arp_probe(bond);
2750 if (bond->params.arp_interval)
2751 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2753 if (should_notify_peers) {
2754 if (!rtnl_trylock())
2756 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2761 /*-------------------------- netdev event handling --------------------------*/
2764 * Change device name
2766 static int bond_event_changename(struct bonding *bond)
2768 bond_remove_proc_entry(bond);
2769 bond_create_proc_entry(bond);
2771 bond_debug_reregister(bond);
2776 static int bond_master_netdev_event(unsigned long event,
2777 struct net_device *bond_dev)
2779 struct bonding *event_bond = netdev_priv(bond_dev);
2782 case NETDEV_CHANGENAME:
2783 return bond_event_changename(event_bond);
2784 case NETDEV_UNREGISTER:
2785 bond_remove_proc_entry(event_bond);
2787 case NETDEV_REGISTER:
2788 bond_create_proc_entry(event_bond);
2790 case NETDEV_NOTIFY_PEERS:
2791 if (event_bond->send_peer_notif)
2792 event_bond->send_peer_notif--;
2801 static int bond_slave_netdev_event(unsigned long event,
2802 struct net_device *slave_dev)
2804 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2805 struct bonding *bond;
2806 struct net_device *bond_dev;
2810 /* A netdev event can be generated while enslaving a device
2811 * before netdev_rx_handler_register is called in which case
2812 * slave will be NULL
2816 bond_dev = slave->bond->dev;
2820 case NETDEV_UNREGISTER:
2821 if (bond_dev->type != ARPHRD_ETHER)
2822 bond_release_and_destroy(bond_dev, slave_dev);
2824 bond_release(bond_dev, slave_dev);
2828 old_speed = slave->speed;
2829 old_duplex = slave->duplex;
2831 bond_update_speed_duplex(slave);
2833 if (bond->params.mode == BOND_MODE_8023AD) {
2834 if (old_speed != slave->speed)
2835 bond_3ad_adapter_speed_changed(slave);
2836 if (old_duplex != slave->duplex)
2837 bond_3ad_adapter_duplex_changed(slave);
2842 * ... Or is it this?
2845 case NETDEV_CHANGEMTU:
2847 * TODO: Should slaves be allowed to
2848 * independently alter their MTU? For
2849 * an active-backup bond, slaves need
2850 * not be the same type of device, so
2851 * MTUs may vary. For other modes,
2852 * slaves arguably should have the
2853 * same MTUs. To do this, we'd need to
2854 * take over the slave's change_mtu
2855 * function for the duration of their
2859 case NETDEV_CHANGENAME:
2861 * TODO: handle changing the primary's name
2864 case NETDEV_FEAT_CHANGE:
2865 bond_compute_features(bond);
2867 case NETDEV_RESEND_IGMP:
2868 /* Propagate to master device */
2869 call_netdevice_notifiers(event, slave->bond->dev);
2879 * bond_netdev_event: handle netdev notifier chain events.
2881 * This function receives events for the netdev chain. The caller (an
2882 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2883 * locks for us to safely manipulate the slave devices (RTNL lock,
2886 static int bond_netdev_event(struct notifier_block *this,
2887 unsigned long event, void *ptr)
2889 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2891 pr_debug("event_dev: %s, event: %lx\n",
2892 event_dev ? event_dev->name : "None",
2895 if (!(event_dev->priv_flags & IFF_BONDING))
2898 if (event_dev->flags & IFF_MASTER) {
2899 pr_debug("IFF_MASTER\n");
2900 return bond_master_netdev_event(event, event_dev);
2903 if (event_dev->flags & IFF_SLAVE) {
2904 pr_debug("IFF_SLAVE\n");
2905 return bond_slave_netdev_event(event, event_dev);
2911 static struct notifier_block bond_netdev_notifier = {
2912 .notifier_call = bond_netdev_event,
2915 /*---------------------------- Hashing Policies -----------------------------*/
2917 /* L2 hash helper */
2918 static inline u32 bond_eth_hash(struct sk_buff *skb)
2920 struct ethhdr *data = (struct ethhdr *)skb->data;
2922 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2923 return data->h_dest[5] ^ data->h_source[5];
2928 /* Extract the appropriate headers based on bond's xmit policy */
2929 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2930 struct flow_keys *fk)
2932 const struct ipv6hdr *iph6;
2933 const struct iphdr *iph;
2934 int noff, proto = -1;
2936 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2937 return skb_flow_dissect(skb, fk);
2940 noff = skb_network_offset(skb);
2941 if (skb->protocol == htons(ETH_P_IP)) {
2942 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2945 fk->src = iph->saddr;
2946 fk->dst = iph->daddr;
2947 noff += iph->ihl << 2;
2948 if (!ip_is_fragment(iph))
2949 proto = iph->protocol;
2950 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2951 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2953 iph6 = ipv6_hdr(skb);
2954 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2955 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2956 noff += sizeof(*iph6);
2957 proto = iph6->nexthdr;
2961 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2962 fk->ports = skb_flow_get_ports(skb, noff, proto);
2968 * bond_xmit_hash - generate a hash value based on the xmit policy
2969 * @bond: bonding device
2970 * @skb: buffer to use for headers
2971 * @count: modulo value
2973 * This function will extract the necessary headers from the skb buffer and use
2974 * them to generate a hash based on the xmit_policy set in the bonding device
2975 * which will be reduced modulo count before returning.
