2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.h>
82 #include <linux/reciprocal_div.h>
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 #define BOND_LINK_MON_INTERV 0
91 #define BOND_LINK_ARP_INTERV 0
93 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
94 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
95 static int num_peer_notif = 1;
96 static int miimon = BOND_LINK_MON_INTERV;
99 static int use_carrier = 1;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static int min_links;
105 static char *ad_select;
106 static char *xmit_hash_policy;
107 static int arp_interval = BOND_LINK_ARP_INTERV;
108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
109 static char *arp_validate;
110 static char *arp_all_targets;
111 static char *fail_over_mac;
112 static int all_slaves_active;
113 static struct bond_params bonding_defaults;
114 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
115 static int packets_per_slave = 1;
116 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
118 module_param(max_bonds, int, 0);
119 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
120 module_param(tx_queues, int, 0);
121 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
122 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
124 "failover event (alias of num_unsol_na)");
125 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
126 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
127 "failover event (alias of num_grat_arp)");
128 module_param(miimon, int, 0);
129 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
130 module_param(updelay, int, 0);
131 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
132 module_param(downdelay, int, 0);
133 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
135 module_param(use_carrier, int, 0);
136 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
137 "0 for off, 1 for on (default)");
138 module_param(mode, charp, 0);
139 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
140 "1 for active-backup, 2 for balance-xor, "
141 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
142 "6 for balance-alb");
143 module_param(primary, charp, 0);
144 MODULE_PARM_DESC(primary, "Primary network device to use");
145 module_param(primary_reselect, charp, 0);
146 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
148 "0 for always (default), "
149 "1 for only if speed of primary is "
151 "2 for only on active slave "
153 module_param(lacp_rate, charp, 0);
154 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
155 "0 for slow, 1 for fast");
156 module_param(ad_select, charp, 0);
157 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
158 "0 for stable (default), 1 for bandwidth, "
160 module_param(min_links, int, 0);
161 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
163 module_param(xmit_hash_policy, charp, 0);
164 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
165 "0 for layer 2 (default), 1 for layer 3+4, "
166 "2 for layer 2+3, 3 for encap layer 2+3, "
167 "4 for encap layer 3+4");
168 module_param(arp_interval, int, 0);
169 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
170 module_param_array(arp_ip_target, charp, NULL, 0);
171 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
172 module_param(arp_validate, charp, 0);
173 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
174 "0 for none (default), 1 for active, "
175 "2 for backup, 3 for all");
176 module_param(arp_all_targets, charp, 0);
177 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
178 module_param(fail_over_mac, charp, 0);
179 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
180 "the same MAC; 0 for none (default), "
181 "1 for active, 2 for follow");
182 module_param(all_slaves_active, int, 0);
183 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
184 "by setting active flag for all slaves; "
185 "0 for never (default), 1 for always.");
186 module_param(resend_igmp, int, 0);
187 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
189 module_param(packets_per_slave, int, 0);
190 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
191 "mode; 0 for a random slave, 1 packet per "
192 "slave (default), >1 packets per slave.");
193 module_param(lp_interval, uint, 0);
194 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
195 "the bonding driver sends learning packets to "
196 "each slaves peer switch. The default is 1.");
198 /*----------------------------- Global variables ----------------------------*/
200 #ifdef CONFIG_NET_POLL_CONTROLLER
201 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
204 int bond_net_id __read_mostly;
206 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
207 static int arp_ip_count;
208 static int bond_mode = BOND_MODE_ROUNDROBIN;
209 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
210 static int lacp_fast;
212 const struct bond_parm_tbl bond_lacp_tbl[] = {
213 { "slow", AD_LACP_SLOW},
214 { "fast", AD_LACP_FAST},
218 const struct bond_parm_tbl bond_mode_tbl[] = {
219 { "balance-rr", BOND_MODE_ROUNDROBIN},
220 { "active-backup", BOND_MODE_ACTIVEBACKUP},
221 { "balance-xor", BOND_MODE_XOR},
222 { "broadcast", BOND_MODE_BROADCAST},
223 { "802.3ad", BOND_MODE_8023AD},
224 { "balance-tlb", BOND_MODE_TLB},
225 { "balance-alb", BOND_MODE_ALB},
229 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
230 { "layer2", BOND_XMIT_POLICY_LAYER2},
231 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
232 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
233 { "encap2+3", BOND_XMIT_POLICY_ENCAP23},
234 { "encap3+4", BOND_XMIT_POLICY_ENCAP34},
238 const struct bond_parm_tbl arp_all_targets_tbl[] = {
239 { "any", BOND_ARP_TARGETS_ANY},
240 { "all", BOND_ARP_TARGETS_ALL},
244 const struct bond_parm_tbl arp_validate_tbl[] = {
245 { "none", BOND_ARP_VALIDATE_NONE},
246 { "active", BOND_ARP_VALIDATE_ACTIVE},
247 { "backup", BOND_ARP_VALIDATE_BACKUP},
248 { "all", BOND_ARP_VALIDATE_ALL},
252 const struct bond_parm_tbl fail_over_mac_tbl[] = {
253 { "none", BOND_FOM_NONE},
254 { "active", BOND_FOM_ACTIVE},
255 { "follow", BOND_FOM_FOLLOW},
259 const struct bond_parm_tbl pri_reselect_tbl[] = {
260 { "always", BOND_PRI_RESELECT_ALWAYS},
261 { "better", BOND_PRI_RESELECT_BETTER},
262 { "failure", BOND_PRI_RESELECT_FAILURE},
266 struct bond_parm_tbl ad_select_tbl[] = {
267 { "stable", BOND_AD_STABLE},
268 { "bandwidth", BOND_AD_BANDWIDTH},
269 { "count", BOND_AD_COUNT},
273 /*-------------------------- Forward declarations ---------------------------*/
275 static int bond_init(struct net_device *bond_dev);
276 static void bond_uninit(struct net_device *bond_dev);
278 /*---------------------------- General routines -----------------------------*/
280 const char *bond_mode_name(int mode)
282 static const char *names[] = {
283 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
284 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
285 [BOND_MODE_XOR] = "load balancing (xor)",
286 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
287 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
288 [BOND_MODE_TLB] = "transmit load balancing",
289 [BOND_MODE_ALB] = "adaptive load balancing",
292 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
298 /*---------------------------------- VLAN -----------------------------------*/
301 * bond_dev_queue_xmit - Prepare skb for xmit.
303 * @bond: bond device that got this skb for tx.
304 * @skb: hw accel VLAN tagged skb to transmit
305 * @slave_dev: slave that is supposed to xmit this skbuff
307 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
308 struct net_device *slave_dev)
310 skb->dev = slave_dev;
312 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
313 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
314 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
316 if (unlikely(netpoll_tx_running(bond->dev)))
317 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
325 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
326 * We don't protect the slave list iteration with a lock because:
327 * a. This operation is performed in IOCTL context,
328 * b. The operation is protected by the RTNL semaphore in the 8021q code,
329 * c. Holding a lock with BH disabled while directly calling a base driver
330 * entry point is generally a BAD idea.
332 * The design of synchronization/protection for this operation in the 8021q
333 * module is good for one or more VLAN devices over a single physical device
334 * and cannot be extended for a teaming solution like bonding, so there is a
335 * potential race condition here where a net device from the vlan group might
336 * be referenced (either by a base driver or the 8021q code) while it is being
337 * removed from the system. However, it turns out we're not making matters
338 * worse, and if it works for regular VLAN usage it will work here too.
342 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
343 * @bond_dev: bonding net device that got called
344 * @vid: vlan id being added
346 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
347 __be16 proto, u16 vid)
349 struct bonding *bond = netdev_priv(bond_dev);
350 struct slave *slave, *rollback_slave;
351 struct list_head *iter;
354 bond_for_each_slave(bond, slave, iter) {
355 res = vlan_vid_add(slave->dev, proto, vid);
363 /* unwind to the slave that failed */
364 bond_for_each_slave(bond, rollback_slave, iter) {
365 if (rollback_slave == slave)
368 vlan_vid_del(rollback_slave->dev, proto, vid);
375 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
376 * @bond_dev: bonding net device that got called
377 * @vid: vlan id being removed
379 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
380 __be16 proto, u16 vid)
382 struct bonding *bond = netdev_priv(bond_dev);
383 struct list_head *iter;
386 bond_for_each_slave(bond, slave, iter)
387 vlan_vid_del(slave->dev, proto, vid);
389 if (bond_is_lb(bond))
390 bond_alb_clear_vlan(bond, vid);
395 /*------------------------------- Link status -------------------------------*/
398 * Set the carrier state for the master according to the state of its
399 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
400 * do special 802.3ad magic.
402 * Returns zero if carrier state does not change, nonzero if it does.
404 static int bond_set_carrier(struct bonding *bond)
406 struct list_head *iter;
409 if (!bond_has_slaves(bond))
412 if (bond->params.mode == BOND_MODE_8023AD)
413 return bond_3ad_set_carrier(bond);
415 bond_for_each_slave(bond, slave, iter) {
416 if (slave->link == BOND_LINK_UP) {
417 if (!netif_carrier_ok(bond->dev)) {
418 netif_carrier_on(bond->dev);
426 if (netif_carrier_ok(bond->dev)) {
427 netif_carrier_off(bond->dev);
434 * Get link speed and duplex from the slave's base driver
435 * using ethtool. If for some reason the call fails or the
436 * values are invalid, set speed and duplex to -1,
439 static void bond_update_speed_duplex(struct slave *slave)
441 struct net_device *slave_dev = slave->dev;
442 struct ethtool_cmd ecmd;
446 slave->speed = SPEED_UNKNOWN;
447 slave->duplex = DUPLEX_UNKNOWN;
449 res = __ethtool_get_settings(slave_dev, &ecmd);
453 slave_speed = ethtool_cmd_speed(&ecmd);
454 if (slave_speed == 0 || slave_speed == ((__u32) -1))
457 switch (ecmd.duplex) {
465 slave->speed = slave_speed;
466 slave->duplex = ecmd.duplex;
472 * if <dev> supports MII link status reporting, check its link status.
474 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
475 * depending upon the setting of the use_carrier parameter.
477 * Return either BMSR_LSTATUS, meaning that the link is up (or we
478 * can't tell and just pretend it is), or 0, meaning that the link is
481 * If reporting is non-zero, instead of faking link up, return -1 if
482 * both ETHTOOL and MII ioctls fail (meaning the device does not
483 * support them). If use_carrier is set, return whatever it says.
484 * It'd be nice if there was a good way to tell if a driver supports
485 * netif_carrier, but there really isn't.
487 static int bond_check_dev_link(struct bonding *bond,
488 struct net_device *slave_dev, int reporting)
490 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
491 int (*ioctl)(struct net_device *, struct ifreq *, int);
493 struct mii_ioctl_data *mii;
495 if (!reporting && !netif_running(slave_dev))
498 if (bond->params.use_carrier)
499 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
501 /* Try to get link status using Ethtool first. */
502 if (slave_dev->ethtool_ops->get_link)
503 return slave_dev->ethtool_ops->get_link(slave_dev) ?
506 /* Ethtool can't be used, fallback to MII ioctls. */
507 ioctl = slave_ops->ndo_do_ioctl;
509 /* TODO: set pointer to correct ioctl on a per team member */
510 /* bases to make this more efficient. that is, once */
511 /* we determine the correct ioctl, we will always */
512 /* call it and not the others for that team */
516 * We cannot assume that SIOCGMIIPHY will also read a
517 * register; not all network drivers (e.g., e100)
521 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
522 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
524 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
525 mii->reg_num = MII_BMSR;
526 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
527 return mii->val_out & BMSR_LSTATUS;
532 * If reporting, report that either there's no dev->do_ioctl,
533 * or both SIOCGMIIREG and get_link failed (meaning that we
534 * cannot report link status). If not reporting, pretend
537 return reporting ? -1 : BMSR_LSTATUS;
540 /*----------------------------- Multicast list ------------------------------*/
543 * Push the promiscuity flag down to appropriate slaves
545 static int bond_set_promiscuity(struct bonding *bond, int inc)
547 struct list_head *iter;
550 if (USES_PRIMARY(bond->params.mode)) {
551 /* write lock already acquired */
552 if (bond->curr_active_slave) {
553 err = dev_set_promiscuity(bond->curr_active_slave->dev,
559 bond_for_each_slave(bond, slave, iter) {
560 err = dev_set_promiscuity(slave->dev, inc);
569 * Push the allmulti flag down to all slaves
571 static int bond_set_allmulti(struct bonding *bond, int inc)
573 struct list_head *iter;
576 if (USES_PRIMARY(bond->params.mode)) {
577 /* write lock already acquired */
578 if (bond->curr_active_slave) {
579 err = dev_set_allmulti(bond->curr_active_slave->dev,
585 bond_for_each_slave(bond, slave, iter) {
586 err = dev_set_allmulti(slave->dev, inc);
595 * Retrieve the list of registered multicast addresses for the bonding
596 * device and retransmit an IGMP JOIN request to the current active
599 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
601 struct bonding *bond = container_of(work, struct bonding,
604 if (!rtnl_trylock()) {
605 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
608 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
610 if (bond->igmp_retrans > 1) {
611 bond->igmp_retrans--;
612 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
617 /* Flush bond's hardware addresses from slave
619 static void bond_hw_addr_flush(struct net_device *bond_dev,
620 struct net_device *slave_dev)
622 struct bonding *bond = netdev_priv(bond_dev);
624 dev_uc_unsync(slave_dev, bond_dev);
625 dev_mc_unsync(slave_dev, bond_dev);
627 if (bond->params.mode == BOND_MODE_8023AD) {
628 /* del lacpdu mc addr from mc list */
629 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
631 dev_mc_del(slave_dev, lacpdu_multicast);
635 /*--------------------------- Active slave change ---------------------------*/
637 /* Update the hardware address list and promisc/allmulti for the new and
638 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
639 * slaves up date at all times; only the USES_PRIMARY modes need to call
640 * this function to swap these settings during a failover.
