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>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV 0
90 #define BOND_LINK_ARP_INTERV 0
92 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
93 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
94 static int num_peer_notif = 1;
95 static int miimon = BOND_LINK_MON_INTERV;
98 static int use_carrier = 1;
100 static char *primary;
101 static char *primary_reselect;
102 static char *lacp_rate;
103 static int min_links;
104 static char *ad_select;
105 static char *xmit_hash_policy;
106 static int arp_interval = BOND_LINK_ARP_INTERV;
107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
108 static char *arp_validate;
109 static char *arp_all_targets;
110 static char *fail_over_mac;
111 static int all_slaves_active;
112 static struct bond_params bonding_defaults;
113 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
114 static int packets_per_slave = 1;
115 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
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.");
192 module_param(lp_interval, uint, 0);
193 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
194 "the bonding driver sends learning packets to "
195 "each slaves peer switch. The default is 1.");
197 /*----------------------------- Global variables ----------------------------*/
199 #ifdef CONFIG_NET_POLL_CONTROLLER
200 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
203 int bond_net_id __read_mostly;
205 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
206 static int arp_ip_count;
207 static int bond_mode = BOND_MODE_ROUNDROBIN;
208 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
209 static int lacp_fast;
211 const struct bond_parm_tbl bond_lacp_tbl[] = {
212 { "slow", AD_LACP_SLOW},
213 { "fast", AD_LACP_FAST},
217 const struct bond_parm_tbl arp_all_targets_tbl[] = {
218 { "any", BOND_ARP_TARGETS_ANY},
219 { "all", BOND_ARP_TARGETS_ALL},
223 const struct bond_parm_tbl fail_over_mac_tbl[] = {
224 { "none", BOND_FOM_NONE},
225 { "active", BOND_FOM_ACTIVE},
226 { "follow", BOND_FOM_FOLLOW},
230 const struct bond_parm_tbl pri_reselect_tbl[] = {
231 { "always", BOND_PRI_RESELECT_ALWAYS},
232 { "better", BOND_PRI_RESELECT_BETTER},
233 { "failure", BOND_PRI_RESELECT_FAILURE},
237 struct bond_parm_tbl ad_select_tbl[] = {
238 { "stable", BOND_AD_STABLE},
239 { "bandwidth", BOND_AD_BANDWIDTH},
240 { "count", BOND_AD_COUNT},
244 /*-------------------------- Forward declarations ---------------------------*/
246 static int bond_init(struct net_device *bond_dev);
247 static void bond_uninit(struct net_device *bond_dev);
249 /*---------------------------- General routines -----------------------------*/
251 const char *bond_mode_name(int mode)
253 static const char *names[] = {
254 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
255 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
256 [BOND_MODE_XOR] = "load balancing (xor)",
257 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
258 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
259 [BOND_MODE_TLB] = "transmit load balancing",
260 [BOND_MODE_ALB] = "adaptive load balancing",
263 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
269 /*---------------------------------- VLAN -----------------------------------*/
272 * bond_dev_queue_xmit - Prepare skb for xmit.
274 * @bond: bond device that got this skb for tx.
275 * @skb: hw accel VLAN tagged skb to transmit
276 * @slave_dev: slave that is supposed to xmit this skbuff
278 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
279 struct net_device *slave_dev)
281 skb->dev = slave_dev;
283 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
284 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
285 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
287 if (unlikely(netpoll_tx_running(bond->dev)))
288 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
294 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
295 * We don't protect the slave list iteration with a lock because:
296 * a. This operation is performed in IOCTL context,
297 * b. The operation is protected by the RTNL semaphore in the 8021q code,
298 * c. Holding a lock with BH disabled while directly calling a base driver
299 * entry point is generally a BAD idea.
301 * The design of synchronization/protection for this operation in the 8021q
302 * module is good for one or more VLAN devices over a single physical device
303 * and cannot be extended for a teaming solution like bonding, so there is a
304 * potential race condition here where a net device from the vlan group might
305 * be referenced (either by a base driver or the 8021q code) while it is being
306 * removed from the system. However, it turns out we're not making matters
307 * worse, and if it works for regular VLAN usage it will work here too.
311 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
312 * @bond_dev: bonding net device that got called
313 * @vid: vlan id being added
315 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
316 __be16 proto, u16 vid)
318 struct bonding *bond = netdev_priv(bond_dev);
319 struct slave *slave, *rollback_slave;
320 struct list_head *iter;
323 bond_for_each_slave(bond, slave, iter) {
324 res = vlan_vid_add(slave->dev, proto, vid);
332 /* unwind to the slave that failed */
333 bond_for_each_slave(bond, rollback_slave, iter) {
334 if (rollback_slave == slave)
337 vlan_vid_del(rollback_slave->dev, proto, vid);
344 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
345 * @bond_dev: bonding net device that got called
346 * @vid: vlan id being removed
348 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
349 __be16 proto, u16 vid)
351 struct bonding *bond = netdev_priv(bond_dev);
352 struct list_head *iter;
355 bond_for_each_slave(bond, slave, iter)
356 vlan_vid_del(slave->dev, proto, vid);
358 if (bond_is_lb(bond))
359 bond_alb_clear_vlan(bond, vid);
364 /*------------------------------- Link status -------------------------------*/
367 * Set the carrier state for the master according to the state of its
368 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
369 * do special 802.3ad magic.
371 * Returns zero if carrier state does not change, nonzero if it does.
373 static int bond_set_carrier(struct bonding *bond)
375 struct list_head *iter;
378 if (!bond_has_slaves(bond))
381 if (bond->params.mode == BOND_MODE_8023AD)
382 return bond_3ad_set_carrier(bond);
384 bond_for_each_slave(bond, slave, iter) {
385 if (slave->link == BOND_LINK_UP) {
386 if (!netif_carrier_ok(bond->dev)) {
387 netif_carrier_on(bond->dev);
395 if (netif_carrier_ok(bond->dev)) {
396 netif_carrier_off(bond->dev);
403 * Get link speed and duplex from the slave's base driver
404 * using ethtool. If for some reason the call fails or the
405 * values are invalid, set speed and duplex to -1,
408 static void bond_update_speed_duplex(struct slave *slave)
410 struct net_device *slave_dev = slave->dev;
411 struct ethtool_cmd ecmd;
415 slave->speed = SPEED_UNKNOWN;
416 slave->duplex = DUPLEX_UNKNOWN;
418 res = __ethtool_get_settings(slave_dev, &ecmd);
422 slave_speed = ethtool_cmd_speed(&ecmd);
423 if (slave_speed == 0 || slave_speed == ((__u32) -1))
426 switch (ecmd.duplex) {
434 slave->speed = slave_speed;
435 slave->duplex = ecmd.duplex;
440 const char *bond_slave_link_status(s8 link)
457 * if <dev> supports MII link status reporting, check its link status.
459 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
460 * depending upon the setting of the use_carrier parameter.
462 * Return either BMSR_LSTATUS, meaning that the link is up (or we
463 * can't tell and just pretend it is), or 0, meaning that the link is
466 * If reporting is non-zero, instead of faking link up, return -1 if
467 * both ETHTOOL and MII ioctls fail (meaning the device does not
468 * support them). If use_carrier is set, return whatever it says.
469 * It'd be nice if there was a good way to tell if a driver supports
470 * netif_carrier, but there really isn't.
472 static int bond_check_dev_link(struct bonding *bond,
473 struct net_device *slave_dev, int reporting)
475 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
476 int (*ioctl)(struct net_device *, struct ifreq *, int);
478 struct mii_ioctl_data *mii;
480 if (!reporting && !netif_running(slave_dev))
483 if (bond->params.use_carrier)
484 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
486 /* Try to get link status using Ethtool first. */
487 if (slave_dev->ethtool_ops->get_link)
488 return slave_dev->ethtool_ops->get_link(slave_dev) ?
491 /* Ethtool can't be used, fallback to MII ioctls. */
492 ioctl = slave_ops->ndo_do_ioctl;
494 /* TODO: set pointer to correct ioctl on a per team member */
495 /* bases to make this more efficient. that is, once */
496 /* we determine the correct ioctl, we will always */
497 /* call it and not the others for that team */
501 * We cannot assume that SIOCGMIIPHY will also read a
502 * register; not all network drivers (e.g., e100)
506 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
507 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
509 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
510 mii->reg_num = MII_BMSR;
511 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
512 return mii->val_out & BMSR_LSTATUS;
517 * If reporting, report that either there's no dev->do_ioctl,
518 * or both SIOCGMIIREG and get_link failed (meaning that we
519 * cannot report link status). If not reporting, pretend
522 return reporting ? -1 : BMSR_LSTATUS;
525 /*----------------------------- Multicast list ------------------------------*/
528 * Push the promiscuity flag down to appropriate slaves
530 static int bond_set_promiscuity(struct bonding *bond, int inc)
532 struct list_head *iter;
535 if (USES_PRIMARY(bond->params.mode)) {
536 /* write lock already acquired */
537 if (bond->curr_active_slave) {
538 err = dev_set_promiscuity(bond->curr_active_slave->dev,
544 bond_for_each_slave(bond, slave, iter) {
545 err = dev_set_promiscuity(slave->dev, inc);
554 * Push the allmulti flag down to all slaves
556 static int bond_set_allmulti(struct bonding *bond, int inc)
558 struct list_head *iter;
561 if (USES_PRIMARY(bond->params.mode)) {
562 /* write lock already acquired */
563 if (bond->curr_active_slave) {
564 err = dev_set_allmulti(bond->curr_active_slave->dev,
570 bond_for_each_slave(bond, slave, iter) {
571 err = dev_set_allmulti(slave->dev, inc);
580 * Retrieve the list of registered multicast addresses for the bonding
581 * device and retransmit an IGMP JOIN request to the current active
584 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
586 struct bonding *bond = container_of(work, struct bonding,
589 if (!rtnl_trylock()) {
590 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
593 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
595 if (bond->igmp_retrans > 1) {
596 bond->igmp_retrans--;
597 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
602 /* Flush bond's hardware addresses from slave
604 static void bond_hw_addr_flush(struct net_device *bond_dev,
605 struct net_device *slave_dev)
607 struct bonding *bond = netdev_priv(bond_dev);
609 dev_uc_unsync(slave_dev, bond_dev);
610 dev_mc_unsync(slave_dev, bond_dev);
612 if (bond->params.mode == BOND_MODE_8023AD) {
613 /* del lacpdu mc addr from mc list */
614 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
616 dev_mc_del(slave_dev, lacpdu_multicast);
620 /*--------------------------- Active slave change ---------------------------*/
622 /* Update the hardware address list and promisc/allmulti for the new and
623 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
624 * slaves up date at all times; only the USES_PRIMARY modes need to call
625 * this function to swap these settings during a failover.
627 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
628 struct slave *old_active)
633 if (bond->dev->flags & IFF_PROMISC)
634 dev_set_promiscuity(old_active->dev, -1);
636 if (bond->dev->flags & IFF_ALLMULTI)
637 dev_set_allmulti(old_active->dev, -1);
639 bond_hw_addr_flush(bond->dev, old_active->dev);
643 /* FIXME: Signal errors upstream. */
644 if (bond->dev->flags & IFF_PROMISC)
645 dev_set_promiscuity(new_active->dev, 1);
647 if (bond->dev->flags & IFF_ALLMULTI)
648 dev_set_allmulti(new_active->dev, 1);
650 netif_addr_lock_bh(bond->dev);
651 dev_uc_sync(new_active->dev, bond->dev);
652 dev_mc_sync(new_active->dev, bond->dev);
653 netif_addr_unlock_bh(bond->dev);
658 * bond_set_dev_addr - clone slave's address to bond
659 * @bond_dev: bond net device
660 * @slave_dev: slave net device
662 * Should be called with RTNL held.
664 static void bond_set_dev_addr(struct net_device *bond_dev,
665 struct net_device *slave_dev)
667 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
668 bond_dev, slave_dev, slave_dev->addr_len);
669 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
670 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
671 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
675 * bond_do_fail_over_mac
677 * Perform special MAC address swapping for fail_over_mac settings
679 * Called with RTNL, curr_slave_lock for write_bh.