2977 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
2979 struct flow_keys flow;
2982 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2983 !bond_flow_dissect(bond, skb, &flow))
2984 return bond_eth_hash(skb) % count;
2986 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
2987 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
2988 hash = bond_eth_hash(skb);
2990 hash = (__force u32)flow.ports;
2991 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
2992 hash ^= (hash >> 16);
2993 hash ^= (hash >> 8);
2995 return hash % count;
2998 /*-------------------------- Device entry points ----------------------------*/
3000 static void bond_work_init_all(struct bonding *bond)
3002 INIT_DELAYED_WORK(&bond->mcast_work,
3003 bond_resend_igmp_join_requests_delayed);
3004 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3005 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3006 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3007 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3009 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3010 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3013 static void bond_work_cancel_all(struct bonding *bond)
3015 cancel_delayed_work_sync(&bond->mii_work);
3016 cancel_delayed_work_sync(&bond->arp_work);
3017 cancel_delayed_work_sync(&bond->alb_work);
3018 cancel_delayed_work_sync(&bond->ad_work);
3019 cancel_delayed_work_sync(&bond->mcast_work);
3022 static int bond_open(struct net_device *bond_dev)
3024 struct bonding *bond = netdev_priv(bond_dev);
3025 struct list_head *iter;
3026 struct slave *slave;
3028 /* reset slave->backup and slave->inactive */
3029 read_lock(&bond->lock);
3030 if (bond_has_slaves(bond)) {
3031 read_lock(&bond->curr_slave_lock);
3032 bond_for_each_slave(bond, slave, iter) {
3033 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3034 && (slave != bond->curr_active_slave)) {
3035 bond_set_slave_inactive_flags(slave);
3037 bond_set_slave_active_flags(slave);
3040 read_unlock(&bond->curr_slave_lock);
3042 read_unlock(&bond->lock);
3044 bond_work_init_all(bond);
3046 if (bond_is_lb(bond)) {
3047 /* bond_alb_initialize must be called before the timer
3050 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3052 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3055 if (bond->params.miimon) /* link check interval, in milliseconds. */
3056 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3058 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3059 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3060 if (bond->params.arp_validate)
3061 bond->recv_probe = bond_arp_rcv;
3064 if (bond->params.mode == BOND_MODE_8023AD) {
3065 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3066 /* register to receive LACPDUs */
3067 bond->recv_probe = bond_3ad_lacpdu_recv;
3068 bond_3ad_initiate_agg_selection(bond, 1);
3074 static int bond_close(struct net_device *bond_dev)
3076 struct bonding *bond = netdev_priv(bond_dev);
3078 bond_work_cancel_all(bond);
3079 bond->send_peer_notif = 0;
3080 if (bond_is_lb(bond))
3081 bond_alb_deinitialize(bond);
3082 bond->recv_probe = NULL;
3087 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3088 struct rtnl_link_stats64 *stats)
3090 struct bonding *bond = netdev_priv(bond_dev);
3091 struct rtnl_link_stats64 temp;
3092 struct list_head *iter;
3093 struct slave *slave;
3095 memset(stats, 0, sizeof(*stats));
3097 read_lock_bh(&bond->lock);
3098 bond_for_each_slave(bond, slave, iter) {
3099 const struct rtnl_link_stats64 *sstats =
3100 dev_get_stats(slave->dev, &temp);
3102 stats->rx_packets += sstats->rx_packets;
3103 stats->rx_bytes += sstats->rx_bytes;
3104 stats->rx_errors += sstats->rx_errors;
3105 stats->rx_dropped += sstats->rx_dropped;
3107 stats->tx_packets += sstats->tx_packets;
3108 stats->tx_bytes += sstats->tx_bytes;
3109 stats->tx_errors += sstats->tx_errors;
3110 stats->tx_dropped += sstats->tx_dropped;
3112 stats->multicast += sstats->multicast;
3113 stats->collisions += sstats->collisions;
3115 stats->rx_length_errors += sstats->rx_length_errors;
3116 stats->rx_over_errors += sstats->rx_over_errors;
3117 stats->rx_crc_errors += sstats->rx_crc_errors;
3118 stats->rx_frame_errors += sstats->rx_frame_errors;
3119 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3120 stats->rx_missed_errors += sstats->rx_missed_errors;
3122 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3123 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3124 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3125 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3126 stats->tx_window_errors += sstats->tx_window_errors;
3128 read_unlock_bh(&bond->lock);
3133 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3135 struct bonding *bond = netdev_priv(bond_dev);
3136 struct net_device *slave_dev = NULL;
3137 struct ifbond k_binfo;
3138 struct ifbond __user *u_binfo = NULL;
3139 struct ifslave k_sinfo;
3140 struct ifslave __user *u_sinfo = NULL;
3141 struct mii_ioctl_data *mii = NULL;
3145 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3157 * We do this again just in case we were called by SIOCGMIIREG
3158 * instead of SIOCGMIIPHY.