642 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
643 struct slave *old_active)
648 if (bond->dev->flags & IFF_PROMISC)
649 dev_set_promiscuity(old_active->dev, -1);
651 if (bond->dev->flags & IFF_ALLMULTI)
652 dev_set_allmulti(old_active->dev, -1);
654 bond_hw_addr_flush(bond->dev, old_active->dev);
658 /* FIXME: Signal errors upstream. */
659 if (bond->dev->flags & IFF_PROMISC)
660 dev_set_promiscuity(new_active->dev, 1);
662 if (bond->dev->flags & IFF_ALLMULTI)
663 dev_set_allmulti(new_active->dev, 1);
665 netif_addr_lock_bh(bond->dev);
666 dev_uc_sync(new_active->dev, bond->dev);
667 dev_mc_sync(new_active->dev, bond->dev);
668 netif_addr_unlock_bh(bond->dev);
673 * bond_set_dev_addr - clone slave's address to bond
674 * @bond_dev: bond net device
675 * @slave_dev: slave net device
677 * Should be called with RTNL held.
679 static void bond_set_dev_addr(struct net_device *bond_dev,
680 struct net_device *slave_dev)
682 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
683 bond_dev, slave_dev, slave_dev->addr_len);
684 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
685 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
686 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
690 * bond_do_fail_over_mac
692 * Perform special MAC address swapping for fail_over_mac settings
694 * Called with RTNL, curr_slave_lock for write_bh.
696 static void bond_do_fail_over_mac(struct bonding *bond,
697 struct slave *new_active,
698 struct slave *old_active)
699 __releases(&bond->curr_slave_lock)
700 __acquires(&bond->curr_slave_lock)
702 u8 tmp_mac[ETH_ALEN];
703 struct sockaddr saddr;
706 switch (bond->params.fail_over_mac) {
707 case BOND_FOM_ACTIVE:
709 write_unlock_bh(&bond->curr_slave_lock);
710 bond_set_dev_addr(bond->dev, new_active->dev);
711 write_lock_bh(&bond->curr_slave_lock);
714 case BOND_FOM_FOLLOW:
716 * if new_active && old_active, swap them
717 * if just old_active, do nothing (going to no active slave)
718 * if just new_active, set new_active to bond's MAC
723 write_unlock_bh(&bond->curr_slave_lock);
726 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
727 memcpy(saddr.sa_data, old_active->dev->dev_addr,
729 saddr.sa_family = new_active->dev->type;
731 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
732 saddr.sa_family = bond->dev->type;
735 rv = dev_set_mac_address(new_active->dev, &saddr);
737 pr_err("%s: Error %d setting MAC of slave %s\n",
738 bond->dev->name, -rv, new_active->dev->name);
745 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
746 saddr.sa_family = old_active->dev->type;
748 rv = dev_set_mac_address(old_active->dev, &saddr);
750 pr_err("%s: Error %d setting MAC of slave %s\n",
751 bond->dev->name, -rv, new_active->dev->name);
753 write_lock_bh(&bond->curr_slave_lock);
756 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
757 bond->dev->name, bond->params.fail_over_mac);
763 static bool bond_should_change_active(struct bonding *bond)
765 struct slave *prim = bond->primary_slave;
766 struct slave *curr = bond->curr_active_slave;
768 if (!prim || !curr || curr->link != BOND_LINK_UP)
770 if (bond->force_primary) {
771 bond->force_primary = false;
774 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
775 (prim->speed < curr->speed ||
776 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
778 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
784 * find_best_interface - select the best available slave to be the active one
785 * @bond: our bonding struct
787 static struct slave *bond_find_best_slave(struct bonding *bond)
789 struct slave *slave, *bestslave = NULL;
790 struct list_head *iter;
791 int mintime = bond->params.updelay;
793 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
794 bond_should_change_active(bond))
795 return bond->primary_slave;
797 bond_for_each_slave(bond, slave, iter) {
798 if (slave->link == BOND_LINK_UP)
800 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
801 slave->delay < mintime) {
802 mintime = slave->delay;
810 static bool bond_should_notify_peers(struct bonding *bond)
815 slave = rcu_dereference(bond->curr_active_slave);
818 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
819 bond->dev->name, slave ? slave->dev->name : "NULL");
821 if (!slave || !bond->send_peer_notif ||
822 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
829 * change_active_interface - change the active slave into the specified one
830 * @bond: our bonding struct
831 * @new: the new slave to make the active one
833 * Set the new slave to the bond's settings and unset them on the old
835 * Setting include flags, mc-list, promiscuity, allmulti, etc.
837 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
838 * because it is apparently the best available slave we have, even though its
839 * updelay hasn't timed out yet.
841 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
843 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
845 struct slave *old_active = bond->curr_active_slave;
847 if (old_active == new_active)
851 new_active->jiffies = jiffies;
853 if (new_active->link == BOND_LINK_BACK) {
854 if (USES_PRIMARY(bond->params.mode)) {
855 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
856 bond->dev->name, new_active->dev->name,
857 (bond->params.updelay - new_active->delay) * bond->params.miimon);
860 new_active->delay = 0;
861 new_active->link = BOND_LINK_UP;
863 if (bond->params.mode == BOND_MODE_8023AD)
864 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
866 if (bond_is_lb(bond))
867 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
869 if (USES_PRIMARY(bond->params.mode)) {
870 pr_info("%s: making interface %s the new active one.\n",
871 bond->dev->name, new_active->dev->name);
876 if (USES_PRIMARY(bond->params.mode))
877 bond_hw_addr_swap(bond, new_active, old_active);
879 if (bond_is_lb(bond)) {
880 bond_alb_handle_active_change(bond, new_active);
882 bond_set_slave_inactive_flags(old_active);
884 bond_set_slave_active_flags(new_active);
886 rcu_assign_pointer(bond->curr_active_slave, new_active);
889 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
891 bond_set_slave_inactive_flags(old_active);
894 bool should_notify_peers = false;
896 bond_set_slave_active_flags(new_active);
898 if (bond->params.fail_over_mac)
899 bond_do_fail_over_mac(bond, new_active,
902 if (netif_running(bond->dev)) {
903 bond->send_peer_notif =
904 bond->params.num_peer_notif;
905 should_notify_peers =
906 bond_should_notify_peers(bond);
909 write_unlock_bh(&bond->curr_slave_lock);
911 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
912 if (should_notify_peers)
913 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
916 write_lock_bh(&bond->curr_slave_lock);
920 /* resend IGMP joins since active slave has changed or
921 * all were sent on curr_active_slave.
922 * resend only if bond is brought up with the affected
923 * bonding modes and the retransmission is enabled */
924 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
925 ((USES_PRIMARY(bond->params.mode) && new_active) ||
926 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
927 bond->igmp_retrans = bond->params.resend_igmp;
928 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
933 * bond_select_active_slave - select a new active slave, if needed
934 * @bond: our bonding struct
936 * This functions should be called when one of the following occurs:
937 * - The old curr_active_slave has been released or lost its link.
938 * - The primary_slave has got its link back.
939 * - A slave has got its link back and there's no old curr_active_slave.
941 * Caller must hold curr_slave_lock for write_bh.
943 void bond_select_active_slave(struct bonding *bond)
945 struct slave *best_slave;
948 best_slave = bond_find_best_slave(bond);
949 if (best_slave != bond->curr_active_slave) {
950 bond_change_active_slave(bond, best_slave);
951 rv = bond_set_carrier(bond);
955 if (netif_carrier_ok(bond->dev)) {
956 pr_info("%s: first active interface up!\n",
959 pr_info("%s: now running without any active interface !\n",
965 #ifdef CONFIG_NET_POLL_CONTROLLER
966 static inline int slave_enable_netpoll(struct slave *slave)
971 np = kzalloc(sizeof(*np), GFP_ATOMIC);
976 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
985 static inline void slave_disable_netpoll(struct slave *slave)
987 struct netpoll *np = slave->np;
993 __netpoll_free_async(np);
995 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
997 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
999 if (!slave_dev->netdev_ops->ndo_poll_controller)
1004 static void bond_poll_controller(struct net_device *bond_dev)
1008 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1010 struct bonding *bond = netdev_priv(bond_dev);
1011 struct list_head *iter;
1012 struct slave *slave;
1014 bond_for_each_slave(bond, slave, iter)
1015 if (IS_UP(slave->dev))
1016 slave_disable_netpoll(slave);
1019 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1021 struct bonding *bond = netdev_priv(dev);
1022 struct list_head *iter;
1023 struct slave *slave;
1026 bond_for_each_slave(bond, slave, iter) {
1027 err = slave_enable_netpoll(slave);
1029 bond_netpoll_cleanup(dev);
1036 static inline int slave_enable_netpoll(struct slave *slave)
1040 static inline void slave_disable_netpoll(struct slave *slave)
1043 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1048 /*---------------------------------- IOCTL ----------------------------------*/
1050 static netdev_features_t bond_fix_features(struct net_device *dev,
1051 netdev_features_t features)
1053 struct bonding *bond = netdev_priv(dev);
1054 struct list_head *iter;
1055 netdev_features_t mask;
1056 struct slave *slave;
1058 if (!bond_has_slaves(bond)) {
1059 /* Disable adding VLANs to empty bond. But why? --mq */
1060 features |= NETIF_F_VLAN_CHALLENGED;
1065 features &= ~NETIF_F_ONE_FOR_ALL;
1066 features |= NETIF_F_ALL_FOR_ALL;
1068 bond_for_each_slave(bond, slave, iter) {
1069 features = netdev_increment_features(features,
1070 slave->dev->features,
1073 features = netdev_add_tso_features(features, mask);
1078 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1079 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1080 NETIF_F_HIGHDMA | NETIF_F_LRO)
1082 static void bond_compute_features(struct bonding *bond)
1084 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1085 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1086 struct net_device *bond_dev = bond->dev;
1087 struct list_head *iter;
1088 struct slave *slave;
1089 unsigned short max_hard_header_len = ETH_HLEN;
1090 unsigned int gso_max_size = GSO_MAX_SIZE;
1091 u16 gso_max_segs = GSO_MAX_SEGS;
1093 if (!bond_has_slaves(bond))
1096 bond_for_each_slave(bond, slave, iter) {
1097 vlan_features = netdev_increment_features(vlan_features,
1098 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1100 dst_release_flag &= slave->dev->priv_flags;
1101 if (slave->dev->hard_header_len > max_hard_header_len)
1102 max_hard_header_len = slave->dev->hard_header_len;
1104 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1105 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1109 bond_dev->vlan_features = vlan_features;
1110 bond_dev->hard_header_len = max_hard_header_len;
1111 bond_dev->gso_max_segs = gso_max_segs;
1112 netif_set_gso_max_size(bond_dev, gso_max_size);
1114 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1115 bond_dev->priv_flags = flags | dst_release_flag;
1117 netdev_change_features(bond_dev);
1120 static void bond_setup_by_slave(struct net_device *bond_dev,
1121 struct net_device *slave_dev)
1123 bond_dev->header_ops = slave_dev->header_ops;
1125 bond_dev->type = slave_dev->type;
1126 bond_dev->hard_header_len = slave_dev->hard_header_len;
1127 bond_dev->addr_len = slave_dev->addr_len;
1129 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1130 slave_dev->addr_len);
1133 /* On bonding slaves other than the currently active slave, suppress
1134 * duplicates except for alb non-mcast/bcast.
