2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.h>
82 #include <linux/reciprocal_div.h>
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 #define BOND_LINK_MON_INTERV 0
91 #define BOND_LINK_ARP_INTERV 0
93 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
94 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
95 static int num_peer_notif = 1;
96 static int miimon = BOND_LINK_MON_INTERV;
99 static int use_carrier = 1;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static int min_links;
105 static char *ad_select;
106 static char *xmit_hash_policy;
107 static int arp_interval = BOND_LINK_ARP_INTERV;
108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
109 static char *arp_validate;
110 static char *arp_all_targets;
111 static char *fail_over_mac;
112 static int all_slaves_active;
113 static struct bond_params bonding_defaults;
114 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
115 static int packets_per_slave = 1;
117 module_param(max_bonds, int, 0);
118 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
119 module_param(tx_queues, int, 0);
120 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
121 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
122 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
123 "failover event (alias of num_unsol_na)");
124 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
125 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
126 "failover event (alias of num_grat_arp)");
127 module_param(miimon, int, 0);
128 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
129 module_param(updelay, int, 0);
130 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
131 module_param(downdelay, int, 0);
132 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
134 module_param(use_carrier, int, 0);
135 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
136 "0 for off, 1 for on (default)");
137 module_param(mode, charp, 0);
138 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
139 "1 for active-backup, 2 for balance-xor, "
140 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
141 "6 for balance-alb");
142 module_param(primary, charp, 0);
143 MODULE_PARM_DESC(primary, "Primary network device to use");
144 module_param(primary_reselect, charp, 0);
145 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
147 "0 for always (default), "
148 "1 for only if speed of primary is "
150 "2 for only on active slave "
152 module_param(lacp_rate, charp, 0);
153 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
154 "0 for slow, 1 for fast");
155 module_param(ad_select, charp, 0);
156 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
157 "0 for stable (default), 1 for bandwidth, "
159 module_param(min_links, int, 0);
160 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
162 module_param(xmit_hash_policy, charp, 0);
163 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
164 "0 for layer 2 (default), 1 for layer 3+4, "
165 "2 for layer 2+3, 3 for encap layer 2+3, "
166 "4 for encap layer 3+4");
167 module_param(arp_interval, int, 0);
168 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
169 module_param_array(arp_ip_target, charp, NULL, 0);
170 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
171 module_param(arp_validate, charp, 0);
172 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
173 "0 for none (default), 1 for active, "
174 "2 for backup, 3 for all");
175 module_param(arp_all_targets, charp, 0);
176 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
177 module_param(fail_over_mac, charp, 0);
178 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
179 "the same MAC; 0 for none (default), "
180 "1 for active, 2 for follow");
181 module_param(all_slaves_active, int, 0);
182 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
183 "by setting active flag for all slaves; "
184 "0 for never (default), 1 for always.");
185 module_param(resend_igmp, int, 0);
186 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
188 module_param(packets_per_slave, int, 0);
189 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
190 "mode; 0 for a random slave, 1 packet per "
191 "slave (default), >1 packets per slave.");
193 /*----------------------------- Global variables ----------------------------*/
195 #ifdef CONFIG_NET_POLL_CONTROLLER
196 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
199 int bond_net_id __read_mostly;
201 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
202 static int arp_ip_count;
203 static int bond_mode = BOND_MODE_ROUNDROBIN;
204 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
205 static int lacp_fast;
207 const struct bond_parm_tbl bond_lacp_tbl[] = {
208 { "slow", AD_LACP_SLOW},
209 { "fast", AD_LACP_FAST},
213 const struct bond_parm_tbl bond_mode_tbl[] = {
214 { "balance-rr", BOND_MODE_ROUNDROBIN},
215 { "active-backup", BOND_MODE_ACTIVEBACKUP},
216 { "balance-xor", BOND_MODE_XOR},
217 { "broadcast", BOND_MODE_BROADCAST},
218 { "802.3ad", BOND_MODE_8023AD},
219 { "balance-tlb", BOND_MODE_TLB},
220 { "balance-alb", BOND_MODE_ALB},
224 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
225 { "layer2", BOND_XMIT_POLICY_LAYER2},
226 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
227 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
228 { "encap2+3", BOND_XMIT_POLICY_ENCAP23},
229 { "encap3+4", BOND_XMIT_POLICY_ENCAP34},
233 const struct bond_parm_tbl arp_all_targets_tbl[] = {
234 { "any", BOND_ARP_TARGETS_ANY},
235 { "all", BOND_ARP_TARGETS_ALL},
239 const struct bond_parm_tbl arp_validate_tbl[] = {
240 { "none", BOND_ARP_VALIDATE_NONE},
241 { "active", BOND_ARP_VALIDATE_ACTIVE},
242 { "backup", BOND_ARP_VALIDATE_BACKUP},
243 { "all", BOND_ARP_VALIDATE_ALL},
247 const struct bond_parm_tbl fail_over_mac_tbl[] = {
248 { "none", BOND_FOM_NONE},
249 { "active", BOND_FOM_ACTIVE},
250 { "follow", BOND_FOM_FOLLOW},
254 const struct bond_parm_tbl pri_reselect_tbl[] = {
255 { "always", BOND_PRI_RESELECT_ALWAYS},
256 { "better", BOND_PRI_RESELECT_BETTER},
257 { "failure", BOND_PRI_RESELECT_FAILURE},
261 struct bond_parm_tbl ad_select_tbl[] = {
262 { "stable", BOND_AD_STABLE},
263 { "bandwidth", BOND_AD_BANDWIDTH},
264 { "count", BOND_AD_COUNT},
268 /*-------------------------- Forward declarations ---------------------------*/
270 static int bond_init(struct net_device *bond_dev);
271 static void bond_uninit(struct net_device *bond_dev);
273 /*---------------------------- General routines -----------------------------*/
275 const char *bond_mode_name(int mode)
277 static const char *names[] = {
278 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
279 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
280 [BOND_MODE_XOR] = "load balancing (xor)",
281 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
282 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
283 [BOND_MODE_TLB] = "transmit load balancing",
284 [BOND_MODE_ALB] = "adaptive load balancing",
287 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
293 /*---------------------------------- VLAN -----------------------------------*/
296 * bond_dev_queue_xmit - Prepare skb for xmit.
298 * @bond: bond device that got this skb for tx.
299 * @skb: hw accel VLAN tagged skb to transmit
300 * @slave_dev: slave that is supposed to xmit this skbuff
302 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
303 struct net_device *slave_dev)
305 skb->dev = slave_dev;
307 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
308 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
309 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
311 if (unlikely(netpoll_tx_running(bond->dev)))
312 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
320 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
321 * We don't protect the slave list iteration with a lock because:
322 * a. This operation is performed in IOCTL context,
323 * b. The operation is protected by the RTNL semaphore in the 8021q code,
324 * c. Holding a lock with BH disabled while directly calling a base driver
325 * entry point is generally a BAD idea.
327 * The design of synchronization/protection for this operation in the 8021q
328 * module is good for one or more VLAN devices over a single physical device
329 * and cannot be extended for a teaming solution like bonding, so there is a
330 * potential race condition here where a net device from the vlan group might
331 * be referenced (either by a base driver or the 8021q code) while it is being
332 * removed from the system. However, it turns out we're not making matters
333 * worse, and if it works for regular VLAN usage it will work here too.
337 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
338 * @bond_dev: bonding net device that got called
339 * @vid: vlan id being added
341 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
342 __be16 proto, u16 vid)
344 struct bonding *bond = netdev_priv(bond_dev);
345 struct slave *slave, *rollback_slave;
346 struct list_head *iter;
349 bond_for_each_slave(bond, slave, iter) {
350 res = vlan_vid_add(slave->dev, proto, vid);
358 /* unwind to the slave that failed */
359 bond_for_each_slave(bond, rollback_slave, iter) {
360 if (rollback_slave == slave)
363 vlan_vid_del(rollback_slave->dev, proto, vid);
370 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
371 * @bond_dev: bonding net device that got called
372 * @vid: vlan id being removed
374 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
375 __be16 proto, u16 vid)
377 struct bonding *bond = netdev_priv(bond_dev);
378 struct list_head *iter;
381 bond_for_each_slave(bond, slave, iter)
382 vlan_vid_del(slave->dev, proto, vid);
384 if (bond_is_lb(bond))
385 bond_alb_clear_vlan(bond, vid);
390 /*------------------------------- Link status -------------------------------*/
393 * Set the carrier state for the master according to the state of its
394 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
395 * do special 802.3ad magic.
397 * Returns zero if carrier state does not change, nonzero if it does.
399 static int bond_set_carrier(struct bonding *bond)
401 struct list_head *iter;
404 if (!bond_has_slaves(bond))
407 if (bond->params.mode == BOND_MODE_8023AD)
408 return bond_3ad_set_carrier(bond);
410 bond_for_each_slave(bond, slave, iter) {
411 if (slave->link == BOND_LINK_UP) {
412 if (!netif_carrier_ok(bond->dev)) {
413 netif_carrier_on(bond->dev);
421 if (netif_carrier_ok(bond->dev)) {
422 netif_carrier_off(bond->dev);
429 * Get link speed and duplex from the slave's base driver
430 * using ethtool. If for some reason the call fails or the
431 * values are invalid, set speed and duplex to -1,
434 static void bond_update_speed_duplex(struct slave *slave)
436 struct net_device *slave_dev = slave->dev;
437 struct ethtool_cmd ecmd;
441 slave->speed = SPEED_UNKNOWN;
442 slave->duplex = DUPLEX_UNKNOWN;
444 res = __ethtool_get_settings(slave_dev, &ecmd);
448 slave_speed = ethtool_cmd_speed(&ecmd);
449 if (slave_speed == 0 || slave_speed == ((__u32) -1))
452 switch (ecmd.duplex) {
460 slave->speed = slave_speed;
461 slave->duplex = ecmd.duplex;
467 * if <dev> supports MII link status reporting, check its link status.
469 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
470 * depending upon the setting of the use_carrier parameter.
472 * Return either BMSR_LSTATUS, meaning that the link is up (or we
473 * can't tell and just pretend it is), or 0, meaning that the link is
476 * If reporting is non-zero, instead of faking link up, return -1 if
477 * both ETHTOOL and MII ioctls fail (meaning the device does not
478 * support them). If use_carrier is set, return whatever it says.
479 * It'd be nice if there was a good way to tell if a driver supports
480 * netif_carrier, but there really isn't.
482 static int bond_check_dev_link(struct bonding *bond,
483 struct net_device *slave_dev, int reporting)
485 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
486 int (*ioctl)(struct net_device *, struct ifreq *, int);
488 struct mii_ioctl_data *mii;
490 if (!reporting && !netif_running(slave_dev))
493 if (bond->params.use_carrier)
494 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
496 /* Try to get link status using Ethtool first. */
497 if (slave_dev->ethtool_ops->get_link)
498 return slave_dev->ethtool_ops->get_link(slave_dev) ?
501 /* Ethtool can't be used, fallback to MII ioctls. */
502 ioctl = slave_ops->ndo_do_ioctl;
504 /* TODO: set pointer to correct ioctl on a per team member */
505 /* bases to make this more efficient. that is, once */
506 /* we determine the correct ioctl, we will always */
507 /* call it and not the others for that team */
511 * We cannot assume that SIOCGMIIPHY will also read a
512 * register; not all network drivers (e.g., e100)
516 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
517 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
519 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
520 mii->reg_num = MII_BMSR;
521 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
522 return mii->val_out & BMSR_LSTATUS;
527 * If reporting, report that either there's no dev->do_ioctl,
528 * or both SIOCGMIIREG and get_link failed (meaning that we
529 * cannot report link status). If not reporting, pretend
532 return reporting ? -1 : BMSR_LSTATUS;
535 /*----------------------------- Multicast list ------------------------------*/
538 * Push the promiscuity flag down to appropriate slaves
540 static int bond_set_promiscuity(struct bonding *bond, int inc)
542 struct list_head *iter;
545 if (USES_PRIMARY(bond->params.mode)) {
546 /* write lock already acquired */
547 if (bond->curr_active_slave) {
548 err = dev_set_promiscuity(bond->curr_active_slave->dev,
554 bond_for_each_slave(bond, slave, iter) {
555 err = dev_set_promiscuity(slave->dev, inc);
564 * Push the allmulti flag down to all slaves
566 static int bond_set_allmulti(struct bonding *bond, int inc)
568 struct list_head *iter;
571 if (USES_PRIMARY(bond->params.mode)) {
572 /* write lock already acquired */
573 if (bond->curr_active_slave) {
574 err = dev_set_allmulti(bond->curr_active_slave->dev,
580 bond_for_each_slave(bond, slave, iter) {
581 err = dev_set_allmulti(slave->dev, inc);
590 * Retrieve the list of registered multicast addresses for the bonding
591 * device and retransmit an IGMP JOIN request to the current active
594 static void bond_resend_igmp_join_requests(struct bonding *bond)
596 if (!rtnl_trylock()) {
597 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
600 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
603 /* We use curr_slave_lock to protect against concurrent access to
604 * igmp_retrans from multiple running instances of this function and
605 * bond_change_active_slave
607 write_lock_bh(&bond->curr_slave_lock);
608 if (bond->igmp_retrans > 1) {
609 bond->igmp_retrans--;
610 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
612 write_unlock_bh(&bond->curr_slave_lock);
615 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
617 struct bonding *bond = container_of(work, struct bonding,
620 bond_resend_igmp_join_requests(bond);
623 /* Flush bond's hardware addresses from slave
625 static void bond_hw_addr_flush(struct net_device *bond_dev,
626 struct net_device *slave_dev)
628 struct bonding *bond = netdev_priv(bond_dev);
630 dev_uc_unsync(slave_dev, bond_dev);
631 dev_mc_unsync(slave_dev, bond_dev);
633 if (bond->params.mode == BOND_MODE_8023AD) {
634 /* del lacpdu mc addr from mc list */
635 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
637 dev_mc_del(slave_dev, lacpdu_multicast);
641 /*--------------------------- Active slave change ---------------------------*/
643 /* Update the hardware address list and promisc/allmulti for the new and
644 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
645 * slaves up date at all times; only the USES_PRIMARY modes need to call
646 * this function to swap these settings during a failover.