3165 if (mii->reg_num == 1) {
3167 read_lock(&bond->lock);
3168 read_lock(&bond->curr_slave_lock);
3169 if (netif_carrier_ok(bond->dev))
3170 mii->val_out = BMSR_LSTATUS;
3172 read_unlock(&bond->curr_slave_lock);
3173 read_unlock(&bond->lock);
3177 case BOND_INFO_QUERY_OLD:
3178 case SIOCBONDINFOQUERY:
3179 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3181 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3184 res = bond_info_query(bond_dev, &k_binfo);
3186 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3190 case BOND_SLAVE_INFO_QUERY_OLD:
3191 case SIOCBONDSLAVEINFOQUERY:
3192 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3194 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3197 res = bond_slave_info_query(bond_dev, &k_sinfo);
3199 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3208 net = dev_net(bond_dev);
3210 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3213 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3215 pr_debug("slave_dev=%p:\n", slave_dev);
3220 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3222 case BOND_ENSLAVE_OLD:
3223 case SIOCBONDENSLAVE:
3224 res = bond_enslave(bond_dev, slave_dev);
3226 case BOND_RELEASE_OLD:
3227 case SIOCBONDRELEASE:
3228 res = bond_release(bond_dev, slave_dev);
3230 case BOND_SETHWADDR_OLD:
3231 case SIOCBONDSETHWADDR:
3232 bond_set_dev_addr(bond_dev, slave_dev);
3235 case BOND_CHANGE_ACTIVE_OLD:
3236 case SIOCBONDCHANGEACTIVE:
3237 res = bond_option_active_slave_set(bond, slave_dev);
3249 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3251 struct bonding *bond = netdev_priv(bond_dev);
3253 if (change & IFF_PROMISC)
3254 bond_set_promiscuity(bond,
3255 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3257 if (change & IFF_ALLMULTI)
3258 bond_set_allmulti(bond,
3259 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3262 static void bond_set_rx_mode(struct net_device *bond_dev)
3264 struct bonding *bond = netdev_priv(bond_dev);
3265 struct list_head *iter;
3266 struct slave *slave;
3270 if (USES_PRIMARY(bond->params.mode)) {
3271 slave = rcu_dereference(bond->curr_active_slave);
3273 dev_uc_sync(slave->dev, bond_dev);
3274 dev_mc_sync(slave->dev, bond_dev);
3277 bond_for_each_slave_rcu(bond, slave, iter) {
3278 dev_uc_sync_multiple(slave->dev, bond_dev);
3279 dev_mc_sync_multiple(slave->dev, bond_dev);
3285 static int bond_neigh_init(struct neighbour *n)
3287 struct bonding *bond = netdev_priv(n->dev);
3288 const struct net_device_ops *slave_ops;
3289 struct neigh_parms parms;
3290 struct slave *slave;
3293 slave = bond_first_slave(bond);
3296 slave_ops = slave->dev->netdev_ops;
3297 if (!slave_ops->ndo_neigh_setup)
3300 parms.neigh_setup = NULL;
3301 parms.neigh_cleanup = NULL;
3302 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3307 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3308 * after the last slave has been detached. Assumes that all slaves
3309 * utilize the same neigh_cleanup (true at this writing as only user
3312 n->parms->neigh_cleanup = parms.neigh_cleanup;
3314 if (!parms.neigh_setup)
3317 return parms.neigh_setup(n);
3321 * The bonding ndo_neigh_setup is called at init time beofre any
3322 * slave exists. So we must declare proxy setup function which will
3323 * be used at run time to resolve the actual slave neigh param setup.
3325 * It's also called by master devices (such as vlans) to setup their
3326 * underlying devices. In that case - do nothing, we're already set up from
3329 static int bond_neigh_setup(struct net_device *dev,
3330 struct neigh_parms *parms)
3332 /* modify only our neigh_parms */
3333 if (parms->dev == dev)
3334 parms->neigh_setup = bond_neigh_init;
3340 * Change the MTU of all of a master's slaves to match the master
3342 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3344 struct bonding *bond = netdev_priv(bond_dev);
3345 struct slave *slave, *rollback_slave;
3346 struct list_head *iter;
3349 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3350 (bond_dev ? bond_dev->name : "None"), new_mtu);
3352 /* Can't hold bond->lock with bh disabled here since
3353 * some base drivers panic. On the other hand we can't
3354 * hold bond->lock without bh disabled because we'll
3355 * deadlock. The only solution is to rely on the fact
3356 * that we're under rtnl_lock here, and the slaves
3357 * list won't change. This doesn't solve the problem
3358 * of setting the slave's MTU while it is
3359 * transmitting, but the assumption is that the base
3360 * driver can handle that.
3362 * TODO: figure out a way to safely iterate the slaves
3363 * list, but without holding a lock around the actual
3364 * call to the base driver.
3367 bond_for_each_slave(bond, slave, iter) {
3368 pr_debug("s %p c_m %p\n",
3370 slave->dev->netdev_ops->ndo_change_mtu);
3372 res = dev_set_mtu(slave->dev, new_mtu);
3375 /* If we failed to set the slave's mtu to the new value
3376 * we must abort the operation even in ACTIVE_BACKUP
3377 * mode, because if we allow the backup slaves to have
3378 * different mtu values than the active slave we'll
3379 * need to change their mtu when doing a failover. That
3380 * means changing their mtu from timer context, which
3381 * is probably not a good idea.
3383 pr_debug("err %d %s\n", res, slave->dev->name);
3388 bond_dev->mtu = new_mtu;
3393 /* unwind from head to the slave that failed */
3394 bond_for_each_slave(bond, rollback_slave, iter) {
3397 if (rollback_slave == slave)
3400 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3402 pr_debug("unwind err %d dev %s\n",
3403 tmp_res, rollback_slave->dev->name);
3413 * Note that many devices must be down to change the HW address, and
3414 * downing the master releases all slaves. We can make bonds full of
3415 * bonding devices to test this, however.