681 static void bond_do_fail_over_mac(struct bonding *bond,
682 struct slave *new_active,
683 struct slave *old_active)
684 __releases(&bond->curr_slave_lock)
685 __acquires(&bond->curr_slave_lock)
687 u8 tmp_mac[ETH_ALEN];
688 struct sockaddr saddr;
691 switch (bond->params.fail_over_mac) {
692 case BOND_FOM_ACTIVE:
694 write_unlock_bh(&bond->curr_slave_lock);
695 bond_set_dev_addr(bond->dev, new_active->dev);
696 write_lock_bh(&bond->curr_slave_lock);
699 case BOND_FOM_FOLLOW:
701 * if new_active && old_active, swap them
702 * if just old_active, do nothing (going to no active slave)
703 * if just new_active, set new_active to bond's MAC
708 write_unlock_bh(&bond->curr_slave_lock);
711 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
712 memcpy(saddr.sa_data, old_active->dev->dev_addr,
714 saddr.sa_family = new_active->dev->type;
716 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
717 saddr.sa_family = bond->dev->type;
720 rv = dev_set_mac_address(new_active->dev, &saddr);
722 pr_err("%s: Error %d setting MAC of slave %s\n",
723 bond->dev->name, -rv, new_active->dev->name);
730 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
731 saddr.sa_family = old_active->dev->type;
733 rv = dev_set_mac_address(old_active->dev, &saddr);
735 pr_err("%s: Error %d setting MAC of slave %s\n",
736 bond->dev->name, -rv, new_active->dev->name);
738 write_lock_bh(&bond->curr_slave_lock);
741 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
742 bond->dev->name, bond->params.fail_over_mac);
748 static bool bond_should_change_active(struct bonding *bond)
750 struct slave *prim = bond->primary_slave;
751 struct slave *curr = bond->curr_active_slave;
753 if (!prim || !curr || curr->link != BOND_LINK_UP)
755 if (bond->force_primary) {
756 bond->force_primary = false;
759 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
760 (prim->speed < curr->speed ||
761 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
763 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
769 * find_best_interface - select the best available slave to be the active one
770 * @bond: our bonding struct
772 static struct slave *bond_find_best_slave(struct bonding *bond)
774 struct slave *slave, *bestslave = NULL;
775 struct list_head *iter;
776 int mintime = bond->params.updelay;
778 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
779 bond_should_change_active(bond))
780 return bond->primary_slave;
782 bond_for_each_slave(bond, slave, iter) {
783 if (slave->link == BOND_LINK_UP)
785 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
786 slave->delay < mintime) {
787 mintime = slave->delay;
795 static bool bond_should_notify_peers(struct bonding *bond)
800 slave = rcu_dereference(bond->curr_active_slave);
803 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
804 bond->dev->name, slave ? slave->dev->name : "NULL");
806 if (!slave || !bond->send_peer_notif ||
807 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
814 * change_active_interface - change the active slave into the specified one
815 * @bond: our bonding struct
816 * @new: the new slave to make the active one
818 * Set the new slave to the bond's settings and unset them on the old
820 * Setting include flags, mc-list, promiscuity, allmulti, etc.
822 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
823 * because it is apparently the best available slave we have, even though its
824 * updelay hasn't timed out yet.
826 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
828 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
830 struct slave *old_active = bond->curr_active_slave;
832 if (old_active == new_active)
836 new_active->jiffies = jiffies;
838 if (new_active->link == BOND_LINK_BACK) {
839 if (USES_PRIMARY(bond->params.mode)) {
840 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
841 bond->dev->name, new_active->dev->name,
842 (bond->params.updelay - new_active->delay) * bond->params.miimon);
845 new_active->delay = 0;
846 new_active->link = BOND_LINK_UP;
848 if (bond->params.mode == BOND_MODE_8023AD)
849 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
851 if (bond_is_lb(bond))
852 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
854 if (USES_PRIMARY(bond->params.mode)) {
855 pr_info("%s: making interface %s the new active one.\n",
856 bond->dev->name, new_active->dev->name);
861 if (USES_PRIMARY(bond->params.mode))
862 bond_hw_addr_swap(bond, new_active, old_active);
864 if (bond_is_lb(bond)) {
865 bond_alb_handle_active_change(bond, new_active);
867 bond_set_slave_inactive_flags(old_active);
869 bond_set_slave_active_flags(new_active);
871 rcu_assign_pointer(bond->curr_active_slave, new_active);
874 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
876 bond_set_slave_inactive_flags(old_active);
879 bool should_notify_peers = false;
881 bond_set_slave_active_flags(new_active);
883 if (bond->params.fail_over_mac)
884 bond_do_fail_over_mac(bond, new_active,
887 if (netif_running(bond->dev)) {
888 bond->send_peer_notif =
889 bond->params.num_peer_notif;
890 should_notify_peers =
891 bond_should_notify_peers(bond);
894 write_unlock_bh(&bond->curr_slave_lock);
896 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
897 if (should_notify_peers)
898 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
901 write_lock_bh(&bond->curr_slave_lock);
905 /* resend IGMP joins since active slave has changed or
906 * all were sent on curr_active_slave.
907 * resend only if bond is brought up with the affected
908 * bonding modes and the retransmission is enabled */
909 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
910 ((USES_PRIMARY(bond->params.mode) && new_active) ||
911 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
912 bond->igmp_retrans = bond->params.resend_igmp;
913 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
918 * bond_select_active_slave - select a new active slave, if needed
919 * @bond: our bonding struct
921 * This functions should be called when one of the following occurs:
922 * - The old curr_active_slave has been released or lost its link.
923 * - The primary_slave has got its link back.
924 * - A slave has got its link back and there's no old curr_active_slave.
926 * Caller must hold curr_slave_lock for write_bh.
928 void bond_select_active_slave(struct bonding *bond)
930 struct slave *best_slave;
933 best_slave = bond_find_best_slave(bond);
934 if (best_slave != bond->curr_active_slave) {
935 bond_change_active_slave(bond, best_slave);
936 rv = bond_set_carrier(bond);
940 if (netif_carrier_ok(bond->dev)) {
941 pr_info("%s: first active interface up!\n",
944 pr_info("%s: now running without any active interface !\n",
950 #ifdef CONFIG_NET_POLL_CONTROLLER
951 static inline int slave_enable_netpoll(struct slave *slave)
956 np = kzalloc(sizeof(*np), GFP_ATOMIC);
961 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
970 static inline void slave_disable_netpoll(struct slave *slave)
972 struct netpoll *np = slave->np;
978 __netpoll_free_async(np);
980 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
982 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
984 if (!slave_dev->netdev_ops->ndo_poll_controller)
989 static void bond_poll_controller(struct net_device *bond_dev)
993 static void bond_netpoll_cleanup(struct net_device *bond_dev)
995 struct bonding *bond = netdev_priv(bond_dev);
996 struct list_head *iter;
999 bond_for_each_slave(bond, slave, iter)
1000 if (IS_UP(slave->dev))
1001 slave_disable_netpoll(slave);
1004 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1006 struct bonding *bond = netdev_priv(dev);
1007 struct list_head *iter;
1008 struct slave *slave;
1011 bond_for_each_slave(bond, slave, iter) {
1012 err = slave_enable_netpoll(slave);
1014 bond_netpoll_cleanup(dev);
1021 static inline int slave_enable_netpoll(struct slave *slave)
1025 static inline void slave_disable_netpoll(struct slave *slave)
1028 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1033 /*---------------------------------- IOCTL ----------------------------------*/
1035 static netdev_features_t bond_fix_features(struct net_device *dev,
1036 netdev_features_t features)
1038 struct bonding *bond = netdev_priv(dev);
1039 struct list_head *iter;
1040 netdev_features_t mask;
1041 struct slave *slave;
1043 if (!bond_has_slaves(bond)) {
1044 /* Disable adding VLANs to empty bond. But why? --mq */
1045 features |= NETIF_F_VLAN_CHALLENGED;
1050 features &= ~NETIF_F_ONE_FOR_ALL;
1051 features |= NETIF_F_ALL_FOR_ALL;
1053 bond_for_each_slave(bond, slave, iter) {
1054 features = netdev_increment_features(features,
1055 slave->dev->features,
1058 features = netdev_add_tso_features(features, mask);
1063 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1064 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1065 NETIF_F_HIGHDMA | NETIF_F_LRO)
1067 static void bond_compute_features(struct bonding *bond)
1069 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1070 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1071 struct net_device *bond_dev = bond->dev;
1072 struct list_head *iter;
1073 struct slave *slave;
1074 unsigned short max_hard_header_len = ETH_HLEN;
1075 unsigned int gso_max_size = GSO_MAX_SIZE;
1076 u16 gso_max_segs = GSO_MAX_SEGS;
1078 if (!bond_has_slaves(bond))
1081 bond_for_each_slave(bond, slave, iter) {
1082 vlan_features = netdev_increment_features(vlan_features,
1083 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1085 dst_release_flag &= slave->dev->priv_flags;
1086 if (slave->dev->hard_header_len > max_hard_header_len)
1087 max_hard_header_len = slave->dev->hard_header_len;
1089 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1090 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1094 bond_dev->vlan_features = vlan_features;
1095 bond_dev->hard_header_len = max_hard_header_len;
1096 bond_dev->gso_max_segs = gso_max_segs;
1097 netif_set_gso_max_size(bond_dev, gso_max_size);
1099 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1100 bond_dev->priv_flags = flags | dst_release_flag;
1102 netdev_change_features(bond_dev);
1105 static void bond_setup_by_slave(struct net_device *bond_dev,
1106 struct net_device *slave_dev)
1108 bond_dev->header_ops = slave_dev->header_ops;
1110 bond_dev->type = slave_dev->type;
1111 bond_dev->hard_header_len = slave_dev->hard_header_len;
1112 bond_dev->addr_len = slave_dev->addr_len;
1114 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1115 slave_dev->addr_len);
1118 /* On bonding slaves other than the currently active slave, suppress
1119 * duplicates except for alb non-mcast/bcast.
1121 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1122 struct slave *slave,
1123 struct bonding *bond)
1125 if (bond_is_slave_inactive(slave)) {
1126 if (bond->params.mode == BOND_MODE_ALB &&
1127 skb->pkt_type != PACKET_BROADCAST &&
1128 skb->pkt_type != PACKET_MULTICAST)
1135 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1137 struct sk_buff *skb = *pskb;
1138 struct slave *slave;
1139 struct bonding *bond;
1140 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1142 int ret = RX_HANDLER_ANOTHER;
1144 skb = skb_share_check(skb, GFP_ATOMIC);
1146 return RX_HANDLER_CONSUMED;
1150 slave = bond_slave_get_rcu(skb->dev);
1153 if (bond->params.arp_interval)
1154 slave->dev->last_rx = jiffies;
1156 recv_probe = ACCESS_ONCE(bond->recv_probe);
1158 ret = recv_probe(skb, bond, slave);
1159 if (ret == RX_HANDLER_CONSUMED) {
1165 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1166 return RX_HANDLER_EXACT;
1169 skb->dev = bond->dev;
1171 if (bond->params.mode == BOND_MODE_ALB &&
1172 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1173 skb->pkt_type == PACKET_HOST) {
1175 if (unlikely(skb_cow_head(skb,
1176 skb->data - skb_mac_header(skb)))) {
1178 return RX_HANDLER_CONSUMED;
1180 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1186 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1187 struct net_device *slave_dev,
1188 struct slave *slave)
1192 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1195 slave_dev->flags |= IFF_SLAVE;
1196 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1200 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1201 struct net_device *slave_dev)
1203 netdev_upper_dev_unlink(slave_dev, bond_dev);
1204 slave_dev->flags &= ~IFF_SLAVE;
1205 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1208 /* enslave device <slave> to bond device <master> */
1209 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1211 struct bonding *bond = netdev_priv(bond_dev);
1212 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1213 struct slave *new_slave = NULL, *prev_slave;
1214 struct sockaddr addr;
1218 if (!bond->params.use_carrier &&
1219 slave_dev->ethtool_ops->get_link == NULL &&
1220 slave_ops->ndo_do_ioctl == NULL) {
1221 pr_warning("%s: Warning: no link monitoring support for %s\n",
1222 bond_dev->name, slave_dev->name);
1225 /* already enslaved */
1226 if (slave_dev->flags & IFF_SLAVE) {
1227 pr_debug("Error, Device was already enslaved\n");
1231 /* vlan challenged mutual exclusion */
1232 /* no need to lock since we're protected by rtnl_lock */
1233 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1234 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1235 if (vlan_uses_dev(bond_dev)) {
1236 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1237 bond_dev->name, slave_dev->name, bond_dev->name);
1240 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1241 bond_dev->name, slave_dev->name,
1242 slave_dev->name, bond_dev->name);
1245 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1249 * Old ifenslave binaries are no longer supported. These can
1250 * be identified with moderate accuracy by the state of the slave:
1251 * the current ifenslave will set the interface down prior to
1252 * enslaving it; the old ifenslave will not.
1254 if ((slave_dev->flags & IFF_UP)) {
1255 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1258 goto err_undo_flags;
1261 /* set bonding device ether type by slave - bonding netdevices are
1262 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1263 * there is a need to override some of the type dependent attribs/funcs.
1265 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1266 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1268 if (!bond_has_slaves(bond)) {
1269 if (bond_dev->type != slave_dev->type) {
1270 pr_debug("%s: change device type from %d to %d\n",
1272 bond_dev->type, slave_dev->type);
1274 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1276 res = notifier_to_errno(res);
1278 pr_err("%s: refused to change device type\n",
1281 goto err_undo_flags;
1284 /* Flush unicast and multicast addresses */
1285 dev_uc_flush(bond_dev);
1286 dev_mc_flush(bond_dev);
1288 if (slave_dev->type != ARPHRD_ETHER)
1289 bond_setup_by_slave(bond_dev, slave_dev);
1291 ether_setup(bond_dev);
1292 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1295 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1298 } else if (bond_dev->type != slave_dev->type) {
1299 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1301 slave_dev->type, bond_dev->type);
1303 goto err_undo_flags;
1306 if (slave_ops->ndo_set_mac_address == NULL) {
1307 if (!bond_has_slaves(bond)) {
1308 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1310 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1311 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1312 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",
1315 goto err_undo_flags;
1319 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1321 /* If this is the first slave, then we need to set the master's hardware
1322 * address to be the same as the slave's. */
1323 if (!bond_has_slaves(bond) &&
1324 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1325 bond_set_dev_addr(bond->dev, slave_dev);
1327 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1330 goto err_undo_flags;
1333 * Set the new_slave's queue_id to be zero. Queue ID mapping
1334 * is set via sysfs or module option if desired.