648 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
649 struct slave *old_active)
654 if (bond->dev->flags & IFF_PROMISC)
655 dev_set_promiscuity(old_active->dev, -1);
657 if (bond->dev->flags & IFF_ALLMULTI)
658 dev_set_allmulti(old_active->dev, -1);
660 bond_hw_addr_flush(bond->dev, old_active->dev);
664 /* FIXME: Signal errors upstream. */
665 if (bond->dev->flags & IFF_PROMISC)
666 dev_set_promiscuity(new_active->dev, 1);
668 if (bond->dev->flags & IFF_ALLMULTI)
669 dev_set_allmulti(new_active->dev, 1);
671 netif_addr_lock_bh(bond->dev);
672 dev_uc_sync(new_active->dev, bond->dev);
673 dev_mc_sync(new_active->dev, bond->dev);
674 netif_addr_unlock_bh(bond->dev);
679 * bond_set_dev_addr - clone slave's address to bond
680 * @bond_dev: bond net device
681 * @slave_dev: slave net device
683 * Should be called with RTNL held.
685 static void bond_set_dev_addr(struct net_device *bond_dev,
686 struct net_device *slave_dev)
688 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
689 bond_dev, slave_dev, slave_dev->addr_len);
690 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
691 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
692 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
696 * bond_do_fail_over_mac
698 * Perform special MAC address swapping for fail_over_mac settings
700 * Called with RTNL, curr_slave_lock for write_bh.
702 static void bond_do_fail_over_mac(struct bonding *bond,
703 struct slave *new_active,
704 struct slave *old_active)
705 __releases(&bond->curr_slave_lock)
706 __acquires(&bond->curr_slave_lock)
708 u8 tmp_mac[ETH_ALEN];
709 struct sockaddr saddr;
712 switch (bond->params.fail_over_mac) {
713 case BOND_FOM_ACTIVE:
715 write_unlock_bh(&bond->curr_slave_lock);
716 bond_set_dev_addr(bond->dev, new_active->dev);
717 write_lock_bh(&bond->curr_slave_lock);
720 case BOND_FOM_FOLLOW:
722 * if new_active && old_active, swap them
723 * if just old_active, do nothing (going to no active slave)
724 * if just new_active, set new_active to bond's MAC
729 write_unlock_bh(&bond->curr_slave_lock);
732 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
733 memcpy(saddr.sa_data, old_active->dev->dev_addr,
735 saddr.sa_family = new_active->dev->type;
737 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
738 saddr.sa_family = bond->dev->type;
741 rv = dev_set_mac_address(new_active->dev, &saddr);
743 pr_err("%s: Error %d setting MAC of slave %s\n",
744 bond->dev->name, -rv, new_active->dev->name);
751 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
752 saddr.sa_family = old_active->dev->type;
754 rv = dev_set_mac_address(old_active->dev, &saddr);
756 pr_err("%s: Error %d setting MAC of slave %s\n",
757 bond->dev->name, -rv, new_active->dev->name);
759 write_lock_bh(&bond->curr_slave_lock);
762 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
763 bond->dev->name, bond->params.fail_over_mac);
769 static bool bond_should_change_active(struct bonding *bond)
771 struct slave *prim = bond->primary_slave;
772 struct slave *curr = bond->curr_active_slave;
774 if (!prim || !curr || curr->link != BOND_LINK_UP)
776 if (bond->force_primary) {
777 bond->force_primary = false;
780 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
781 (prim->speed < curr->speed ||
782 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
784 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
790 * find_best_interface - select the best available slave to be the active one
791 * @bond: our bonding struct
793 static struct slave *bond_find_best_slave(struct bonding *bond)
795 struct slave *slave, *bestslave = NULL;
796 struct list_head *iter;
797 int mintime = bond->params.updelay;
799 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
800 bond_should_change_active(bond))
801 return bond->primary_slave;
803 bond_for_each_slave(bond, slave, iter) {
804 if (slave->link == BOND_LINK_UP)
806 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
807 slave->delay < mintime) {
808 mintime = slave->delay;
816 static bool bond_should_notify_peers(struct bonding *bond)
821 slave = rcu_dereference(bond->curr_active_slave);
824 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
825 bond->dev->name, slave ? slave->dev->name : "NULL");
827 if (!slave || !bond->send_peer_notif ||
828 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
835 * change_active_interface - change the active slave into the specified one
836 * @bond: our bonding struct
837 * @new: the new slave to make the active one
839 * Set the new slave to the bond's settings and unset them on the old
841 * Setting include flags, mc-list, promiscuity, allmulti, etc.
843 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
844 * because it is apparently the best available slave we have, even though its
845 * updelay hasn't timed out yet.
847 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
849 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
851 struct slave *old_active = bond->curr_active_slave;
853 if (old_active == new_active)
857 new_active->jiffies = jiffies;
859 if (new_active->link == BOND_LINK_BACK) {
860 if (USES_PRIMARY(bond->params.mode)) {
861 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
862 bond->dev->name, new_active->dev->name,
863 (bond->params.updelay - new_active->delay) * bond->params.miimon);
866 new_active->delay = 0;
867 new_active->link = BOND_LINK_UP;
869 if (bond->params.mode == BOND_MODE_8023AD)
870 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
872 if (bond_is_lb(bond))
873 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
875 if (USES_PRIMARY(bond->params.mode)) {
876 pr_info("%s: making interface %s the new active one.\n",
877 bond->dev->name, new_active->dev->name);
882 if (USES_PRIMARY(bond->params.mode))
883 bond_hw_addr_swap(bond, new_active, old_active);
885 if (bond_is_lb(bond)) {
886 bond_alb_handle_active_change(bond, new_active);
888 bond_set_slave_inactive_flags(old_active);
890 bond_set_slave_active_flags(new_active);
892 rcu_assign_pointer(bond->curr_active_slave, new_active);
895 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
897 bond_set_slave_inactive_flags(old_active);
900 bool should_notify_peers = false;
902 bond_set_slave_active_flags(new_active);
904 if (bond->params.fail_over_mac)
905 bond_do_fail_over_mac(bond, new_active,
908 if (netif_running(bond->dev)) {
909 bond->send_peer_notif =
910 bond->params.num_peer_notif;
911 should_notify_peers =
912 bond_should_notify_peers(bond);
915 write_unlock_bh(&bond->curr_slave_lock);
917 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
918 if (should_notify_peers)
919 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
922 write_lock_bh(&bond->curr_slave_lock);
926 /* resend IGMP joins since active slave has changed or
927 * all were sent on curr_active_slave.
928 * resend only if bond is brought up with the affected
929 * bonding modes and the retransmission is enabled */
930 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
931 ((USES_PRIMARY(bond->params.mode) && new_active) ||
932 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
933 bond->igmp_retrans = bond->params.resend_igmp;
934 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
939 * bond_select_active_slave - select a new active slave, if needed
940 * @bond: our bonding struct
942 * This functions should be called when one of the following occurs:
943 * - The old curr_active_slave has been released or lost its link.
944 * - The primary_slave has got its link back.
945 * - A slave has got its link back and there's no old curr_active_slave.
947 * Caller must hold curr_slave_lock for write_bh.
949 void bond_select_active_slave(struct bonding *bond)
951 struct slave *best_slave;
954 best_slave = bond_find_best_slave(bond);
955 if (best_slave != bond->curr_active_slave) {
956 bond_change_active_slave(bond, best_slave);
957 rv = bond_set_carrier(bond);
961 if (netif_carrier_ok(bond->dev)) {
962 pr_info("%s: first active interface up!\n",
965 pr_info("%s: now running without any active interface !\n",
971 #ifdef CONFIG_NET_POLL_CONTROLLER
972 static inline int slave_enable_netpoll(struct slave *slave)
977 np = kzalloc(sizeof(*np), GFP_ATOMIC);
982 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
991 static inline void slave_disable_netpoll(struct slave *slave)
993 struct netpoll *np = slave->np;
999 __netpoll_free_async(np);
1001 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1003 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1005 if (!slave_dev->netdev_ops->ndo_poll_controller)
1010 static void bond_poll_controller(struct net_device *bond_dev)
1014 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1016 struct bonding *bond = netdev_priv(bond_dev);
1017 struct list_head *iter;
1018 struct slave *slave;
1020 bond_for_each_slave(bond, slave, iter)
1021 if (IS_UP(slave->dev))
1022 slave_disable_netpoll(slave);
1025 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1027 struct bonding *bond = netdev_priv(dev);
1028 struct list_head *iter;
1029 struct slave *slave;
1032 bond_for_each_slave(bond, slave, iter) {
1033 err = slave_enable_netpoll(slave);
1035 bond_netpoll_cleanup(dev);
1042 static inline int slave_enable_netpoll(struct slave *slave)
1046 static inline void slave_disable_netpoll(struct slave *slave)
1049 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1054 /*---------------------------------- IOCTL ----------------------------------*/
1056 static netdev_features_t bond_fix_features(struct net_device *dev,
1057 netdev_features_t features)
1059 struct bonding *bond = netdev_priv(dev);
1060 struct list_head *iter;
1061 netdev_features_t mask;
1062 struct slave *slave;
1064 if (!bond_has_slaves(bond)) {
1065 /* Disable adding VLANs to empty bond. But why? --mq */
1066 features |= NETIF_F_VLAN_CHALLENGED;
1071 features &= ~NETIF_F_ONE_FOR_ALL;
1072 features |= NETIF_F_ALL_FOR_ALL;
1074 bond_for_each_slave(bond, slave, iter) {
1075 features = netdev_increment_features(features,
1076 slave->dev->features,
1079 features = netdev_add_tso_features(features, mask);
1084 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1085 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1086 NETIF_F_HIGHDMA | NETIF_F_LRO)
1088 static void bond_compute_features(struct bonding *bond)
1090 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1091 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1092 struct net_device *bond_dev = bond->dev;
1093 struct list_head *iter;
1094 struct slave *slave;
1095 unsigned short max_hard_header_len = ETH_HLEN;
1096 unsigned int gso_max_size = GSO_MAX_SIZE;
1097 u16 gso_max_segs = GSO_MAX_SEGS;
1099 if (!bond_has_slaves(bond))
1102 bond_for_each_slave(bond, slave, iter) {
1103 vlan_features = netdev_increment_features(vlan_features,
1104 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1106 dst_release_flag &= slave->dev->priv_flags;
1107 if (slave->dev->hard_header_len > max_hard_header_len)
1108 max_hard_header_len = slave->dev->hard_header_len;
1110 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1111 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1115 bond_dev->vlan_features = vlan_features;
1116 bond_dev->hard_header_len = max_hard_header_len;
1117 bond_dev->gso_max_segs = gso_max_segs;
1118 netif_set_gso_max_size(bond_dev, gso_max_size);
1120 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1121 bond_dev->priv_flags = flags | dst_release_flag;
1123 netdev_change_features(bond_dev);
1126 static void bond_setup_by_slave(struct net_device *bond_dev,
1127 struct net_device *slave_dev)
1129 bond_dev->header_ops = slave_dev->header_ops;
1131 bond_dev->type = slave_dev->type;
1132 bond_dev->hard_header_len = slave_dev->hard_header_len;
1133 bond_dev->addr_len = slave_dev->addr_len;
1135 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1136 slave_dev->addr_len);
1139 /* On bonding slaves other than the currently active slave, suppress
1140 * duplicates except for alb non-mcast/bcast.
