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 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 #include "bonding_priv.h"
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 static int packets_per_slave = 1;
114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
116 module_param(max_bonds, int, 0);
117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118 module_param(tx_queues, int, 0);
119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126 module_param(miimon, int, 0);
127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128 module_param(updelay, int, 0);
129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130 module_param(downdelay, int, 0);
131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
133 module_param(use_carrier, int, 0);
134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136 module_param(mode, charp, 0);
137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141 module_param(primary, charp, 0);
142 MODULE_PARM_DESC(primary, "Primary network device to use");
143 module_param(primary_reselect, charp, 0);
144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
149 "2 for only on active slave "
151 module_param(lacp_rate, charp, 0);
152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154 module_param(ad_select, charp, 0);
155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
158 module_param(min_links, int, 0);
159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166 module_param(arp_interval, int, 0);
167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168 module_param_array(arp_ip_target, charp, NULL, 0);
169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170 module_param(arp_validate, charp, 0);
171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174 module_param(arp_all_targets, charp, 0);
175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac, charp, 0);
177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180 module_param(all_slaves_active, int, 0);
181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184 module_param(resend_igmp, int, 0);
185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
187 module_param(packets_per_slave, int, 0);
188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191 module_param(lp_interval, uint, 0);
192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
196 /*----------------------------- Global variables ----------------------------*/
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
202 int bond_net_id __read_mostly;
204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
205 static int arp_ip_count;
206 static int bond_mode = BOND_MODE_ROUNDROBIN;
207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
208 static int lacp_fast;
210 /*-------------------------- Forward declarations ---------------------------*/
212 static int bond_init(struct net_device *bond_dev);
213 static void bond_uninit(struct net_device *bond_dev);
214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
215 struct rtnl_link_stats64 *stats);
216 static void bond_slave_arr_handler(struct work_struct *work);
217 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
220 /*---------------------------- General routines -----------------------------*/
222 const char *bond_mode_name(int mode)
224 static const char *names[] = {
225 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
226 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
227 [BOND_MODE_XOR] = "load balancing (xor)",
228 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
229 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
230 [BOND_MODE_TLB] = "transmit load balancing",
231 [BOND_MODE_ALB] = "adaptive load balancing",
234 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
240 /*---------------------------------- VLAN -----------------------------------*/
243 * bond_dev_queue_xmit - Prepare skb for xmit.
245 * @bond: bond device that got this skb for tx.
246 * @skb: hw accel VLAN tagged skb to transmit
247 * @slave_dev: slave that is supposed to xmit this skbuff
249 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
250 struct net_device *slave_dev)
252 skb->dev = slave_dev;
254 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
255 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
256 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
258 if (unlikely(netpoll_tx_running(bond->dev)))
259 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
264 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
265 * We don't protect the slave list iteration with a lock because:
266 * a. This operation is performed in IOCTL context,
267 * b. The operation is protected by the RTNL semaphore in the 8021q code,
268 * c. Holding a lock with BH disabled while directly calling a base driver
269 * entry point is generally a BAD idea.
271 * The design of synchronization/protection for this operation in the 8021q
272 * module is good for one or more VLAN devices over a single physical device
273 * and cannot be extended for a teaming solution like bonding, so there is a
274 * potential race condition here where a net device from the vlan group might
275 * be referenced (either by a base driver or the 8021q code) while it is being
276 * removed from the system. However, it turns out we're not making matters
277 * worse, and if it works for regular VLAN usage it will work here too.