1136 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1137 struct slave *slave,
1138 struct bonding *bond)
1140 if (bond_is_slave_inactive(slave)) {
1141 if (bond->params.mode == BOND_MODE_ALB &&
1142 skb->pkt_type != PACKET_BROADCAST &&
1143 skb->pkt_type != PACKET_MULTICAST)
1150 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1152 struct sk_buff *skb = *pskb;
1153 struct slave *slave;
1154 struct bonding *bond;
1155 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1157 int ret = RX_HANDLER_ANOTHER;
1159 skb = skb_share_check(skb, GFP_ATOMIC);
1161 return RX_HANDLER_CONSUMED;
1165 slave = bond_slave_get_rcu(skb->dev);
1168 if (bond->params.arp_interval)
1169 slave->dev->last_rx = jiffies;
1171 recv_probe = ACCESS_ONCE(bond->recv_probe);
1173 ret = recv_probe(skb, bond, slave);
1174 if (ret == RX_HANDLER_CONSUMED) {
1180 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1181 return RX_HANDLER_EXACT;
1184 skb->dev = bond->dev;
1186 if (bond->params.mode == BOND_MODE_ALB &&
1187 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1188 skb->pkt_type == PACKET_HOST) {
1190 if (unlikely(skb_cow_head(skb,
1191 skb->data - skb_mac_header(skb)))) {
1193 return RX_HANDLER_CONSUMED;
1195 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1201 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1202 struct net_device *slave_dev,
1203 struct slave *slave)
1207 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1210 slave_dev->flags |= IFF_SLAVE;
1211 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1215 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1216 struct net_device *slave_dev)
1218 netdev_upper_dev_unlink(slave_dev, bond_dev);
1219 slave_dev->flags &= ~IFF_SLAVE;
1220 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1223 /* enslave device <slave> to bond device <master> */
1224 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1226 struct bonding *bond = netdev_priv(bond_dev);
1227 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1228 struct slave *new_slave = NULL, *prev_slave;
1229 struct sockaddr addr;
1233 if (!bond->params.use_carrier &&
1234 slave_dev->ethtool_ops->get_link == NULL &&
1235 slave_ops->ndo_do_ioctl == NULL) {
1236 pr_warning("%s: Warning: no link monitoring support for %s\n",
1237 bond_dev->name, slave_dev->name);
1240 /* already enslaved */
1241 if (slave_dev->flags & IFF_SLAVE) {
1242 pr_debug("Error, Device was already enslaved\n");
1246 /* vlan challenged mutual exclusion */
1247 /* no need to lock since we're protected by rtnl_lock */
1248 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1249 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1250 if (vlan_uses_dev(bond_dev)) {
1251 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1252 bond_dev->name, slave_dev->name, bond_dev->name);
1255 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1256 bond_dev->name, slave_dev->name,
1257 slave_dev->name, bond_dev->name);
1260 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1264 * Old ifenslave binaries are no longer supported. These can
1265 * be identified with moderate accuracy by the state of the slave:
1266 * the current ifenslave will set the interface down prior to
1267 * enslaving it; the old ifenslave will not.
1269 if ((slave_dev->flags & IFF_UP)) {
1270 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1273 goto err_undo_flags;
1276 /* set bonding device ether type by slave - bonding netdevices are
1277 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1278 * there is a need to override some of the type dependent attribs/funcs.
1280 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1281 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1283 if (!bond_has_slaves(bond)) {
1284 if (bond_dev->type != slave_dev->type) {
1285 pr_debug("%s: change device type from %d to %d\n",
1287 bond_dev->type, slave_dev->type);
1289 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1291 res = notifier_to_errno(res);
1293 pr_err("%s: refused to change device type\n",
1296 goto err_undo_flags;
1299 /* Flush unicast and multicast addresses */
1300 dev_uc_flush(bond_dev);
1301 dev_mc_flush(bond_dev);
1303 if (slave_dev->type != ARPHRD_ETHER)
1304 bond_setup_by_slave(bond_dev, slave_dev);
1306 ether_setup(bond_dev);
1307 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1310 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1313 } else if (bond_dev->type != slave_dev->type) {
1314 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1316 slave_dev->type, bond_dev->type);
1318 goto err_undo_flags;
1321 if (slave_ops->ndo_set_mac_address == NULL) {
1322 if (!bond_has_slaves(bond)) {
1323 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1325 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1326 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1327 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",
1330 goto err_undo_flags;
1334 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1336 /* If this is the first slave, then we need to set the master's hardware
1337 * address to be the same as the slave's. */
1338 if (!bond_has_slaves(bond) &&
1339 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1340 bond_set_dev_addr(bond->dev, slave_dev);
1342 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1345 goto err_undo_flags;
1348 * Set the new_slave's queue_id to be zero. Queue ID mapping
1349 * is set via sysfs or module option if desired.
1351 new_slave->queue_id = 0;
1353 /* Save slave's original mtu and then set it to match the bond */
1354 new_slave->original_mtu = slave_dev->mtu;
1355 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1357 pr_debug("Error %d calling dev_set_mtu\n", res);
1362 * Save slave's original ("permanent") mac address for modes
1363 * that need it, and for restoring it upon release, and then
1364 * set it to the master's address
1366 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1368 if (!bond->params.fail_over_mac) {
1370 * Set slave to master's mac address. The application already
1371 * set the master's mac address to that of the first slave
1373 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1374 addr.sa_family = slave_dev->type;
1375 res = dev_set_mac_address(slave_dev, &addr);
1377 pr_debug("Error %d calling set_mac_address\n", res);
1378 goto err_restore_mtu;
1382 /* open the slave since the application closed it */
1383 res = dev_open(slave_dev);
1385 pr_debug("Opening slave %s failed\n", slave_dev->name);
1386 goto err_restore_mac;
1389 new_slave->bond = bond;
1390 new_slave->dev = slave_dev;
1391 slave_dev->priv_flags |= IFF_BONDING;
1393 if (bond_is_lb(bond)) {
1394 /* bond_alb_init_slave() must be called before all other stages since
1395 * it might fail and we do not want to have to undo everything
1397 res = bond_alb_init_slave(bond, new_slave);
1402 /* If the mode USES_PRIMARY, then the following is handled by
1403 * bond_change_active_slave().
1405 if (!USES_PRIMARY(bond->params.mode)) {
1406 /* set promiscuity level to new slave */
1407 if (bond_dev->flags & IFF_PROMISC) {
1408 res = dev_set_promiscuity(slave_dev, 1);
1413 /* set allmulti level to new slave */
1414 if (bond_dev->flags & IFF_ALLMULTI) {
1415 res = dev_set_allmulti(slave_dev, 1);
1420 netif_addr_lock_bh(bond_dev);
1422 dev_mc_sync_multiple(slave_dev, bond_dev);
1423 dev_uc_sync_multiple(slave_dev, bond_dev);
1425 netif_addr_unlock_bh(bond_dev);
1428 if (bond->params.mode == BOND_MODE_8023AD) {
1429 /* add lacpdu mc addr to mc list */
1430 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1432 dev_mc_add(slave_dev, lacpdu_multicast);
1435 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1437 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1438 bond_dev->name, slave_dev->name);
1442 prev_slave = bond_last_slave(bond);
1444 new_slave->delay = 0;
1445 new_slave->link_failure_count = 0;
1447 bond_update_speed_duplex(new_slave);
1449 new_slave->last_arp_rx = jiffies -
1450 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1451 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1452 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1454 if (bond->params.miimon && !bond->params.use_carrier) {
1455 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1457 if ((link_reporting == -1) && !bond->params.arp_interval) {
1459 * miimon is set but a bonded network driver
1460 * does not support ETHTOOL/MII and
1461 * arp_interval is not set. Note: if
1462 * use_carrier is enabled, we will never go
1463 * here (because netif_carrier is always
1464 * supported); thus, we don't need to change
1465 * the messages for netif_carrier.
1467 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",
1468 bond_dev->name, slave_dev->name);
1469 } else if (link_reporting == -1) {
1470 /* unable get link status using mii/ethtool */
1471 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",
1472 bond_dev->name, slave_dev->name);
1476 /* check for initial state */
1477 if (bond->params.miimon) {
1478 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1479 if (bond->params.updelay) {
1480 new_slave->link = BOND_LINK_BACK;
1481 new_slave->delay = bond->params.updelay;
1483 new_slave->link = BOND_LINK_UP;
1486 new_slave->link = BOND_LINK_DOWN;
1488 } else if (bond->params.arp_interval) {
1489 new_slave->link = (netif_carrier_ok(slave_dev) ?
1490 BOND_LINK_UP : BOND_LINK_DOWN);
1492 new_slave->link = BOND_LINK_UP;
1495 if (new_slave->link != BOND_LINK_DOWN)
1496 new_slave->jiffies = jiffies;
1497 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1498 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1499 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1501 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1502 /* if there is a primary slave, remember it */
1503 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1504 bond->primary_slave = new_slave;
1505 bond->force_primary = true;
1509 switch (bond->params.mode) {
1510 case BOND_MODE_ACTIVEBACKUP:
1511 bond_set_slave_inactive_flags(new_slave);
1513 case BOND_MODE_8023AD:
1514 /* in 802.3ad mode, the internal mechanism
1515 * will activate the slaves in the selected
1518 bond_set_slave_inactive_flags(new_slave);
1519 /* if this is the first slave */
1521 SLAVE_AD_INFO(new_slave).id = 1;
1522 /* Initialize AD with the number of times that the AD timer is called in 1 second
1523 * can be called only after the mac address of the bond is set
1525 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1527 SLAVE_AD_INFO(new_slave).id =
1528 SLAVE_AD_INFO(prev_slave).id + 1;
1531 bond_3ad_bind_slave(new_slave);
1535 bond_set_active_slave(new_slave);
1536 bond_set_slave_inactive_flags(new_slave);
1539 pr_debug("This slave is always active in trunk mode\n");
1541 /* always active in trunk mode */
1542 bond_set_active_slave(new_slave);
1544 /* In trunking mode there is little meaning to curr_active_slave
1545 * anyway (it holds no special properties of the bond device),
1546 * so we can change it without calling change_active_interface()
1548 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1549 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1552 } /* switch(bond_mode) */
1554 #ifdef CONFIG_NET_POLL_CONTROLLER
1555 slave_dev->npinfo = bond->dev->npinfo;
1556 if (slave_dev->npinfo) {
1557 if (slave_enable_netpoll(new_slave)) {
1558 read_unlock(&bond->lock);
1559 pr_info("Error, %s: master_dev is using netpoll, "
1560 "but new slave device does not support netpoll.\n",
1568 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1571 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1575 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1577 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1578 goto err_unregister;
1582 bond_compute_features(bond);
1583 bond_set_carrier(bond);
1585 if (USES_PRIMARY(bond->params.mode)) {
1586 write_lock_bh(&bond->curr_slave_lock);
1587 bond_select_active_slave(bond);
1588 write_unlock_bh(&bond->curr_slave_lock);
1591 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1592 bond_dev->name, slave_dev->name,
1593 bond_is_active_slave(new_slave) ? "n active" : " backup",
1594 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1596 /* enslave is successful */
1599 /* Undo stages on error */
1601 netdev_rx_handler_unregister(slave_dev);
1604 if (!USES_PRIMARY(bond->params.mode))
1605 bond_hw_addr_flush(bond_dev, slave_dev);
1607 vlan_vids_del_by_dev(slave_dev, bond_dev);
1608 if (bond->primary_slave == new_slave)
1609 bond->primary_slave = NULL;
1610 if (bond->curr_active_slave == new_slave) {
1611 write_lock_bh(&bond->curr_slave_lock);
1612 bond_change_active_slave(bond, NULL);
1613 bond_select_active_slave(bond);
1614 write_unlock_bh(&bond->curr_slave_lock);
1616 slave_disable_netpoll(new_slave);
1619 slave_dev->priv_flags &= ~IFF_BONDING;
1620 dev_close(slave_dev);
1623 if (!bond->params.fail_over_mac) {
1624 /* XXX TODO - fom follow mode needs to change master's
1625 * MAC if this slave's MAC is in use by the bond, or at
1626 * least print a warning.
1628 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1629 addr.sa_family = slave_dev->type;
1630 dev_set_mac_address(slave_dev, &addr);
1634 dev_set_mtu(slave_dev, new_slave->original_mtu);
1640 /* Enslave of first slave has failed and we need to fix master's mac */
1641 if (!bond_has_slaves(bond) &&
1642 ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1643 eth_hw_addr_random(bond_dev);
1649 * Try to release the slave device <slave> from the bond device <master>
1650 * It is legal to access curr_active_slave without a lock because all the function
1651 * is write-locked. If "all" is true it means that the function is being called
1652 * while destroying a bond interface and all slaves are being released.
1654 * The rules for slave state should be:
1655 * for Active/Backup:
1656 * Active stays on all backups go down
1657 * for Bonded connections:
1658 * The first up interface should be left on and all others downed.
1660 static int __bond_release_one(struct net_device *bond_dev,
1661 struct net_device *slave_dev,
1664 struct bonding *bond = netdev_priv(bond_dev);
1665 struct slave *slave, *oldcurrent;
1666 struct sockaddr addr;
1667 int old_flags = bond_dev->flags;
1668 netdev_features_t old_features = bond_dev->features;
1670 /* slave is not a slave or master is not master of this slave */
1671 if (!(slave_dev->flags & IFF_SLAVE) ||
1672 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1673 pr_err("%s: Error: cannot release %s.\n",
1674 bond_dev->name, slave_dev->name);
1680 slave = bond_get_slave_by_dev(bond, slave_dev);
1682 /* not a slave of this bond */
1683 pr_info("%s: %s not enslaved\n",
1684 bond_dev->name, slave_dev->name);
1685 unblock_netpoll_tx();
1689 /* release the slave from its bond */
1692 bond_upper_dev_unlink(bond_dev, slave_dev);
1693 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1694 * for this slave anymore.
1696 netdev_rx_handler_unregister(slave_dev);
1697 write_lock_bh(&bond->lock);
1699 /* Inform AD package of unbinding of slave. */
1700 if (bond->params.mode == BOND_MODE_8023AD)
1701 bond_3ad_unbind_slave(slave);
1703 write_unlock_bh(&bond->lock);
1705 pr_info("%s: releasing %s interface %s\n",
1707 bond_is_active_slave(slave) ? "active" : "backup",
1710 oldcurrent = bond->curr_active_slave;
1712 bond->current_arp_slave = NULL;
1714 if (!all && !bond->params.fail_over_mac) {
1715 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1716 bond_has_slaves(bond))
1717 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",
1718 bond_dev->name, slave_dev->name,
1720 bond_dev->name, slave_dev->name);
1723 if (bond->primary_slave == slave)
1724 bond->primary_slave = NULL;
1726 if (oldcurrent == slave) {
1727 write_lock_bh(&bond->curr_slave_lock);
1728 bond_change_active_slave(bond, NULL);
1729 write_unlock_bh(&bond->curr_slave_lock);
1732 if (bond_is_lb(bond)) {
1733 /* Must be called only after the slave has been
1734 * detached from the list and the curr_active_slave
1735 * has been cleared (if our_slave == old_current),
1736 * but before a new active slave is selected.