1142 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1143 struct slave *slave,
1144 struct bonding *bond)
1146 if (bond_is_slave_inactive(slave)) {
1147 if (bond->params.mode == BOND_MODE_ALB &&
1148 skb->pkt_type != PACKET_BROADCAST &&
1149 skb->pkt_type != PACKET_MULTICAST)
1156 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1158 struct sk_buff *skb = *pskb;
1159 struct slave *slave;
1160 struct bonding *bond;
1161 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1163 int ret = RX_HANDLER_ANOTHER;
1165 skb = skb_share_check(skb, GFP_ATOMIC);
1167 return RX_HANDLER_CONSUMED;
1171 slave = bond_slave_get_rcu(skb->dev);
1174 if (bond->params.arp_interval)
1175 slave->dev->last_rx = jiffies;
1177 recv_probe = ACCESS_ONCE(bond->recv_probe);
1179 ret = recv_probe(skb, bond, slave);
1180 if (ret == RX_HANDLER_CONSUMED) {
1186 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1187 return RX_HANDLER_EXACT;
1190 skb->dev = bond->dev;
1192 if (bond->params.mode == BOND_MODE_ALB &&
1193 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1194 skb->pkt_type == PACKET_HOST) {
1196 if (unlikely(skb_cow_head(skb,
1197 skb->data - skb_mac_header(skb)))) {
1199 return RX_HANDLER_CONSUMED;
1201 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1207 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1208 struct net_device *slave_dev,
1209 struct slave *slave)
1213 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1216 slave_dev->flags |= IFF_SLAVE;
1217 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1221 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1222 struct net_device *slave_dev)
1224 netdev_upper_dev_unlink(slave_dev, bond_dev);
1225 slave_dev->flags &= ~IFF_SLAVE;
1226 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1229 /* enslave device <slave> to bond device <master> */
1230 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1232 struct bonding *bond = netdev_priv(bond_dev);
1233 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1234 struct slave *new_slave = NULL, *prev_slave;
1235 struct sockaddr addr;
1239 if (!bond->params.use_carrier &&
1240 slave_dev->ethtool_ops->get_link == NULL &&
1241 slave_ops->ndo_do_ioctl == NULL) {
1242 pr_warning("%s: Warning: no link monitoring support for %s\n",
1243 bond_dev->name, slave_dev->name);
1246 /* already enslaved */
1247 if (slave_dev->flags & IFF_SLAVE) {
1248 pr_debug("Error, Device was already enslaved\n");
1252 /* vlan challenged mutual exclusion */
1253 /* no need to lock since we're protected by rtnl_lock */
1254 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1255 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1256 if (vlan_uses_dev(bond_dev)) {
1257 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1258 bond_dev->name, slave_dev->name, bond_dev->name);
1261 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1262 bond_dev->name, slave_dev->name,
1263 slave_dev->name, bond_dev->name);
1266 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1270 * Old ifenslave binaries are no longer supported. These can
1271 * be identified with moderate accuracy by the state of the slave:
1272 * the current ifenslave will set the interface down prior to
1273 * enslaving it; the old ifenslave will not.
1275 if ((slave_dev->flags & IFF_UP)) {
1276 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1279 goto err_undo_flags;
1282 /* set bonding device ether type by slave - bonding netdevices are
1283 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1284 * there is a need to override some of the type dependent attribs/funcs.
1286 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1287 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1289 if (!bond_has_slaves(bond)) {
1290 if (bond_dev->type != slave_dev->type) {
1291 pr_debug("%s: change device type from %d to %d\n",
1293 bond_dev->type, slave_dev->type);
1295 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1297 res = notifier_to_errno(res);
1299 pr_err("%s: refused to change device type\n",
1302 goto err_undo_flags;
1305 /* Flush unicast and multicast addresses */
1306 dev_uc_flush(bond_dev);
1307 dev_mc_flush(bond_dev);
1309 if (slave_dev->type != ARPHRD_ETHER)
1310 bond_setup_by_slave(bond_dev, slave_dev);
1312 ether_setup(bond_dev);
1313 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1316 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1319 } else if (bond_dev->type != slave_dev->type) {
1320 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1322 slave_dev->type, bond_dev->type);
1324 goto err_undo_flags;
1327 if (slave_ops->ndo_set_mac_address == NULL) {
1328 if (!bond_has_slaves(bond)) {
1329 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1331 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1332 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1333 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",
1336 goto err_undo_flags;
1340 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1342 /* If this is the first slave, then we need to set the master's hardware
1343 * address to be the same as the slave's. */
1344 if (!bond_has_slaves(bond) &&
1345 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1346 bond_set_dev_addr(bond->dev, slave_dev);
1348 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1351 goto err_undo_flags;
1354 * Set the new_slave's queue_id to be zero. Queue ID mapping
1355 * is set via sysfs or module option if desired.
1357 new_slave->queue_id = 0;
1359 /* Save slave's original mtu and then set it to match the bond */
1360 new_slave->original_mtu = slave_dev->mtu;
1361 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1363 pr_debug("Error %d calling dev_set_mtu\n", res);
1368 * Save slave's original ("permanent") mac address for modes
1369 * that need it, and for restoring it upon release, and then
1370 * set it to the master's address
1372 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1374 if (!bond->params.fail_over_mac) {
1376 * Set slave to master's mac address. The application already
1377 * set the master's mac address to that of the first slave
1379 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1380 addr.sa_family = slave_dev->type;
1381 res = dev_set_mac_address(slave_dev, &addr);
1383 pr_debug("Error %d calling set_mac_address\n", res);
1384 goto err_restore_mtu;
1388 /* open the slave since the application closed it */
1389 res = dev_open(slave_dev);
1391 pr_debug("Opening slave %s failed\n", slave_dev->name);
1392 goto err_restore_mac;
1395 new_slave->bond = bond;
1396 new_slave->dev = slave_dev;
1397 slave_dev->priv_flags |= IFF_BONDING;
1399 if (bond_is_lb(bond)) {
1400 /* bond_alb_init_slave() must be called before all other stages since
1401 * it might fail and we do not want to have to undo everything
1403 res = bond_alb_init_slave(bond, new_slave);
1408 /* If the mode USES_PRIMARY, then the following is handled by
1409 * bond_change_active_slave().
1411 if (!USES_PRIMARY(bond->params.mode)) {
1412 /* set promiscuity level to new slave */
1413 if (bond_dev->flags & IFF_PROMISC) {
1414 res = dev_set_promiscuity(slave_dev, 1);
1419 /* set allmulti level to new slave */
1420 if (bond_dev->flags & IFF_ALLMULTI) {
1421 res = dev_set_allmulti(slave_dev, 1);
1426 netif_addr_lock_bh(bond_dev);
1428 dev_mc_sync_multiple(slave_dev, bond_dev);
1429 dev_uc_sync_multiple(slave_dev, bond_dev);
1431 netif_addr_unlock_bh(bond_dev);
1434 if (bond->params.mode == BOND_MODE_8023AD) {
1435 /* add lacpdu mc addr to mc list */
1436 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1438 dev_mc_add(slave_dev, lacpdu_multicast);
1441 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1443 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1444 bond_dev->name, slave_dev->name);
1448 prev_slave = bond_last_slave(bond);
1450 new_slave->delay = 0;
1451 new_slave->link_failure_count = 0;
1453 bond_update_speed_duplex(new_slave);
1455 new_slave->last_arp_rx = jiffies -
1456 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1457 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1458 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1460 if (bond->params.miimon && !bond->params.use_carrier) {
1461 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1463 if ((link_reporting == -1) && !bond->params.arp_interval) {
1465 * miimon is set but a bonded network driver
1466 * does not support ETHTOOL/MII and
1467 * arp_interval is not set. Note: if
1468 * use_carrier is enabled, we will never go
1469 * here (because netif_carrier is always
1470 * supported); thus, we don't need to change
1471 * the messages for netif_carrier.
1473 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",
1474 bond_dev->name, slave_dev->name);
1475 } else if (link_reporting == -1) {
1476 /* unable get link status using mii/ethtool */
1477 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",
1478 bond_dev->name, slave_dev->name);
1482 /* check for initial state */
1483 if (bond->params.miimon) {
1484 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1485 if (bond->params.updelay) {
1486 new_slave->link = BOND_LINK_BACK;
1487 new_slave->delay = bond->params.updelay;
1489 new_slave->link = BOND_LINK_UP;
1492 new_slave->link = BOND_LINK_DOWN;
1494 } else if (bond->params.arp_interval) {
1495 new_slave->link = (netif_carrier_ok(slave_dev) ?
1496 BOND_LINK_UP : BOND_LINK_DOWN);
1498 new_slave->link = BOND_LINK_UP;
1501 if (new_slave->link != BOND_LINK_DOWN)
1502 new_slave->jiffies = jiffies;
1503 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1504 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1505 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1507 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1508 /* if there is a primary slave, remember it */
1509 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1510 bond->primary_slave = new_slave;
1511 bond->force_primary = true;
1515 switch (bond->params.mode) {
1516 case BOND_MODE_ACTIVEBACKUP:
1517 bond_set_slave_inactive_flags(new_slave);
1519 case BOND_MODE_8023AD:
1520 /* in 802.3ad mode, the internal mechanism
1521 * will activate the slaves in the selected
1524 bond_set_slave_inactive_flags(new_slave);
1525 /* if this is the first slave */
1527 SLAVE_AD_INFO(new_slave).id = 1;
1528 /* Initialize AD with the number of times that the AD timer is called in 1 second
1529 * can be called only after the mac address of the bond is set
1531 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1533 SLAVE_AD_INFO(new_slave).id =
1534 SLAVE_AD_INFO(prev_slave).id + 1;
1537 bond_3ad_bind_slave(new_slave);
1541 bond_set_active_slave(new_slave);
1542 bond_set_slave_inactive_flags(new_slave);
1545 pr_debug("This slave is always active in trunk mode\n");
1547 /* always active in trunk mode */
1548 bond_set_active_slave(new_slave);
1550 /* In trunking mode there is little meaning to curr_active_slave
1551 * anyway (it holds no special properties of the bond device),
1552 * so we can change it without calling change_active_interface()
1554 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1555 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1558 } /* switch(bond_mode) */
1560 #ifdef CONFIG_NET_POLL_CONTROLLER
1561 slave_dev->npinfo = bond->dev->npinfo;
1562 if (slave_dev->npinfo) {
1563 if (slave_enable_netpoll(new_slave)) {
1564 read_unlock(&bond->lock);
1565 pr_info("Error, %s: master_dev is using netpoll, "
1566 "but new slave device does not support netpoll.\n",
1574 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1577 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1581 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1583 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1584 goto err_unregister;
1588 bond_compute_features(bond);
1589 bond_set_carrier(bond);
1591 if (USES_PRIMARY(bond->params.mode)) {
1592 write_lock_bh(&bond->curr_slave_lock);
1593 bond_select_active_slave(bond);
1594 write_unlock_bh(&bond->curr_slave_lock);
1597 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1598 bond_dev->name, slave_dev->name,
1599 bond_is_active_slave(new_slave) ? "n active" : " backup",
1600 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1602 /* enslave is successful */
1605 /* Undo stages on error */
1607 netdev_rx_handler_unregister(slave_dev);
1610 if (!USES_PRIMARY(bond->params.mode))
1611 bond_hw_addr_flush(bond_dev, slave_dev);
1613 vlan_vids_del_by_dev(slave_dev, bond_dev);
1614 if (bond->primary_slave == new_slave)
1615 bond->primary_slave = NULL;
1616 if (bond->curr_active_slave == new_slave) {
1617 write_lock_bh(&bond->curr_slave_lock);
1618 bond_change_active_slave(bond, NULL);
1619 bond_select_active_slave(bond);
1620 write_unlock_bh(&bond->curr_slave_lock);
1622 slave_disable_netpoll(new_slave);
1625 slave_dev->priv_flags &= ~IFF_BONDING;
1626 dev_close(slave_dev);
1629 if (!bond->params.fail_over_mac) {
1630 /* XXX TODO - fom follow mode needs to change master's
1631 * MAC if this slave's MAC is in use by the bond, or at
1632 * least print a warning.
1634 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1635 addr.sa_family = slave_dev->type;
1636 dev_set_mac_address(slave_dev, &addr);
1640 dev_set_mtu(slave_dev, new_slave->original_mtu);
1646 /* Enslave of first slave has failed and we need to fix master's mac */
1647 if (!bond_has_slaves(bond) &&
1648 ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1649 eth_hw_addr_random(bond_dev);
1655 * Try to release the slave device <slave> from the bond device <master>
1656 * It is legal to access curr_active_slave without a lock because all the function
1657 * is write-locked. If "all" is true it means that the function is being called
1658 * while destroying a bond interface and all slaves are being released.
1660 * The rules for slave state should be:
1661 * for Active/Backup:
1662 * Active stays on all backups go down
1663 * for Bonded connections:
1664 * The first up interface should be left on and all others downed.
1666 static int __bond_release_one(struct net_device *bond_dev,
1667 struct net_device *slave_dev,
1670 struct bonding *bond = netdev_priv(bond_dev);
1671 struct slave *slave, *oldcurrent;
1672 struct sockaddr addr;
1673 int old_flags = bond_dev->flags;
1674 netdev_features_t old_features = bond_dev->features;
1676 /* slave is not a slave or master is not master of this slave */
1677 if (!(slave_dev->flags & IFF_SLAVE) ||
1678 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1679 pr_err("%s: Error: cannot release %s.\n",
1680 bond_dev->name, slave_dev->name);
1686 slave = bond_get_slave_by_dev(bond, slave_dev);
1688 /* not a slave of this bond */
1689 pr_info("%s: %s not enslaved\n",
1690 bond_dev->name, slave_dev->name);
1691 unblock_netpoll_tx();
1695 /* release the slave from its bond */
1698 bond_upper_dev_unlink(bond_dev, slave_dev);
1699 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1700 * for this slave anymore.