281 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
282 * @bond_dev: bonding net device that got called
283 * @vid: vlan id being added
285 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
286 __be16 proto, u16 vid)
288 struct bonding *bond = netdev_priv(bond_dev);
289 struct slave *slave, *rollback_slave;
290 struct list_head *iter;
293 bond_for_each_slave(bond, slave, iter) {
294 res = vlan_vid_add(slave->dev, proto, vid);
302 /* unwind to the slave that failed */
303 bond_for_each_slave(bond, rollback_slave, iter) {
304 if (rollback_slave == slave)
307 vlan_vid_del(rollback_slave->dev, proto, vid);
314 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
315 * @bond_dev: bonding net device that got called
316 * @vid: vlan id being removed
318 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
319 __be16 proto, u16 vid)
321 struct bonding *bond = netdev_priv(bond_dev);
322 struct list_head *iter;
325 bond_for_each_slave(bond, slave, iter)
326 vlan_vid_del(slave->dev, proto, vid);
328 if (bond_is_lb(bond))
329 bond_alb_clear_vlan(bond, vid);
334 /*------------------------------- Link status -------------------------------*/
336 /* Set the carrier state for the master according to the state of its
337 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
338 * do special 802.3ad magic.
340 * Returns zero if carrier state does not change, nonzero if it does.
342 int bond_set_carrier(struct bonding *bond)
344 struct list_head *iter;
347 if (!bond_has_slaves(bond))
350 if (BOND_MODE(bond) == BOND_MODE_8023AD)
351 return bond_3ad_set_carrier(bond);
353 bond_for_each_slave(bond, slave, iter) {
354 if (slave->link == BOND_LINK_UP) {
355 if (!netif_carrier_ok(bond->dev)) {
356 netif_carrier_on(bond->dev);
364 if (netif_carrier_ok(bond->dev)) {
365 netif_carrier_off(bond->dev);
371 /* Get link speed and duplex from the slave's base driver
372 * using ethtool. If for some reason the call fails or the
373 * values are invalid, set speed and duplex to -1,
376 static void bond_update_speed_duplex(struct slave *slave)
378 struct net_device *slave_dev = slave->dev;
379 struct ethtool_cmd ecmd;
383 slave->speed = SPEED_UNKNOWN;
384 slave->duplex = DUPLEX_UNKNOWN;
386 res = __ethtool_get_settings(slave_dev, &ecmd);
390 slave_speed = ethtool_cmd_speed(&ecmd);
391 if (slave_speed == 0 || slave_speed == ((__u32) -1))
394 switch (ecmd.duplex) {
402 slave->speed = slave_speed;
403 slave->duplex = ecmd.duplex;
408 const char *bond_slave_link_status(s8 link)
424 /* if <dev> supports MII link status reporting, check its link status.
426 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
427 * depending upon the setting of the use_carrier parameter.
429 * Return either BMSR_LSTATUS, meaning that the link is up (or we
430 * can't tell and just pretend it is), or 0, meaning that the link is
433 * If reporting is non-zero, instead of faking link up, return -1 if
434 * both ETHTOOL and MII ioctls fail (meaning the device does not
435 * support them). If use_carrier is set, return whatever it says.
436 * It'd be nice if there was a good way to tell if a driver supports
437 * netif_carrier, but there really isn't.
439 static int bond_check_dev_link(struct bonding *bond,
440 struct net_device *slave_dev, int reporting)
442 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
443 int (*ioctl)(struct net_device *, struct ifreq *, int);
445 struct mii_ioctl_data *mii;
447 if (!reporting && !netif_running(slave_dev))
450 if (bond->params.use_carrier)
451 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
453 /* Try to get link status using Ethtool first. */
454 if (slave_dev->ethtool_ops->get_link)
455 return slave_dev->ethtool_ops->get_link(slave_dev) ?