1738 bond_alb_deinit_slave(bond, slave);
1742 rcu_assign_pointer(bond->curr_active_slave, NULL);
1743 } else if (oldcurrent == slave) {
1745 * Note that we hold RTNL over this sequence, so there
1746 * is no concern that another slave add/remove event
1749 write_lock_bh(&bond->curr_slave_lock);
1751 bond_select_active_slave(bond);
1753 write_unlock_bh(&bond->curr_slave_lock);
1756 if (!bond_has_slaves(bond)) {
1757 bond_set_carrier(bond);
1758 eth_hw_addr_random(bond_dev);
1760 if (vlan_uses_dev(bond_dev)) {
1761 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1762 bond_dev->name, bond_dev->name);
1763 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1768 unblock_netpoll_tx();
1771 if (!bond_has_slaves(bond)) {
1772 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1773 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1776 bond_compute_features(bond);
1777 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1778 (old_features & NETIF_F_VLAN_CHALLENGED))
1779 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1780 bond_dev->name, slave_dev->name, bond_dev->name);
1782 /* must do this from outside any spinlocks */
1783 vlan_vids_del_by_dev(slave_dev, bond_dev);
1785 /* If the mode USES_PRIMARY, then this cases was handled above by
1786 * bond_change_active_slave(..., NULL)
1788 if (!USES_PRIMARY(bond->params.mode)) {
1789 /* unset promiscuity level from slave
1790 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1791 * of the IFF_PROMISC flag in the bond_dev, but we need the
1792 * value of that flag before that change, as that was the value
1793 * when this slave was attached, so we cache at the start of the
1794 * function and use it here. Same goes for ALLMULTI below
1796 if (old_flags & IFF_PROMISC)
1797 dev_set_promiscuity(slave_dev, -1);
1799 /* unset allmulti level from slave */
1800 if (old_flags & IFF_ALLMULTI)
1801 dev_set_allmulti(slave_dev, -1);
1803 bond_hw_addr_flush(bond_dev, slave_dev);
1806 slave_disable_netpoll(slave);
1808 /* close slave before restoring its mac address */
1809 dev_close(slave_dev);
1811 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1812 /* restore original ("permanent") mac address */
1813 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1814 addr.sa_family = slave_dev->type;
1815 dev_set_mac_address(slave_dev, &addr);
1818 dev_set_mtu(slave_dev, slave->original_mtu);
1820 slave_dev->priv_flags &= ~IFF_BONDING;
1824 return 0; /* deletion OK */
1827 /* A wrapper used because of ndo_del_link */
1828 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1830 return __bond_release_one(bond_dev, slave_dev, false);
1834 * First release a slave and then destroy the bond if no more slaves are left.
1835 * Must be under rtnl_lock when this function is called.
1837 static int bond_release_and_destroy(struct net_device *bond_dev,
1838 struct net_device *slave_dev)
1840 struct bonding *bond = netdev_priv(bond_dev);
1843 ret = bond_release(bond_dev, slave_dev);
1844 if (ret == 0 && !bond_has_slaves(bond)) {
1845 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1846 pr_info("%s: destroying bond %s.\n",
1847 bond_dev->name, bond_dev->name);
1848 unregister_netdevice(bond_dev);
1853 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1855 struct bonding *bond = netdev_priv(bond_dev);
1857 info->bond_mode = bond->params.mode;
1858 info->miimon = bond->params.miimon;
1860 read_lock(&bond->lock);
1861 info->num_slaves = bond->slave_cnt;
1862 read_unlock(&bond->lock);
1867 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1869 struct bonding *bond = netdev_priv(bond_dev);
1870 struct list_head *iter;
1871 int i = 0, res = -ENODEV;
1872 struct slave *slave;
1874 read_lock(&bond->lock);
1875 bond_for_each_slave(bond, slave, iter) {
1876 if (i++ == (int)info->slave_id) {
1878 strcpy(info->slave_name, slave->dev->name);
1879 info->link = slave->link;
1880 info->state = bond_slave_state(slave);
1881 info->link_failure_count = slave->link_failure_count;
1885 read_unlock(&bond->lock);
1890 /*-------------------------------- Monitoring -------------------------------*/
1893 static int bond_miimon_inspect(struct bonding *bond)
1895 int link_state, commit = 0;
1896 struct list_head *iter;
1897 struct slave *slave;
1898 bool ignore_updelay;
1900 ignore_updelay = !bond->curr_active_slave ? true : false;
1902 bond_for_each_slave_rcu(bond, slave, iter) {
1903 slave->new_link = BOND_LINK_NOCHANGE;
1905 link_state = bond_check_dev_link(bond, slave->dev, 0);
1907 switch (slave->link) {
1912 slave->link = BOND_LINK_FAIL;
1913 slave->delay = bond->params.downdelay;
1915 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1917 (bond->params.mode ==
1918 BOND_MODE_ACTIVEBACKUP) ?
1919 (bond_is_active_slave(slave) ?
1920 "active " : "backup ") : "",
1922 bond->params.downdelay * bond->params.miimon);
1925 case BOND_LINK_FAIL:
1928 * recovered before downdelay expired
1930 slave->link = BOND_LINK_UP;
1931 slave->jiffies = jiffies;
1932 pr_info("%s: link status up again after %d ms for interface %s.\n",
1934 (bond->params.downdelay - slave->delay) *
1935 bond->params.miimon,
1940 if (slave->delay <= 0) {
1941 slave->new_link = BOND_LINK_DOWN;
1949 case BOND_LINK_DOWN:
1953 slave->link = BOND_LINK_BACK;
1954 slave->delay = bond->params.updelay;
1957 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1958 bond->dev->name, slave->dev->name,
1959 ignore_updelay ? 0 :
1960 bond->params.updelay *
1961 bond->params.miimon);
1964 case BOND_LINK_BACK:
1966 slave->link = BOND_LINK_DOWN;
1967 pr_info("%s: link status down again after %d ms for interface %s.\n",
1969 (bond->params.updelay - slave->delay) *
1970 bond->params.miimon,
1979 if (slave->delay <= 0) {
1980 slave->new_link = BOND_LINK_UP;
1982 ignore_updelay = false;
1994 static void bond_miimon_commit(struct bonding *bond)
1996 struct list_head *iter;
1997 struct slave *slave;
1999 bond_for_each_slave(bond, slave, iter) {
2000 switch (slave->new_link) {
2001 case BOND_LINK_NOCHANGE:
2005 slave->link = BOND_LINK_UP;
2006 slave->jiffies = jiffies;
2008 if (bond->params.mode == BOND_MODE_8023AD) {
2009 /* prevent it from being the active one */
2010 bond_set_backup_slave(slave);
2011 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2012 /* make it immediately active */
2013 bond_set_active_slave(slave);
2014 } else if (slave != bond->primary_slave) {
2015 /* prevent it from being the active one */
2016 bond_set_backup_slave(slave);
2019 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2020 bond->dev->name, slave->dev->name,
2021 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2022 slave->duplex ? "full" : "half");
2024 /* notify ad that the link status has changed */
2025 if (bond->params.mode == BOND_MODE_8023AD)
2026 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2028 if (bond_is_lb(bond))
2029 bond_alb_handle_link_change(bond, slave,
2032 if (!bond->curr_active_slave ||
2033 (slave == bond->primary_slave))
2038 case BOND_LINK_DOWN:
2039 if (slave->link_failure_count < UINT_MAX)
2040 slave->link_failure_count++;
2042 slave->link = BOND_LINK_DOWN;
2044 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2045 bond->params.mode == BOND_MODE_8023AD)
2046 bond_set_slave_inactive_flags(slave);
2048 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2049 bond->dev->name, slave->dev->name);
2051 if (bond->params.mode == BOND_MODE_8023AD)
2052 bond_3ad_handle_link_change(slave,
2055 if (bond_is_lb(bond))
2056 bond_alb_handle_link_change(bond, slave,
2059 if (slave == bond->curr_active_slave)
2065 pr_err("%s: invalid new link %d on slave %s\n",
2066 bond->dev->name, slave->new_link,
2068 slave->new_link = BOND_LINK_NOCHANGE;
2076 write_lock_bh(&bond->curr_slave_lock);
2077 bond_select_active_slave(bond);
2078 write_unlock_bh(&bond->curr_slave_lock);
2079 unblock_netpoll_tx();
2082 bond_set_carrier(bond);
2088 * Really a wrapper that splits the mii monitor into two phases: an
2089 * inspection, then (if inspection indicates something needs to be done)
2090 * an acquisition of appropriate locks followed by a commit phase to
2091 * implement whatever link state changes are indicated.
2093 void bond_mii_monitor(struct work_struct *work)
2095 struct bonding *bond = container_of(work, struct bonding,
2097 bool should_notify_peers = false;
2098 unsigned long delay;
2100 delay = msecs_to_jiffies(bond->params.miimon);
2102 if (!bond_has_slaves(bond))
2107 should_notify_peers = bond_should_notify_peers(bond);
2109 if (bond_miimon_inspect(bond)) {
2112 /* Race avoidance with bond_close cancel of workqueue */
2113 if (!rtnl_trylock()) {
2115 should_notify_peers = false;
2119 bond_miimon_commit(bond);
2121 rtnl_unlock(); /* might sleep, hold no other locks */
2126 if (bond->params.miimon)
2127 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2129 if (should_notify_peers) {
2130 if (!rtnl_trylock())
2132 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2137 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2139 struct net_device *upper;
2140 struct list_head *iter;
2143 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2147 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2148 if (ip == bond_confirm_addr(upper, 0, ip)) {
2159 * We go to the (large) trouble of VLAN tagging ARP frames because
2160 * switches in VLAN mode (especially if ports are configured as
2161 * "native" to a VLAN) might not pass non-tagged frames.
2163 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2165 struct sk_buff *skb;
2167 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2168 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2170 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2171 NULL, slave_dev->dev_addr, NULL);
2174 pr_err("ARP packet allocation failed\n");
2178 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2180 pr_err("failed to insert VLAN tag\n");
2188 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2190 struct net_device *upper, *vlan_upper;
2191 struct list_head *iter, *vlan_iter;
2193 __be32 *targets = bond->params.arp_targets, addr;
2196 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2197 pr_debug("basa: target %pI4\n", &targets[i]);
2199 /* Find out through which dev should the packet go */
2200 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2203 pr_debug("%s: no route to arp_ip_target %pI4\n",
2204 bond->dev->name, &targets[i]);
2210 /* bond device itself */
2211 if (rt->dst.dev == bond->dev)
2215 /* first we search only for vlan devices. for every vlan
2216 * found we verify its upper dev list, searching for the
2217 * rt->dst.dev. If found we save the tag of the vlan and
2218 * proceed to send the packet.
2222 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2224 if (!is_vlan_dev(vlan_upper))
2226 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2228 if (upper == rt->dst.dev) {
2229 vlan_id = vlan_dev_vlan_id(vlan_upper);
2236 /* if the device we're looking for is not on top of any of
2237 * our upper vlans, then just search for any dev that
2238 * matches, and in case it's a vlan - save the id
2240 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2241 if (upper == rt->dst.dev) {
2242 /* if it's a vlan - get its VID */
2243 if (is_vlan_dev(upper))
2244 vlan_id = vlan_dev_vlan_id(upper);
2252 /* Not our device - skip */
2253 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2254 bond->dev->name, &targets[i],
2255 rt->dst.dev ? rt->dst.dev->name : "NULL");
2261 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2263 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2268 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2272 if (!sip || !bond_has_this_ip(bond, tip)) {
2273 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2277 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2279 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2282 slave->last_arp_rx = jiffies;
2283 slave->target_last_arp_rx[i] = jiffies;
2286 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2287 struct slave *slave)
2289 struct arphdr *arp = (struct arphdr *)skb->data;
2290 unsigned char *arp_ptr;
2294 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2295 return RX_HANDLER_ANOTHER;
2297 read_lock(&bond->lock);
2299 if (!slave_do_arp_validate(bond, slave))
2302 alen = arp_hdr_len(bond->dev);
2304 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2305 bond->dev->name, skb->dev->name);
2307 if (alen > skb_headlen(skb)) {
2308 arp = kmalloc(alen, GFP_ATOMIC);
2311 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2315 if (arp->ar_hln != bond->dev->addr_len ||
2316 skb->pkt_type == PACKET_OTHERHOST ||
2317 skb->pkt_type == PACKET_LOOPBACK ||
2318 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2319 arp->ar_pro != htons(ETH_P_IP) ||
2323 arp_ptr = (unsigned char *)(arp + 1);
2324 arp_ptr += bond->dev->addr_len;
2325 memcpy(&sip, arp_ptr, 4);
2326 arp_ptr += 4 + bond->dev->addr_len;
2327 memcpy(&tip, arp_ptr, 4);
2329 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2330 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2331 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2335 * Backup slaves won't see the ARP reply, but do come through
2336 * here for each ARP probe (so we swap the sip/tip to validate
2337 * the probe). In a "redundant switch, common router" type of
2338 * configuration, the ARP probe will (hopefully) travel from
2339 * the active, through one switch, the router, then the other
2340 * switch before reaching the backup.