1702 netdev_rx_handler_unregister(slave_dev);
1703 write_lock_bh(&bond->lock);
1705 /* Inform AD package of unbinding of slave. */
1706 if (bond->params.mode == BOND_MODE_8023AD) {
1707 /* must be called before the slave is
1708 * detached from the list
1710 bond_3ad_unbind_slave(slave);
1712 write_unlock_bh(&bond->lock);
1714 pr_info("%s: releasing %s interface %s\n",
1716 bond_is_active_slave(slave) ? "active" : "backup",
1719 oldcurrent = bond->curr_active_slave;
1721 bond->current_arp_slave = NULL;
1723 if (!all && !bond->params.fail_over_mac) {
1724 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1725 bond_has_slaves(bond))
1726 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",
1727 bond_dev->name, slave_dev->name,
1729 bond_dev->name, slave_dev->name);
1732 if (bond->primary_slave == slave)
1733 bond->primary_slave = NULL;
1735 if (oldcurrent == slave) {
1736 write_lock_bh(&bond->curr_slave_lock);
1737 bond_change_active_slave(bond, NULL);
1738 write_unlock_bh(&bond->curr_slave_lock);
1741 if (bond_is_lb(bond)) {
1742 /* Must be called only after the slave has been
1743 * detached from the list and the curr_active_slave
1744 * has been cleared (if our_slave == old_current),
1745 * but before a new active slave is selected.
1747 bond_alb_deinit_slave(bond, slave);
1751 rcu_assign_pointer(bond->curr_active_slave, NULL);
1752 } else if (oldcurrent == slave) {
1754 * Note that we hold RTNL over this sequence, so there
1755 * is no concern that another slave add/remove event
1758 write_lock_bh(&bond->curr_slave_lock);
1760 bond_select_active_slave(bond);
1762 write_unlock_bh(&bond->curr_slave_lock);
1765 if (!bond_has_slaves(bond)) {
1766 bond_set_carrier(bond);
1767 eth_hw_addr_random(bond_dev);
1769 if (vlan_uses_dev(bond_dev)) {
1770 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1771 bond_dev->name, bond_dev->name);
1772 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1777 unblock_netpoll_tx();
1780 if (!bond_has_slaves(bond)) {
1781 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1782 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1785 bond_compute_features(bond);
1786 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1787 (old_features & NETIF_F_VLAN_CHALLENGED))
1788 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1789 bond_dev->name, slave_dev->name, bond_dev->name);
1791 /* must do this from outside any spinlocks */
1792 vlan_vids_del_by_dev(slave_dev, bond_dev);
1794 /* If the mode USES_PRIMARY, then this cases was handled above by
1795 * bond_change_active_slave(..., NULL)
1797 if (!USES_PRIMARY(bond->params.mode)) {
1798 /* unset promiscuity level from slave
1799 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1800 * of the IFF_PROMISC flag in the bond_dev, but we need the
1801 * value of that flag before that change, as that was the value
1802 * when this slave was attached, so we cache at the start of the
1803 * function and use it here. Same goes for ALLMULTI below
1805 if (old_flags & IFF_PROMISC)
1806 dev_set_promiscuity(slave_dev, -1);
1808 /* unset allmulti level from slave */
1809 if (old_flags & IFF_ALLMULTI)
1810 dev_set_allmulti(slave_dev, -1);
1812 bond_hw_addr_flush(bond_dev, slave_dev);
1815 slave_disable_netpoll(slave);
1817 /* close slave before restoring its mac address */
1818 dev_close(slave_dev);
1820 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1821 /* restore original ("permanent") mac address */
1822 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1823 addr.sa_family = slave_dev->type;
1824 dev_set_mac_address(slave_dev, &addr);
1827 dev_set_mtu(slave_dev, slave->original_mtu);
1829 slave_dev->priv_flags &= ~IFF_BONDING;
1833 return 0; /* deletion OK */
1836 /* A wrapper used because of ndo_del_link */
1837 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1839 return __bond_release_one(bond_dev, slave_dev, false);
1843 * First release a slave and then destroy the bond if no more slaves are left.
1844 * Must be under rtnl_lock when this function is called.
1846 static int bond_release_and_destroy(struct net_device *bond_dev,
1847 struct net_device *slave_dev)
1849 struct bonding *bond = netdev_priv(bond_dev);
1852 ret = bond_release(bond_dev, slave_dev);
1853 if (ret == 0 && !bond_has_slaves(bond)) {
1854 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1855 pr_info("%s: destroying bond %s.\n",
1856 bond_dev->name, bond_dev->name);
1857 unregister_netdevice(bond_dev);
1862 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1864 struct bonding *bond = netdev_priv(bond_dev);
1866 info->bond_mode = bond->params.mode;
1867 info->miimon = bond->params.miimon;
1869 read_lock(&bond->lock);
1870 info->num_slaves = bond->slave_cnt;
1871 read_unlock(&bond->lock);
1876 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1878 struct bonding *bond = netdev_priv(bond_dev);
1879 struct list_head *iter;
1880 int i = 0, res = -ENODEV;
1881 struct slave *slave;
1883 read_lock(&bond->lock);
1884 bond_for_each_slave(bond, slave, iter) {
1885 if (i++ == (int)info->slave_id) {
1887 strcpy(info->slave_name, slave->dev->name);
1888 info->link = slave->link;
1889 info->state = bond_slave_state(slave);
1890 info->link_failure_count = slave->link_failure_count;
1894 read_unlock(&bond->lock);
1899 /*-------------------------------- Monitoring -------------------------------*/
1902 static int bond_miimon_inspect(struct bonding *bond)
1904 int link_state, commit = 0;
1905 struct list_head *iter;
1906 struct slave *slave;
1907 bool ignore_updelay;
1909 ignore_updelay = !bond->curr_active_slave ? true : false;
1911 bond_for_each_slave_rcu(bond, slave, iter) {
1912 slave->new_link = BOND_LINK_NOCHANGE;
1914 link_state = bond_check_dev_link(bond, slave->dev, 0);
1916 switch (slave->link) {
1921 slave->link = BOND_LINK_FAIL;
1922 slave->delay = bond->params.downdelay;
1924 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1926 (bond->params.mode ==
1927 BOND_MODE_ACTIVEBACKUP) ?
1928 (bond_is_active_slave(slave) ?
1929 "active " : "backup ") : "",
1931 bond->params.downdelay * bond->params.miimon);
1934 case BOND_LINK_FAIL:
1937 * recovered before downdelay expired
1939 slave->link = BOND_LINK_UP;
1940 slave->jiffies = jiffies;
1941 pr_info("%s: link status up again after %d ms for interface %s.\n",
1943 (bond->params.downdelay - slave->delay) *
1944 bond->params.miimon,
1949 if (slave->delay <= 0) {
1950 slave->new_link = BOND_LINK_DOWN;
1958 case BOND_LINK_DOWN:
1962 slave->link = BOND_LINK_BACK;
1963 slave->delay = bond->params.updelay;
1966 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1967 bond->dev->name, slave->dev->name,
1968 ignore_updelay ? 0 :
1969 bond->params.updelay *
1970 bond->params.miimon);
1973 case BOND_LINK_BACK:
1975 slave->link = BOND_LINK_DOWN;
1976 pr_info("%s: link status down again after %d ms for interface %s.\n",
1978 (bond->params.updelay - slave->delay) *
1979 bond->params.miimon,
1988 if (slave->delay <= 0) {
1989 slave->new_link = BOND_LINK_UP;
1991 ignore_updelay = false;
2003 static void bond_miimon_commit(struct bonding *bond)
2005 struct list_head *iter;
2006 struct slave *slave;
2008 bond_for_each_slave(bond, slave, iter) {
2009 switch (slave->new_link) {
2010 case BOND_LINK_NOCHANGE:
2014 slave->link = BOND_LINK_UP;
2015 slave->jiffies = jiffies;
2017 if (bond->params.mode == BOND_MODE_8023AD) {
2018 /* prevent it from being the active one */
2019 bond_set_backup_slave(slave);
2020 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2021 /* make it immediately active */
2022 bond_set_active_slave(slave);
2023 } else if (slave != bond->primary_slave) {
2024 /* prevent it from being the active one */
2025 bond_set_backup_slave(slave);
2028 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2029 bond->dev->name, slave->dev->name,
2030 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2031 slave->duplex ? "full" : "half");
2033 /* notify ad that the link status has changed */
2034 if (bond->params.mode == BOND_MODE_8023AD)
2035 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2037 if (bond_is_lb(bond))
2038 bond_alb_handle_link_change(bond, slave,
2041 if (!bond->curr_active_slave ||
2042 (slave == bond->primary_slave))
2047 case BOND_LINK_DOWN:
2048 if (slave->link_failure_count < UINT_MAX)
2049 slave->link_failure_count++;
2051 slave->link = BOND_LINK_DOWN;
2053 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2054 bond->params.mode == BOND_MODE_8023AD)
2055 bond_set_slave_inactive_flags(slave);
2057 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2058 bond->dev->name, slave->dev->name);
2060 if (bond->params.mode == BOND_MODE_8023AD)
2061 bond_3ad_handle_link_change(slave,
2064 if (bond_is_lb(bond))
2065 bond_alb_handle_link_change(bond, slave,
2068 if (slave == bond->curr_active_slave)
2074 pr_err("%s: invalid new link %d on slave %s\n",
2075 bond->dev->name, slave->new_link,
2077 slave->new_link = BOND_LINK_NOCHANGE;
2085 write_lock_bh(&bond->curr_slave_lock);
2086 bond_select_active_slave(bond);
2087 write_unlock_bh(&bond->curr_slave_lock);
2088 unblock_netpoll_tx();
2091 bond_set_carrier(bond);
2097 * Really a wrapper that splits the mii monitor into two phases: an
2098 * inspection, then (if inspection indicates something needs to be done)
2099 * an acquisition of appropriate locks followed by a commit phase to
2100 * implement whatever link state changes are indicated.
2102 void bond_mii_monitor(struct work_struct *work)
2104 struct bonding *bond = container_of(work, struct bonding,
2106 bool should_notify_peers = false;
2107 unsigned long delay;
2109 delay = msecs_to_jiffies(bond->params.miimon);
2111 if (!bond_has_slaves(bond))
2116 should_notify_peers = bond_should_notify_peers(bond);
2118 if (bond_miimon_inspect(bond)) {
2121 /* Race avoidance with bond_close cancel of workqueue */
2122 if (!rtnl_trylock()) {
2124 should_notify_peers = false;
2128 bond_miimon_commit(bond);
2130 rtnl_unlock(); /* might sleep, hold no other locks */
2135 if (bond->params.miimon)
2136 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2138 if (should_notify_peers) {
2139 if (!rtnl_trylock())
2141 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2146 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2148 struct net_device *upper;
2149 struct list_head *iter;
2152 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2156 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2157 if (ip == bond_confirm_addr(upper, 0, ip)) {
2168 * We go to the (large) trouble of VLAN tagging ARP frames because
2169 * switches in VLAN mode (especially if ports are configured as
2170 * "native" to a VLAN) might not pass non-tagged frames.
2172 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2174 struct sk_buff *skb;
2176 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2177 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2179 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2180 NULL, slave_dev->dev_addr, NULL);
2183 pr_err("ARP packet allocation failed\n");
2187 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2189 pr_err("failed to insert VLAN tag\n");
2197 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2199 struct net_device *upper, *vlan_upper;
2200 struct list_head *iter, *vlan_iter;
2202 __be32 *targets = bond->params.arp_targets, addr;
2205 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2206 pr_debug("basa: target %pI4\n", &targets[i]);
2208 /* Find out through which dev should the packet go */
2209 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2212 pr_debug("%s: no route to arp_ip_target %pI4\n",
2213 bond->dev->name, &targets[i]);
2219 /* bond device itself */
2220 if (rt->dst.dev == bond->dev)
2224 /* first we search only for vlan devices. for every vlan
2225 * found we verify its upper dev list, searching for the
2226 * rt->dst.dev. If found we save the tag of the vlan and
2227 * proceed to send the packet.
2231 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2233 if (!is_vlan_dev(vlan_upper))
2235 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2237 if (upper == rt->dst.dev) {
2238 vlan_id = vlan_dev_vlan_id(vlan_upper);
2245 /* if the device we're looking for is not on top of any of
2246 * our upper vlans, then just search for any dev that
2247 * matches, and in case it's a vlan - save the id
2249 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2250 if (upper == rt->dst.dev) {
2251 /* if it's a vlan - get its VID */
2252 if (is_vlan_dev(upper))
2253 vlan_id = vlan_dev_vlan_id(upper);
2261 /* Not our device - skip */
2262 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2263 bond->dev->name, &targets[i],
2264 rt->dst.dev ? rt->dst.dev->name : "NULL");
2270 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2272 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2277 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2281 if (!sip || !bond_has_this_ip(bond, tip)) {
2282 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2286 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2288 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2291 slave->last_arp_rx = jiffies;
2292 slave->target_last_arp_rx[i] = jiffies;
2295 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2296 struct slave *slave)
2298 struct arphdr *arp = (struct arphdr *)skb->data;
2299 unsigned char *arp_ptr;
2303 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2304 return RX_HANDLER_ANOTHER;
2306 read_lock(&bond->lock);
2308 if (!slave_do_arp_validate(bond, slave))
2311 alen = arp_hdr_len(bond->dev);
2313 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2314 bond->dev->name, skb->dev->name);
2316 if (alen > skb_headlen(skb)) {
2317 arp = kmalloc(alen, GFP_ATOMIC);
2320 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2324 if (arp->ar_hln != bond->dev->addr_len ||
2325 skb->pkt_type == PACKET_OTHERHOST ||
2326 skb->pkt_type == PACKET_LOOPBACK ||
2327 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2328 arp->ar_pro != htons(ETH_P_IP) ||
2332 arp_ptr = (unsigned char *)(arp + 1);
2333 arp_ptr += bond->dev->addr_len;
2334 memcpy(&sip, arp_ptr, 4);
2335 arp_ptr += 4 + bond->dev->addr_len;
2336 memcpy(&tip, arp_ptr, 4);
2338 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2339 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2340 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2344 * Backup slaves won't see the ARP reply, but do come through
2345 * here for each ARP probe (so we swap the sip/tip to validate
2346 * the probe). In a "redundant switch, common router" type of
2347 * configuration, the ARP probe will (hopefully) travel from
2348 * the active, through one switch, the router, then the other
2349 * switch before reaching the backup.