1336 new_slave->queue_id = 0;
1338 /* Save slave's original mtu and then set it to match the bond */
1339 new_slave->original_mtu = slave_dev->mtu;
1340 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1342 pr_debug("Error %d calling dev_set_mtu\n", res);
1347 * Save slave's original ("permanent") mac address for modes
1348 * that need it, and for restoring it upon release, and then
1349 * set it to the master's address
1351 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1353 if (!bond->params.fail_over_mac) {
1355 * Set slave to master's mac address. The application already
1356 * set the master's mac address to that of the first slave
1358 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1359 addr.sa_family = slave_dev->type;
1360 res = dev_set_mac_address(slave_dev, &addr);
1362 pr_debug("Error %d calling set_mac_address\n", res);
1363 goto err_restore_mtu;
1367 /* open the slave since the application closed it */
1368 res = dev_open(slave_dev);
1370 pr_debug("Opening slave %s failed\n", slave_dev->name);
1371 goto err_restore_mac;
1374 new_slave->bond = bond;
1375 new_slave->dev = slave_dev;
1376 slave_dev->priv_flags |= IFF_BONDING;
1378 if (bond_is_lb(bond)) {
1379 /* bond_alb_init_slave() must be called before all other stages since
1380 * it might fail and we do not want to have to undo everything
1382 res = bond_alb_init_slave(bond, new_slave);
1387 /* If the mode USES_PRIMARY, then the following is handled by
1388 * bond_change_active_slave().
1390 if (!USES_PRIMARY(bond->params.mode)) {
1391 /* set promiscuity level to new slave */
1392 if (bond_dev->flags & IFF_PROMISC) {
1393 res = dev_set_promiscuity(slave_dev, 1);
1398 /* set allmulti level to new slave */
1399 if (bond_dev->flags & IFF_ALLMULTI) {
1400 res = dev_set_allmulti(slave_dev, 1);
1405 netif_addr_lock_bh(bond_dev);
1407 dev_mc_sync_multiple(slave_dev, bond_dev);
1408 dev_uc_sync_multiple(slave_dev, bond_dev);
1410 netif_addr_unlock_bh(bond_dev);
1413 if (bond->params.mode == BOND_MODE_8023AD) {
1414 /* add lacpdu mc addr to mc list */
1415 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1417 dev_mc_add(slave_dev, lacpdu_multicast);
1420 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1422 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1423 bond_dev->name, slave_dev->name);
1427 prev_slave = bond_last_slave(bond);
1429 new_slave->delay = 0;
1430 new_slave->link_failure_count = 0;
1432 bond_update_speed_duplex(new_slave);
1434 new_slave->last_arp_rx = jiffies -
1435 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1436 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1437 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1439 if (bond->params.miimon && !bond->params.use_carrier) {
1440 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1442 if ((link_reporting == -1) && !bond->params.arp_interval) {
1444 * miimon is set but a bonded network driver
1445 * does not support ETHTOOL/MII and
1446 * arp_interval is not set. Note: if
1447 * use_carrier is enabled, we will never go
1448 * here (because netif_carrier is always
1449 * supported); thus, we don't need to change
1450 * the messages for netif_carrier.
1452 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",
1453 bond_dev->name, slave_dev->name);
1454 } else if (link_reporting == -1) {
1455 /* unable get link status using mii/ethtool */
1456 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",
1457 bond_dev->name, slave_dev->name);
1461 /* check for initial state */
1462 if (bond->params.miimon) {
1463 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1464 if (bond->params.updelay) {
1465 new_slave->link = BOND_LINK_BACK;
1466 new_slave->delay = bond->params.updelay;
1468 new_slave->link = BOND_LINK_UP;
1471 new_slave->link = BOND_LINK_DOWN;
1473 } else if (bond->params.arp_interval) {
1474 new_slave->link = (netif_carrier_ok(slave_dev) ?
1475 BOND_LINK_UP : BOND_LINK_DOWN);
1477 new_slave->link = BOND_LINK_UP;
1480 if (new_slave->link != BOND_LINK_DOWN)
1481 new_slave->jiffies = jiffies;
1482 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1483 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1484 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1486 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1487 /* if there is a primary slave, remember it */
1488 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1489 bond->primary_slave = new_slave;
1490 bond->force_primary = true;
1494 switch (bond->params.mode) {
1495 case BOND_MODE_ACTIVEBACKUP:
1496 bond_set_slave_inactive_flags(new_slave);
1498 case BOND_MODE_8023AD:
1499 /* in 802.3ad mode, the internal mechanism
1500 * will activate the slaves in the selected
1503 bond_set_slave_inactive_flags(new_slave);
1504 /* if this is the first slave */
1506 SLAVE_AD_INFO(new_slave).id = 1;
1507 /* Initialize AD with the number of times that the AD timer is called in 1 second
1508 * can be called only after the mac address of the bond is set
1510 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1512 SLAVE_AD_INFO(new_slave).id =
1513 SLAVE_AD_INFO(prev_slave).id + 1;
1516 bond_3ad_bind_slave(new_slave);
1520 bond_set_active_slave(new_slave);
1521 bond_set_slave_inactive_flags(new_slave);
1524 pr_debug("This slave is always active in trunk mode\n");
1526 /* always active in trunk mode */
1527 bond_set_active_slave(new_slave);
1529 /* In trunking mode there is little meaning to curr_active_slave
1530 * anyway (it holds no special properties of the bond device),
1531 * so we can change it without calling change_active_interface()
1533 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1534 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1537 } /* switch(bond_mode) */
1539 #ifdef CONFIG_NET_POLL_CONTROLLER
1540 slave_dev->npinfo = bond->dev->npinfo;
1541 if (slave_dev->npinfo) {
1542 if (slave_enable_netpoll(new_slave)) {
1543 read_unlock(&bond->lock);
1544 pr_info("Error, %s: master_dev is using netpoll, "
1545 "but new slave device does not support netpoll.\n",
1553 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1556 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1560 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1562 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1563 goto err_unregister;
1566 res = bond_sysfs_slave_add(new_slave);
1568 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1569 goto err_upper_unlink;
1573 bond_compute_features(bond);
1574 bond_set_carrier(bond);
1576 if (USES_PRIMARY(bond->params.mode)) {
1577 write_lock_bh(&bond->curr_slave_lock);
1578 bond_select_active_slave(bond);
1579 write_unlock_bh(&bond->curr_slave_lock);
1582 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1583 bond_dev->name, slave_dev->name,
1584 bond_is_active_slave(new_slave) ? "n active" : " backup",
1585 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1587 /* enslave is successful */
1590 /* Undo stages on error */
1592 bond_upper_dev_unlink(bond_dev, slave_dev);
1595 netdev_rx_handler_unregister(slave_dev);
1598 if (!USES_PRIMARY(bond->params.mode))
1599 bond_hw_addr_flush(bond_dev, slave_dev);
1601 vlan_vids_del_by_dev(slave_dev, bond_dev);
1602 if (bond->primary_slave == new_slave)
1603 bond->primary_slave = NULL;
1604 if (bond->curr_active_slave == new_slave) {
1605 write_lock_bh(&bond->curr_slave_lock);
1606 bond_change_active_slave(bond, NULL);
1607 bond_select_active_slave(bond);
1608 write_unlock_bh(&bond->curr_slave_lock);
1610 slave_disable_netpoll(new_slave);
1613 slave_dev->priv_flags &= ~IFF_BONDING;
1614 dev_close(slave_dev);
1617 if (!bond->params.fail_over_mac) {
1618 /* XXX TODO - fom follow mode needs to change master's
1619 * MAC if this slave's MAC is in use by the bond, or at
1620 * least print a warning.
1622 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1623 addr.sa_family = slave_dev->type;
1624 dev_set_mac_address(slave_dev, &addr);
1628 dev_set_mtu(slave_dev, new_slave->original_mtu);
1634 /* Enslave of first slave has failed and we need to fix master's mac */
1635 if (!bond_has_slaves(bond) &&
1636 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1637 eth_hw_addr_random(bond_dev);
1643 * Try to release the slave device <slave> from the bond device <master>
1644 * It is legal to access curr_active_slave without a lock because all the function
1645 * is write-locked. If "all" is true it means that the function is being called
1646 * while destroying a bond interface and all slaves are being released.
1648 * The rules for slave state should be:
1649 * for Active/Backup:
1650 * Active stays on all backups go down
1651 * for Bonded connections:
1652 * The first up interface should be left on and all others downed.
1654 static int __bond_release_one(struct net_device *bond_dev,
1655 struct net_device *slave_dev,
1658 struct bonding *bond = netdev_priv(bond_dev);
1659 struct slave *slave, *oldcurrent;
1660 struct sockaddr addr;
1661 int old_flags = bond_dev->flags;
1662 netdev_features_t old_features = bond_dev->features;
1664 /* slave is not a slave or master is not master of this slave */
1665 if (!(slave_dev->flags & IFF_SLAVE) ||
1666 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1667 pr_err("%s: Error: cannot release %s.\n",
1668 bond_dev->name, slave_dev->name);
1674 slave = bond_get_slave_by_dev(bond, slave_dev);
1676 /* not a slave of this bond */
1677 pr_info("%s: %s not enslaved\n",
1678 bond_dev->name, slave_dev->name);
1679 unblock_netpoll_tx();
1683 /* release the slave from its bond */
1686 bond_sysfs_slave_del(slave);
1688 bond_upper_dev_unlink(bond_dev, slave_dev);
1689 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1690 * for this slave anymore.
1692 netdev_rx_handler_unregister(slave_dev);
1693 write_lock_bh(&bond->lock);
1695 /* Inform AD package of unbinding of slave. */
1696 if (bond->params.mode == BOND_MODE_8023AD)
1697 bond_3ad_unbind_slave(slave);
1699 write_unlock_bh(&bond->lock);
1701 pr_info("%s: releasing %s interface %s\n",
1703 bond_is_active_slave(slave) ? "active" : "backup",
1706 oldcurrent = bond->curr_active_slave;
1708 bond->current_arp_slave = NULL;
1710 if (!all && !bond->params.fail_over_mac) {
1711 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1712 bond_has_slaves(bond))
1713 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",
1714 bond_dev->name, slave_dev->name,
1716 bond_dev->name, slave_dev->name);
1719 if (bond->primary_slave == slave)
1720 bond->primary_slave = NULL;
1722 if (oldcurrent == slave) {
1723 write_lock_bh(&bond->curr_slave_lock);
1724 bond_change_active_slave(bond, NULL);
1725 write_unlock_bh(&bond->curr_slave_lock);
1728 if (bond_is_lb(bond)) {
1729 /* Must be called only after the slave has been
1730 * detached from the list and the curr_active_slave
1731 * has been cleared (if our_slave == old_current),
1732 * but before a new active slave is selected.
1734 bond_alb_deinit_slave(bond, slave);
1738 rcu_assign_pointer(bond->curr_active_slave, NULL);
1739 } else if (oldcurrent == slave) {
1741 * Note that we hold RTNL over this sequence, so there
1742 * is no concern that another slave add/remove event
1745 write_lock_bh(&bond->curr_slave_lock);
1747 bond_select_active_slave(bond);
1749 write_unlock_bh(&bond->curr_slave_lock);
1752 if (!bond_has_slaves(bond)) {
1753 bond_set_carrier(bond);
1754 eth_hw_addr_random(bond_dev);
1756 if (vlan_uses_dev(bond_dev)) {
1757 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1758 bond_dev->name, bond_dev->name);
1759 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1764 unblock_netpoll_tx();
1767 if (!bond_has_slaves(bond)) {
1768 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1769 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1772 bond_compute_features(bond);
1773 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1774 (old_features & NETIF_F_VLAN_CHALLENGED))
1775 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1776 bond_dev->name, slave_dev->name, bond_dev->name);
1778 /* must do this from outside any spinlocks */
1779 vlan_vids_del_by_dev(slave_dev, bond_dev);
1781 /* If the mode USES_PRIMARY, then this cases was handled above by
1782 * bond_change_active_slave(..., NULL)
1784 if (!USES_PRIMARY(bond->params.mode)) {
1785 /* unset promiscuity level from slave
1786 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1787 * of the IFF_PROMISC flag in the bond_dev, but we need the
1788 * value of that flag before that change, as that was the value
1789 * when this slave was attached, so we cache at the start of the
1790 * function and use it here. Same goes for ALLMULTI below
1792 if (old_flags & IFF_PROMISC)
1793 dev_set_promiscuity(slave_dev, -1);
1795 /* unset allmulti level from slave */
1796 if (old_flags & IFF_ALLMULTI)
1797 dev_set_allmulti(slave_dev, -1);
1799 bond_hw_addr_flush(bond_dev, slave_dev);
1802 slave_disable_netpoll(slave);
1804 /* close slave before restoring its mac address */
1805 dev_close(slave_dev);
1807 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1808 /* restore original ("permanent") mac address */
1809 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1810 addr.sa_family = slave_dev->type;
1811 dev_set_mac_address(slave_dev, &addr);
1814 dev_set_mtu(slave_dev, slave->original_mtu);
1816 slave_dev->priv_flags &= ~IFF_BONDING;
1820 return 0; /* deletion OK */
1823 /* A wrapper used because of ndo_del_link */
1824 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1826 return __bond_release_one(bond_dev, slave_dev, false);
1830 * First release a slave and then destroy the bond if no more slaves are left.