2342 * We 'trust' the arp requests if there is an active slave and
2343 * it received valid arp reply(s) after it became active. This
2344 * is done to avoid endless looping when we can't reach the
2345 * arp_ip_target and fool ourselves with our own arp requests.
2347 if (bond_is_active_slave(slave))
2348 bond_validate_arp(bond, slave, sip, tip);
2349 else if (bond->curr_active_slave &&
2350 time_after(slave_last_rx(bond, bond->curr_active_slave),
2351 bond->curr_active_slave->jiffies))
2352 bond_validate_arp(bond, slave, tip, sip);
2355 read_unlock(&bond->lock);
2356 if (arp != (struct arphdr *)skb->data)
2358 return RX_HANDLER_ANOTHER;
2361 /* function to verify if we're in the arp_interval timeslice, returns true if
2362 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2363 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2365 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2368 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2370 return time_in_range(jiffies,
2371 last_act - delta_in_ticks,
2372 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2376 * this function is called regularly to monitor each slave's link
2377 * ensuring that traffic is being sent and received when arp monitoring
2378 * is used in load-balancing mode. if the adapter has been dormant, then an
2379 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2380 * arp monitoring in active backup mode.
2382 void bond_loadbalance_arp_mon(struct work_struct *work)
2384 struct bonding *bond = container_of(work, struct bonding,
2386 struct slave *slave, *oldcurrent;
2387 struct list_head *iter;
2388 int do_failover = 0;
2390 if (!bond_has_slaves(bond))
2395 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2396 /* see if any of the previous devices are up now (i.e. they have
2397 * xmt and rcv traffic). the curr_active_slave does not come into
2398 * the picture unless it is null. also, slave->jiffies is not needed
2399 * here because we send an arp on each slave and give a slave as
2400 * long as it needs to get the tx/rx within the delta.
2401 * TODO: what about up/down delay in arp mode? it wasn't here before
2404 bond_for_each_slave_rcu(bond, slave, iter) {
2405 unsigned long trans_start = dev_trans_start(slave->dev);
2407 if (slave->link != BOND_LINK_UP) {
2408 if (bond_time_in_interval(bond, trans_start, 1) &&
2409 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2411 slave->link = BOND_LINK_UP;
2412 bond_set_active_slave(slave);
2414 /* primary_slave has no meaning in round-robin
2415 * mode. the window of a slave being up and
2416 * curr_active_slave being null after enslaving
2420 pr_info("%s: link status definitely up for interface %s, ",
2425 pr_info("%s: interface %s is now up\n",
2431 /* slave->link == BOND_LINK_UP */
2433 /* not all switches will respond to an arp request
2434 * when the source ip is 0, so don't take the link down
2435 * if we don't know our ip yet
2437 if (!bond_time_in_interval(bond, trans_start, 2) ||
2438 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2440 slave->link = BOND_LINK_DOWN;
2441 bond_set_backup_slave(slave);
2443 if (slave->link_failure_count < UINT_MAX)
2444 slave->link_failure_count++;
2446 pr_info("%s: interface %s is now down.\n",
2450 if (slave == oldcurrent)
2455 /* note: if switch is in round-robin mode, all links
2456 * must tx arp to ensure all links rx an arp - otherwise
2457 * links may oscillate or not come up at all; if switch is
2458 * in something like xor mode, there is nothing we can
2459 * do - all replies will be rx'ed on same link causing slaves
2460 * to be unstable during low/no traffic periods
2462 if (IS_UP(slave->dev))
2463 bond_arp_send_all(bond, slave);
2469 /* the bond_select_active_slave must hold RTNL
2470 * and curr_slave_lock for write.
2472 if (!rtnl_trylock())
2475 write_lock_bh(&bond->curr_slave_lock);
2477 bond_select_active_slave(bond);
2479 write_unlock_bh(&bond->curr_slave_lock);
2480 unblock_netpoll_tx();
2485 if (bond->params.arp_interval)
2486 queue_delayed_work(bond->wq, &bond->arp_work,
2487 msecs_to_jiffies(bond->params.arp_interval));
2491 * Called to inspect slaves for active-backup mode ARP monitor link state
2492 * changes. Sets new_link in slaves to specify what action should take
2493 * place for the slave. Returns 0 if no changes are found, >0 if changes
2494 * to link states must be committed.
2496 * Called with rcu_read_lock hold.
2498 static int bond_ab_arp_inspect(struct bonding *bond)
2500 unsigned long trans_start, last_rx;
2501 struct list_head *iter;
2502 struct slave *slave;
2505 bond_for_each_slave_rcu(bond, slave, iter) {
2506 slave->new_link = BOND_LINK_NOCHANGE;
2507 last_rx = slave_last_rx(bond, slave);
2509 if (slave->link != BOND_LINK_UP) {
2510 if (bond_time_in_interval(bond, last_rx, 1)) {
2511 slave->new_link = BOND_LINK_UP;
2518 * Give slaves 2*delta after being enslaved or made
2519 * active. This avoids bouncing, as the last receive
2520 * times need a full ARP monitor cycle to be updated.
2522 if (bond_time_in_interval(bond, slave->jiffies, 2))
2526 * Backup slave is down if:
2527 * - No current_arp_slave AND
2528 * - more than 3*delta since last receive AND
2529 * - the bond has an IP address
2531 * Note: a non-null current_arp_slave indicates
2532 * the curr_active_slave went down and we are
2533 * searching for a new one; under this condition
2534 * we only take the curr_active_slave down - this
2535 * gives each slave a chance to tx/rx traffic
2536 * before being taken out
2538 if (!bond_is_active_slave(slave) &&
2539 !bond->current_arp_slave &&
2540 !bond_time_in_interval(bond, last_rx, 3)) {
2541 slave->new_link = BOND_LINK_DOWN;
2546 * Active slave is down if:
2547 * - more than 2*delta since transmitting OR
2548 * - (more than 2*delta since receive AND
2549 * the bond has an IP address)
2551 trans_start = dev_trans_start(slave->dev);
2552 if (bond_is_active_slave(slave) &&
2553 (!bond_time_in_interval(bond, trans_start, 2) ||
2554 !bond_time_in_interval(bond, last_rx, 2))) {
2555 slave->new_link = BOND_LINK_DOWN;
2564 * Called to commit link state changes noted by inspection step of
2565 * active-backup mode ARP monitor.
2567 * Called with RTNL hold.
2569 static void bond_ab_arp_commit(struct bonding *bond)
2571 unsigned long trans_start;
2572 struct list_head *iter;
2573 struct slave *slave;
2575 bond_for_each_slave(bond, slave, iter) {
2576 switch (slave->new_link) {
2577 case BOND_LINK_NOCHANGE:
2581 trans_start = dev_trans_start(slave->dev);
2582 if (bond->curr_active_slave != slave ||
2583 (!bond->curr_active_slave &&
2584 bond_time_in_interval(bond, trans_start, 1))) {
2585 slave->link = BOND_LINK_UP;
2586 if (bond->current_arp_slave) {
2587 bond_set_slave_inactive_flags(
2588 bond->current_arp_slave);
2589 bond->current_arp_slave = NULL;
2592 pr_info("%s: link status definitely up for interface %s.\n",
2593 bond->dev->name, slave->dev->name);
2595 if (!bond->curr_active_slave ||
2596 (slave == bond->primary_slave))
2603 case BOND_LINK_DOWN:
2604 if (slave->link_failure_count < UINT_MAX)
2605 slave->link_failure_count++;
2607 slave->link = BOND_LINK_DOWN;
2608 bond_set_slave_inactive_flags(slave);
2610 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2611 bond->dev->name, slave->dev->name);
2613 if (slave == bond->curr_active_slave) {
2614 bond->current_arp_slave = NULL;
2621 pr_err("%s: impossible: new_link %d on slave %s\n",
2622 bond->dev->name, slave->new_link,
2630 write_lock_bh(&bond->curr_slave_lock);
2631 bond_select_active_slave(bond);
2632 write_unlock_bh(&bond->curr_slave_lock);
2633 unblock_netpoll_tx();
2636 bond_set_carrier(bond);
2640 * Send ARP probes for active-backup mode ARP monitor.
2642 * Called with rcu_read_lock hold.
2644 static void bond_ab_arp_probe(struct bonding *bond)
2646 struct slave *slave, *before = NULL, *new_slave = NULL,
2647 *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2648 struct list_head *iter;
2651 read_lock(&bond->curr_slave_lock);
2653 if (curr_arp_slave && bond->curr_active_slave)
2654 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2655 curr_arp_slave->dev->name,
2656 bond->curr_active_slave->dev->name);
2658 if (bond->curr_active_slave) {
2659 bond_arp_send_all(bond, bond->curr_active_slave);
2660 read_unlock(&bond->curr_slave_lock);
2664 read_unlock(&bond->curr_slave_lock);
2666 /* if we don't have a curr_active_slave, search for the next available
2667 * backup slave from the current_arp_slave and make it the candidate
2668 * for becoming the curr_active_slave
2671 if (!curr_arp_slave) {
2672 curr_arp_slave = bond_first_slave_rcu(bond);
2673 if (!curr_arp_slave)
2677 bond_set_slave_inactive_flags(curr_arp_slave);
2679 bond_for_each_slave_rcu(bond, slave, iter) {
2680 if (!found && !before && IS_UP(slave->dev))
2683 if (found && !new_slave && IS_UP(slave->dev))
2685 /* if the link state is up at this point, we
2686 * mark it down - this can happen if we have
2687 * simultaneous link failures and
2688 * reselect_active_interface doesn't make this
2689 * one the current slave so it is still marked
2690 * up when it is actually down
2692 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2693 slave->link = BOND_LINK_DOWN;
2694 if (slave->link_failure_count < UINT_MAX)
2695 slave->link_failure_count++;
2697 bond_set_slave_inactive_flags(slave);
2699 pr_info("%s: backup interface %s is now down.\n",
2700 bond->dev->name, slave->dev->name);
2702 if (slave == curr_arp_slave)
2706 if (!new_slave && before)
2712 new_slave->link = BOND_LINK_BACK;
2713 bond_set_slave_active_flags(new_slave);
2714 bond_arp_send_all(bond, new_slave);
2715 new_slave->jiffies = jiffies;
2716 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2719 void bond_activebackup_arp_mon(struct work_struct *work)
2721 struct bonding *bond = container_of(work, struct bonding,
2723 bool should_notify_peers = false;
2726 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2728 if (!bond_has_slaves(bond))
2733 should_notify_peers = bond_should_notify_peers(bond);
2735 if (bond_ab_arp_inspect(bond)) {
2738 /* Race avoidance with bond_close flush of workqueue */
2739 if (!rtnl_trylock()) {
2741 should_notify_peers = false;
2745 bond_ab_arp_commit(bond);
2751 bond_ab_arp_probe(bond);
2755 if (bond->params.arp_interval)
2756 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2758 if (should_notify_peers) {
2759 if (!rtnl_trylock())
2761 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2766 /*-------------------------- netdev event handling --------------------------*/
2769 * Change device name
2771 static int bond_event_changename(struct bonding *bond)
2773 bond_remove_proc_entry(bond);
2774 bond_create_proc_entry(bond);
2776 bond_debug_reregister(bond);
2781 static int bond_master_netdev_event(unsigned long event,
2782 struct net_device *bond_dev)
2784 struct bonding *event_bond = netdev_priv(bond_dev);
2787 case NETDEV_CHANGENAME:
2788 return bond_event_changename(event_bond);
2789 case NETDEV_UNREGISTER:
2790 bond_remove_proc_entry(event_bond);
2792 case NETDEV_REGISTER:
2793 bond_create_proc_entry(event_bond);
2795 case NETDEV_NOTIFY_PEERS:
2796 if (event_bond->send_peer_notif)
2797 event_bond->send_peer_notif--;
2806 static int bond_slave_netdev_event(unsigned long event,
2807 struct net_device *slave_dev)
2809 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2810 struct bonding *bond;
2811 struct net_device *bond_dev;
2815 /* A netdev event can be generated while enslaving a device
2816 * before netdev_rx_handler_register is called in which case
2817 * slave will be NULL
2821 bond_dev = slave->bond->dev;
2825 case NETDEV_UNREGISTER:
2826 if (bond_dev->type != ARPHRD_ETHER)
2827 bond_release_and_destroy(bond_dev, slave_dev);
2829 bond_release(bond_dev, slave_dev);
2833 old_speed = slave->speed;
2834 old_duplex = slave->duplex;
2836 bond_update_speed_duplex(slave);
2838 if (bond->params.mode == BOND_MODE_8023AD) {
2839 if (old_speed != slave->speed)
2840 bond_3ad_adapter_speed_changed(slave);
2841 if (old_duplex != slave->duplex)
2842 bond_3ad_adapter_duplex_changed(slave);
2847 * ... Or is it this?