3417 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3419 struct bonding *bond = netdev_priv(bond_dev);
3420 struct slave *slave, *rollback_slave;
3421 struct sockaddr *sa = addr, tmp_sa;
3422 struct list_head *iter;
3425 if (bond->params.mode == BOND_MODE_ALB)
3426 return bond_alb_set_mac_address(bond_dev, addr);
3429 pr_debug("bond=%p, name=%s\n",
3430 bond, bond_dev ? bond_dev->name : "None");
3432 /* If fail_over_mac is enabled, do nothing and return success.
3433 * Returning an error causes ifenslave to fail.
3435 if (bond->params.fail_over_mac)
3438 if (!is_valid_ether_addr(sa->sa_data))
3439 return -EADDRNOTAVAIL;
3441 /* Can't hold bond->lock with bh disabled here since
3442 * some base drivers panic. On the other hand we can't
3443 * hold bond->lock without bh disabled because we'll
3444 * deadlock. The only solution is to rely on the fact
3445 * that we're under rtnl_lock here, and the slaves
3446 * list won't change. This doesn't solve the problem
3447 * of setting the slave's hw address while it is
3448 * transmitting, but the assumption is that the base
3449 * driver can handle that.
3451 * TODO: figure out a way to safely iterate the slaves
3452 * list, but without holding a lock around the actual
3453 * call to the base driver.
3456 bond_for_each_slave(bond, slave, iter) {
3457 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3458 pr_debug("slave %p %s\n", slave, slave->dev->name);
3460 if (slave_ops->ndo_set_mac_address == NULL) {
3462 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3466 res = dev_set_mac_address(slave->dev, addr);
3468 /* TODO: consider downing the slave
3470 * User should expect communications
3471 * breakage anyway until ARP finish
3474 pr_debug("err %d %s\n", res, slave->dev->name);
3480 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3484 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3485 tmp_sa.sa_family = bond_dev->type;
3487 /* unwind from head to the slave that failed */
3488 bond_for_each_slave(bond, rollback_slave, iter) {
3491 if (rollback_slave == slave)
3494 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3496 pr_debug("unwind err %d dev %s\n",
3497 tmp_res, rollback_slave->dev->name);
3505 * bond_xmit_slave_id - transmit skb through slave with slave_id
3506 * @bond: bonding device that is transmitting
3507 * @skb: buffer to transmit
3508 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3510 * This function tries to transmit through slave with slave_id but in case
3511 * it fails, it tries to find the first available slave for transmission.
3512 * The skb is consumed in all cases, thus the function is void.
3514 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3516 struct list_head *iter;
3517 struct slave *slave;
3520 /* Here we start from the slave with slave_id */
3521 bond_for_each_slave_rcu(bond, slave, iter) {
3523 if (slave_can_tx(slave)) {
3524 bond_dev_queue_xmit(bond, skb, slave->dev);
3530 /* Here we start from the first slave up to slave_id */
3532 bond_for_each_slave_rcu(bond, slave, iter) {
3535 if (slave_can_tx(slave)) {
3536 bond_dev_queue_xmit(bond, skb, slave->dev);
3540 /* no slave that can tx has been found */
3545 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3546 * @bond: bonding device to use
3548 * Based on the value of the bonding device's packets_per_slave parameter
3549 * this function generates a slave id, which is usually used as the next
3550 * slave to transmit through.
3552 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3554 int packets_per_slave = bond->params.packets_per_slave;
3557 switch (packets_per_slave) {
3559 slave_id = prandom_u32();
3562 slave_id = bond->rr_tx_counter;
3565 slave_id = reciprocal_divide(bond->rr_tx_counter,
3569 bond->rr_tx_counter++;
3574 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3576 struct bonding *bond = netdev_priv(bond_dev);
3577 struct iphdr *iph = ip_hdr(skb);
3578 struct slave *slave;
3581 /* Start with the curr_active_slave that joined the bond as the
3582 * default for sending IGMP traffic. For failover purposes one
3583 * needs to maintain some consistency for the interface that will
3584 * send the join/membership reports. The curr_active_slave found
3585 * will send all of this type of traffic.
3587 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3588 slave = rcu_dereference(bond->curr_active_slave);
3589 if (slave && slave_can_tx(slave))
3590 bond_dev_queue_xmit(bond, skb, slave->dev);
3592 bond_xmit_slave_id(bond, skb, 0);
3594 slave_id = bond_rr_gen_slave_id(bond);
3595 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3598 return NETDEV_TX_OK;
3602 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3603 * the bond has a usable interface.
3605 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3607 struct bonding *bond = netdev_priv(bond_dev);
3608 struct slave *slave;
3610 slave = rcu_dereference(bond->curr_active_slave);
3612 bond_dev_queue_xmit(bond, skb, slave->dev);
3616 return NETDEV_TX_OK;
3619 /* In bond_xmit_xor() , we determine the output device by using a pre-
3620 * determined xmit_hash_policy(), If the selected device is not enabled,
3621 * find the next active slave.