1831 * Must be under rtnl_lock when this function is called.
1833 static int bond_release_and_destroy(struct net_device *bond_dev,
1834 struct net_device *slave_dev)
1836 struct bonding *bond = netdev_priv(bond_dev);
1839 ret = bond_release(bond_dev, slave_dev);
1840 if (ret == 0 && !bond_has_slaves(bond)) {
1841 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1842 pr_info("%s: destroying bond %s.\n",
1843 bond_dev->name, bond_dev->name);
1844 unregister_netdevice(bond_dev);
1849 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1851 struct bonding *bond = netdev_priv(bond_dev);
1853 info->bond_mode = bond->params.mode;
1854 info->miimon = bond->params.miimon;
1856 read_lock(&bond->lock);
1857 info->num_slaves = bond->slave_cnt;
1858 read_unlock(&bond->lock);
1863 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1865 struct bonding *bond = netdev_priv(bond_dev);
1866 struct list_head *iter;
1867 int i = 0, res = -ENODEV;
1868 struct slave *slave;
1870 read_lock(&bond->lock);
1871 bond_for_each_slave(bond, slave, iter) {
1872 if (i++ == (int)info->slave_id) {
1874 strcpy(info->slave_name, slave->dev->name);
1875 info->link = slave->link;
1876 info->state = bond_slave_state(slave);
1877 info->link_failure_count = slave->link_failure_count;
1881 read_unlock(&bond->lock);
1886 /*-------------------------------- Monitoring -------------------------------*/
1889 static int bond_miimon_inspect(struct bonding *bond)
1891 int link_state, commit = 0;
1892 struct list_head *iter;
1893 struct slave *slave;
1894 bool ignore_updelay;
1896 ignore_updelay = !bond->curr_active_slave ? true : false;
1898 bond_for_each_slave_rcu(bond, slave, iter) {
1899 slave->new_link = BOND_LINK_NOCHANGE;
1901 link_state = bond_check_dev_link(bond, slave->dev, 0);
1903 switch (slave->link) {
1908 slave->link = BOND_LINK_FAIL;
1909 slave->delay = bond->params.downdelay;
1911 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1913 (bond->params.mode ==
1914 BOND_MODE_ACTIVEBACKUP) ?
1915 (bond_is_active_slave(slave) ?
1916 "active " : "backup ") : "",
1918 bond->params.downdelay * bond->params.miimon);
1921 case BOND_LINK_FAIL:
1924 * recovered before downdelay expired
1926 slave->link = BOND_LINK_UP;
1927 slave->jiffies = jiffies;
1928 pr_info("%s: link status up again after %d ms for interface %s.\n",
1930 (bond->params.downdelay - slave->delay) *
1931 bond->params.miimon,
1936 if (slave->delay <= 0) {
1937 slave->new_link = BOND_LINK_DOWN;
1945 case BOND_LINK_DOWN:
1949 slave->link = BOND_LINK_BACK;
1950 slave->delay = bond->params.updelay;
1953 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1954 bond->dev->name, slave->dev->name,
1955 ignore_updelay ? 0 :
1956 bond->params.updelay *
1957 bond->params.miimon);
1960 case BOND_LINK_BACK:
1962 slave->link = BOND_LINK_DOWN;
1963 pr_info("%s: link status down again after %d ms for interface %s.\n",
1965 (bond->params.updelay - slave->delay) *
1966 bond->params.miimon,
1975 if (slave->delay <= 0) {
1976 slave->new_link = BOND_LINK_UP;
1978 ignore_updelay = false;
1990 static void bond_miimon_commit(struct bonding *bond)
1992 struct list_head *iter;
1993 struct slave *slave;
1995 bond_for_each_slave(bond, slave, iter) {
1996 switch (slave->new_link) {
1997 case BOND_LINK_NOCHANGE:
2001 slave->link = BOND_LINK_UP;
2002 slave->jiffies = jiffies;
2004 if (bond->params.mode == BOND_MODE_8023AD) {
2005 /* prevent it from being the active one */
2006 bond_set_backup_slave(slave);
2007 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2008 /* make it immediately active */
2009 bond_set_active_slave(slave);
2010 } else if (slave != bond->primary_slave) {
2011 /* prevent it from being the active one */
2012 bond_set_backup_slave(slave);
2015 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2016 bond->dev->name, slave->dev->name,
2017 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2018 slave->duplex ? "full" : "half");
2020 /* notify ad that the link status has changed */
2021 if (bond->params.mode == BOND_MODE_8023AD)
2022 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2024 if (bond_is_lb(bond))
2025 bond_alb_handle_link_change(bond, slave,
2028 if (!bond->curr_active_slave ||
2029 (slave == bond->primary_slave))
2034 case BOND_LINK_DOWN:
2035 if (slave->link_failure_count < UINT_MAX)
2036 slave->link_failure_count++;
2038 slave->link = BOND_LINK_DOWN;
2040 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2041 bond->params.mode == BOND_MODE_8023AD)
2042 bond_set_slave_inactive_flags(slave);
2044 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2045 bond->dev->name, slave->dev->name);
2047 if (bond->params.mode == BOND_MODE_8023AD)
2048 bond_3ad_handle_link_change(slave,
2051 if (bond_is_lb(bond))
2052 bond_alb_handle_link_change(bond, slave,
2055 if (slave == bond->curr_active_slave)
2061 pr_err("%s: invalid new link %d on slave %s\n",
2062 bond->dev->name, slave->new_link,
2064 slave->new_link = BOND_LINK_NOCHANGE;
2072 write_lock_bh(&bond->curr_slave_lock);
2073 bond_select_active_slave(bond);
2074 write_unlock_bh(&bond->curr_slave_lock);
2075 unblock_netpoll_tx();
2078 bond_set_carrier(bond);
2084 * Really a wrapper that splits the mii monitor into two phases: an
2085 * inspection, then (if inspection indicates something needs to be done)
2086 * an acquisition of appropriate locks followed by a commit phase to
2087 * implement whatever link state changes are indicated.
2089 static void bond_mii_monitor(struct work_struct *work)
2091 struct bonding *bond = container_of(work, struct bonding,
2093 bool should_notify_peers = false;
2094 unsigned long delay;
2096 delay = msecs_to_jiffies(bond->params.miimon);
2098 if (!bond_has_slaves(bond))
2103 should_notify_peers = bond_should_notify_peers(bond);
2105 if (bond_miimon_inspect(bond)) {
2108 /* Race avoidance with bond_close cancel of workqueue */
2109 if (!rtnl_trylock()) {
2111 should_notify_peers = false;
2115 bond_miimon_commit(bond);
2117 rtnl_unlock(); /* might sleep, hold no other locks */
2122 if (bond->params.miimon)
2123 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2125 if (should_notify_peers) {
2126 if (!rtnl_trylock())
2128 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2133 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2135 struct net_device *upper;
2136 struct list_head *iter;
2139 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2143 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2144 if (ip == bond_confirm_addr(upper, 0, ip)) {
2155 * We go to the (large) trouble of VLAN tagging ARP frames because
2156 * switches in VLAN mode (especially if ports are configured as
2157 * "native" to a VLAN) might not pass non-tagged frames.
2159 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2161 struct sk_buff *skb;
2163 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2164 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2166 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2167 NULL, slave_dev->dev_addr, NULL);
2170 pr_err("ARP packet allocation failed\n");
2174 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2176 pr_err("failed to insert VLAN tag\n");
2184 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2186 struct net_device *upper, *vlan_upper;
2187 struct list_head *iter, *vlan_iter;
2189 __be32 *targets = bond->params.arp_targets, addr;
2192 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2193 pr_debug("basa: target %pI4\n", &targets[i]);
2195 /* Find out through which dev should the packet go */
2196 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2199 pr_debug("%s: no route to arp_ip_target %pI4\n",
2200 bond->dev->name, &targets[i]);
2206 /* bond device itself */
2207 if (rt->dst.dev == bond->dev)
2211 /* first we search only for vlan devices. for every vlan
2212 * found we verify its upper dev list, searching for the
2213 * rt->dst.dev. If found we save the tag of the vlan and
2214 * proceed to send the packet.
2218 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2220 if (!is_vlan_dev(vlan_upper))
2222 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2224 if (upper == rt->dst.dev) {
2225 vlan_id = vlan_dev_vlan_id(vlan_upper);
2232 /* if the device we're looking for is not on top of any of
2233 * our upper vlans, then just search for any dev that
2234 * matches, and in case it's a vlan - save the id
2236 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2237 if (upper == rt->dst.dev) {
2238 /* if it's a vlan - get its VID */
2239 if (is_vlan_dev(upper))
2240 vlan_id = vlan_dev_vlan_id(upper);
2248 /* Not our device - skip */
2249 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2250 bond->dev->name, &targets[i],
2251 rt->dst.dev ? rt->dst.dev->name : "NULL");
2257 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2259 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2264 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2268 if (!sip || !bond_has_this_ip(bond, tip)) {
2269 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2273 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2275 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2278 slave->last_arp_rx = jiffies;
2279 slave->target_last_arp_rx[i] = jiffies;
2282 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2283 struct slave *slave)
2285 struct arphdr *arp = (struct arphdr *)skb->data;
2286 unsigned char *arp_ptr;
2290 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2291 return RX_HANDLER_ANOTHER;
2293 read_lock(&bond->lock);
2295 if (!slave_do_arp_validate(bond, slave))
2298 alen = arp_hdr_len(bond->dev);
2300 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2301 bond->dev->name, skb->dev->name);
2303 if (alen > skb_headlen(skb)) {
2304 arp = kmalloc(alen, GFP_ATOMIC);
2307 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2311 if (arp->ar_hln != bond->dev->addr_len ||
2312 skb->pkt_type == PACKET_OTHERHOST ||
2313 skb->pkt_type == PACKET_LOOPBACK ||
2314 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2315 arp->ar_pro != htons(ETH_P_IP) ||
2319 arp_ptr = (unsigned char *)(arp + 1);
2320 arp_ptr += bond->dev->addr_len;
2321 memcpy(&sip, arp_ptr, 4);
2322 arp_ptr += 4 + bond->dev->addr_len;
2323 memcpy(&tip, arp_ptr, 4);
2325 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2326 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2327 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2331 * Backup slaves won't see the ARP reply, but do come through
2332 * here for each ARP probe (so we swap the sip/tip to validate
2333 * the probe). In a "redundant switch, common router" type of
2334 * configuration, the ARP probe will (hopefully) travel from
2335 * the active, through one switch, the router, then the other
2336 * switch before reaching the backup.
2338 * We 'trust' the arp requests if there is an active slave and
2339 * it received valid arp reply(s) after it became active. This
2340 * is done to avoid endless looping when we can't reach the
2341 * arp_ip_target and fool ourselves with our own arp requests.
2343 if (bond_is_active_slave(slave))
2344 bond_validate_arp(bond, slave, sip, tip);
2345 else if (bond->curr_active_slave &&
2346 time_after(slave_last_rx(bond, bond->curr_active_slave),
2347 bond->curr_active_slave->jiffies))
2348 bond_validate_arp(bond, slave, tip, sip);
2351 read_unlock(&bond->lock);
2352 if (arp != (struct arphdr *)skb->data)
2354 return RX_HANDLER_ANOTHER;
2357 /* function to verify if we're in the arp_interval timeslice, returns true if
2358 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2359 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2361 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2364 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2366 return time_in_range(jiffies,
2367 last_act - delta_in_ticks,
2368 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2372 * this function is called regularly to monitor each slave's link
2373 * ensuring that traffic is being sent and received when arp monitoring
2374 * is used in load-balancing mode. if the adapter has been dormant, then an
2375 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2376 * arp monitoring in active backup mode.
2378 static void bond_loadbalance_arp_mon(struct work_struct *work)
2380 struct bonding *bond = container_of(work, struct bonding,
2382 struct slave *slave, *oldcurrent;
2383 struct list_head *iter;
2384 int do_failover = 0;
2386 if (!bond_has_slaves(bond))
2391 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2392 /* see if any of the previous devices are up now (i.e. they have
2393 * xmt and rcv traffic). the curr_active_slave does not come into
2394 * the picture unless it is null. also, slave->jiffies is not needed
2395 * here because we send an arp on each slave and give a slave as
2396 * long as it needs to get the tx/rx within the delta.