2351 * We 'trust' the arp requests if there is an active slave and
2352 * it received valid arp reply(s) after it became active. This
2353 * is done to avoid endless looping when we can't reach the
2354 * arp_ip_target and fool ourselves with our own arp requests.
2356 if (bond_is_active_slave(slave))
2357 bond_validate_arp(bond, slave, sip, tip);
2358 else if (bond->curr_active_slave &&
2359 time_after(slave_last_rx(bond, bond->curr_active_slave),
2360 bond->curr_active_slave->jiffies))
2361 bond_validate_arp(bond, slave, tip, sip);
2364 read_unlock(&bond->lock);
2365 if (arp != (struct arphdr *)skb->data)
2367 return RX_HANDLER_ANOTHER;
2370 /* function to verify if we're in the arp_interval timeslice, returns true if
2371 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2372 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2374 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2377 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2379 return time_in_range(jiffies,
2380 last_act - delta_in_ticks,
2381 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2385 * this function is called regularly to monitor each slave's link
2386 * ensuring that traffic is being sent and received when arp monitoring
2387 * is used in load-balancing mode. if the adapter has been dormant, then an
2388 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2389 * arp monitoring in active backup mode.
2391 void bond_loadbalance_arp_mon(struct work_struct *work)
2393 struct bonding *bond = container_of(work, struct bonding,
2395 struct slave *slave, *oldcurrent;
2396 struct list_head *iter;
2397 int do_failover = 0;
2399 if (!bond_has_slaves(bond))
2404 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2405 /* see if any of the previous devices are up now (i.e. they have
2406 * xmt and rcv traffic). the curr_active_slave does not come into
2407 * the picture unless it is null. also, slave->jiffies is not needed
2408 * here because we send an arp on each slave and give a slave as
2409 * long as it needs to get the tx/rx within the delta.
2410 * TODO: what about up/down delay in arp mode? it wasn't here before
2413 bond_for_each_slave_rcu(bond, slave, iter) {
2414 unsigned long trans_start = dev_trans_start(slave->dev);
2416 if (slave->link != BOND_LINK_UP) {
2417 if (bond_time_in_interval(bond, trans_start, 1) &&
2418 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2420 slave->link = BOND_LINK_UP;
2421 bond_set_active_slave(slave);
2423 /* primary_slave has no meaning in round-robin
2424 * mode. the window of a slave being up and
2425 * curr_active_slave being null after enslaving
2429 pr_info("%s: link status definitely up for interface %s, ",
2434 pr_info("%s: interface %s is now up\n",
2440 /* slave->link == BOND_LINK_UP */
2442 /* not all switches will respond to an arp request
2443 * when the source ip is 0, so don't take the link down
2444 * if we don't know our ip yet
2446 if (!bond_time_in_interval(bond, trans_start, 2) ||
2447 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2449 slave->link = BOND_LINK_DOWN;
2450 bond_set_backup_slave(slave);
2452 if (slave->link_failure_count < UINT_MAX)
2453 slave->link_failure_count++;
2455 pr_info("%s: interface %s is now down.\n",
2459 if (slave == oldcurrent)
2464 /* note: if switch is in round-robin mode, all links
2465 * must tx arp to ensure all links rx an arp - otherwise
2466 * links may oscillate or not come up at all; if switch is
2467 * in something like xor mode, there is nothing we can
2468 * do - all replies will be rx'ed on same link causing slaves
2469 * to be unstable during low/no traffic periods
2471 if (IS_UP(slave->dev))
2472 bond_arp_send_all(bond, slave);
2478 /* the bond_select_active_slave must hold RTNL
2479 * and curr_slave_lock for write.
2481 if (!rtnl_trylock())
2484 write_lock_bh(&bond->curr_slave_lock);
2486 bond_select_active_slave(bond);
2488 write_unlock_bh(&bond->curr_slave_lock);
2489 unblock_netpoll_tx();
2494 if (bond->params.arp_interval)
2495 queue_delayed_work(bond->wq, &bond->arp_work,
2496 msecs_to_jiffies(bond->params.arp_interval));
2500 * Called to inspect slaves for active-backup mode ARP monitor link state
2501 * changes. Sets new_link in slaves to specify what action should take
2502 * place for the slave. Returns 0 if no changes are found, >0 if changes
2503 * to link states must be committed.
2505 * Called with rcu_read_lock hold.
2507 static int bond_ab_arp_inspect(struct bonding *bond)
2509 unsigned long trans_start, last_rx;
2510 struct list_head *iter;
2511 struct slave *slave;
2514 bond_for_each_slave_rcu(bond, slave, iter) {
2515 slave->new_link = BOND_LINK_NOCHANGE;
2516 last_rx = slave_last_rx(bond, slave);
2518 if (slave->link != BOND_LINK_UP) {
2519 if (bond_time_in_interval(bond, last_rx, 1)) {
2520 slave->new_link = BOND_LINK_UP;
2527 * Give slaves 2*delta after being enslaved or made
2528 * active. This avoids bouncing, as the last receive
2529 * times need a full ARP monitor cycle to be updated.
2531 if (bond_time_in_interval(bond, slave->jiffies, 2))
2535 * Backup slave is down if:
2536 * - No current_arp_slave AND
2537 * - more than 3*delta since last receive AND
2538 * - the bond has an IP address
2540 * Note: a non-null current_arp_slave indicates
2541 * the curr_active_slave went down and we are
2542 * searching for a new one; under this condition
2543 * we only take the curr_active_slave down - this
2544 * gives each slave a chance to tx/rx traffic
2545 * before being taken out
2547 if (!bond_is_active_slave(slave) &&
2548 !bond->current_arp_slave &&
2549 !bond_time_in_interval(bond, last_rx, 3)) {
2550 slave->new_link = BOND_LINK_DOWN;
2555 * Active slave is down if:
2556 * - more than 2*delta since transmitting OR
2557 * - (more than 2*delta since receive AND
2558 * the bond has an IP address)
2560 trans_start = dev_trans_start(slave->dev);
2561 if (bond_is_active_slave(slave) &&
2562 (!bond_time_in_interval(bond, trans_start, 2) ||
2563 !bond_time_in_interval(bond, last_rx, 2))) {
2564 slave->new_link = BOND_LINK_DOWN;
2573 * Called to commit link state changes noted by inspection step of
2574 * active-backup mode ARP monitor.
2576 * Called with RTNL hold.
2578 static void bond_ab_arp_commit(struct bonding *bond)
2580 unsigned long trans_start;
2581 struct list_head *iter;
2582 struct slave *slave;
2584 bond_for_each_slave(bond, slave, iter) {
2585 switch (slave->new_link) {
2586 case BOND_LINK_NOCHANGE:
2590 trans_start = dev_trans_start(slave->dev);
2591 if (bond->curr_active_slave != slave ||
2592 (!bond->curr_active_slave &&
2593 bond_time_in_interval(bond, trans_start, 1))) {
2594 slave->link = BOND_LINK_UP;
2595 if (bond->current_arp_slave) {
2596 bond_set_slave_inactive_flags(
2597 bond->current_arp_slave);
2598 bond->current_arp_slave = NULL;
2601 pr_info("%s: link status definitely up for interface %s.\n",
2602 bond->dev->name, slave->dev->name);
2604 if (!bond->curr_active_slave ||
2605 (slave == bond->primary_slave))
2612 case BOND_LINK_DOWN:
2613 if (slave->link_failure_count < UINT_MAX)
2614 slave->link_failure_count++;
2616 slave->link = BOND_LINK_DOWN;
2617 bond_set_slave_inactive_flags(slave);
2619 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2620 bond->dev->name, slave->dev->name);
2622 if (slave == bond->curr_active_slave) {
2623 bond->current_arp_slave = NULL;
2630 pr_err("%s: impossible: new_link %d on slave %s\n",
2631 bond->dev->name, slave->new_link,
2639 write_lock_bh(&bond->curr_slave_lock);
2640 bond_select_active_slave(bond);
2641 write_unlock_bh(&bond->curr_slave_lock);
2642 unblock_netpoll_tx();
2645 bond_set_carrier(bond);
2649 * Send ARP probes for active-backup mode ARP monitor.
2651 * Called with rcu_read_lock hold.
2653 static void bond_ab_arp_probe(struct bonding *bond)
2655 struct slave *slave, *before = NULL, *new_slave = NULL,
2656 *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2657 struct list_head *iter;
2660 read_lock(&bond->curr_slave_lock);
2662 if (curr_arp_slave && bond->curr_active_slave)
2663 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2664 curr_arp_slave->dev->name,
2665 bond->curr_active_slave->dev->name);
2667 if (bond->curr_active_slave) {
2668 bond_arp_send_all(bond, bond->curr_active_slave);
2669 read_unlock(&bond->curr_slave_lock);
2673 read_unlock(&bond->curr_slave_lock);
2675 /* if we don't have a curr_active_slave, search for the next available
2676 * backup slave from the current_arp_slave and make it the candidate
2677 * for becoming the curr_active_slave
2680 if (!curr_arp_slave) {
2681 curr_arp_slave = bond_first_slave_rcu(bond);
2682 if (!curr_arp_slave)
2686 bond_set_slave_inactive_flags(curr_arp_slave);
2688 bond_for_each_slave_rcu(bond, slave, iter) {
2689 if (!found && !before && IS_UP(slave->dev))
2692 if (found && !new_slave && IS_UP(slave->dev))
2694 /* if the link state is up at this point, we
2695 * mark it down - this can happen if we have
2696 * simultaneous link failures and
2697 * reselect_active_interface doesn't make this
2698 * one the current slave so it is still marked
2699 * up when it is actually down
2701 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2702 slave->link = BOND_LINK_DOWN;
2703 if (slave->link_failure_count < UINT_MAX)
2704 slave->link_failure_count++;
2706 bond_set_slave_inactive_flags(slave);
2708 pr_info("%s: backup interface %s is now down.\n",
2709 bond->dev->name, slave->dev->name);
2711 if (slave == curr_arp_slave)
2715 if (!new_slave && before)
2721 new_slave->link = BOND_LINK_BACK;
2722 bond_set_slave_active_flags(new_slave);
2723 bond_arp_send_all(bond, new_slave);
2724 new_slave->jiffies = jiffies;
2725 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2728 void bond_activebackup_arp_mon(struct work_struct *work)
2730 struct bonding *bond = container_of(work, struct bonding,
2732 bool should_notify_peers = false;
2735 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2737 if (!bond_has_slaves(bond))
2742 should_notify_peers = bond_should_notify_peers(bond);
2744 if (bond_ab_arp_inspect(bond)) {
2747 /* Race avoidance with bond_close flush of workqueue */
2748 if (!rtnl_trylock()) {
2750 should_notify_peers = false;
2754 bond_ab_arp_commit(bond);
2760 bond_ab_arp_probe(bond);
2764 if (bond->params.arp_interval)
2765 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2767 if (should_notify_peers) {
2768 if (!rtnl_trylock())
2770 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2775 /*-------------------------- netdev event handling --------------------------*/
2778 * Change device name
2780 static int bond_event_changename(struct bonding *bond)
2782 bond_remove_proc_entry(bond);
2783 bond_create_proc_entry(bond);
2785 bond_debug_reregister(bond);
2790 static int bond_master_netdev_event(unsigned long event,
2791 struct net_device *bond_dev)
2793 struct bonding *event_bond = netdev_priv(bond_dev);
2796 case NETDEV_CHANGENAME:
2797 return bond_event_changename(event_bond);
2798 case NETDEV_UNREGISTER:
2799 bond_remove_proc_entry(event_bond);
2801 case NETDEV_REGISTER:
2802 bond_create_proc_entry(event_bond);
2804 case NETDEV_NOTIFY_PEERS:
2805 if (event_bond->send_peer_notif)
2806 event_bond->send_peer_notif--;
2815 static int bond_slave_netdev_event(unsigned long event,
2816 struct net_device *slave_dev)
2818 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2819 struct bonding *bond;
2820 struct net_device *bond_dev;
2824 /* A netdev event can be generated while enslaving a device
2825 * before netdev_rx_handler_register is called in which case
2826 * slave will be NULL
2830 bond_dev = slave->bond->dev;
2834 case NETDEV_UNREGISTER:
2835 if (bond_dev->type != ARPHRD_ETHER)
2836 bond_release_and_destroy(bond_dev, slave_dev);
2838 bond_release(bond_dev, slave_dev);
2842 old_speed = slave->speed;
2843 old_duplex = slave->duplex;
2845 bond_update_speed_duplex(slave);
2847 if (bond->params.mode == BOND_MODE_8023AD) {
2848 if (old_speed != slave->speed)
2849 bond_3ad_adapter_speed_changed(slave);
2850 if (old_duplex != slave->duplex)
2851 bond_3ad_adapter_duplex_changed(slave);
2856 * ... Or is it this?