3623 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3625 struct bonding *bond = netdev_priv(bond_dev);
3627 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3629 return NETDEV_TX_OK;
3632 /* in broadcast mode, we send everything to all usable interfaces. */
3633 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3635 struct bonding *bond = netdev_priv(bond_dev);
3636 struct slave *slave = NULL;
3637 struct list_head *iter;
3639 bond_for_each_slave_rcu(bond, slave, iter) {
3640 if (bond_is_last_slave(bond, slave))
3642 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3643 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3646 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3650 /* bond_dev_queue_xmit always returns 0 */
3651 bond_dev_queue_xmit(bond, skb2, slave->dev);
3654 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3655 bond_dev_queue_xmit(bond, skb, slave->dev);
3659 return NETDEV_TX_OK;
3662 /*------------------------- Device initialization ---------------------------*/
3665 * Lookup the slave that corresponds to a qid
3667 static inline int bond_slave_override(struct bonding *bond,
3668 struct sk_buff *skb)
3670 struct slave *slave = NULL;
3671 struct slave *check_slave;
3672 struct list_head *iter;
3675 if (!skb->queue_mapping)
3678 /* Find out if any slaves have the same mapping as this skb. */
3679 bond_for_each_slave_rcu(bond, check_slave, iter) {
3680 if (check_slave->queue_id == skb->queue_mapping) {
3681 slave = check_slave;
3686 /* If the slave isn't UP, use default transmit policy. */
3687 if (slave && slave->queue_id && IS_UP(slave->dev) &&
3688 (slave->link == BOND_LINK_UP)) {
3689 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3696 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3699 * This helper function exists to help dev_pick_tx get the correct
3700 * destination queue. Using a helper function skips a call to
3701 * skb_tx_hash and will put the skbs in the queue we expect on their
3702 * way down to the bonding driver.
3704 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3707 * Save the original txq to restore before passing to the driver
3709 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3711 if (unlikely(txq >= dev->real_num_tx_queues)) {
3713 txq -= dev->real_num_tx_queues;
3714 } while (txq >= dev->real_num_tx_queues);
3719 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3721 struct bonding *bond = netdev_priv(dev);
3723 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3724 if (!bond_slave_override(bond, skb))
3725 return NETDEV_TX_OK;
3728 switch (bond->params.mode) {
3729 case BOND_MODE_ROUNDROBIN:
3730 return bond_xmit_roundrobin(skb, dev);
3731 case BOND_MODE_ACTIVEBACKUP:
3732 return bond_xmit_activebackup(skb, dev);
3734 return bond_xmit_xor(skb, dev);
3735 case BOND_MODE_BROADCAST:
3736 return bond_xmit_broadcast(skb, dev);
3737 case BOND_MODE_8023AD:
3738 return bond_3ad_xmit_xor(skb, dev);
3741 return bond_alb_xmit(skb, dev);
3743 /* Should never happen, mode already checked */
3744 pr_err("%s: Error: Unknown bonding mode %d\n",
3745 dev->name, bond->params.mode);
3748 return NETDEV_TX_OK;
3752 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3754 struct bonding *bond = netdev_priv(dev);
3755 netdev_tx_t ret = NETDEV_TX_OK;
3758 * If we risk deadlock from transmitting this in the
3759 * netpoll path, tell netpoll to queue the frame for later tx
3761 if (is_netpoll_tx_blocked(dev))
3762 return NETDEV_TX_BUSY;
3765 if (bond_has_slaves(bond))
3766 ret = __bond_start_xmit(skb, dev);
3774 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3775 struct ethtool_cmd *ecmd)
3777 struct bonding *bond = netdev_priv(bond_dev);
3778 unsigned long speed = 0;
3779 struct list_head *iter;
3780 struct slave *slave;
3782 ecmd->duplex = DUPLEX_UNKNOWN;
3783 ecmd->port = PORT_OTHER;
3785 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3786 * do not need to check mode. Though link speed might not represent
3787 * the true receive or transmit bandwidth (not all modes are symmetric)
3788 * this is an accurate maximum.