2397 * TODO: what about up/down delay in arp mode? it wasn't here before
2400 bond_for_each_slave_rcu(bond, slave, iter) {
2401 unsigned long trans_start = dev_trans_start(slave->dev);
2403 if (slave->link != BOND_LINK_UP) {
2404 if (bond_time_in_interval(bond, trans_start, 1) &&
2405 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2407 slave->link = BOND_LINK_UP;
2408 bond_set_active_slave(slave);
2410 /* primary_slave has no meaning in round-robin
2411 * mode. the window of a slave being up and
2412 * curr_active_slave being null after enslaving
2416 pr_info("%s: link status definitely up for interface %s, ",
2421 pr_info("%s: interface %s is now up\n",
2427 /* slave->link == BOND_LINK_UP */
2429 /* not all switches will respond to an arp request
2430 * when the source ip is 0, so don't take the link down
2431 * if we don't know our ip yet
2433 if (!bond_time_in_interval(bond, trans_start, 2) ||
2434 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2436 slave->link = BOND_LINK_DOWN;
2437 bond_set_backup_slave(slave);
2439 if (slave->link_failure_count < UINT_MAX)
2440 slave->link_failure_count++;
2442 pr_info("%s: interface %s is now down.\n",
2446 if (slave == oldcurrent)
2451 /* note: if switch is in round-robin mode, all links
2452 * must tx arp to ensure all links rx an arp - otherwise
2453 * links may oscillate or not come up at all; if switch is
2454 * in something like xor mode, there is nothing we can
2455 * do - all replies will be rx'ed on same link causing slaves
2456 * to be unstable during low/no traffic periods
2458 if (IS_UP(slave->dev))
2459 bond_arp_send_all(bond, slave);
2465 /* the bond_select_active_slave must hold RTNL
2466 * and curr_slave_lock for write.
2468 if (!rtnl_trylock())
2471 write_lock_bh(&bond->curr_slave_lock);
2473 bond_select_active_slave(bond);
2475 write_unlock_bh(&bond->curr_slave_lock);
2476 unblock_netpoll_tx();
2481 if (bond->params.arp_interval)
2482 queue_delayed_work(bond->wq, &bond->arp_work,
2483 msecs_to_jiffies(bond->params.arp_interval));
2487 * Called to inspect slaves for active-backup mode ARP monitor link state
2488 * changes. Sets new_link in slaves to specify what action should take
2489 * place for the slave. Returns 0 if no changes are found, >0 if changes
2490 * to link states must be committed.
2492 * Called with rcu_read_lock hold.
2494 static int bond_ab_arp_inspect(struct bonding *bond)
2496 unsigned long trans_start, last_rx;
2497 struct list_head *iter;
2498 struct slave *slave;
2501 bond_for_each_slave_rcu(bond, slave, iter) {
2502 slave->new_link = BOND_LINK_NOCHANGE;
2503 last_rx = slave_last_rx(bond, slave);
2505 if (slave->link != BOND_LINK_UP) {
2506 if (bond_time_in_interval(bond, last_rx, 1)) {
2507 slave->new_link = BOND_LINK_UP;
2514 * Give slaves 2*delta after being enslaved or made
2515 * active. This avoids bouncing, as the last receive
2516 * times need a full ARP monitor cycle to be updated.
2518 if (bond_time_in_interval(bond, slave->jiffies, 2))
2522 * Backup slave is down if:
2523 * - No current_arp_slave AND
2524 * - more than 3*delta since last receive AND
2525 * - the bond has an IP address
2527 * Note: a non-null current_arp_slave indicates
2528 * the curr_active_slave went down and we are
2529 * searching for a new one; under this condition
2530 * we only take the curr_active_slave down - this
2531 * gives each slave a chance to tx/rx traffic
2532 * before being taken out
2534 if (!bond_is_active_slave(slave) &&
2535 !bond->current_arp_slave &&
2536 !bond_time_in_interval(bond, last_rx, 3)) {
2537 slave->new_link = BOND_LINK_DOWN;
2542 * Active slave is down if:
2543 * - more than 2*delta since transmitting OR
2544 * - (more than 2*delta since receive AND
2545 * the bond has an IP address)
2547 trans_start = dev_trans_start(slave->dev);
2548 if (bond_is_active_slave(slave) &&
2549 (!bond_time_in_interval(bond, trans_start, 2) ||
2550 !bond_time_in_interval(bond, last_rx, 2))) {
2551 slave->new_link = BOND_LINK_DOWN;
2560 * Called to commit link state changes noted by inspection step of
2561 * active-backup mode ARP monitor.
2563 * Called with RTNL hold.
2565 static void bond_ab_arp_commit(struct bonding *bond)
2567 unsigned long trans_start;
2568 struct list_head *iter;
2569 struct slave *slave;
2571 bond_for_each_slave(bond, slave, iter) {
2572 switch (slave->new_link) {
2573 case BOND_LINK_NOCHANGE:
2577 trans_start = dev_trans_start(slave->dev);
2578 if (bond->curr_active_slave != slave ||
2579 (!bond->curr_active_slave &&
2580 bond_time_in_interval(bond, trans_start, 1))) {
2581 slave->link = BOND_LINK_UP;
2582 if (bond->current_arp_slave) {
2583 bond_set_slave_inactive_flags(
2584 bond->current_arp_slave);
2585 bond->current_arp_slave = NULL;
2588 pr_info("%s: link status definitely up for interface %s.\n",
2589 bond->dev->name, slave->dev->name);
2591 if (!bond->curr_active_slave ||
2592 (slave == bond->primary_slave))
2599 case BOND_LINK_DOWN:
2600 if (slave->link_failure_count < UINT_MAX)
2601 slave->link_failure_count++;
2603 slave->link = BOND_LINK_DOWN;
2604 bond_set_slave_inactive_flags(slave);
2606 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2607 bond->dev->name, slave->dev->name);
2609 if (slave == bond->curr_active_slave) {
2610 bond->current_arp_slave = NULL;
2617 pr_err("%s: impossible: new_link %d on slave %s\n",
2618 bond->dev->name, slave->new_link,
2626 write_lock_bh(&bond->curr_slave_lock);
2627 bond_select_active_slave(bond);
2628 write_unlock_bh(&bond->curr_slave_lock);
2629 unblock_netpoll_tx();
2632 bond_set_carrier(bond);
2636 * Send ARP probes for active-backup mode ARP monitor.
2638 * Called with rcu_read_lock hold.
2640 static void bond_ab_arp_probe(struct bonding *bond)
2642 struct slave *slave, *before = NULL, *new_slave = NULL,
2643 *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2644 struct list_head *iter;
2647 read_lock(&bond->curr_slave_lock);
2649 if (curr_arp_slave && bond->curr_active_slave)
2650 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2651 curr_arp_slave->dev->name,
2652 bond->curr_active_slave->dev->name);
2654 if (bond->curr_active_slave) {
2655 bond_arp_send_all(bond, bond->curr_active_slave);
2656 read_unlock(&bond->curr_slave_lock);
2660 read_unlock(&bond->curr_slave_lock);
2662 /* if we don't have a curr_active_slave, search for the next available
2663 * backup slave from the current_arp_slave and make it the candidate
2664 * for becoming the curr_active_slave
2667 if (!curr_arp_slave) {
2668 curr_arp_slave = bond_first_slave_rcu(bond);
2669 if (!curr_arp_slave)
2673 bond_set_slave_inactive_flags(curr_arp_slave);
2675 bond_for_each_slave_rcu(bond, slave, iter) {
2676 if (!found && !before && IS_UP(slave->dev))
2679 if (found && !new_slave && IS_UP(slave->dev))
2681 /* if the link state is up at this point, we
2682 * mark it down - this can happen if we have
2683 * simultaneous link failures and
2684 * reselect_active_interface doesn't make this
2685 * one the current slave so it is still marked
2686 * up when it is actually down
2688 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2689 slave->link = BOND_LINK_DOWN;
2690 if (slave->link_failure_count < UINT_MAX)
2691 slave->link_failure_count++;
2693 bond_set_slave_inactive_flags(slave);
2695 pr_info("%s: backup interface %s is now down.\n",
2696 bond->dev->name, slave->dev->name);
2698 if (slave == curr_arp_slave)
2702 if (!new_slave && before)
2708 new_slave->link = BOND_LINK_BACK;
2709 bond_set_slave_active_flags(new_slave);
2710 bond_arp_send_all(bond, new_slave);
2711 new_slave->jiffies = jiffies;
2712 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2715 static void bond_activebackup_arp_mon(struct work_struct *work)
2717 struct bonding *bond = container_of(work, struct bonding,
2719 bool should_notify_peers = false;
2722 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2724 if (!bond_has_slaves(bond))
2729 should_notify_peers = bond_should_notify_peers(bond);
2731 if (bond_ab_arp_inspect(bond)) {
2734 /* Race avoidance with bond_close flush of workqueue */
2735 if (!rtnl_trylock()) {
2737 should_notify_peers = false;
2741 bond_ab_arp_commit(bond);
2747 bond_ab_arp_probe(bond);
2751 if (bond->params.arp_interval)
2752 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2754 if (should_notify_peers) {
2755 if (!rtnl_trylock())
2757 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2762 /*-------------------------- netdev event handling --------------------------*/
2765 * Change device name
2767 static int bond_event_changename(struct bonding *bond)
2769 bond_remove_proc_entry(bond);
2770 bond_create_proc_entry(bond);
2772 bond_debug_reregister(bond);
2777 static int bond_master_netdev_event(unsigned long event,
2778 struct net_device *bond_dev)
2780 struct bonding *event_bond = netdev_priv(bond_dev);
2783 case NETDEV_CHANGENAME:
2784 return bond_event_changename(event_bond);
2785 case NETDEV_UNREGISTER:
2786 bond_remove_proc_entry(event_bond);
2788 case NETDEV_REGISTER:
2789 bond_create_proc_entry(event_bond);
2791 case NETDEV_NOTIFY_PEERS:
2792 if (event_bond->send_peer_notif)
2793 event_bond->send_peer_notif--;
2802 static int bond_slave_netdev_event(unsigned long event,
2803 struct net_device *slave_dev)
2805 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2806 struct bonding *bond;
2807 struct net_device *bond_dev;
2811 /* A netdev event can be generated while enslaving a device
2812 * before netdev_rx_handler_register is called in which case
2813 * slave will be NULL
2817 bond_dev = slave->bond->dev;
2821 case NETDEV_UNREGISTER:
2822 if (bond_dev->type != ARPHRD_ETHER)
2823 bond_release_and_destroy(bond_dev, slave_dev);
2825 bond_release(bond_dev, slave_dev);
2829 old_speed = slave->speed;
2830 old_duplex = slave->duplex;
2832 bond_update_speed_duplex(slave);
2834 if (bond->params.mode == BOND_MODE_8023AD) {
2835 if (old_speed != slave->speed)
2836 bond_3ad_adapter_speed_changed(slave);
2837 if (old_duplex != slave->duplex)
2838 bond_3ad_adapter_duplex_changed(slave);
2843 * ... Or is it this?
2846 case NETDEV_CHANGEMTU:
2848 * TODO: Should slaves be allowed to
2849 * independently alter their MTU? For
2850 * an active-backup bond, slaves need
2851 * not be the same type of device, so
2852 * MTUs may vary. For other modes,
2853 * slaves arguably should have the
2854 * same MTUs. To do this, we'd need to
2855 * take over the slave's change_mtu
2856 * function for the duration of their
2860 case NETDEV_CHANGENAME:
2861 /* we don't care if we don't have primary set */
2862 if (!USES_PRIMARY(bond->params.mode) ||
2863 !bond->params.primary[0])
2866 if (slave == bond->primary_slave) {
2867 /* slave's name changed - he's no longer primary */
2868 bond->primary_slave = NULL;
2869 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2870 /* we have a new primary slave */
2871 bond->primary_slave = slave;
2872 } else { /* we didn't change primary - exit */
2876 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2877 bond->dev->name, bond->primary_slave ? slave_dev->name :
2879 write_lock_bh(&bond->curr_slave_lock);
2880 bond_select_active_slave(bond);
2881 write_unlock_bh(&bond->curr_slave_lock);
2883 case NETDEV_FEAT_CHANGE:
2884 bond_compute_features(bond);
2886 case NETDEV_RESEND_IGMP:
2887 /* Propagate to master device */
2888 call_netdevice_notifiers(event, slave->bond->dev);
2898 * bond_netdev_event: handle netdev notifier chain events.