458 /* Ethtool can't be used, fallback to MII ioctls. */
459 ioctl = slave_ops->ndo_do_ioctl;
461 /* TODO: set pointer to correct ioctl on a per team member
462 * bases to make this more efficient. that is, once
463 * we determine the correct ioctl, we will always
464 * call it and not the others for that team
468 /* We cannot assume that SIOCGMIIPHY will also read a
469 * register; not all network drivers (e.g., e100)
473 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
474 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
476 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
477 mii->reg_num = MII_BMSR;
478 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
479 return mii->val_out & BMSR_LSTATUS;
483 /* If reporting, report that either there's no dev->do_ioctl,
484 * or both SIOCGMIIREG and get_link failed (meaning that we
485 * cannot report link status). If not reporting, pretend
488 return reporting ? -1 : BMSR_LSTATUS;
491 /*----------------------------- Multicast list ------------------------------*/
493 /* Push the promiscuity flag down to appropriate slaves */
494 static int bond_set_promiscuity(struct bonding *bond, int inc)
496 struct list_head *iter;
499 if (bond_uses_primary(bond)) {
500 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
503 err = dev_set_promiscuity(curr_active->dev, inc);
507 bond_for_each_slave(bond, slave, iter) {
508 err = dev_set_promiscuity(slave->dev, inc);
516 /* Push the allmulti flag down to all slaves */
517 static int bond_set_allmulti(struct bonding *bond, int inc)
519 struct list_head *iter;
522 if (bond_uses_primary(bond)) {
523 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
526 err = dev_set_allmulti(curr_active->dev, inc);
530 bond_for_each_slave(bond, slave, iter) {
531 err = dev_set_allmulti(slave->dev, inc);
539 /* Retrieve the list of registered multicast addresses for the bonding
540 * device and retransmit an IGMP JOIN request to the current active
543 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
545 struct bonding *bond = container_of(work, struct bonding,
548 if (!rtnl_trylock()) {
549 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
552 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
554 if (bond->igmp_retrans > 1) {
555 bond->igmp_retrans--;
556 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
561 /* Flush bond's hardware addresses from slave */
562 static void bond_hw_addr_flush(struct net_device *bond_dev,
563 struct net_device *slave_dev)
565 struct bonding *bond = netdev_priv(bond_dev);
567 dev_uc_unsync(slave_dev, bond_dev);
568 dev_mc_unsync(slave_dev, bond_dev);
570 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
571 /* del lacpdu mc addr from mc list */
572 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
574 dev_mc_del(slave_dev, lacpdu_multicast);
578 /*--------------------------- Active slave change ---------------------------*/
580 /* Update the hardware address list and promisc/allmulti for the new and
581 * old active slaves (if any). Modes that are not using primary keep all
582 * slaves up date at all times; only the modes that use primary need to call
583 * this function to swap these settings during a failover.
585 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
586 struct slave *old_active)
589 if (bond->dev->flags & IFF_PROMISC)
590 dev_set_promiscuity(old_active->dev, -1);
592 if (bond->dev->flags & IFF_ALLMULTI)
593 dev_set_allmulti(old_active->dev, -1);
595 bond_hw_addr_flush(bond->dev, old_active->dev);
599 /* FIXME: Signal errors upstream. */
600 if (bond->dev->flags & IFF_PROMISC)
601 dev_set_promiscuity(new_active->dev, 1);
603 if (bond->dev->flags & IFF_ALLMULTI)
604 dev_set_allmulti(new_active->dev, 1);
606 netif_addr_lock_bh(bond->dev);
607 dev_uc_sync(new_active->dev, bond->dev);
608 dev_mc_sync(new_active->dev, bond->dev);
609 netif_addr_unlock_bh(bond->dev);
614 * bond_set_dev_addr - clone slave's address to bond
615 * @bond_dev: bond net device
616 * @slave_dev: slave net device
618 * Should be called with RTNL held.