3790 read_lock(&bond->lock);
3791 bond_for_each_slave(bond, slave, iter) {
3792 if (SLAVE_IS_OK(slave)) {
3793 if (slave->speed != SPEED_UNKNOWN)
3794 speed += slave->speed;
3795 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3796 slave->duplex != DUPLEX_UNKNOWN)
3797 ecmd->duplex = slave->duplex;
3800 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3801 read_unlock(&bond->lock);
3806 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3807 struct ethtool_drvinfo *drvinfo)
3809 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3810 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3811 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3815 static const struct ethtool_ops bond_ethtool_ops = {
3816 .get_drvinfo = bond_ethtool_get_drvinfo,
3817 .get_settings = bond_ethtool_get_settings,
3818 .get_link = ethtool_op_get_link,
3821 static const struct net_device_ops bond_netdev_ops = {
3822 .ndo_init = bond_init,
3823 .ndo_uninit = bond_uninit,
3824 .ndo_open = bond_open,
3825 .ndo_stop = bond_close,
3826 .ndo_start_xmit = bond_start_xmit,
3827 .ndo_select_queue = bond_select_queue,
3828 .ndo_get_stats64 = bond_get_stats,
3829 .ndo_do_ioctl = bond_do_ioctl,
3830 .ndo_change_rx_flags = bond_change_rx_flags,
3831 .ndo_set_rx_mode = bond_set_rx_mode,
3832 .ndo_change_mtu = bond_change_mtu,
3833 .ndo_set_mac_address = bond_set_mac_address,
3834 .ndo_neigh_setup = bond_neigh_setup,
3835 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3836 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3837 #ifdef CONFIG_NET_POLL_CONTROLLER
3838 .ndo_netpoll_setup = bond_netpoll_setup,
3839 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3840 .ndo_poll_controller = bond_poll_controller,
3842 .ndo_add_slave = bond_enslave,
3843 .ndo_del_slave = bond_release,
3844 .ndo_fix_features = bond_fix_features,
3847 static const struct device_type bond_type = {
3851 static void bond_destructor(struct net_device *bond_dev)
3853 struct bonding *bond = netdev_priv(bond_dev);
3855 destroy_workqueue(bond->wq);
3856 free_netdev(bond_dev);
3859 void bond_setup(struct net_device *bond_dev)
3861 struct bonding *bond = netdev_priv(bond_dev);
3863 /* initialize rwlocks */
3864 rwlock_init(&bond->lock);
3865 rwlock_init(&bond->curr_slave_lock);
3866 bond->params = bonding_defaults;
3868 /* Initialize pointers */
3869 bond->dev = bond_dev;
3871 /* Initialize the device entry points */
3872 ether_setup(bond_dev);
3873 bond_dev->netdev_ops = &bond_netdev_ops;
3874 bond_dev->ethtool_ops = &bond_ethtool_ops;
3876 bond_dev->destructor = bond_destructor;
3878 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3880 /* Initialize the device options */
3881 bond_dev->tx_queue_len = 0;
3882 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3883 bond_dev->priv_flags |= IFF_BONDING;
3884 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3886 /* At first, we block adding VLANs. That's the only way to
3887 * prevent problems that occur when adding VLANs over an
3888 * empty bond. The block will be removed once non-challenged
3889 * slaves are enslaved.
3891 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3893 /* don't acquire bond device's netif_tx_lock when
3895 bond_dev->features |= NETIF_F_LLTX;
3897 /* By default, we declare the bond to be fully
3898 * VLAN hardware accelerated capable. Special
3899 * care is taken in the various xmit functions
3900 * when there are slaves that are not hw accel
3904 bond_dev->hw_features = BOND_VLAN_FEATURES |
3905 NETIF_F_HW_VLAN_CTAG_TX |
3906 NETIF_F_HW_VLAN_CTAG_RX |
3907 NETIF_F_HW_VLAN_CTAG_FILTER;
3909 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3910 bond_dev->features |= bond_dev->hw_features;
3914 * Destroy a bonding device.
3915 * Must be under rtnl_lock when this function is called.
3917 static void bond_uninit(struct net_device *bond_dev)
3919 struct bonding *bond = netdev_priv(bond_dev);
3920 struct list_head *iter;
3921 struct slave *slave;
3923 bond_netpoll_cleanup(bond_dev);
3925 /* Release the bonded slaves */
3926 bond_for_each_slave(bond, slave, iter)
3927 __bond_release_one(bond_dev, slave->dev, true);
3928 pr_info("%s: released all slaves\n", bond_dev->name);
3930 list_del(&bond->bond_list);
3932 bond_debug_unregister(bond);
3935 /*------------------------- Module initialization ---------------------------*/
3938 * Convert string input module parms. Accept either the
3939 * number of the mode or its string name. A bit complicated because
3940 * some mode names are substrings of other names, and calls from sysfs
3941 * may have whitespace in the name (trailing newlines, for example).
3943 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3945 int modeint = -1, i, rv;
3946 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
3948 for (p = (char *)buf; *p; p++)
3949 if (!(isdigit(*p) || isspace(*p)))
3953 rv = sscanf(buf, "%20s", modestr);
3955 rv = sscanf(buf, "%d", &modeint);
3960 for (i = 0; tbl[i].modename; i++) {
3961 if (modeint == tbl[i].mode)
3963 if (strcmp(modestr, tbl[i].modename) == 0)
3970 static int bond_check_params(struct bond_params *params)
3972 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3973 int arp_all_targets_value;
3976 * Convert string parameters.