2900 * This function receives events for the netdev chain. The caller (an
2901 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2902 * locks for us to safely manipulate the slave devices (RTNL lock,
2905 static int bond_netdev_event(struct notifier_block *this,
2906 unsigned long event, void *ptr)
2908 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2910 pr_debug("event_dev: %s, event: %lx\n",
2911 event_dev ? event_dev->name : "None",
2914 if (!(event_dev->priv_flags & IFF_BONDING))
2917 if (event_dev->flags & IFF_MASTER) {
2918 pr_debug("IFF_MASTER\n");
2919 return bond_master_netdev_event(event, event_dev);
2922 if (event_dev->flags & IFF_SLAVE) {
2923 pr_debug("IFF_SLAVE\n");
2924 return bond_slave_netdev_event(event, event_dev);
2930 static struct notifier_block bond_netdev_notifier = {
2931 .notifier_call = bond_netdev_event,
2934 /*---------------------------- Hashing Policies -----------------------------*/
2936 /* L2 hash helper */
2937 static inline u32 bond_eth_hash(struct sk_buff *skb)
2939 struct ethhdr *data = (struct ethhdr *)skb->data;
2941 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2942 return data->h_dest[5] ^ data->h_source[5];
2947 /* Extract the appropriate headers based on bond's xmit policy */
2948 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2949 struct flow_keys *fk)
2951 const struct ipv6hdr *iph6;
2952 const struct iphdr *iph;
2953 int noff, proto = -1;
2955 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2956 return skb_flow_dissect(skb, fk);
2959 noff = skb_network_offset(skb);
2960 if (skb->protocol == htons(ETH_P_IP)) {
2961 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2964 fk->src = iph->saddr;
2965 fk->dst = iph->daddr;
2966 noff += iph->ihl << 2;
2967 if (!ip_is_fragment(iph))
2968 proto = iph->protocol;
2969 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2970 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2972 iph6 = ipv6_hdr(skb);
2973 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2974 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2975 noff += sizeof(*iph6);
2976 proto = iph6->nexthdr;
2980 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2981 fk->ports = skb_flow_get_ports(skb, noff, proto);
2987 * bond_xmit_hash - generate a hash value based on the xmit policy
2988 * @bond: bonding device
2989 * @skb: buffer to use for headers
2990 * @count: modulo value
2992 * This function will extract the necessary headers from the skb buffer and use
2993 * them to generate a hash based on the xmit_policy set in the bonding device
2994 * which will be reduced modulo count before returning.
2996 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
2998 struct flow_keys flow;
3001 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3002 !bond_flow_dissect(bond, skb, &flow))
3003 return bond_eth_hash(skb) % count;
3005 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3006 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3007 hash = bond_eth_hash(skb);
3009 hash = (__force u32)flow.ports;
3010 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3011 hash ^= (hash >> 16);
3012 hash ^= (hash >> 8);
3014 return hash % count;
3017 /*-------------------------- Device entry points ----------------------------*/
3019 static void bond_work_init_all(struct bonding *bond)
3021 INIT_DELAYED_WORK(&bond->mcast_work,
3022 bond_resend_igmp_join_requests_delayed);
3023 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3024 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3025 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3026 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3028 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3029 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3032 static void bond_work_cancel_all(struct bonding *bond)
3034 cancel_delayed_work_sync(&bond->mii_work);
3035 cancel_delayed_work_sync(&bond->arp_work);
3036 cancel_delayed_work_sync(&bond->alb_work);
3037 cancel_delayed_work_sync(&bond->ad_work);
3038 cancel_delayed_work_sync(&bond->mcast_work);
3041 static int bond_open(struct net_device *bond_dev)
3043 struct bonding *bond = netdev_priv(bond_dev);
3044 struct list_head *iter;
3045 struct slave *slave;
3047 /* reset slave->backup and slave->inactive */
3048 read_lock(&bond->lock);
3049 if (bond_has_slaves(bond)) {
3050 read_lock(&bond->curr_slave_lock);
3051 bond_for_each_slave(bond, slave, iter) {
3052 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3053 && (slave != bond->curr_active_slave)) {
3054 bond_set_slave_inactive_flags(slave);
3056 bond_set_slave_active_flags(slave);
3059 read_unlock(&bond->curr_slave_lock);
3061 read_unlock(&bond->lock);
3063 bond_work_init_all(bond);
3065 if (bond_is_lb(bond)) {
3066 /* bond_alb_initialize must be called before the timer
3069 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3071 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3074 if (bond->params.miimon) /* link check interval, in milliseconds. */
3075 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3077 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3078 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3079 if (bond->params.arp_validate)
3080 bond->recv_probe = bond_arp_rcv;
3083 if (bond->params.mode == BOND_MODE_8023AD) {
3084 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3085 /* register to receive LACPDUs */
3086 bond->recv_probe = bond_3ad_lacpdu_recv;
3087 bond_3ad_initiate_agg_selection(bond, 1);
3093 static int bond_close(struct net_device *bond_dev)
3095 struct bonding *bond = netdev_priv(bond_dev);
3097 bond_work_cancel_all(bond);
3098 bond->send_peer_notif = 0;
3099 if (bond_is_lb(bond))
3100 bond_alb_deinitialize(bond);
3101 bond->recv_probe = NULL;
3106 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3107 struct rtnl_link_stats64 *stats)
3109 struct bonding *bond = netdev_priv(bond_dev);
3110 struct rtnl_link_stats64 temp;
3111 struct list_head *iter;
3112 struct slave *slave;
3114 memset(stats, 0, sizeof(*stats));
3116 read_lock_bh(&bond->lock);
3117 bond_for_each_slave(bond, slave, iter) {
3118 const struct rtnl_link_stats64 *sstats =
3119 dev_get_stats(slave->dev, &temp);
3121 stats->rx_packets += sstats->rx_packets;
3122 stats->rx_bytes += sstats->rx_bytes;
3123 stats->rx_errors += sstats->rx_errors;
3124 stats->rx_dropped += sstats->rx_dropped;
3126 stats->tx_packets += sstats->tx_packets;
3127 stats->tx_bytes += sstats->tx_bytes;
3128 stats->tx_errors += sstats->tx_errors;
3129 stats->tx_dropped += sstats->tx_dropped;
3131 stats->multicast += sstats->multicast;
3132 stats->collisions += sstats->collisions;
3134 stats->rx_length_errors += sstats->rx_length_errors;
3135 stats->rx_over_errors += sstats->rx_over_errors;
3136 stats->rx_crc_errors += sstats->rx_crc_errors;
3137 stats->rx_frame_errors += sstats->rx_frame_errors;
3138 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3139 stats->rx_missed_errors += sstats->rx_missed_errors;
3141 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3142 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3143 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3144 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3145 stats->tx_window_errors += sstats->tx_window_errors;
3147 read_unlock_bh(&bond->lock);
3152 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3154 struct bonding *bond = netdev_priv(bond_dev);
3155 struct net_device *slave_dev = NULL;
3156 struct ifbond k_binfo;
3157 struct ifbond __user *u_binfo = NULL;
3158 struct ifslave k_sinfo;
3159 struct ifslave __user *u_sinfo = NULL;
3160 struct mii_ioctl_data *mii = NULL;
3164 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3176 * We do this again just in case we were called by SIOCGMIIREG
3177 * instead of SIOCGMIIPHY.
3184 if (mii->reg_num == 1) {
3186 read_lock(&bond->lock);
3187 read_lock(&bond->curr_slave_lock);
3188 if (netif_carrier_ok(bond->dev))
3189 mii->val_out = BMSR_LSTATUS;
3191 read_unlock(&bond->curr_slave_lock);
3192 read_unlock(&bond->lock);
3196 case BOND_INFO_QUERY_OLD:
3197 case SIOCBONDINFOQUERY:
3198 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3200 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3203 res = bond_info_query(bond_dev, &k_binfo);
3205 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3209 case BOND_SLAVE_INFO_QUERY_OLD:
3210 case SIOCBONDSLAVEINFOQUERY:
3211 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3213 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3216 res = bond_slave_info_query(bond_dev, &k_sinfo);
3218 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3227 net = dev_net(bond_dev);
3229 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3232 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3234 pr_debug("slave_dev=%p:\n", slave_dev);
3239 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3241 case BOND_ENSLAVE_OLD:
3242 case SIOCBONDENSLAVE:
3243 res = bond_enslave(bond_dev, slave_dev);
3245 case BOND_RELEASE_OLD:
3246 case SIOCBONDRELEASE:
3247 res = bond_release(bond_dev, slave_dev);
3249 case BOND_SETHWADDR_OLD:
3250 case SIOCBONDSETHWADDR:
3251 bond_set_dev_addr(bond_dev, slave_dev);
3254 case BOND_CHANGE_ACTIVE_OLD:
3255 case SIOCBONDCHANGEACTIVE:
3256 res = bond_option_active_slave_set(bond, slave_dev);
3265 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3267 struct bonding *bond = netdev_priv(bond_dev);
3269 if (change & IFF_PROMISC)
3270 bond_set_promiscuity(bond,
3271 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3273 if (change & IFF_ALLMULTI)
3274 bond_set_allmulti(bond,
3275 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3278 static void bond_set_rx_mode(struct net_device *bond_dev)
3280 struct bonding *bond = netdev_priv(bond_dev);
3281 struct list_head *iter;
3282 struct slave *slave;
3286 if (USES_PRIMARY(bond->params.mode)) {
3287 slave = rcu_dereference(bond->curr_active_slave);
3289 dev_uc_sync(slave->dev, bond_dev);
3290 dev_mc_sync(slave->dev, bond_dev);
3293 bond_for_each_slave_rcu(bond, slave, iter) {
3294 dev_uc_sync_multiple(slave->dev, bond_dev);
3295 dev_mc_sync_multiple(slave->dev, bond_dev);
3301 static int bond_neigh_init(struct neighbour *n)
3303 struct bonding *bond = netdev_priv(n->dev);
3304 const struct net_device_ops *slave_ops;
3305 struct neigh_parms parms;
3306 struct slave *slave;
3309 slave = bond_first_slave(bond);
3312 slave_ops = slave->dev->netdev_ops;
3313 if (!slave_ops->ndo_neigh_setup)
3316 parms.neigh_setup = NULL;
3317 parms.neigh_cleanup = NULL;
3318 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3323 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3324 * after the last slave has been detached. Assumes that all slaves
3325 * utilize the same neigh_cleanup (true at this writing as only user
3328 n->parms->neigh_cleanup = parms.neigh_cleanup;
3330 if (!parms.neigh_setup)
3333 return parms.neigh_setup(n);
3337 * The bonding ndo_neigh_setup is called at init time beofre any
3338 * slave exists. So we must declare proxy setup function which will
3339 * be used at run time to resolve the actual slave neigh param setup.
3341 * It's also called by master devices (such as vlans) to setup their
3342 * underlying devices. In that case - do nothing, we're already set up from
3345 static int bond_neigh_setup(struct net_device *dev,
3346 struct neigh_parms *parms)
3348 /* modify only our neigh_parms */
3349 if (parms->dev == dev)
3350 parms->neigh_setup = bond_neigh_init;
3356 * Change the MTU of all of a master's slaves to match the master
3358 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3360 struct bonding *bond = netdev_priv(bond_dev);
3361 struct slave *slave, *rollback_slave;
3362 struct list_head *iter;
3365 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3366 (bond_dev ? bond_dev->name : "None"), new_mtu);
3368 /* Can't hold bond->lock with bh disabled here since
3369 * some base drivers panic. On the other hand we can't
3370 * hold bond->lock without bh disabled because we'll
3371 * deadlock. The only solution is to rely on the fact
3372 * that we're under rtnl_lock here, and the slaves
3373 * list won't change. This doesn't solve the problem
3374 * of setting the slave's MTU while it is
3375 * transmitting, but the assumption is that the base
3376 * driver can handle that.
3378 * TODO: figure out a way to safely iterate the slaves
3379 * list, but without holding a lock around the actual
3380 * call to the base driver.
3383 bond_for_each_slave(bond, slave, iter) {
3384 pr_debug("s %p c_m %p\n",
3386 slave->dev->netdev_ops->ndo_change_mtu);
3388 res = dev_set_mtu(slave->dev, new_mtu);
3391 /* If we failed to set the slave's mtu to the new value
3392 * we must abort the operation even in ACTIVE_BACKUP
3393 * mode, because if we allow the backup slaves to have
3394 * different mtu values than the active slave we'll
3395 * need to change their mtu when doing a failover. That
3396 * means changing their mtu from timer context, which
3397 * is probably not a good idea.
3399 pr_debug("err %d %s\n", res, slave->dev->name);
3404 bond_dev->mtu = new_mtu;
3409 /* unwind from head to the slave that failed */
3410 bond_for_each_slave(bond, rollback_slave, iter) {
3413 if (rollback_slave == slave)
3416 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3418 pr_debug("unwind err %d dev %s\n",
3419 tmp_res, rollback_slave->dev->name);
3429 * Note that many devices must be down to change the HW address, and
3430 * downing the master releases all slaves. We can make bonds full of
3431 * bonding devices to test this, however.
3433 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3435 struct bonding *bond = netdev_priv(bond_dev);
3436 struct slave *slave, *rollback_slave;
3437 struct sockaddr *sa = addr, tmp_sa;
3438 struct list_head *iter;
3441 if (bond->params.mode == BOND_MODE_ALB)
3442 return bond_alb_set_mac_address(bond_dev, addr);
3445 pr_debug("bond=%p, name=%s\n",
3446 bond, bond_dev ? bond_dev->name : "None");
3448 /* If fail_over_mac is enabled, do nothing and return success.