2859 case NETDEV_CHANGEMTU:
2861 * TODO: Should slaves be allowed to
2862 * independently alter their MTU? For
2863 * an active-backup bond, slaves need
2864 * not be the same type of device, so
2865 * MTUs may vary. For other modes,
2866 * slaves arguably should have the
2867 * same MTUs. To do this, we'd need to
2868 * take over the slave's change_mtu
2869 * function for the duration of their
2873 case NETDEV_CHANGENAME:
2875 * TODO: handle changing the primary's name
2878 case NETDEV_FEAT_CHANGE:
2879 bond_compute_features(bond);
2881 case NETDEV_RESEND_IGMP:
2882 /* Propagate to master device */
2883 call_netdevice_notifiers(event, slave->bond->dev);
2893 * bond_netdev_event: handle netdev notifier chain events.
2895 * This function receives events for the netdev chain. The caller (an
2896 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2897 * locks for us to safely manipulate the slave devices (RTNL lock,
2900 static int bond_netdev_event(struct notifier_block *this,
2901 unsigned long event, void *ptr)
2903 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2905 pr_debug("event_dev: %s, event: %lx\n",
2906 event_dev ? event_dev->name : "None",
2909 if (!(event_dev->priv_flags & IFF_BONDING))
2912 if (event_dev->flags & IFF_MASTER) {
2913 pr_debug("IFF_MASTER\n");
2914 return bond_master_netdev_event(event, event_dev);
2917 if (event_dev->flags & IFF_SLAVE) {
2918 pr_debug("IFF_SLAVE\n");
2919 return bond_slave_netdev_event(event, event_dev);
2925 static struct notifier_block bond_netdev_notifier = {
2926 .notifier_call = bond_netdev_event,
2929 /*---------------------------- Hashing Policies -----------------------------*/
2931 /* L2 hash helper */
2932 static inline u32 bond_eth_hash(struct sk_buff *skb)
2934 struct ethhdr *data = (struct ethhdr *)skb->data;
2936 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2937 return data->h_dest[5] ^ data->h_source[5];
2942 /* Extract the appropriate headers based on bond's xmit policy */
2943 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2944 struct flow_keys *fk)
2946 const struct ipv6hdr *iph6;
2947 const struct iphdr *iph;
2948 int noff, proto = -1;
2950 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2951 return skb_flow_dissect(skb, fk);
2954 noff = skb_network_offset(skb);
2955 if (skb->protocol == htons(ETH_P_IP)) {
2956 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2959 fk->src = iph->saddr;
2960 fk->dst = iph->daddr;
2961 noff += iph->ihl << 2;
2962 if (!ip_is_fragment(iph))
2963 proto = iph->protocol;
2964 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2965 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2967 iph6 = ipv6_hdr(skb);
2968 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2969 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2970 noff += sizeof(*iph6);
2971 proto = iph6->nexthdr;
2975 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2976 fk->ports = skb_flow_get_ports(skb, noff, proto);
2982 * bond_xmit_hash - generate a hash value based on the xmit policy
2983 * @bond: bonding device
2984 * @skb: buffer to use for headers
2985 * @count: modulo value
2987 * This function will extract the necessary headers from the skb buffer and use
2988 * them to generate a hash based on the xmit_policy set in the bonding device
2989 * which will be reduced modulo count before returning.
2991 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
2993 struct flow_keys flow;
2996 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2997 !bond_flow_dissect(bond, skb, &flow))
2998 return bond_eth_hash(skb) % count;
3000 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3001 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3002 hash = bond_eth_hash(skb);
3004 hash = (__force u32)flow.ports;
3005 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3006 hash ^= (hash >> 16);
3007 hash ^= (hash >> 8);
3009 return hash % count;
3012 /*-------------------------- Device entry points ----------------------------*/
3014 static void bond_work_init_all(struct bonding *bond)
3016 INIT_DELAYED_WORK(&bond->mcast_work,
3017 bond_resend_igmp_join_requests_delayed);
3018 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3019 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3020 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3021 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3023 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3024 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3027 static void bond_work_cancel_all(struct bonding *bond)
3029 cancel_delayed_work_sync(&bond->mii_work);
3030 cancel_delayed_work_sync(&bond->arp_work);
3031 cancel_delayed_work_sync(&bond->alb_work);
3032 cancel_delayed_work_sync(&bond->ad_work);
3033 cancel_delayed_work_sync(&bond->mcast_work);
3036 static int bond_open(struct net_device *bond_dev)
3038 struct bonding *bond = netdev_priv(bond_dev);
3039 struct list_head *iter;
3040 struct slave *slave;
3042 /* reset slave->backup and slave->inactive */
3043 read_lock(&bond->lock);
3044 if (bond_has_slaves(bond)) {
3045 read_lock(&bond->curr_slave_lock);
3046 bond_for_each_slave(bond, slave, iter) {
3047 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3048 && (slave != bond->curr_active_slave)) {
3049 bond_set_slave_inactive_flags(slave);
3051 bond_set_slave_active_flags(slave);
3054 read_unlock(&bond->curr_slave_lock);
3056 read_unlock(&bond->lock);
3058 bond_work_init_all(bond);
3060 if (bond_is_lb(bond)) {
3061 /* bond_alb_initialize must be called before the timer
3064 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3066 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3069 if (bond->params.miimon) /* link check interval, in milliseconds. */
3070 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3072 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3073 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3074 if (bond->params.arp_validate)
3075 bond->recv_probe = bond_arp_rcv;
3078 if (bond->params.mode == BOND_MODE_8023AD) {
3079 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3080 /* register to receive LACPDUs */
3081 bond->recv_probe = bond_3ad_lacpdu_recv;
3082 bond_3ad_initiate_agg_selection(bond, 1);
3088 static int bond_close(struct net_device *bond_dev)
3090 struct bonding *bond = netdev_priv(bond_dev);
3092 bond_work_cancel_all(bond);
3093 bond->send_peer_notif = 0;
3094 if (bond_is_lb(bond))
3095 bond_alb_deinitialize(bond);
3096 bond->recv_probe = NULL;
3101 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3102 struct rtnl_link_stats64 *stats)
3104 struct bonding *bond = netdev_priv(bond_dev);
3105 struct rtnl_link_stats64 temp;
3106 struct list_head *iter;
3107 struct slave *slave;
3109 memset(stats, 0, sizeof(*stats));
3111 read_lock_bh(&bond->lock);
3112 bond_for_each_slave(bond, slave, iter) {
3113 const struct rtnl_link_stats64 *sstats =
3114 dev_get_stats(slave->dev, &temp);
3116 stats->rx_packets += sstats->rx_packets;
3117 stats->rx_bytes += sstats->rx_bytes;
3118 stats->rx_errors += sstats->rx_errors;
3119 stats->rx_dropped += sstats->rx_dropped;
3121 stats->tx_packets += sstats->tx_packets;
3122 stats->tx_bytes += sstats->tx_bytes;
3123 stats->tx_errors += sstats->tx_errors;
3124 stats->tx_dropped += sstats->tx_dropped;
3126 stats->multicast += sstats->multicast;
3127 stats->collisions += sstats->collisions;
3129 stats->rx_length_errors += sstats->rx_length_errors;
3130 stats->rx_over_errors += sstats->rx_over_errors;
3131 stats->rx_crc_errors += sstats->rx_crc_errors;
3132 stats->rx_frame_errors += sstats->rx_frame_errors;
3133 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3134 stats->rx_missed_errors += sstats->rx_missed_errors;
3136 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3137 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3138 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3139 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3140 stats->tx_window_errors += sstats->tx_window_errors;
3142 read_unlock_bh(&bond->lock);
3147 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3149 struct bonding *bond = netdev_priv(bond_dev);
3150 struct net_device *slave_dev = NULL;
3151 struct ifbond k_binfo;
3152 struct ifbond __user *u_binfo = NULL;
3153 struct ifslave k_sinfo;
3154 struct ifslave __user *u_sinfo = NULL;
3155 struct mii_ioctl_data *mii = NULL;
3159 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3171 * We do this again just in case we were called by SIOCGMIIREG
3172 * instead of SIOCGMIIPHY.
3179 if (mii->reg_num == 1) {
3181 read_lock(&bond->lock);
3182 read_lock(&bond->curr_slave_lock);
3183 if (netif_carrier_ok(bond->dev))
3184 mii->val_out = BMSR_LSTATUS;
3186 read_unlock(&bond->curr_slave_lock);
3187 read_unlock(&bond->lock);
3191 case BOND_INFO_QUERY_OLD:
3192 case SIOCBONDINFOQUERY:
3193 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3195 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3198 res = bond_info_query(bond_dev, &k_binfo);
3200 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3204 case BOND_SLAVE_INFO_QUERY_OLD:
3205 case SIOCBONDSLAVEINFOQUERY:
3206 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3208 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3211 res = bond_slave_info_query(bond_dev, &k_sinfo);
3213 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3222 net = dev_net(bond_dev);
3224 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3227 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3229 pr_debug("slave_dev=%p:\n", slave_dev);
3234 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3236 case BOND_ENSLAVE_OLD:
3237 case SIOCBONDENSLAVE:
3238 res = bond_enslave(bond_dev, slave_dev);
3240 case BOND_RELEASE_OLD:
3241 case SIOCBONDRELEASE:
3242 res = bond_release(bond_dev, slave_dev);
3244 case BOND_SETHWADDR_OLD:
3245 case SIOCBONDSETHWADDR:
3246 bond_set_dev_addr(bond_dev, slave_dev);
3249 case BOND_CHANGE_ACTIVE_OLD:
3250 case SIOCBONDCHANGEACTIVE:
3251 res = bond_option_active_slave_set(bond, slave_dev);
3263 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3265 struct bonding *bond = netdev_priv(bond_dev);
3267 if (change & IFF_PROMISC)
3268 bond_set_promiscuity(bond,
3269 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3271 if (change & IFF_ALLMULTI)
3272 bond_set_allmulti(bond,
3273 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3276 static void bond_set_rx_mode(struct net_device *bond_dev)
3278 struct bonding *bond = netdev_priv(bond_dev);
3279 struct list_head *iter;
3280 struct slave *slave;
3284 if (USES_PRIMARY(bond->params.mode)) {
3285 slave = rcu_dereference(bond->curr_active_slave);
3287 dev_uc_sync(slave->dev, bond_dev);
3288 dev_mc_sync(slave->dev, bond_dev);
3291 bond_for_each_slave_rcu(bond, slave, iter) {
3292 dev_uc_sync_multiple(slave->dev, bond_dev);
3293 dev_mc_sync_multiple(slave->dev, bond_dev);
3299 static int bond_neigh_init(struct neighbour *n)
3301 struct bonding *bond = netdev_priv(n->dev);
3302 const struct net_device_ops *slave_ops;
3303 struct neigh_parms parms;
3304 struct slave *slave;
3307 slave = bond_first_slave(bond);
3310 slave_ops = slave->dev->netdev_ops;
3311 if (!slave_ops->ndo_neigh_setup)
3314 parms.neigh_setup = NULL;
3315 parms.neigh_cleanup = NULL;
3316 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3321 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3322 * after the last slave has been detached. Assumes that all slaves
3323 * utilize the same neigh_cleanup (true at this writing as only user
3326 n->parms->neigh_cleanup = parms.neigh_cleanup;
3328 if (!parms.neigh_setup)
3331 return parms.neigh_setup(n);
3335 * The bonding ndo_neigh_setup is called at init time beofre any
3336 * slave exists. So we must declare proxy setup function which will
3337 * be used at run time to resolve the actual slave neigh param setup.
3339 * It's also called by master devices (such as vlans) to setup their
3340 * underlying devices. In that case - do nothing, we're already set up from
3343 static int bond_neigh_setup(struct net_device *dev,
3344 struct neigh_parms *parms)
3346 /* modify only our neigh_parms */
3347 if (parms->dev == dev)
3348 parms->neigh_setup = bond_neigh_init;
3354 * Change the MTU of all of a master's slaves to match the master
3356 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3358 struct bonding *bond = netdev_priv(bond_dev);
3359 struct slave *slave, *rollback_slave;
3360 struct list_head *iter;
3363 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3364 (bond_dev ? bond_dev->name : "None"), new_mtu);
3366 /* Can't hold bond->lock with bh disabled here since
3367 * some base drivers panic. On the other hand we can't
3368 * hold bond->lock without bh disabled because we'll
3369 * deadlock. The only solution is to rely on the fact
3370 * that we're under rtnl_lock here, and the slaves
3371 * list won't change. This doesn't solve the problem
3372 * of setting the slave's MTU while it is
3373 * transmitting, but the assumption is that the base
3374 * driver can handle that.
3376 * TODO: figure out a way to safely iterate the slaves
3377 * list, but without holding a lock around the actual
3378 * call to the base driver.
3381 bond_for_each_slave(bond, slave, iter) {
3382 pr_debug("s %p c_m %p\n",
3384 slave->dev->netdev_ops->ndo_change_mtu);
3386 res = dev_set_mtu(slave->dev, new_mtu);
3389 /* If we failed to set the slave's mtu to the new value
3390 * we must abort the operation even in ACTIVE_BACKUP
3391 * mode, because if we allow the backup slaves to have
3392 * different mtu values than the active slave we'll
3393 * need to change their mtu when doing a failover. That
3394 * means changing their mtu from timer context, which
3395 * is probably not a good idea.