2850 case NETDEV_CHANGEMTU:
2852 * TODO: Should slaves be allowed to
2853 * independently alter their MTU? For
2854 * an active-backup bond, slaves need
2855 * not be the same type of device, so
2856 * MTUs may vary. For other modes,
2857 * slaves arguably should have the
2858 * same MTUs. To do this, we'd need to
2859 * take over the slave's change_mtu
2860 * function for the duration of their
2864 case NETDEV_CHANGENAME:
2866 * TODO: handle changing the primary's name
2869 case NETDEV_FEAT_CHANGE:
2870 bond_compute_features(bond);
2872 case NETDEV_RESEND_IGMP:
2873 /* Propagate to master device */
2874 call_netdevice_notifiers(event, slave->bond->dev);
2884 * bond_netdev_event: handle netdev notifier chain events.
2886 * This function receives events for the netdev chain. The caller (an
2887 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2888 * locks for us to safely manipulate the slave devices (RTNL lock,
2891 static int bond_netdev_event(struct notifier_block *this,
2892 unsigned long event, void *ptr)
2894 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2896 pr_debug("event_dev: %s, event: %lx\n",
2897 event_dev ? event_dev->name : "None",
2900 if (!(event_dev->priv_flags & IFF_BONDING))
2903 if (event_dev->flags & IFF_MASTER) {
2904 pr_debug("IFF_MASTER\n");
2905 return bond_master_netdev_event(event, event_dev);
2908 if (event_dev->flags & IFF_SLAVE) {
2909 pr_debug("IFF_SLAVE\n");
2910 return bond_slave_netdev_event(event, event_dev);
2916 static struct notifier_block bond_netdev_notifier = {
2917 .notifier_call = bond_netdev_event,
2920 /*---------------------------- Hashing Policies -----------------------------*/
2922 /* L2 hash helper */
2923 static inline u32 bond_eth_hash(struct sk_buff *skb)
2925 struct ethhdr *data = (struct ethhdr *)skb->data;
2927 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2928 return data->h_dest[5] ^ data->h_source[5];
2933 /* Extract the appropriate headers based on bond's xmit policy */
2934 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2935 struct flow_keys *fk)
2937 const struct ipv6hdr *iph6;
2938 const struct iphdr *iph;
2939 int noff, proto = -1;
2941 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2942 return skb_flow_dissect(skb, fk);
2945 noff = skb_network_offset(skb);
2946 if (skb->protocol == htons(ETH_P_IP)) {
2947 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2950 fk->src = iph->saddr;
2951 fk->dst = iph->daddr;
2952 noff += iph->ihl << 2;
2953 if (!ip_is_fragment(iph))
2954 proto = iph->protocol;
2955 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2956 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2958 iph6 = ipv6_hdr(skb);
2959 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2960 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2961 noff += sizeof(*iph6);
2962 proto = iph6->nexthdr;
2966 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2967 fk->ports = skb_flow_get_ports(skb, noff, proto);
2973 * bond_xmit_hash - generate a hash value based on the xmit policy
2974 * @bond: bonding device
2975 * @skb: buffer to use for headers
2976 * @count: modulo value
2978 * This function will extract the necessary headers from the skb buffer and use
2979 * them to generate a hash based on the xmit_policy set in the bonding device
2980 * which will be reduced modulo count before returning.
2982 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
2984 struct flow_keys flow;
2987 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2988 !bond_flow_dissect(bond, skb, &flow))
2989 return bond_eth_hash(skb) % count;
2991 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
2992 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
2993 hash = bond_eth_hash(skb);
2995 hash = (__force u32)flow.ports;
2996 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
2997 hash ^= (hash >> 16);
2998 hash ^= (hash >> 8);
3000 return hash % count;
3003 /*-------------------------- Device entry points ----------------------------*/
3005 static void bond_work_init_all(struct bonding *bond)
3007 INIT_DELAYED_WORK(&bond->mcast_work,
3008 bond_resend_igmp_join_requests_delayed);
3009 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3010 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3011 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3012 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3014 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3015 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3018 static void bond_work_cancel_all(struct bonding *bond)
3020 cancel_delayed_work_sync(&bond->mii_work);
3021 cancel_delayed_work_sync(&bond->arp_work);
3022 cancel_delayed_work_sync(&bond->alb_work);
3023 cancel_delayed_work_sync(&bond->ad_work);
3024 cancel_delayed_work_sync(&bond->mcast_work);
3027 static int bond_open(struct net_device *bond_dev)
3029 struct bonding *bond = netdev_priv(bond_dev);
3030 struct list_head *iter;
3031 struct slave *slave;
3033 /* reset slave->backup and slave->inactive */
3034 read_lock(&bond->lock);
3035 if (bond_has_slaves(bond)) {
3036 read_lock(&bond->curr_slave_lock);
3037 bond_for_each_slave(bond, slave, iter) {
3038 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3039 && (slave != bond->curr_active_slave)) {
3040 bond_set_slave_inactive_flags(slave);
3042 bond_set_slave_active_flags(slave);
3045 read_unlock(&bond->curr_slave_lock);
3047 read_unlock(&bond->lock);
3049 bond_work_init_all(bond);
3051 if (bond_is_lb(bond)) {
3052 /* bond_alb_initialize must be called before the timer
3055 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3057 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3060 if (bond->params.miimon) /* link check interval, in milliseconds. */
3061 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3063 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3064 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3065 if (bond->params.arp_validate)
3066 bond->recv_probe = bond_arp_rcv;
3069 if (bond->params.mode == BOND_MODE_8023AD) {
3070 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3071 /* register to receive LACPDUs */
3072 bond->recv_probe = bond_3ad_lacpdu_recv;
3073 bond_3ad_initiate_agg_selection(bond, 1);
3079 static int bond_close(struct net_device *bond_dev)
3081 struct bonding *bond = netdev_priv(bond_dev);
3083 bond_work_cancel_all(bond);
3084 bond->send_peer_notif = 0;
3085 if (bond_is_lb(bond))
3086 bond_alb_deinitialize(bond);
3087 bond->recv_probe = NULL;
3092 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3093 struct rtnl_link_stats64 *stats)
3095 struct bonding *bond = netdev_priv(bond_dev);
3096 struct rtnl_link_stats64 temp;
3097 struct list_head *iter;
3098 struct slave *slave;
3100 memset(stats, 0, sizeof(*stats));
3102 read_lock_bh(&bond->lock);
3103 bond_for_each_slave(bond, slave, iter) {
3104 const struct rtnl_link_stats64 *sstats =
3105 dev_get_stats(slave->dev, &temp);
3107 stats->rx_packets += sstats->rx_packets;
3108 stats->rx_bytes += sstats->rx_bytes;
3109 stats->rx_errors += sstats->rx_errors;
3110 stats->rx_dropped += sstats->rx_dropped;
3112 stats->tx_packets += sstats->tx_packets;
3113 stats->tx_bytes += sstats->tx_bytes;
3114 stats->tx_errors += sstats->tx_errors;
3115 stats->tx_dropped += sstats->tx_dropped;
3117 stats->multicast += sstats->multicast;
3118 stats->collisions += sstats->collisions;
3120 stats->rx_length_errors += sstats->rx_length_errors;
3121 stats->rx_over_errors += sstats->rx_over_errors;
3122 stats->rx_crc_errors += sstats->rx_crc_errors;
3123 stats->rx_frame_errors += sstats->rx_frame_errors;
3124 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3125 stats->rx_missed_errors += sstats->rx_missed_errors;
3127 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3128 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3129 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3130 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3131 stats->tx_window_errors += sstats->tx_window_errors;
3133 read_unlock_bh(&bond->lock);
3138 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3140 struct bonding *bond = netdev_priv(bond_dev);
3141 struct net_device *slave_dev = NULL;
3142 struct ifbond k_binfo;
3143 struct ifbond __user *u_binfo = NULL;
3144 struct ifslave k_sinfo;
3145 struct ifslave __user *u_sinfo = NULL;
3146 struct mii_ioctl_data *mii = NULL;
3150 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3162 * We do this again just in case we were called by SIOCGMIIREG
3163 * instead of SIOCGMIIPHY.
3170 if (mii->reg_num == 1) {
3172 read_lock(&bond->lock);
3173 read_lock(&bond->curr_slave_lock);
3174 if (netif_carrier_ok(bond->dev))
3175 mii->val_out = BMSR_LSTATUS;
3177 read_unlock(&bond->curr_slave_lock);
3178 read_unlock(&bond->lock);
3182 case BOND_INFO_QUERY_OLD:
3183 case SIOCBONDINFOQUERY:
3184 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3186 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3189 res = bond_info_query(bond_dev, &k_binfo);
3191 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3195 case BOND_SLAVE_INFO_QUERY_OLD:
3196 case SIOCBONDSLAVEINFOQUERY:
3197 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3199 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3202 res = bond_slave_info_query(bond_dev, &k_sinfo);
3204 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3213 net = dev_net(bond_dev);
3215 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3218 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3220 pr_debug("slave_dev=%p:\n", slave_dev);
3225 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3227 case BOND_ENSLAVE_OLD:
3228 case SIOCBONDENSLAVE:
3229 res = bond_enslave(bond_dev, slave_dev);
3231 case BOND_RELEASE_OLD:
3232 case SIOCBONDRELEASE:
3233 res = bond_release(bond_dev, slave_dev);
3235 case BOND_SETHWADDR_OLD:
3236 case SIOCBONDSETHWADDR:
3237 bond_set_dev_addr(bond_dev, slave_dev);
3240 case BOND_CHANGE_ACTIVE_OLD:
3241 case SIOCBONDCHANGEACTIVE:
3242 res = bond_option_active_slave_set(bond, slave_dev);
3254 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3256 struct bonding *bond = netdev_priv(bond_dev);
3258 if (change & IFF_PROMISC)
3259 bond_set_promiscuity(bond,
3260 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3262 if (change & IFF_ALLMULTI)
3263 bond_set_allmulti(bond,
3264 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3267 static void bond_set_rx_mode(struct net_device *bond_dev)
3269 struct bonding *bond = netdev_priv(bond_dev);
3270 struct list_head *iter;
3271 struct slave *slave;
3275 if (USES_PRIMARY(bond->params.mode)) {
3276 slave = rcu_dereference(bond->curr_active_slave);
3278 dev_uc_sync(slave->dev, bond_dev);
3279 dev_mc_sync(slave->dev, bond_dev);
3282 bond_for_each_slave_rcu(bond, slave, iter) {
3283 dev_uc_sync_multiple(slave->dev, bond_dev);
3284 dev_mc_sync_multiple(slave->dev, bond_dev);
3290 static int bond_neigh_init(struct neighbour *n)
3292 struct bonding *bond = netdev_priv(n->dev);
3293 const struct net_device_ops *slave_ops;
3294 struct neigh_parms parms;
3295 struct slave *slave;
3298 slave = bond_first_slave(bond);
3301 slave_ops = slave->dev->netdev_ops;
3302 if (!slave_ops->ndo_neigh_setup)
3305 parms.neigh_setup = NULL;
3306 parms.neigh_cleanup = NULL;
3307 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3312 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3313 * after the last slave has been detached. Assumes that all slaves
3314 * utilize the same neigh_cleanup (true at this writing as only user
3317 n->parms->neigh_cleanup = parms.neigh_cleanup;
3319 if (!parms.neigh_setup)
3322 return parms.neigh_setup(n);
3326 * The bonding ndo_neigh_setup is called at init time beofre any
3327 * slave exists. So we must declare proxy setup function which will
3328 * be used at run time to resolve the actual slave neigh param setup.
3330 * It's also called by master devices (such as vlans) to setup their
3331 * underlying devices. In that case - do nothing, we're already set up from
3334 static int bond_neigh_setup(struct net_device *dev,
3335 struct neigh_parms *parms)
3337 /* modify only our neigh_parms */
3338 if (parms->dev == dev)
3339 parms->neigh_setup = bond_neigh_init;
3345 * Change the MTU of all of a master's slaves to match the master
3347 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3349 struct bonding *bond = netdev_priv(bond_dev);
3350 struct slave *slave, *rollback_slave;
3351 struct list_head *iter;
3354 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3355 (bond_dev ? bond_dev->name : "None"), new_mtu);
3357 /* Can't hold bond->lock with bh disabled here since
3358 * some base drivers panic. On the other hand we can't
3359 * hold bond->lock without bh disabled because we'll
3360 * deadlock. The only solution is to rely on the fact
3361 * that we're under rtnl_lock here, and the slaves
3362 * list won't change. This doesn't solve the problem
3363 * of setting the slave's MTU while it is
3364 * transmitting, but the assumption is that the base
3365 * driver can handle that.
3367 * TODO: figure out a way to safely iterate the slaves
3368 * list, but without holding a lock around the actual
3369 * call to the base driver.
3372 bond_for_each_slave(bond, slave, iter) {
3373 pr_debug("s %p c_m %p\n",
3375 slave->dev->netdev_ops->ndo_change_mtu);
3377 res = dev_set_mtu(slave->dev, new_mtu);
3380 /* If we failed to set the slave's mtu to the new value
3381 * we must abort the operation even in ACTIVE_BACKUP
3382 * mode, because if we allow the backup slaves to have
3383 * different mtu values than the active slave we'll
3384 * need to change their mtu when doing a failover. That
3385 * means changing their mtu from timer context, which
3386 * is probably not a good idea.