3449 * Returning an error causes ifenslave to fail.
3451 if (bond->params.fail_over_mac)
3454 if (!is_valid_ether_addr(sa->sa_data))
3455 return -EADDRNOTAVAIL;
3457 /* Can't hold bond->lock with bh disabled here since
3458 * some base drivers panic. On the other hand we can't
3459 * hold bond->lock without bh disabled because we'll
3460 * deadlock. The only solution is to rely on the fact
3461 * that we're under rtnl_lock here, and the slaves
3462 * list won't change. This doesn't solve the problem
3463 * of setting the slave's hw address while it is
3464 * transmitting, but the assumption is that the base
3465 * driver can handle that.
3467 * TODO: figure out a way to safely iterate the slaves
3468 * list, but without holding a lock around the actual
3469 * call to the base driver.
3472 bond_for_each_slave(bond, slave, iter) {
3473 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3474 pr_debug("slave %p %s\n", slave, slave->dev->name);
3476 if (slave_ops->ndo_set_mac_address == NULL) {
3478 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3482 res = dev_set_mac_address(slave->dev, addr);
3484 /* TODO: consider downing the slave
3486 * User should expect communications
3487 * breakage anyway until ARP finish
3490 pr_debug("err %d %s\n", res, slave->dev->name);
3496 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3500 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3501 tmp_sa.sa_family = bond_dev->type;
3503 /* unwind from head to the slave that failed */
3504 bond_for_each_slave(bond, rollback_slave, iter) {
3507 if (rollback_slave == slave)
3510 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3512 pr_debug("unwind err %d dev %s\n",
3513 tmp_res, rollback_slave->dev->name);
3521 * bond_xmit_slave_id - transmit skb through slave with slave_id
3522 * @bond: bonding device that is transmitting
3523 * @skb: buffer to transmit
3524 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3526 * This function tries to transmit through slave with slave_id but in case
3527 * it fails, it tries to find the first available slave for transmission.
3528 * The skb is consumed in all cases, thus the function is void.
3530 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3532 struct list_head *iter;
3533 struct slave *slave;
3536 /* Here we start from the slave with slave_id */
3537 bond_for_each_slave_rcu(bond, slave, iter) {
3539 if (slave_can_tx(slave)) {
3540 bond_dev_queue_xmit(bond, skb, slave->dev);
3546 /* Here we start from the first slave up to slave_id */
3548 bond_for_each_slave_rcu(bond, slave, iter) {
3551 if (slave_can_tx(slave)) {
3552 bond_dev_queue_xmit(bond, skb, slave->dev);
3556 /* no slave that can tx has been found */
3561 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3562 * @bond: bonding device to use
3564 * Based on the value of the bonding device's packets_per_slave parameter
3565 * this function generates a slave id, which is usually used as the next
3566 * slave to transmit through.
3568 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3571 struct reciprocal_value reciprocal_packets_per_slave;
3572 int packets_per_slave = bond->params.packets_per_slave;
3574 switch (packets_per_slave) {
3576 slave_id = prandom_u32();
3579 slave_id = bond->rr_tx_counter;
3582 reciprocal_packets_per_slave =
3583 bond->params.reciprocal_packets_per_slave;
3584 slave_id = reciprocal_divide(bond->rr_tx_counter,
3585 reciprocal_packets_per_slave);
3588 bond->rr_tx_counter++;
3593 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3595 struct bonding *bond = netdev_priv(bond_dev);
3596 struct iphdr *iph = ip_hdr(skb);
3597 struct slave *slave;
3600 /* Start with the curr_active_slave that joined the bond as the
3601 * default for sending IGMP traffic. For failover purposes one
3602 * needs to maintain some consistency for the interface that will
3603 * send the join/membership reports. The curr_active_slave found
3604 * will send all of this type of traffic.
3606 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3607 slave = rcu_dereference(bond->curr_active_slave);
3608 if (slave && slave_can_tx(slave))
3609 bond_dev_queue_xmit(bond, skb, slave->dev);
3611 bond_xmit_slave_id(bond, skb, 0);
3613 slave_id = bond_rr_gen_slave_id(bond);
3614 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3617 return NETDEV_TX_OK;
3621 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3622 * the bond has a usable interface.
3624 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3626 struct bonding *bond = netdev_priv(bond_dev);
3627 struct slave *slave;
3629 slave = rcu_dereference(bond->curr_active_slave);
3631 bond_dev_queue_xmit(bond, skb, slave->dev);
3635 return NETDEV_TX_OK;
3638 /* In bond_xmit_xor() , we determine the output device by using a pre-
3639 * determined xmit_hash_policy(), If the selected device is not enabled,
3640 * find the next active slave.
3642 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3644 struct bonding *bond = netdev_priv(bond_dev);
3646 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3648 return NETDEV_TX_OK;
3651 /* in broadcast mode, we send everything to all usable interfaces. */
3652 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3654 struct bonding *bond = netdev_priv(bond_dev);
3655 struct slave *slave = NULL;
3656 struct list_head *iter;
3658 bond_for_each_slave_rcu(bond, slave, iter) {
3659 if (bond_is_last_slave(bond, slave))
3661 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3662 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3665 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3669 /* bond_dev_queue_xmit always returns 0 */
3670 bond_dev_queue_xmit(bond, skb2, slave->dev);
3673 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3674 bond_dev_queue_xmit(bond, skb, slave->dev);
3678 return NETDEV_TX_OK;
3681 /*------------------------- Device initialization ---------------------------*/
3684 * Lookup the slave that corresponds to a qid
3686 static inline int bond_slave_override(struct bonding *bond,
3687 struct sk_buff *skb)
3689 struct slave *slave = NULL;
3690 struct list_head *iter;
3692 if (!skb->queue_mapping)
3695 /* Find out if any slaves have the same mapping as this skb. */
3696 bond_for_each_slave_rcu(bond, slave, iter) {
3697 if (slave->queue_id == skb->queue_mapping) {
3698 if (slave_can_tx(slave)) {
3699 bond_dev_queue_xmit(bond, skb, slave->dev);
3702 /* If the slave isn't UP, use default transmit policy. */
3711 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3715 * This helper function exists to help dev_pick_tx get the correct
3716 * destination queue. Using a helper function skips a call to
3717 * skb_tx_hash and will put the skbs in the queue we expect on their
3718 * way down to the bonding driver.
3720 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3723 * Save the original txq to restore before passing to the driver
3725 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3727 if (unlikely(txq >= dev->real_num_tx_queues)) {
3729 txq -= dev->real_num_tx_queues;
3730 } while (txq >= dev->real_num_tx_queues);
3735 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3737 struct bonding *bond = netdev_priv(dev);
3739 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3740 if (!bond_slave_override(bond, skb))
3741 return NETDEV_TX_OK;
3744 switch (bond->params.mode) {
3745 case BOND_MODE_ROUNDROBIN:
3746 return bond_xmit_roundrobin(skb, dev);
3747 case BOND_MODE_ACTIVEBACKUP:
3748 return bond_xmit_activebackup(skb, dev);
3750 return bond_xmit_xor(skb, dev);
3751 case BOND_MODE_BROADCAST:
3752 return bond_xmit_broadcast(skb, dev);
3753 case BOND_MODE_8023AD:
3754 return bond_3ad_xmit_xor(skb, dev);
3757 return bond_alb_xmit(skb, dev);
3759 /* Should never happen, mode already checked */
3760 pr_err("%s: Error: Unknown bonding mode %d\n",
3761 dev->name, bond->params.mode);
3764 return NETDEV_TX_OK;
3768 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3770 struct bonding *bond = netdev_priv(dev);
3771 netdev_tx_t ret = NETDEV_TX_OK;
3774 * If we risk deadlock from transmitting this in the
3775 * netpoll path, tell netpoll to queue the frame for later tx
3777 if (is_netpoll_tx_blocked(dev))
3778 return NETDEV_TX_BUSY;
3781 if (bond_has_slaves(bond))
3782 ret = __bond_start_xmit(skb, dev);
3790 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3791 struct ethtool_cmd *ecmd)
3793 struct bonding *bond = netdev_priv(bond_dev);
3794 unsigned long speed = 0;
3795 struct list_head *iter;
3796 struct slave *slave;
3798 ecmd->duplex = DUPLEX_UNKNOWN;
3799 ecmd->port = PORT_OTHER;
3801 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3802 * do not need to check mode. Though link speed might not represent
3803 * the true receive or transmit bandwidth (not all modes are symmetric)
3804 * this is an accurate maximum.
3806 read_lock(&bond->lock);
3807 bond_for_each_slave(bond, slave, iter) {
3808 if (SLAVE_IS_OK(slave)) {
3809 if (slave->speed != SPEED_UNKNOWN)
3810 speed += slave->speed;
3811 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3812 slave->duplex != DUPLEX_UNKNOWN)
3813 ecmd->duplex = slave->duplex;
3816 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3817 read_unlock(&bond->lock);
3822 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3823 struct ethtool_drvinfo *drvinfo)
3825 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3826 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3827 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3831 static const struct ethtool_ops bond_ethtool_ops = {
3832 .get_drvinfo = bond_ethtool_get_drvinfo,
3833 .get_settings = bond_ethtool_get_settings,
3834 .get_link = ethtool_op_get_link,
3837 static const struct net_device_ops bond_netdev_ops = {
3838 .ndo_init = bond_init,
3839 .ndo_uninit = bond_uninit,
3840 .ndo_open = bond_open,
3841 .ndo_stop = bond_close,
3842 .ndo_start_xmit = bond_start_xmit,
3843 .ndo_select_queue = bond_select_queue,
3844 .ndo_get_stats64 = bond_get_stats,
3845 .ndo_do_ioctl = bond_do_ioctl,
3846 .ndo_change_rx_flags = bond_change_rx_flags,
3847 .ndo_set_rx_mode = bond_set_rx_mode,
3848 .ndo_change_mtu = bond_change_mtu,
3849 .ndo_set_mac_address = bond_set_mac_address,
3850 .ndo_neigh_setup = bond_neigh_setup,
3851 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3852 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3853 #ifdef CONFIG_NET_POLL_CONTROLLER
3854 .ndo_netpoll_setup = bond_netpoll_setup,
3855 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3856 .ndo_poll_controller = bond_poll_controller,
3858 .ndo_add_slave = bond_enslave,
3859 .ndo_del_slave = bond_release,
3860 .ndo_get_slave = bond_get_slave,
3861 .ndo_fix_features = bond_fix_features,
3864 static const struct device_type bond_type = {
3868 static void bond_destructor(struct net_device *bond_dev)
3870 struct bonding *bond = netdev_priv(bond_dev);
3872 destroy_workqueue(bond->wq);
3873 free_netdev(bond_dev);
3876 void bond_setup(struct net_device *bond_dev)
3878 struct bonding *bond = netdev_priv(bond_dev);
3880 /* initialize rwlocks */
3881 rwlock_init(&bond->lock);
3882 rwlock_init(&bond->curr_slave_lock);
3883 bond->params = bonding_defaults;
3885 /* Initialize pointers */
3886 bond->dev = bond_dev;
3888 /* Initialize the device entry points */
3889 ether_setup(bond_dev);
3890 bond_dev->netdev_ops = &bond_netdev_ops;
3891 bond_dev->ethtool_ops = &bond_ethtool_ops;
3893 bond_dev->destructor = bond_destructor;
3895 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3897 /* Initialize the device options */
3898 bond_dev->tx_queue_len = 0;
3899 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3900 bond_dev->priv_flags |= IFF_BONDING;
3901 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3903 /* At first, we block adding VLANs. That's the only way to
3904 * prevent problems that occur when adding VLANs over an
3905 * empty bond. The block will be removed once non-challenged
3906 * slaves are enslaved.
3908 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3910 /* don't acquire bond device's netif_tx_lock when
3912 bond_dev->features |= NETIF_F_LLTX;
3914 /* By default, we declare the bond to be fully
3915 * VLAN hardware accelerated capable. Special
3916 * care is taken in the various xmit functions
3917 * when there are slaves that are not hw accel
3921 bond_dev->hw_features = BOND_VLAN_FEATURES |
3922 NETIF_F_HW_VLAN_CTAG_TX |
3923 NETIF_F_HW_VLAN_CTAG_RX |
3924 NETIF_F_HW_VLAN_CTAG_FILTER;
3926 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3927 bond_dev->features |= bond_dev->hw_features;
3931 * Destroy a bonding device.
3932 * Must be under rtnl_lock when this function is called.