620 static void bond_set_dev_addr(struct net_device *bond_dev,
621 struct net_device *slave_dev)
623 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
624 bond_dev, slave_dev, slave_dev->addr_len);
625 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
626 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
627 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
630 static struct slave *bond_get_old_active(struct bonding *bond,
631 struct slave *new_active)
634 struct list_head *iter;
636 bond_for_each_slave(bond, slave, iter) {
637 if (slave == new_active)
640 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
647 /* bond_do_fail_over_mac
649 * Perform special MAC address swapping for fail_over_mac settings
653 static void bond_do_fail_over_mac(struct bonding *bond,
654 struct slave *new_active,
655 struct slave *old_active)
657 u8 tmp_mac[ETH_ALEN];
658 struct sockaddr saddr;
661 switch (bond->params.fail_over_mac) {
662 case BOND_FOM_ACTIVE:
664 bond_set_dev_addr(bond->dev, new_active->dev);
666 case BOND_FOM_FOLLOW:
667 /* if new_active && old_active, swap them
668 * if just old_active, do nothing (going to no active slave)
669 * if just new_active, set new_active to bond's MAC
675 old_active = bond_get_old_active(bond, new_active);
678 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
679 ether_addr_copy(saddr.sa_data,
680 old_active->dev->dev_addr);
681 saddr.sa_family = new_active->dev->type;
683 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
684 saddr.sa_family = bond->dev->type;
687 rv = dev_set_mac_address(new_active->dev, &saddr);
689 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
690 -rv, new_active->dev->name);
697 ether_addr_copy(saddr.sa_data, tmp_mac);
698 saddr.sa_family = old_active->dev->type;
700 rv = dev_set_mac_address(old_active->dev, &saddr);
702 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
703 -rv, new_active->dev->name);
707 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
708 bond->params.fail_over_mac);
714 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
716 struct slave *prim = rtnl_dereference(bond->primary_slave);
717 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
719 if (!prim || prim->link != BOND_LINK_UP) {
720 if (!curr || curr->link != BOND_LINK_UP)
725 if (bond->force_primary) {
726 bond->force_primary = false;
730 if (!curr || curr->link != BOND_LINK_UP)
733 /* At this point, prim and curr are both up */
734 switch (bond->params.primary_reselect) {
735 case BOND_PRI_RESELECT_ALWAYS:
737 case BOND_PRI_RESELECT_BETTER:
738 if (prim->speed < curr->speed)
740 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
743 case BOND_PRI_RESELECT_FAILURE:
746 netdev_err(bond->dev, "impossible primary_reselect %d\n",
747 bond->params.primary_reselect);
753 * bond_find_best_slave - select the best available slave to be the active one
754 * @bond: our bonding struct
756 static struct slave *bond_find_best_slave(struct bonding *bond)
758 struct slave *slave, *bestslave = NULL;
759 struct list_head *iter;
760 int mintime = bond->params.updelay;
762 slave = bond_choose_primary_or_current(bond);
766 bond_for_each_slave(bond, slave, iter) {
767 if (slave->link == BOND_LINK_UP)
769 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
770 slave->delay < mintime) {
771 mintime = slave->delay;
779 static bool bond_should_notify_peers(struct bonding *bond)
784 slave = rcu_dereference(bond->curr_active_slave);
787 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
788 slave ? slave->dev->name : "NULL");
790 if (!slave || !bond->send_peer_notif ||
791 !netif_carrier_ok(bond->dev) ||
792 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
799 * change_active_interface - change the active slave into the specified one
800 * @bond: our bonding struct
801 * @new: the new slave to make the active one
803 * Set the new slave to the bond's settings and unset them on the old
805 * Setting include flags, mc-list, promiscuity, allmulti, etc.
807 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
808 * because it is apparently the best available slave we have, even though its
809 * updelay hasn't timed out yet.
811 * Caller must hold RTNL.