3388 pr_debug("err %d %s\n", res, slave->dev->name);
3393 bond_dev->mtu = new_mtu;
3398 /* unwind from head to the slave that failed */
3399 bond_for_each_slave(bond, rollback_slave, iter) {
3402 if (rollback_slave == slave)
3405 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3407 pr_debug("unwind err %d dev %s\n",
3408 tmp_res, rollback_slave->dev->name);
3418 * Note that many devices must be down to change the HW address, and
3419 * downing the master releases all slaves. We can make bonds full of
3420 * bonding devices to test this, however.
3422 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3424 struct bonding *bond = netdev_priv(bond_dev);
3425 struct slave *slave, *rollback_slave;
3426 struct sockaddr *sa = addr, tmp_sa;
3427 struct list_head *iter;
3430 if (bond->params.mode == BOND_MODE_ALB)
3431 return bond_alb_set_mac_address(bond_dev, addr);
3434 pr_debug("bond=%p, name=%s\n",
3435 bond, bond_dev ? bond_dev->name : "None");
3437 /* If fail_over_mac is enabled, do nothing and return success.
3438 * Returning an error causes ifenslave to fail.
3440 if (bond->params.fail_over_mac)
3443 if (!is_valid_ether_addr(sa->sa_data))
3444 return -EADDRNOTAVAIL;
3446 /* Can't hold bond->lock with bh disabled here since
3447 * some base drivers panic. On the other hand we can't
3448 * hold bond->lock without bh disabled because we'll
3449 * deadlock. The only solution is to rely on the fact
3450 * that we're under rtnl_lock here, and the slaves
3451 * list won't change. This doesn't solve the problem
3452 * of setting the slave's hw address while it is
3453 * transmitting, but the assumption is that the base
3454 * driver can handle that.
3456 * TODO: figure out a way to safely iterate the slaves
3457 * list, but without holding a lock around the actual
3458 * call to the base driver.
3461 bond_for_each_slave(bond, slave, iter) {
3462 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3463 pr_debug("slave %p %s\n", slave, slave->dev->name);
3465 if (slave_ops->ndo_set_mac_address == NULL) {
3467 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3471 res = dev_set_mac_address(slave->dev, addr);
3473 /* TODO: consider downing the slave
3475 * User should expect communications
3476 * breakage anyway until ARP finish
3479 pr_debug("err %d %s\n", res, slave->dev->name);
3485 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3489 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3490 tmp_sa.sa_family = bond_dev->type;
3492 /* unwind from head to the slave that failed */
3493 bond_for_each_slave(bond, rollback_slave, iter) {
3496 if (rollback_slave == slave)
3499 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3501 pr_debug("unwind err %d dev %s\n",
3502 tmp_res, rollback_slave->dev->name);
3510 * bond_xmit_slave_id - transmit skb through slave with slave_id
3511 * @bond: bonding device that is transmitting
3512 * @skb: buffer to transmit
3513 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3515 * This function tries to transmit through slave with slave_id but in case
3516 * it fails, it tries to find the first available slave for transmission.
3517 * The skb is consumed in all cases, thus the function is void.
3519 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3521 struct list_head *iter;
3522 struct slave *slave;
3525 /* Here we start from the slave with slave_id */
3526 bond_for_each_slave_rcu(bond, slave, iter) {
3528 if (slave_can_tx(slave)) {
3529 bond_dev_queue_xmit(bond, skb, slave->dev);
3535 /* Here we start from the first slave up to slave_id */
3537 bond_for_each_slave_rcu(bond, slave, iter) {
3540 if (slave_can_tx(slave)) {
3541 bond_dev_queue_xmit(bond, skb, slave->dev);
3545 /* no slave that can tx has been found */
3550 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3551 * @bond: bonding device to use
3553 * Based on the value of the bonding device's packets_per_slave parameter
3554 * this function generates a slave id, which is usually used as the next
3555 * slave to transmit through.
3557 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3559 int packets_per_slave = bond->params.packets_per_slave;
3562 switch (packets_per_slave) {
3564 slave_id = prandom_u32();
3567 slave_id = bond->rr_tx_counter;
3570 slave_id = reciprocal_divide(bond->rr_tx_counter,
3574 bond->rr_tx_counter++;
3579 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3581 struct bonding *bond = netdev_priv(bond_dev);
3582 struct iphdr *iph = ip_hdr(skb);
3583 struct slave *slave;
3586 /* Start with the curr_active_slave that joined the bond as the
3587 * default for sending IGMP traffic. For failover purposes one
3588 * needs to maintain some consistency for the interface that will
3589 * send the join/membership reports. The curr_active_slave found
3590 * will send all of this type of traffic.
3592 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3593 slave = rcu_dereference(bond->curr_active_slave);
3594 if (slave && slave_can_tx(slave))
3595 bond_dev_queue_xmit(bond, skb, slave->dev);
3597 bond_xmit_slave_id(bond, skb, 0);
3599 slave_id = bond_rr_gen_slave_id(bond);
3600 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3603 return NETDEV_TX_OK;
3607 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3608 * the bond has a usable interface.
3610 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3612 struct bonding *bond = netdev_priv(bond_dev);
3613 struct slave *slave;
3615 slave = rcu_dereference(bond->curr_active_slave);
3617 bond_dev_queue_xmit(bond, skb, slave->dev);
3621 return NETDEV_TX_OK;
3624 /* In bond_xmit_xor() , we determine the output device by using a pre-
3625 * determined xmit_hash_policy(), If the selected device is not enabled,
3626 * find the next active slave.
3628 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3630 struct bonding *bond = netdev_priv(bond_dev);
3632 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3634 return NETDEV_TX_OK;
3637 /* in broadcast mode, we send everything to all usable interfaces. */
3638 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3640 struct bonding *bond = netdev_priv(bond_dev);
3641 struct slave *slave = NULL;
3642 struct list_head *iter;
3644 bond_for_each_slave_rcu(bond, slave, iter) {
3645 if (bond_is_last_slave(bond, slave))
3647 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3648 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3651 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3655 /* bond_dev_queue_xmit always returns 0 */
3656 bond_dev_queue_xmit(bond, skb2, slave->dev);
3659 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3660 bond_dev_queue_xmit(bond, skb, slave->dev);
3664 return NETDEV_TX_OK;
3667 /*------------------------- Device initialization ---------------------------*/
3670 * Lookup the slave that corresponds to a qid
3672 static inline int bond_slave_override(struct bonding *bond,
3673 struct sk_buff *skb)
3675 struct slave *slave = NULL;
3676 struct slave *check_slave;
3677 struct list_head *iter;
3680 if (!skb->queue_mapping)
3683 /* Find out if any slaves have the same mapping as this skb. */
3684 bond_for_each_slave_rcu(bond, check_slave, iter) {
3685 if (check_slave->queue_id == skb->queue_mapping) {
3686 slave = check_slave;
3691 /* If the slave isn't UP, use default transmit policy. */
3692 if (slave && slave->queue_id && IS_UP(slave->dev) &&
3693 (slave->link == BOND_LINK_UP)) {
3694 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3701 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3704 * This helper function exists to help dev_pick_tx get the correct
3705 * destination queue. Using a helper function skips a call to
3706 * skb_tx_hash and will put the skbs in the queue we expect on their
3707 * way down to the bonding driver.
3709 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3712 * Save the original txq to restore before passing to the driver
3714 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3716 if (unlikely(txq >= dev->real_num_tx_queues)) {
3718 txq -= dev->real_num_tx_queues;
3719 } while (txq >= dev->real_num_tx_queues);
3724 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3726 struct bonding *bond = netdev_priv(dev);
3728 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3729 if (!bond_slave_override(bond, skb))
3730 return NETDEV_TX_OK;
3733 switch (bond->params.mode) {
3734 case BOND_MODE_ROUNDROBIN:
3735 return bond_xmit_roundrobin(skb, dev);
3736 case BOND_MODE_ACTIVEBACKUP:
3737 return bond_xmit_activebackup(skb, dev);
3739 return bond_xmit_xor(skb, dev);
3740 case BOND_MODE_BROADCAST:
3741 return bond_xmit_broadcast(skb, dev);
3742 case BOND_MODE_8023AD:
3743 return bond_3ad_xmit_xor(skb, dev);
3746 return bond_alb_xmit(skb, dev);
3748 /* Should never happen, mode already checked */
3749 pr_err("%s: Error: Unknown bonding mode %d\n",
3750 dev->name, bond->params.mode);
3753 return NETDEV_TX_OK;
3757 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3759 struct bonding *bond = netdev_priv(dev);
3760 netdev_tx_t ret = NETDEV_TX_OK;
3763 * If we risk deadlock from transmitting this in the
3764 * netpoll path, tell netpoll to queue the frame for later tx
3766 if (is_netpoll_tx_blocked(dev))
3767 return NETDEV_TX_BUSY;
3770 if (bond_has_slaves(bond))
3771 ret = __bond_start_xmit(skb, dev);
3779 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3780 struct ethtool_cmd *ecmd)
3782 struct bonding *bond = netdev_priv(bond_dev);
3783 unsigned long speed = 0;
3784 struct list_head *iter;
3785 struct slave *slave;
3787 ecmd->duplex = DUPLEX_UNKNOWN;
3788 ecmd->port = PORT_OTHER;
3790 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3791 * do not need to check mode. Though link speed might not represent
3792 * the true receive or transmit bandwidth (not all modes are symmetric)
3793 * this is an accurate maximum.
3795 read_lock(&bond->lock);
3796 bond_for_each_slave(bond, slave, iter) {
3797 if (SLAVE_IS_OK(slave)) {
3798 if (slave->speed != SPEED_UNKNOWN)
3799 speed += slave->speed;
3800 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3801 slave->duplex != DUPLEX_UNKNOWN)
3802 ecmd->duplex = slave->duplex;
3805 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3806 read_unlock(&bond->lock);
3811 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3812 struct ethtool_drvinfo *drvinfo)
3814 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3815 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3816 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3820 static const struct ethtool_ops bond_ethtool_ops = {
3821 .get_drvinfo = bond_ethtool_get_drvinfo,
3822 .get_settings = bond_ethtool_get_settings,
3823 .get_link = ethtool_op_get_link,
3826 static const struct net_device_ops bond_netdev_ops = {
3827 .ndo_init = bond_init,
3828 .ndo_uninit = bond_uninit,
3829 .ndo_open = bond_open,
3830 .ndo_stop = bond_close,
3831 .ndo_start_xmit = bond_start_xmit,
3832 .ndo_select_queue = bond_select_queue,
3833 .ndo_get_stats64 = bond_get_stats,
3834 .ndo_do_ioctl = bond_do_ioctl,
3835 .ndo_change_rx_flags = bond_change_rx_flags,
3836 .ndo_set_rx_mode = bond_set_rx_mode,
3837 .ndo_change_mtu = bond_change_mtu,
3838 .ndo_set_mac_address = bond_set_mac_address,
3839 .ndo_neigh_setup = bond_neigh_setup,
3840 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3841 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3842 #ifdef CONFIG_NET_POLL_CONTROLLER
3843 .ndo_netpoll_setup = bond_netpoll_setup,
3844 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3845 .ndo_poll_controller = bond_poll_controller,
3847 .ndo_add_slave = bond_enslave,
3848 .ndo_del_slave = bond_release,
3849 .ndo_fix_features = bond_fix_features,
3852 static const struct device_type bond_type = {
3856 static void bond_destructor(struct net_device *bond_dev)
3858 struct bonding *bond = netdev_priv(bond_dev);
3860 destroy_workqueue(bond->wq);
3861 free_netdev(bond_dev);
3864 void bond_setup(struct net_device *bond_dev)
3866 struct bonding *bond = netdev_priv(bond_dev);
3868 /* initialize rwlocks */
3869 rwlock_init(&bond->lock);
3870 rwlock_init(&bond->curr_slave_lock);
3871 bond->params = bonding_defaults;
3873 /* Initialize pointers */
3874 bond->dev = bond_dev;
3876 /* Initialize the device entry points */
3877 ether_setup(bond_dev);
3878 bond_dev->netdev_ops = &bond_netdev_ops;
3879 bond_dev->ethtool_ops = &bond_ethtool_ops;
3881 bond_dev->destructor = bond_destructor;
3883 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3885 /* Initialize the device options */
3886 bond_dev->tx_queue_len = 0;
3887 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3888 bond_dev->priv_flags |= IFF_BONDING;
3889 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3891 /* At first, we block adding VLANs. That's the only way to
3892 * prevent problems that occur when adding VLANs over an
3893 * empty bond. The block will be removed once non-challenged
3894 * slaves are enslaved.
3896 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3898 /* don't acquire bond device's netif_tx_lock when
3900 bond_dev->features |= NETIF_F_LLTX;
3902 /* By default, we declare the bond to be fully
3903 * VLAN hardware accelerated capable. Special
3904 * care is taken in the various xmit functions
3905 * when there are slaves that are not hw accel
3909 bond_dev->hw_features = BOND_VLAN_FEATURES |
3910 NETIF_F_HW_VLAN_CTAG_TX |
3911 NETIF_F_HW_VLAN_CTAG_RX |
3912 NETIF_F_HW_VLAN_CTAG_FILTER;
3914 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3915 bond_dev->features |= bond_dev->hw_features;
3919 * Destroy a bonding device.
3920 * Must be under rtnl_lock when this function is called.