3934 static void bond_uninit(struct net_device *bond_dev)
3936 struct bonding *bond = netdev_priv(bond_dev);
3937 struct list_head *iter;
3938 struct slave *slave;
3940 bond_netpoll_cleanup(bond_dev);
3942 /* Release the bonded slaves */
3943 bond_for_each_slave(bond, slave, iter)
3944 __bond_release_one(bond_dev, slave->dev, true);
3945 pr_info("%s: released all slaves\n", bond_dev->name);
3947 list_del(&bond->bond_list);
3949 bond_debug_unregister(bond);
3952 /*------------------------- Module initialization ---------------------------*/
3954 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3958 for (i = 0; tbl[i].modename; i++)
3959 if (mode == tbl[i].mode)
3965 static int bond_parm_tbl_lookup_name(const char *modename,
3966 const struct bond_parm_tbl *tbl)
3970 for (i = 0; tbl[i].modename; i++)
3971 if (strcmp(modename, tbl[i].modename) == 0)
3978 * Convert string input module parms. Accept either the
3979 * number of the mode or its string name. A bit complicated because
3980 * some mode names are substrings of other names, and calls from sysfs
3981 * may have whitespace in the name (trailing newlines, for example).
3983 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3986 char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
3988 for (p = (char *)buf; *p; p++)
3989 if (!(isdigit(*p) || isspace(*p)))
3992 if (*p && sscanf(buf, "%20s", modestr) != 0)
3993 return bond_parm_tbl_lookup_name(modestr, tbl);
3994 else if (sscanf(buf, "%d", &modeint) != 0)
3995 return bond_parm_tbl_lookup(modeint, tbl);
4000 static int bond_check_params(struct bond_params *params)
4002 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4003 struct bond_opt_value newval, *valptr;
4004 int arp_all_targets_value;
4007 * Convert string parameters.
4010 bond_opt_initstr(&newval, mode);
4011 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4013 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4016 bond_mode = valptr->value;
4019 if (xmit_hash_policy) {
4020 if ((bond_mode != BOND_MODE_XOR) &&
4021 (bond_mode != BOND_MODE_8023AD)) {
4022 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4023 bond_mode_name(bond_mode));
4025 bond_opt_initstr(&newval, xmit_hash_policy);
4026 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4029 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4033 xmit_hashtype = valptr->value;
4038 if (bond_mode != BOND_MODE_8023AD) {
4039 pr_info("lacp_rate param is irrelevant in mode %s\n",
4040 bond_mode_name(bond_mode));
4042 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4043 if (lacp_fast == -1) {
4044 pr_err("Error: Invalid lacp rate \"%s\"\n",
4045 lacp_rate == NULL ? "NULL" : lacp_rate);
4052 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4053 if (params->ad_select == -1) {
4054 pr_err("Error: Invalid ad_select \"%s\"\n",
4055 ad_select == NULL ? "NULL" : ad_select);
4059 if (bond_mode != BOND_MODE_8023AD) {
4060 pr_warning("ad_select param only affects 802.3ad mode\n");
4063 params->ad_select = BOND_AD_STABLE;
4066 if (max_bonds < 0) {
4067 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4068 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4069 max_bonds = BOND_DEFAULT_MAX_BONDS;
4073 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4074 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4075 miimon = BOND_LINK_MON_INTERV;
4079 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4084 if (downdelay < 0) {
4085 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4086 downdelay, INT_MAX);
4090 if ((use_carrier != 0) && (use_carrier != 1)) {
4091 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4096 if (num_peer_notif < 0 || num_peer_notif > 255) {
4097 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4102 /* reset values for 802.3ad/TLB/ALB */
4103 if (BOND_NO_USES_ARP(bond_mode)) {
4105 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");
4106 pr_warning("Forcing miimon to 100msec\n");
4107 miimon = BOND_DEFAULT_MIIMON;
4111 if (tx_queues < 1 || tx_queues > 255) {
4112 pr_warning("Warning: tx_queues (%d) should be between "
4113 "1 and 255, resetting to %d\n",
4114 tx_queues, BOND_DEFAULT_TX_QUEUES);
4115 tx_queues = BOND_DEFAULT_TX_QUEUES;
4118 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4119 pr_warning("Warning: all_slaves_active module parameter (%d), "
4120 "not of valid value (0/1), so it was set to "
4121 "0\n", all_slaves_active);
4122 all_slaves_active = 0;
4125 if (resend_igmp < 0 || resend_igmp > 255) {
4126 pr_warning("Warning: resend_igmp (%d) should be between "
4127 "0 and 255, resetting to %d\n",
4128 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4129 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4132 bond_opt_initval(&newval, packets_per_slave);
4133 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4134 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4135 packets_per_slave, USHRT_MAX);
4136 packets_per_slave = 1;
4139 if (bond_mode == BOND_MODE_ALB) {
4140 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",
4145 if (updelay || downdelay) {
4146 /* just warn the user the up/down delay will have
4147 * no effect since miimon is zero...
4149 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",
4150 updelay, downdelay);
4153 /* don't allow arp monitoring */
4155 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4156 miimon, arp_interval);
4160 if ((updelay % miimon) != 0) {
4161 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4163 (updelay / miimon) * miimon);
4168 if ((downdelay % miimon) != 0) {
4169 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4171 (downdelay / miimon) * miimon);
4174 downdelay /= miimon;
4177 if (arp_interval < 0) {
4178 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4179 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4180 arp_interval = BOND_LINK_ARP_INTERV;
4183 for (arp_ip_count = 0, i = 0;
4184 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4185 /* not complete check, but should be good enough to
4188 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4189 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4190 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4194 if (bond_get_targets_ip(arp_target, ip) == -1)
4195 arp_target[arp_ip_count++] = ip;
4197 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4202 if (arp_interval && !arp_ip_count) {
4203 /* don't allow arping if no arp_ip_target given... */
4204 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4210 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4211 pr_err("arp_validate only supported in active-backup mode\n");
4214 if (!arp_interval) {
4215 pr_err("arp_validate requires arp_interval\n");
4219 bond_opt_initstr(&newval, arp_validate);
4220 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4223 pr_err("Error: invalid arp_validate \"%s\"\n",
4227 arp_validate_value = valptr->value;
4229 arp_validate_value = 0;
4232 arp_all_targets_value = 0;
4233 if (arp_all_targets) {
4234 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4235 arp_all_targets_tbl);
4237 if (arp_all_targets_value == -1) {
4238 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4240 arp_all_targets_value = 0;
4245 pr_info("MII link monitoring set to %d ms\n", miimon);
4246 } else if (arp_interval) {
4247 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4248 arp_validate_value);
4249 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4250 arp_interval, valptr->string, arp_ip_count);
4252 for (i = 0; i < arp_ip_count; i++)
4253 pr_info(" %s", arp_ip_target[i]);
4257 } else if (max_bonds) {
4258 /* miimon and arp_interval not set, we need one so things
4259 * work as expected, see bonding.txt for details
4261 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");
4264 if (primary && !USES_PRIMARY(bond_mode)) {
4265 /* currently, using a primary only makes sense
4266 * in active backup, TLB or ALB modes
4268 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4269 primary, bond_mode_name(bond_mode));
4273 if (primary && primary_reselect) {
4274 primary_reselect_value = bond_parse_parm(primary_reselect,
4276 if (primary_reselect_value == -1) {
4277 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4279 NULL ? "NULL" : primary_reselect);
4283 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4286 if (fail_over_mac) {
4287 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4289 if (fail_over_mac_value == -1) {
4290 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4291 arp_validate == NULL ? "NULL" : arp_validate);
4295 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4296 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4298 fail_over_mac_value = BOND_FOM_NONE;
4301 if (lp_interval == 0) {
4302 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4303 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4304 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4307 /* fill params struct with the proper values */
4308 params->mode = bond_mode;
4309 params->xmit_policy = xmit_hashtype;
4310 params->miimon = miimon;
4311 params->num_peer_notif = num_peer_notif;
4312 params->arp_interval = arp_interval;
4313 params->arp_validate = arp_validate_value;
4314 params->arp_all_targets = arp_all_targets_value;
4315 params->updelay = updelay;
4316 params->downdelay = downdelay;
4317 params->use_carrier = use_carrier;
4318 params->lacp_fast = lacp_fast;
4319 params->primary[0] = 0;
4320 params->primary_reselect = primary_reselect_value;
4321 params->fail_over_mac = fail_over_mac_value;
4322 params->tx_queues = tx_queues;
4323 params->all_slaves_active = all_slaves_active;
4324 params->resend_igmp = resend_igmp;
4325 params->min_links = min_links;
4326 params->lp_interval = lp_interval;
4327 params->packets_per_slave = packets_per_slave;
4328 if (packets_per_slave > 0) {
4329 params->reciprocal_packets_per_slave =
4330 reciprocal_value(packets_per_slave);
4332 /* reciprocal_packets_per_slave is unused if
4333 * packets_per_slave is 0 or 1, just initialize it
4335 params->reciprocal_packets_per_slave =
4336 (struct reciprocal_value) { 0 };
4340 strncpy(params->primary, primary, IFNAMSIZ);
4341 params->primary[IFNAMSIZ - 1] = 0;
4344 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4349 static struct lock_class_key bonding_netdev_xmit_lock_key;
4350 static struct lock_class_key bonding_netdev_addr_lock_key;
4351 static struct lock_class_key bonding_tx_busylock_key;
4353 static void bond_set_lockdep_class_one(struct net_device *dev,
4354 struct netdev_queue *txq,
4357 lockdep_set_class(&txq->_xmit_lock,
4358 &bonding_netdev_xmit_lock_key);
4361 static void bond_set_lockdep_class(struct net_device *dev)
4363 lockdep_set_class(&dev->addr_list_lock,
4364 &bonding_netdev_addr_lock_key);
4365 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4366 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4370 * Called from registration process
4372 static int bond_init(struct net_device *bond_dev)
4374 struct bonding *bond = netdev_priv(bond_dev);
4375 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4376 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4378 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4381 * Initialize locks that may be required during
4382 * en/deslave operations. All of the bond_open work
4383 * (of which this is part) should really be moved to
4384 * a phase prior to dev_open
4386 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4387 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4389 bond->wq = create_singlethread_workqueue(bond_dev->name);
4393 bond_set_lockdep_class(bond_dev);
4395 list_add_tail(&bond->bond_list, &bn->dev_list);
4397 bond_prepare_sysfs_group(bond);
4399 bond_debug_register(bond);
4401 /* Ensure valid dev_addr */
4402 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4403 bond_dev->addr_assign_type == NET_ADDR_PERM)
4404 eth_hw_addr_random(bond_dev);
4409 unsigned int bond_get_num_tx_queues(void)
4414 /* Create a new bond based on the specified name and bonding parameters.
4415 * If name is NULL, obtain a suitable "bond%d" name for us.
4416 * Caller must NOT hold rtnl_lock; we need to release it here before we
4417 * set up our sysfs entries.
4419 int bond_create(struct net *net, const char *name)
4421 struct net_device *bond_dev;
4426 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4427 name ? name : "bond%d",
4428 bond_setup, tx_queues);
4430 pr_err("%s: eek! can't alloc netdev!\n", name);
4435 dev_net_set(bond_dev, net);
4436 bond_dev->rtnl_link_ops = &bond_link_ops;
4438 res = register_netdevice(bond_dev);
4440 netif_carrier_off(bond_dev);
4444 bond_destructor(bond_dev);
4448 static int __net_init bond_net_init(struct net *net)
4450 struct bond_net *bn = net_generic(net, bond_net_id);
4453 INIT_LIST_HEAD(&bn->dev_list);
4455 bond_create_proc_dir(bn);
4456 bond_create_sysfs(bn);
4461 static void __net_exit bond_net_exit(struct net *net)
4463 struct bond_net *bn = net_generic(net, bond_net_id);
4464 struct bonding *bond, *tmp_bond;
4467 bond_destroy_sysfs(bn);
4468 bond_destroy_proc_dir(bn);
4470 /* Kill off any bonds created after unregistering bond rtnl ops */
4472 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4473 unregister_netdevice_queue(bond->dev, &list);
4474 unregister_netdevice_many(&list);
4478 static struct pernet_operations bond_net_ops = {
4479 .init = bond_net_init,
4480 .exit = bond_net_exit,
4482 .size = sizeof(struct bond_net),
4485 static int __init bonding_init(void)
4490 pr_info("%s", bond_version);
4492 res = bond_check_params(&bonding_defaults);
4496 res = register_pernet_subsys(&bond_net_ops);
4500 res = bond_netlink_init();
4504 bond_create_debugfs();
4506 for (i = 0; i < max_bonds; i++) {
4507 res = bond_create(&init_net, NULL);
4512 register_netdevice_notifier(&bond_netdev_notifier);
4516 bond_netlink_fini();
4518 unregister_pernet_subsys(&bond_net_ops);
4523 static void __exit bonding_exit(void)
4525 unregister_netdevice_notifier(&bond_netdev_notifier);
4527 bond_destroy_debugfs();
4529 bond_netlink_fini();
4530 unregister_pernet_subsys(&bond_net_ops);
4532 #ifdef CONFIG_NET_POLL_CONTROLLER
4534 * Make sure we don't have an imbalance on our netpoll blocking
4536 WARN_ON(atomic_read(&netpoll_block_tx));
4540 module_init(bonding_init);
4541 module_exit(bonding_exit);
4542 MODULE_LICENSE("GPL");
4543 MODULE_VERSION(DRV_VERSION);
4544 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4545 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");