3979 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
3980 if (bond_mode == -1) {
3981 pr_err("Error: Invalid bonding mode \"%s\"\n",
3982 mode == NULL ? "NULL" : mode);
3987 if (xmit_hash_policy) {
3988 if ((bond_mode != BOND_MODE_XOR) &&
3989 (bond_mode != BOND_MODE_8023AD)) {
3990 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3991 bond_mode_name(bond_mode));
3993 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
3995 if (xmit_hashtype == -1) {
3996 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
3997 xmit_hash_policy == NULL ? "NULL" :
4005 if (bond_mode != BOND_MODE_8023AD) {
4006 pr_info("lacp_rate param is irrelevant in mode %s\n",
4007 bond_mode_name(bond_mode));
4009 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4010 if (lacp_fast == -1) {
4011 pr_err("Error: Invalid lacp rate \"%s\"\n",
4012 lacp_rate == NULL ? "NULL" : lacp_rate);
4019 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4020 if (params->ad_select == -1) {
4021 pr_err("Error: Invalid ad_select \"%s\"\n",
4022 ad_select == NULL ? "NULL" : ad_select);
4026 if (bond_mode != BOND_MODE_8023AD) {
4027 pr_warning("ad_select param only affects 802.3ad mode\n");
4030 params->ad_select = BOND_AD_STABLE;
4033 if (max_bonds < 0) {
4034 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4035 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4036 max_bonds = BOND_DEFAULT_MAX_BONDS;
4040 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4041 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4042 miimon = BOND_LINK_MON_INTERV;
4046 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4051 if (downdelay < 0) {
4052 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4053 downdelay, INT_MAX);
4057 if ((use_carrier != 0) && (use_carrier != 1)) {
4058 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4063 if (num_peer_notif < 0 || num_peer_notif > 255) {
4064 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4069 /* reset values for 802.3ad */
4070 if (bond_mode == BOND_MODE_8023AD) {
4072 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");
4073 pr_warning("Forcing miimon to 100msec\n");
4074 miimon = BOND_DEFAULT_MIIMON;
4078 if (tx_queues < 1 || tx_queues > 255) {
4079 pr_warning("Warning: tx_queues (%d) should be between "
4080 "1 and 255, resetting to %d\n",
4081 tx_queues, BOND_DEFAULT_TX_QUEUES);
4082 tx_queues = BOND_DEFAULT_TX_QUEUES;
4085 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4086 pr_warning("Warning: all_slaves_active module parameter (%d), "
4087 "not of valid value (0/1), so it was set to "
4088 "0\n", all_slaves_active);
4089 all_slaves_active = 0;
4092 if (resend_igmp < 0 || resend_igmp > 255) {
4093 pr_warning("Warning: resend_igmp (%d) should be between "
4094 "0 and 255, resetting to %d\n",
4095 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4096 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4099 if (packets_per_slave < 0 || packets_per_slave > USHRT_MAX) {
4100 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4101 packets_per_slave, USHRT_MAX);
4102 packets_per_slave = 1;
4105 /* reset values for TLB/ALB */
4106 if ((bond_mode == BOND_MODE_TLB) ||
4107 (bond_mode == BOND_MODE_ALB)) {
4109 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4110 pr_warning("Forcing miimon to 100msec\n");
4111 miimon = BOND_DEFAULT_MIIMON;
4115 if (bond_mode == BOND_MODE_ALB) {
4116 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",
4121 if (updelay || downdelay) {
4122 /* just warn the user the up/down delay will have
4123 * no effect since miimon is zero...
4125 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",
4126 updelay, downdelay);
4129 /* don't allow arp monitoring */
4131 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4132 miimon, arp_interval);
4136 if ((updelay % miimon) != 0) {
4137 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4139 (updelay / miimon) * miimon);
4144 if ((downdelay % miimon) != 0) {
4145 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4147 (downdelay / miimon) * miimon);
4150 downdelay /= miimon;
4153 if (arp_interval < 0) {
4154 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4155 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4156 arp_interval = BOND_LINK_ARP_INTERV;
4159 for (arp_ip_count = 0, i = 0;
4160 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4161 /* not complete check, but should be good enough to
4164 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4165 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4166 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4170 if (bond_get_targets_ip(arp_target, ip) == -1)
4171 arp_target[arp_ip_count++] = ip;
4173 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4178 if (arp_interval && !arp_ip_count) {
4179 /* don't allow arping if no arp_ip_target given... */
4180 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4186 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4187 pr_err("arp_validate only supported in active-backup mode\n");
4190 if (!arp_interval) {
4191 pr_err("arp_validate requires arp_interval\n");
4195 arp_validate_value = bond_parse_parm(arp_validate,
4197 if (arp_validate_value == -1) {
4198 pr_err("Error: invalid arp_validate \"%s\"\n",
4199 arp_validate == NULL ? "NULL" : arp_validate);
4203 arp_validate_value = 0;
4205 arp_all_targets_value = 0;
4206 if (arp_all_targets) {
4207 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4208 arp_all_targets_tbl);
4210 if (arp_all_targets_value == -1) {
4211 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4213 arp_all_targets_value = 0;
4218 pr_info("MII link monitoring set to %d ms\n", miimon);
4219 } else if (arp_interval) {
4220 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4222 arp_validate_tbl[arp_validate_value].modename,
4225 for (i = 0; i < arp_ip_count; i++)
4226 pr_info(" %s", arp_ip_target[i]);
4230 } else if (max_bonds) {
4231 /* miimon and arp_interval not set, we need one so things
4232 * work as expected, see bonding.txt for details
4234 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");
4237 if (primary && !USES_PRIMARY(bond_mode)) {
4238 /* currently, using a primary only makes sense
4239 * in active backup, TLB or ALB modes