3397 pr_debug("err %d %s\n", res, slave->dev->name);
3402 bond_dev->mtu = new_mtu;
3407 /* unwind from head to the slave that failed */
3408 bond_for_each_slave(bond, rollback_slave, iter) {
3411 if (rollback_slave == slave)
3414 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3416 pr_debug("unwind err %d dev %s\n",
3417 tmp_res, rollback_slave->dev->name);
3427 * Note that many devices must be down to change the HW address, and
3428 * downing the master releases all slaves. We can make bonds full of
3429 * bonding devices to test this, however.
3431 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3433 struct bonding *bond = netdev_priv(bond_dev);
3434 struct slave *slave, *rollback_slave;
3435 struct sockaddr *sa = addr, tmp_sa;
3436 struct list_head *iter;
3439 if (bond->params.mode == BOND_MODE_ALB)
3440 return bond_alb_set_mac_address(bond_dev, addr);
3443 pr_debug("bond=%p, name=%s\n",
3444 bond, bond_dev ? bond_dev->name : "None");
3446 /* If fail_over_mac is enabled, do nothing and return success.
3447 * Returning an error causes ifenslave to fail.
3449 if (bond->params.fail_over_mac)
3452 if (!is_valid_ether_addr(sa->sa_data))
3453 return -EADDRNOTAVAIL;
3455 /* Can't hold bond->lock with bh disabled here since
3456 * some base drivers panic. On the other hand we can't
3457 * hold bond->lock without bh disabled because we'll
3458 * deadlock. The only solution is to rely on the fact
3459 * that we're under rtnl_lock here, and the slaves
3460 * list won't change. This doesn't solve the problem
3461 * of setting the slave's hw address while it is
3462 * transmitting, but the assumption is that the base
3463 * driver can handle that.
3465 * TODO: figure out a way to safely iterate the slaves
3466 * list, but without holding a lock around the actual
3467 * call to the base driver.
3470 bond_for_each_slave(bond, slave, iter) {
3471 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3472 pr_debug("slave %p %s\n", slave, slave->dev->name);
3474 if (slave_ops->ndo_set_mac_address == NULL) {
3476 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3480 res = dev_set_mac_address(slave->dev, addr);
3482 /* TODO: consider downing the slave
3484 * User should expect communications
3485 * breakage anyway until ARP finish
3488 pr_debug("err %d %s\n", res, slave->dev->name);
3494 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3498 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3499 tmp_sa.sa_family = bond_dev->type;
3501 /* unwind from head to the slave that failed */
3502 bond_for_each_slave(bond, rollback_slave, iter) {
3505 if (rollback_slave == slave)
3508 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3510 pr_debug("unwind err %d dev %s\n",
3511 tmp_res, rollback_slave->dev->name);
3519 * bond_xmit_slave_id - transmit skb through slave with slave_id
3520 * @bond: bonding device that is transmitting
3521 * @skb: buffer to transmit
3522 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3524 * This function tries to transmit through slave with slave_id but in case
3525 * it fails, it tries to find the first available slave for transmission.
3526 * The skb is consumed in all cases, thus the function is void.
3528 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3530 struct list_head *iter;
3531 struct slave *slave;
3534 /* Here we start from the slave with slave_id */
3535 bond_for_each_slave_rcu(bond, slave, iter) {
3537 if (slave_can_tx(slave)) {
3538 bond_dev_queue_xmit(bond, skb, slave->dev);
3544 /* Here we start from the first slave up to slave_id */
3546 bond_for_each_slave_rcu(bond, slave, iter) {
3549 if (slave_can_tx(slave)) {
3550 bond_dev_queue_xmit(bond, skb, slave->dev);
3554 /* no slave that can tx has been found */
3559 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3560 * @bond: bonding device to use
3562 * Based on the value of the bonding device's packets_per_slave parameter
3563 * this function generates a slave id, which is usually used as the next
3564 * slave to transmit through.
3566 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3568 int packets_per_slave = bond->params.packets_per_slave;
3571 switch (packets_per_slave) {
3573 slave_id = prandom_u32();
3576 slave_id = bond->rr_tx_counter;
3579 slave_id = reciprocal_divide(bond->rr_tx_counter,
3583 bond->rr_tx_counter++;
3588 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3590 struct bonding *bond = netdev_priv(bond_dev);
3591 struct iphdr *iph = ip_hdr(skb);
3592 struct slave *slave;
3595 /* Start with the curr_active_slave that joined the bond as the
3596 * default for sending IGMP traffic. For failover purposes one
3597 * needs to maintain some consistency for the interface that will
3598 * send the join/membership reports. The curr_active_slave found
3599 * will send all of this type of traffic.
3601 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3602 slave = rcu_dereference(bond->curr_active_slave);
3603 if (slave && slave_can_tx(slave))
3604 bond_dev_queue_xmit(bond, skb, slave->dev);
3606 bond_xmit_slave_id(bond, skb, 0);
3608 slave_id = bond_rr_gen_slave_id(bond);
3609 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3612 return NETDEV_TX_OK;
3616 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3617 * the bond has a usable interface.
3619 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3621 struct bonding *bond = netdev_priv(bond_dev);
3622 struct slave *slave;
3624 slave = rcu_dereference(bond->curr_active_slave);
3626 bond_dev_queue_xmit(bond, skb, slave->dev);
3630 return NETDEV_TX_OK;
3633 /* In bond_xmit_xor() , we determine the output device by using a pre-
3634 * determined xmit_hash_policy(), If the selected device is not enabled,
3635 * find the next active slave.
3637 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3639 struct bonding *bond = netdev_priv(bond_dev);
3641 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3643 return NETDEV_TX_OK;
3646 /* in broadcast mode, we send everything to all usable interfaces. */
3647 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3649 struct bonding *bond = netdev_priv(bond_dev);
3650 struct slave *slave = NULL;
3651 struct list_head *iter;
3653 bond_for_each_slave_rcu(bond, slave, iter) {
3654 if (bond_is_last_slave(bond, slave))
3656 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3657 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3660 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3664 /* bond_dev_queue_xmit always returns 0 */
3665 bond_dev_queue_xmit(bond, skb2, slave->dev);
3668 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3669 bond_dev_queue_xmit(bond, skb, slave->dev);
3673 return NETDEV_TX_OK;
3676 /*------------------------- Device initialization ---------------------------*/
3679 * Lookup the slave that corresponds to a qid
3681 static inline int bond_slave_override(struct bonding *bond,
3682 struct sk_buff *skb)
3684 struct slave *slave = NULL;
3685 struct slave *check_slave;
3686 struct list_head *iter;
3689 if (!skb->queue_mapping)
3692 /* Find out if any slaves have the same mapping as this skb. */
3693 bond_for_each_slave_rcu(bond, check_slave, iter) {
3694 if (check_slave->queue_id == skb->queue_mapping) {
3695 slave = check_slave;
3700 /* If the slave isn't UP, use default transmit policy. */
3701 if (slave && slave->queue_id && IS_UP(slave->dev) &&
3702 (slave->link == BOND_LINK_UP)) {
3703 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3710 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3713 * This helper function exists to help dev_pick_tx get the correct
3714 * destination queue. Using a helper function skips a call to
3715 * skb_tx_hash and will put the skbs in the queue we expect on their
3716 * way down to the bonding driver.
3718 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3721 * Save the original txq to restore before passing to the driver
3723 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3725 if (unlikely(txq >= dev->real_num_tx_queues)) {
3727 txq -= dev->real_num_tx_queues;
3728 } while (txq >= dev->real_num_tx_queues);
3733 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3735 struct bonding *bond = netdev_priv(dev);
3737 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3738 if (!bond_slave_override(bond, skb))
3739 return NETDEV_TX_OK;
3742 switch (bond->params.mode) {
3743 case BOND_MODE_ROUNDROBIN:
3744 return bond_xmit_roundrobin(skb, dev);
3745 case BOND_MODE_ACTIVEBACKUP:
3746 return bond_xmit_activebackup(skb, dev);
3748 return bond_xmit_xor(skb, dev);
3749 case BOND_MODE_BROADCAST:
3750 return bond_xmit_broadcast(skb, dev);
3751 case BOND_MODE_8023AD:
3752 return bond_3ad_xmit_xor(skb, dev);
3755 return bond_alb_xmit(skb, dev);
3757 /* Should never happen, mode already checked */
3758 pr_err("%s: Error: Unknown bonding mode %d\n",
3759 dev->name, bond->params.mode);
3762 return NETDEV_TX_OK;
3766 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3768 struct bonding *bond = netdev_priv(dev);
3769 netdev_tx_t ret = NETDEV_TX_OK;
3772 * If we risk deadlock from transmitting this in the
3773 * netpoll path, tell netpoll to queue the frame for later tx
3775 if (is_netpoll_tx_blocked(dev))
3776 return NETDEV_TX_BUSY;
3779 if (bond_has_slaves(bond))
3780 ret = __bond_start_xmit(skb, dev);
3788 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3789 struct ethtool_cmd *ecmd)
3791 struct bonding *bond = netdev_priv(bond_dev);
3792 unsigned long speed = 0;
3793 struct list_head *iter;
3794 struct slave *slave;
3796 ecmd->duplex = DUPLEX_UNKNOWN;
3797 ecmd->port = PORT_OTHER;
3799 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3800 * do not need to check mode. Though link speed might not represent
3801 * the true receive or transmit bandwidth (not all modes are symmetric)
3802 * this is an accurate maximum.
3804 read_lock(&bond->lock);
3805 bond_for_each_slave(bond, slave, iter) {
3806 if (SLAVE_IS_OK(slave)) {
3807 if (slave->speed != SPEED_UNKNOWN)
3808 speed += slave->speed;
3809 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3810 slave->duplex != DUPLEX_UNKNOWN)
3811 ecmd->duplex = slave->duplex;
3814 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3815 read_unlock(&bond->lock);
3820 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3821 struct ethtool_drvinfo *drvinfo)
3823 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3824 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3825 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3829 static const struct ethtool_ops bond_ethtool_ops = {
3830 .get_drvinfo = bond_ethtool_get_drvinfo,
3831 .get_settings = bond_ethtool_get_settings,
3832 .get_link = ethtool_op_get_link,
3835 static const struct net_device_ops bond_netdev_ops = {
3836 .ndo_init = bond_init,
3837 .ndo_uninit = bond_uninit,
3838 .ndo_open = bond_open,
3839 .ndo_stop = bond_close,
3840 .ndo_start_xmit = bond_start_xmit,
3841 .ndo_select_queue = bond_select_queue,
3842 .ndo_get_stats64 = bond_get_stats,
3843 .ndo_do_ioctl = bond_do_ioctl,
3844 .ndo_change_rx_flags = bond_change_rx_flags,
3845 .ndo_set_rx_mode = bond_set_rx_mode,
3846 .ndo_change_mtu = bond_change_mtu,
3847 .ndo_set_mac_address = bond_set_mac_address,
3848 .ndo_neigh_setup = bond_neigh_setup,
3849 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3850 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3851 #ifdef CONFIG_NET_POLL_CONTROLLER
3852 .ndo_netpoll_setup = bond_netpoll_setup,
3853 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3854 .ndo_poll_controller = bond_poll_controller,
3856 .ndo_add_slave = bond_enslave,
3857 .ndo_del_slave = bond_release,
3858 .ndo_fix_features = bond_fix_features,
3861 static const struct device_type bond_type = {
3865 static void bond_destructor(struct net_device *bond_dev)
3867 struct bonding *bond = netdev_priv(bond_dev);
3869 destroy_workqueue(bond->wq);
3870 free_netdev(bond_dev);
3873 void bond_setup(struct net_device *bond_dev)
3875 struct bonding *bond = netdev_priv(bond_dev);
3877 /* initialize rwlocks */
3878 rwlock_init(&bond->lock);
3879 rwlock_init(&bond->curr_slave_lock);
3880 bond->params = bonding_defaults;
3882 /* Initialize pointers */
3883 bond->dev = bond_dev;
3885 /* Initialize the device entry points */
3886 ether_setup(bond_dev);
3887 bond_dev->netdev_ops = &bond_netdev_ops;
3888 bond_dev->ethtool_ops = &bond_ethtool_ops;
3890 bond_dev->destructor = bond_destructor;
3892 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3894 /* Initialize the device options */
3895 bond_dev->tx_queue_len = 0;
3896 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3897 bond_dev->priv_flags |= IFF_BONDING;
3898 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3900 /* At first, we block adding VLANs. That's the only way to
3901 * prevent problems that occur when adding VLANs over an
3902 * empty bond. The block will be removed once non-challenged
3903 * slaves are enslaved.
3905 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3907 /* don't acquire bond device's netif_tx_lock when
3909 bond_dev->features |= NETIF_F_LLTX;
3911 /* By default, we declare the bond to be fully
3912 * VLAN hardware accelerated capable. Special
3913 * care is taken in the various xmit functions
3914 * when there are slaves that are not hw accel
3918 bond_dev->hw_features = BOND_VLAN_FEATURES |
3919 NETIF_F_HW_VLAN_CTAG_TX |
3920 NETIF_F_HW_VLAN_CTAG_RX |
3921 NETIF_F_HW_VLAN_CTAG_FILTER;
3923 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3924 bond_dev->features |= bond_dev->hw_features;
3928 * Destroy a bonding device.