3922 static void bond_uninit(struct net_device *bond_dev)
3924 struct bonding *bond = netdev_priv(bond_dev);
3925 struct list_head *iter;
3926 struct slave *slave;
3928 bond_netpoll_cleanup(bond_dev);
3930 /* Release the bonded slaves */
3931 bond_for_each_slave(bond, slave, iter)
3932 __bond_release_one(bond_dev, slave->dev, true);
3933 pr_info("%s: released all slaves\n", bond_dev->name);
3935 list_del(&bond->bond_list);
3937 bond_debug_unregister(bond);
3940 /*------------------------- Module initialization ---------------------------*/
3943 * Convert string input module parms. Accept either the
3944 * number of the mode or its string name. A bit complicated because
3945 * some mode names are substrings of other names, and calls from sysfs
3946 * may have whitespace in the name (trailing newlines, for example).
3948 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3950 int modeint = -1, i, rv;
3951 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
3953 for (p = (char *)buf; *p; p++)
3954 if (!(isdigit(*p) || isspace(*p)))
3958 rv = sscanf(buf, "%20s", modestr);
3960 rv = sscanf(buf, "%d", &modeint);
3965 for (i = 0; tbl[i].modename; i++) {
3966 if (modeint == tbl[i].mode)
3968 if (strcmp(modestr, tbl[i].modename) == 0)
3975 static int bond_check_params(struct bond_params *params)
3977 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3978 int arp_all_targets_value;
3981 * Convert string parameters.
3984 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
3985 if (bond_mode == -1) {
3986 pr_err("Error: Invalid bonding mode \"%s\"\n",
3987 mode == NULL ? "NULL" : mode);
3992 if (xmit_hash_policy) {
3993 if ((bond_mode != BOND_MODE_XOR) &&
3994 (bond_mode != BOND_MODE_8023AD)) {
3995 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3996 bond_mode_name(bond_mode));
3998 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4000 if (xmit_hashtype == -1) {
4001 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4002 xmit_hash_policy == NULL ? "NULL" :
4010 if (bond_mode != BOND_MODE_8023AD) {
4011 pr_info("lacp_rate param is irrelevant in mode %s\n",
4012 bond_mode_name(bond_mode));
4014 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4015 if (lacp_fast == -1) {
4016 pr_err("Error: Invalid lacp rate \"%s\"\n",
4017 lacp_rate == NULL ? "NULL" : lacp_rate);
4024 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4025 if (params->ad_select == -1) {
4026 pr_err("Error: Invalid ad_select \"%s\"\n",
4027 ad_select == NULL ? "NULL" : ad_select);
4031 if (bond_mode != BOND_MODE_8023AD) {
4032 pr_warning("ad_select param only affects 802.3ad mode\n");
4035 params->ad_select = BOND_AD_STABLE;
4038 if (max_bonds < 0) {
4039 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4040 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4041 max_bonds = BOND_DEFAULT_MAX_BONDS;
4045 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4046 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4047 miimon = BOND_LINK_MON_INTERV;
4051 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4056 if (downdelay < 0) {
4057 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4058 downdelay, INT_MAX);
4062 if ((use_carrier != 0) && (use_carrier != 1)) {
4063 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4068 if (num_peer_notif < 0 || num_peer_notif > 255) {
4069 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4074 /* reset values for 802.3ad */
4075 if (bond_mode == BOND_MODE_8023AD) {
4077 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");
4078 pr_warning("Forcing miimon to 100msec\n");
4079 miimon = BOND_DEFAULT_MIIMON;
4083 if (tx_queues < 1 || tx_queues > 255) {
4084 pr_warning("Warning: tx_queues (%d) should be between "
4085 "1 and 255, resetting to %d\n",
4086 tx_queues, BOND_DEFAULT_TX_QUEUES);
4087 tx_queues = BOND_DEFAULT_TX_QUEUES;
4090 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4091 pr_warning("Warning: all_slaves_active module parameter (%d), "
4092 "not of valid value (0/1), so it was set to "
4093 "0\n", all_slaves_active);
4094 all_slaves_active = 0;
4097 if (resend_igmp < 0 || resend_igmp > 255) {
4098 pr_warning("Warning: resend_igmp (%d) should be between "
4099 "0 and 255, resetting to %d\n",
4100 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4101 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4104 if (packets_per_slave < 0 || packets_per_slave > USHRT_MAX) {
4105 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4106 packets_per_slave, USHRT_MAX);
4107 packets_per_slave = 1;
4110 /* reset values for TLB/ALB */
4111 if ((bond_mode == BOND_MODE_TLB) ||
4112 (bond_mode == BOND_MODE_ALB)) {
4114 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");
4115 pr_warning("Forcing miimon to 100msec\n");
4116 miimon = BOND_DEFAULT_MIIMON;
4120 if (bond_mode == BOND_MODE_ALB) {
4121 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",
4126 if (updelay || downdelay) {
4127 /* just warn the user the up/down delay will have
4128 * no effect since miimon is zero...
4130 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",
4131 updelay, downdelay);
4134 /* don't allow arp monitoring */
4136 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4137 miimon, arp_interval);
4141 if ((updelay % miimon) != 0) {
4142 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4144 (updelay / miimon) * miimon);
4149 if ((downdelay % miimon) != 0) {
4150 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4152 (downdelay / miimon) * miimon);
4155 downdelay /= miimon;
4158 if (arp_interval < 0) {
4159 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4160 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4161 arp_interval = BOND_LINK_ARP_INTERV;
4164 for (arp_ip_count = 0, i = 0;
4165 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4166 /* not complete check, but should be good enough to
4169 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4170 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4171 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4175 if (bond_get_targets_ip(arp_target, ip) == -1)
4176 arp_target[arp_ip_count++] = ip;
4178 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4183 if (arp_interval && !arp_ip_count) {
4184 /* don't allow arping if no arp_ip_target given... */
4185 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4191 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4192 pr_err("arp_validate only supported in active-backup mode\n");
4195 if (!arp_interval) {
4196 pr_err("arp_validate requires arp_interval\n");
4200 arp_validate_value = bond_parse_parm(arp_validate,
4202 if (arp_validate_value == -1) {
4203 pr_err("Error: invalid arp_validate \"%s\"\n",
4204 arp_validate == NULL ? "NULL" : arp_validate);
4208 arp_validate_value = 0;
4210 arp_all_targets_value = 0;
4211 if (arp_all_targets) {
4212 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4213 arp_all_targets_tbl);
4215 if (arp_all_targets_value == -1) {
4216 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4218 arp_all_targets_value = 0;
4223 pr_info("MII link monitoring set to %d ms\n", miimon);
4224 } else if (arp_interval) {
4225 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4227 arp_validate_tbl[arp_validate_value].modename,
4230 for (i = 0; i < arp_ip_count; i++)
4231 pr_info(" %s", arp_ip_target[i]);
4235 } else if (max_bonds) {
4236 /* miimon and arp_interval not set, we need one so things
4237 * work as expected, see bonding.txt for details
4239 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");
4242 if (primary && !USES_PRIMARY(bond_mode)) {
4243 /* currently, using a primary only makes sense
4244 * in active backup, TLB or ALB modes
4246 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4247 primary, bond_mode_name(bond_mode));
4251 if (primary && primary_reselect) {
4252 primary_reselect_value = bond_parse_parm(primary_reselect,
4254 if (primary_reselect_value == -1) {
4255 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4257 NULL ? "NULL" : primary_reselect);
4261 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4264 if (fail_over_mac) {
4265 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4267 if (fail_over_mac_value == -1) {
4268 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4269 arp_validate == NULL ? "NULL" : arp_validate);
4273 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4274 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4276 fail_over_mac_value = BOND_FOM_NONE;
4279 if (lp_interval == 0) {
4280 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4281 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4282 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4285 /* fill params struct with the proper values */
4286 params->mode = bond_mode;
4287 params->xmit_policy = xmit_hashtype;
4288 params->miimon = miimon;
4289 params->num_peer_notif = num_peer_notif;
4290 params->arp_interval = arp_interval;
4291 params->arp_validate = arp_validate_value;
4292 params->arp_all_targets = arp_all_targets_value;
4293 params->updelay = updelay;
4294 params->downdelay = downdelay;
4295 params->use_carrier = use_carrier;
4296 params->lacp_fast = lacp_fast;
4297 params->primary[0] = 0;
4298 params->primary_reselect = primary_reselect_value;
4299 params->fail_over_mac = fail_over_mac_value;
4300 params->tx_queues = tx_queues;
4301 params->all_slaves_active = all_slaves_active;
4302 params->resend_igmp = resend_igmp;
4303 params->min_links = min_links;
4304 params->lp_interval = lp_interval;
4305 if (packets_per_slave > 1)
4306 params->packets_per_slave = reciprocal_value(packets_per_slave);
4308 params->packets_per_slave = packets_per_slave;
4310 strncpy(params->primary, primary, IFNAMSIZ);
4311 params->primary[IFNAMSIZ - 1] = 0;
4314 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4319 static struct lock_class_key bonding_netdev_xmit_lock_key;
4320 static struct lock_class_key bonding_netdev_addr_lock_key;
4321 static struct lock_class_key bonding_tx_busylock_key;
4323 static void bond_set_lockdep_class_one(struct net_device *dev,
4324 struct netdev_queue *txq,
4327 lockdep_set_class(&txq->_xmit_lock,
4328 &bonding_netdev_xmit_lock_key);
4331 static void bond_set_lockdep_class(struct net_device *dev)
4333 lockdep_set_class(&dev->addr_list_lock,
4334 &bonding_netdev_addr_lock_key);
4335 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4336 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4340 * Called from registration process
4342 static int bond_init(struct net_device *bond_dev)
4344 struct bonding *bond = netdev_priv(bond_dev);
4345 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4346 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4348 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4351 * Initialize locks that may be required during
4352 * en/deslave operations. All of the bond_open work
4353 * (of which this is part) should really be moved to
4354 * a phase prior to dev_open
4356 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4357 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4359 bond->wq = create_singlethread_workqueue(bond_dev->name);
4363 bond_set_lockdep_class(bond_dev);
4365 list_add_tail(&bond->bond_list, &bn->dev_list);
4367 bond_prepare_sysfs_group(bond);
4369 bond_debug_register(bond);
4371 /* Ensure valid dev_addr */
4372 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4373 bond_dev->addr_assign_type == NET_ADDR_PERM)
4374 eth_hw_addr_random(bond_dev);
4379 unsigned int bond_get_num_tx_queues(void)
4384 /* Create a new bond based on the specified name and bonding parameters.
4385 * If name is NULL, obtain a suitable "bond%d" name for us.
4386 * Caller must NOT hold rtnl_lock; we need to release it here before we
4387 * set up our sysfs entries.
4389 int bond_create(struct net *net, const char *name)
4391 struct net_device *bond_dev;
4396 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4397 name ? name : "bond%d",
4398 bond_setup, tx_queues);
4400 pr_err("%s: eek! can't alloc netdev!\n", name);
4405 dev_net_set(bond_dev, net);
4406 bond_dev->rtnl_link_ops = &bond_link_ops;
4408 res = register_netdevice(bond_dev);
4410 netif_carrier_off(bond_dev);
4414 bond_destructor(bond_dev);
4418 static int __net_init bond_net_init(struct net *net)
4420 struct bond_net *bn = net_generic(net, bond_net_id);
4423 INIT_LIST_HEAD(&bn->dev_list);
4425 bond_create_proc_dir(bn);
4426 bond_create_sysfs(bn);
4431 static void __net_exit bond_net_exit(struct net *net)
4433 struct bond_net *bn = net_generic(net, bond_net_id);
4434 struct bonding *bond, *tmp_bond;
4437 bond_destroy_sysfs(bn);
4438 bond_destroy_proc_dir(bn);
4440 /* Kill off any bonds created after unregistering bond rtnl ops */
4442 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4443 unregister_netdevice_queue(bond->dev, &list);
4444 unregister_netdevice_many(&list);
4448 static struct pernet_operations bond_net_ops = {
4449 .init = bond_net_init,
4450 .exit = bond_net_exit,
4452 .size = sizeof(struct bond_net),
4455 static int __init bonding_init(void)
4460 pr_info("%s", bond_version);
4462 res = bond_check_params(&bonding_defaults);
4466 res = register_pernet_subsys(&bond_net_ops);
4470 res = bond_netlink_init();
4474 bond_create_debugfs();
4476 for (i = 0; i < max_bonds; i++) {
4477 res = bond_create(&init_net, NULL);
4482 register_netdevice_notifier(&bond_netdev_notifier);
4486 bond_netlink_fini();
4488 unregister_pernet_subsys(&bond_net_ops);
4493 static void __exit bonding_exit(void)
4495 unregister_netdevice_notifier(&bond_netdev_notifier);
4497 bond_destroy_debugfs();
4499 bond_netlink_fini();
4500 unregister_pernet_subsys(&bond_net_ops);
4502 #ifdef CONFIG_NET_POLL_CONTROLLER
4504 * Make sure we don't have an imbalance on our netpoll blocking
4506 WARN_ON(atomic_read(&netpoll_block_tx));
4510 module_init(bonding_init);
4511 module_exit(bonding_exit);
4512 MODULE_LICENSE("GPL");
4513 MODULE_VERSION(DRV_VERSION);
4514 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4515 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");