4241 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4242 primary, bond_mode_name(bond_mode));
4246 if (primary && primary_reselect) {
4247 primary_reselect_value = bond_parse_parm(primary_reselect,
4249 if (primary_reselect_value == -1) {
4250 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4252 NULL ? "NULL" : primary_reselect);
4256 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4259 if (fail_over_mac) {
4260 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4262 if (fail_over_mac_value == -1) {
4263 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4264 arp_validate == NULL ? "NULL" : arp_validate);
4268 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4269 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4271 fail_over_mac_value = BOND_FOM_NONE;
4274 /* fill params struct with the proper values */
4275 params->mode = bond_mode;
4276 params->xmit_policy = xmit_hashtype;
4277 params->miimon = miimon;
4278 params->num_peer_notif = num_peer_notif;
4279 params->arp_interval = arp_interval;
4280 params->arp_validate = arp_validate_value;
4281 params->arp_all_targets = arp_all_targets_value;
4282 params->updelay = updelay;
4283 params->downdelay = downdelay;
4284 params->use_carrier = use_carrier;
4285 params->lacp_fast = lacp_fast;
4286 params->primary[0] = 0;
4287 params->primary_reselect = primary_reselect_value;
4288 params->fail_over_mac = fail_over_mac_value;
4289 params->tx_queues = tx_queues;
4290 params->all_slaves_active = all_slaves_active;
4291 params->resend_igmp = resend_igmp;
4292 params->min_links = min_links;
4293 params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4294 if (packets_per_slave > 1)
4295 params->packets_per_slave = reciprocal_value(packets_per_slave);
4297 params->packets_per_slave = packets_per_slave;
4299 strncpy(params->primary, primary, IFNAMSIZ);
4300 params->primary[IFNAMSIZ - 1] = 0;
4303 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4308 static struct lock_class_key bonding_netdev_xmit_lock_key;
4309 static struct lock_class_key bonding_netdev_addr_lock_key;
4310 static struct lock_class_key bonding_tx_busylock_key;
4312 static void bond_set_lockdep_class_one(struct net_device *dev,
4313 struct netdev_queue *txq,
4316 lockdep_set_class(&txq->_xmit_lock,
4317 &bonding_netdev_xmit_lock_key);
4320 static void bond_set_lockdep_class(struct net_device *dev)
4322 lockdep_set_class(&dev->addr_list_lock,
4323 &bonding_netdev_addr_lock_key);
4324 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4325 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4329 * Called from registration process
4331 static int bond_init(struct net_device *bond_dev)
4333 struct bonding *bond = netdev_priv(bond_dev);
4334 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4335 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4337 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4340 * Initialize locks that may be required during
4341 * en/deslave operations. All of the bond_open work
4342 * (of which this is part) should really be moved to
4343 * a phase prior to dev_open
4345 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4346 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4348 bond->wq = create_singlethread_workqueue(bond_dev->name);
4352 bond_set_lockdep_class(bond_dev);
4354 list_add_tail(&bond->bond_list, &bn->dev_list);
4356 bond_prepare_sysfs_group(bond);
4358 bond_debug_register(bond);
4360 /* Ensure valid dev_addr */
4361 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4362 bond_dev->addr_assign_type == NET_ADDR_PERM)
4363 eth_hw_addr_random(bond_dev);
4368 unsigned int bond_get_num_tx_queues(void)
4373 /* Create a new bond based on the specified name and bonding parameters.
4374 * If name is NULL, obtain a suitable "bond%d" name for us.
4375 * Caller must NOT hold rtnl_lock; we need to release it here before we
4376 * set up our sysfs entries.
4378 int bond_create(struct net *net, const char *name)
4380 struct net_device *bond_dev;
4385 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4386 name ? name : "bond%d",
4387 bond_setup, tx_queues);
4389 pr_err("%s: eek! can't alloc netdev!\n", name);
4394 dev_net_set(bond_dev, net);
4395 bond_dev->rtnl_link_ops = &bond_link_ops;
4397 res = register_netdevice(bond_dev);
4399 netif_carrier_off(bond_dev);
4403 bond_destructor(bond_dev);
4407 static int __net_init bond_net_init(struct net *net)
4409 struct bond_net *bn = net_generic(net, bond_net_id);
4412 INIT_LIST_HEAD(&bn->dev_list);
4414 bond_create_proc_dir(bn);
4415 bond_create_sysfs(bn);
4420 static void __net_exit bond_net_exit(struct net *net)
4422 struct bond_net *bn = net_generic(net, bond_net_id);
4423 struct bonding *bond, *tmp_bond;
4426 bond_destroy_sysfs(bn);
4427 bond_destroy_proc_dir(bn);
4429 /* Kill off any bonds created after unregistering bond rtnl ops */
4431 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4432 unregister_netdevice_queue(bond->dev, &list);
4433 unregister_netdevice_many(&list);
4437 static struct pernet_operations bond_net_ops = {
4438 .init = bond_net_init,
4439 .exit = bond_net_exit,
4441 .size = sizeof(struct bond_net),
4444 static int __init bonding_init(void)
4449 pr_info("%s", bond_version);
4451 res = bond_check_params(&bonding_defaults);
4455 res = register_pernet_subsys(&bond_net_ops);
4459 res = bond_netlink_init();
4463 bond_create_debugfs();
4465 for (i = 0; i < max_bonds; i++) {
4466 res = bond_create(&init_net, NULL);
4471 register_netdevice_notifier(&bond_netdev_notifier);
4475 bond_netlink_fini();
4477 unregister_pernet_subsys(&bond_net_ops);
4482 static void __exit bonding_exit(void)
4484 unregister_netdevice_notifier(&bond_netdev_notifier);
4486 bond_destroy_debugfs();
4488 bond_netlink_fini();
4489 unregister_pernet_subsys(&bond_net_ops);
4491 #ifdef CONFIG_NET_POLL_CONTROLLER
4493 * Make sure we don't have an imbalance on our netpoll blocking
4495 WARN_ON(atomic_read(&netpoll_block_tx));
4499 module_init(bonding_init);
4500 module_exit(bonding_exit);
4501 MODULE_LICENSE("GPL");
4502 MODULE_VERSION(DRV_VERSION);
4503 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4504 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");