3929 * Must be under rtnl_lock when this function is called.
3931 static void bond_uninit(struct net_device *bond_dev)
3933 struct bonding *bond = netdev_priv(bond_dev);
3934 struct list_head *iter;
3935 struct slave *slave;
3937 bond_netpoll_cleanup(bond_dev);
3939 /* Release the bonded slaves */
3940 bond_for_each_slave(bond, slave, iter)
3941 __bond_release_one(bond_dev, slave->dev, true);
3942 pr_info("%s: released all slaves\n", bond_dev->name);
3944 list_del(&bond->bond_list);
3946 bond_debug_unregister(bond);
3949 /*------------------------- Module initialization ---------------------------*/
3952 * Convert string input module parms. Accept either the
3953 * number of the mode or its string name. A bit complicated because
3954 * some mode names are substrings of other names, and calls from sysfs
3955 * may have whitespace in the name (trailing newlines, for example).
3957 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3959 int modeint = -1, i, rv;
3960 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
3962 for (p = (char *)buf; *p; p++)
3963 if (!(isdigit(*p) || isspace(*p)))
3967 rv = sscanf(buf, "%20s", modestr);
3969 rv = sscanf(buf, "%d", &modeint);
3974 for (i = 0; tbl[i].modename; i++) {
3975 if (modeint == tbl[i].mode)
3977 if (strcmp(modestr, tbl[i].modename) == 0)
3984 static int bond_check_params(struct bond_params *params)
3986 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3987 int arp_all_targets_value;
3990 * Convert string parameters.
3993 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
3994 if (bond_mode == -1) {
3995 pr_err("Error: Invalid bonding mode \"%s\"\n",
3996 mode == NULL ? "NULL" : mode);
4001 if (xmit_hash_policy) {
4002 if ((bond_mode != BOND_MODE_XOR) &&
4003 (bond_mode != BOND_MODE_8023AD)) {
4004 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4005 bond_mode_name(bond_mode));
4007 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4009 if (xmit_hashtype == -1) {
4010 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4011 xmit_hash_policy == NULL ? "NULL" :
4019 if (bond_mode != BOND_MODE_8023AD) {
4020 pr_info("lacp_rate param is irrelevant in mode %s\n",
4021 bond_mode_name(bond_mode));
4023 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4024 if (lacp_fast == -1) {
4025 pr_err("Error: Invalid lacp rate \"%s\"\n",
4026 lacp_rate == NULL ? "NULL" : lacp_rate);
4033 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4034 if (params->ad_select == -1) {
4035 pr_err("Error: Invalid ad_select \"%s\"\n",
4036 ad_select == NULL ? "NULL" : ad_select);
4040 if (bond_mode != BOND_MODE_8023AD) {
4041 pr_warning("ad_select param only affects 802.3ad mode\n");
4044 params->ad_select = BOND_AD_STABLE;
4047 if (max_bonds < 0) {
4048 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4049 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4050 max_bonds = BOND_DEFAULT_MAX_BONDS;
4054 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4055 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4056 miimon = BOND_LINK_MON_INTERV;
4060 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4065 if (downdelay < 0) {
4066 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4067 downdelay, INT_MAX);
4071 if ((use_carrier != 0) && (use_carrier != 1)) {
4072 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4077 if (num_peer_notif < 0 || num_peer_notif > 255) {
4078 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4083 /* reset values for 802.3ad */
4084 if (bond_mode == BOND_MODE_8023AD) {
4086 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");
4087 pr_warning("Forcing miimon to 100msec\n");
4088 miimon = BOND_DEFAULT_MIIMON;
4092 if (tx_queues < 1 || tx_queues > 255) {
4093 pr_warning("Warning: tx_queues (%d) should be between "
4094 "1 and 255, resetting to %d\n",
4095 tx_queues, BOND_DEFAULT_TX_QUEUES);
4096 tx_queues = BOND_DEFAULT_TX_QUEUES;
4099 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4100 pr_warning("Warning: all_slaves_active module parameter (%d), "
4101 "not of valid value (0/1), so it was set to "
4102 "0\n", all_slaves_active);
4103 all_slaves_active = 0;
4106 if (resend_igmp < 0 || resend_igmp > 255) {
4107 pr_warning("Warning: resend_igmp (%d) should be between "
4108 "0 and 255, resetting to %d\n",
4109 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4110 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4113 if (packets_per_slave < 0 || packets_per_slave > USHRT_MAX) {
4114 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4115 packets_per_slave, USHRT_MAX);
4116 packets_per_slave = 1;
4119 /* reset values for TLB/ALB */
4120 if ((bond_mode == BOND_MODE_TLB) ||
4121 (bond_mode == BOND_MODE_ALB)) {
4123 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");
4124 pr_warning("Forcing miimon to 100msec\n");
4125 miimon = BOND_DEFAULT_MIIMON;
4129 if (bond_mode == BOND_MODE_ALB) {
4130 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",
4135 if (updelay || downdelay) {
4136 /* just warn the user the up/down delay will have
4137 * no effect since miimon is zero...
4139 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",
4140 updelay, downdelay);
4143 /* don't allow arp monitoring */
4145 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4146 miimon, arp_interval);
4150 if ((updelay % miimon) != 0) {
4151 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4153 (updelay / miimon) * miimon);
4158 if ((downdelay % miimon) != 0) {
4159 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4161 (downdelay / miimon) * miimon);
4164 downdelay /= miimon;
4167 if (arp_interval < 0) {
4168 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4169 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4170 arp_interval = BOND_LINK_ARP_INTERV;
4173 for (arp_ip_count = 0, i = 0;
4174 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4175 /* not complete check, but should be good enough to
4178 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4179 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4180 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4184 if (bond_get_targets_ip(arp_target, ip) == -1)
4185 arp_target[arp_ip_count++] = ip;
4187 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4192 if (arp_interval && !arp_ip_count) {
4193 /* don't allow arping if no arp_ip_target given... */
4194 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4200 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4201 pr_err("arp_validate only supported in active-backup mode\n");
4204 if (!arp_interval) {
4205 pr_err("arp_validate requires arp_interval\n");
4209 arp_validate_value = bond_parse_parm(arp_validate,
4211 if (arp_validate_value == -1) {
4212 pr_err("Error: invalid arp_validate \"%s\"\n",
4213 arp_validate == NULL ? "NULL" : arp_validate);
4217 arp_validate_value = 0;
4219 arp_all_targets_value = 0;
4220 if (arp_all_targets) {
4221 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4222 arp_all_targets_tbl);
4224 if (arp_all_targets_value == -1) {
4225 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4227 arp_all_targets_value = 0;
4232 pr_info("MII link monitoring set to %d ms\n", miimon);
4233 } else if (arp_interval) {
4234 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4236 arp_validate_tbl[arp_validate_value].modename,
4239 for (i = 0; i < arp_ip_count; i++)
4240 pr_info(" %s", arp_ip_target[i]);
4244 } else if (max_bonds) {
4245 /* miimon and arp_interval not set, we need one so things
4246 * work as expected, see bonding.txt for details
4248 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");
4251 if (primary && !USES_PRIMARY(bond_mode)) {
4252 /* currently, using a primary only makes sense
4253 * in active backup, TLB or ALB modes
4255 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4256 primary, bond_mode_name(bond_mode));
4260 if (primary && primary_reselect) {
4261 primary_reselect_value = bond_parse_parm(primary_reselect,
4263 if (primary_reselect_value == -1) {
4264 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4266 NULL ? "NULL" : primary_reselect);
4270 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4273 if (fail_over_mac) {
4274 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4276 if (fail_over_mac_value == -1) {
4277 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4278 arp_validate == NULL ? "NULL" : arp_validate);
4282 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4283 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4285 fail_over_mac_value = BOND_FOM_NONE;
4288 /* fill params struct with the proper values */
4289 params->mode = bond_mode;
4290 params->xmit_policy = xmit_hashtype;
4291 params->miimon = miimon;
4292 params->num_peer_notif = num_peer_notif;
4293 params->arp_interval = arp_interval;
4294 params->arp_validate = arp_validate_value;
4295 params->arp_all_targets = arp_all_targets_value;
4296 params->updelay = updelay;
4297 params->downdelay = downdelay;
4298 params->use_carrier = use_carrier;
4299 params->lacp_fast = lacp_fast;
4300 params->primary[0] = 0;
4301 params->primary_reselect = primary_reselect_value;
4302 params->fail_over_mac = fail_over_mac_value;
4303 params->tx_queues = tx_queues;
4304 params->all_slaves_active = all_slaves_active;
4305 params->resend_igmp = resend_igmp;
4306 params->min_links = min_links;
4307 params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4308 if (packets_per_slave > 1)
4309 params->packets_per_slave = reciprocal_value(packets_per_slave);
4311 params->packets_per_slave = packets_per_slave;
4313 strncpy(params->primary, primary, IFNAMSIZ);
4314 params->primary[IFNAMSIZ - 1] = 0;
4317 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4322 static struct lock_class_key bonding_netdev_xmit_lock_key;
4323 static struct lock_class_key bonding_netdev_addr_lock_key;
4324 static struct lock_class_key bonding_tx_busylock_key;
4326 static void bond_set_lockdep_class_one(struct net_device *dev,
4327 struct netdev_queue *txq,
4330 lockdep_set_class(&txq->_xmit_lock,
4331 &bonding_netdev_xmit_lock_key);
4334 static void bond_set_lockdep_class(struct net_device *dev)
4336 lockdep_set_class(&dev->addr_list_lock,
4337 &bonding_netdev_addr_lock_key);
4338 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4339 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4343 * Called from registration process
4345 static int bond_init(struct net_device *bond_dev)
4347 struct bonding *bond = netdev_priv(bond_dev);
4348 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4349 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4351 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4354 * Initialize locks that may be required during
4355 * en/deslave operations. All of the bond_open work
4356 * (of which this is part) should really be moved to
4357 * a phase prior to dev_open
4359 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4360 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4362 bond->wq = create_singlethread_workqueue(bond_dev->name);
4366 bond_set_lockdep_class(bond_dev);
4368 list_add_tail(&bond->bond_list, &bn->dev_list);
4370 bond_prepare_sysfs_group(bond);
4372 bond_debug_register(bond);
4374 /* Ensure valid dev_addr */
4375 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4376 bond_dev->addr_assign_type == NET_ADDR_PERM)
4377 eth_hw_addr_random(bond_dev);
4382 unsigned int bond_get_num_tx_queues(void)
4387 /* Create a new bond based on the specified name and bonding parameters.
4388 * If name is NULL, obtain a suitable "bond%d" name for us.
4389 * Caller must NOT hold rtnl_lock; we need to release it here before we
4390 * set up our sysfs entries.
4392 int bond_create(struct net *net, const char *name)
4394 struct net_device *bond_dev;
4399 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4400 name ? name : "bond%d",
4401 bond_setup, tx_queues);
4403 pr_err("%s: eek! can't alloc netdev!\n", name);
4408 dev_net_set(bond_dev, net);
4409 bond_dev->rtnl_link_ops = &bond_link_ops;
4411 res = register_netdevice(bond_dev);
4413 netif_carrier_off(bond_dev);
4417 bond_destructor(bond_dev);
4421 static int __net_init bond_net_init(struct net *net)
4423 struct bond_net *bn = net_generic(net, bond_net_id);
4426 INIT_LIST_HEAD(&bn->dev_list);
4428 bond_create_proc_dir(bn);
4429 bond_create_sysfs(bn);
4434 static void __net_exit bond_net_exit(struct net *net)
4436 struct bond_net *bn = net_generic(net, bond_net_id);
4437 struct bonding *bond, *tmp_bond;
4440 bond_destroy_sysfs(bn);
4441 bond_destroy_proc_dir(bn);
4443 /* Kill off any bonds created after unregistering bond rtnl ops */
4445 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4446 unregister_netdevice_queue(bond->dev, &list);
4447 unregister_netdevice_many(&list);
4451 static struct pernet_operations bond_net_ops = {
4452 .init = bond_net_init,
4453 .exit = bond_net_exit,
4455 .size = sizeof(struct bond_net),
4458 static int __init bonding_init(void)
4463 pr_info("%s", bond_version);
4465 res = bond_check_params(&bonding_defaults);
4469 res = register_pernet_subsys(&bond_net_ops);
4473 res = bond_netlink_init();
4477 bond_create_debugfs();
4479 for (i = 0; i < max_bonds; i++) {
4480 res = bond_create(&init_net, NULL);
4485 register_netdevice_notifier(&bond_netdev_notifier);
4489 bond_netlink_fini();
4491 unregister_pernet_subsys(&bond_net_ops);
4496 static void __exit bonding_exit(void)
4498 unregister_netdevice_notifier(&bond_netdev_notifier);
4500 bond_destroy_debugfs();
4502 bond_netlink_fini();
4503 unregister_pernet_subsys(&bond_net_ops);
4505 #ifdef CONFIG_NET_POLL_CONTROLLER
4507 * Make sure we don't have an imbalance on our netpoll blocking
4509 WARN_ON(atomic_read(&netpoll_block_tx));
4513 module_init(bonding_init);
4514 module_exit(bonding_exit);
4515 MODULE_LICENSE("GPL");
4516 MODULE_VERSION(DRV_VERSION);
4517 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4518 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");