813 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
815 struct slave *old_active;
819 old_active = rtnl_dereference(bond->curr_active_slave);
821 if (old_active == new_active)
825 new_active->last_link_up = jiffies;
827 if (new_active->link == BOND_LINK_BACK) {
828 if (bond_uses_primary(bond)) {
829 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
830 new_active->dev->name,
831 (bond->params.updelay - new_active->delay) * bond->params.miimon);
834 new_active->delay = 0;
835 bond_set_slave_link_state(new_active, BOND_LINK_UP);
837 if (BOND_MODE(bond) == BOND_MODE_8023AD)
838 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
840 if (bond_is_lb(bond))
841 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
843 if (bond_uses_primary(bond)) {
844 netdev_info(bond->dev, "making interface %s the new active one\n",
845 new_active->dev->name);
850 if (bond_uses_primary(bond))
851 bond_hw_addr_swap(bond, new_active, old_active);
853 if (bond_is_lb(bond)) {
854 bond_alb_handle_active_change(bond, new_active);
856 bond_set_slave_inactive_flags(old_active,
857 BOND_SLAVE_NOTIFY_NOW);
859 bond_set_slave_active_flags(new_active,
860 BOND_SLAVE_NOTIFY_NOW);
862 rcu_assign_pointer(bond->curr_active_slave, new_active);
865 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
867 bond_set_slave_inactive_flags(old_active,
868 BOND_SLAVE_NOTIFY_NOW);
871 bool should_notify_peers = false;
873 bond_set_slave_active_flags(new_active,
874 BOND_SLAVE_NOTIFY_NOW);
876 if (bond->params.fail_over_mac)
877 bond_do_fail_over_mac(bond, new_active,
880 if (netif_running(bond->dev)) {
881 bond->send_peer_notif =
882 bond->params.num_peer_notif;
883 should_notify_peers =
884 bond_should_notify_peers(bond);
887 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
888 if (should_notify_peers)
889 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
894 /* resend IGMP joins since active slave has changed or
895 * all were sent on curr_active_slave.
896 * resend only if bond is brought up with the affected
897 * bonding modes and the retransmission is enabled
899 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
900 ((bond_uses_primary(bond) && new_active) ||
901 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
902 bond->igmp_retrans = bond->params.resend_igmp;
903 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
908 * bond_select_active_slave - select a new active slave, if needed
909 * @bond: our bonding struct
911 * This functions should be called when one of the following occurs:
912 * - The old curr_active_slave has been released or lost its link.
913 * - The primary_slave has got its link back.
914 * - A slave has got its link back and there's no old curr_active_slave.
916 * Caller must hold RTNL.
918 void bond_select_active_slave(struct bonding *bond)
920 struct slave *best_slave;
925 best_slave = bond_find_best_slave(bond);
926 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
927 bond_change_active_slave(bond, best_slave);
928 rv = bond_set_carrier(bond);
932 if (netif_carrier_ok(bond->dev)) {
933 netdev_info(bond->dev, "first active interface up!\n");
935 netdev_info(bond->dev, "now running without any active interface!\n");
940 #ifdef CONFIG_NET_POLL_CONTROLLER
941 static inline int slave_enable_netpoll(struct slave *slave)
946 np = kzalloc(sizeof(*np), GFP_KERNEL);
951 err = __netpoll_setup(np, slave->dev);
960 static inline void slave_disable_netpoll(struct slave *slave)
962 struct netpoll *np = slave->np;
968 __netpoll_free_async(np);
971 static void bond_poll_controller(struct net_device *bond_dev)
973 struct bonding *bond = netdev_priv(bond_dev);
974 struct slave *slave = NULL;
975 struct list_head *iter;
976 struct ad_info ad_info;
977 struct netpoll_info *ni;
978 const struct net_device_ops *ops;
980 if (BOND_MODE(bond) == BOND_MODE_8023AD)
981 if (bond_3ad_get_active_agg_info(bond, &ad_info))
984 bond_for_each_slave_rcu(bond, slave, iter) {
985 ops = slave->dev->netdev_ops;
986 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
989 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
990 struct aggregator *agg =
991 SLAVE_AD_INFO(slave)->port.aggregator;
994 agg->aggregator_identifier != ad_info.aggregator_id)
998 ni = rcu_dereference_bh(slave->dev->npinfo);
999 if (down_trylock(&ni->dev_lock))
1001 ops->ndo_poll_controller(slave->dev);
1006 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1008 struct bonding *bond = netdev_priv(bond_dev);
1009 struct list_head *iter;
1010 struct slave *slave;
1012 bond_for_each_slave(bond, slave, iter)
1013 if (bond_slave_is_up(slave))
1014 slave_disable_netpoll(slave);
1017 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1019 struct bonding *bond = netdev_priv(dev);
1020 struct list_head *iter;
1021 struct slave *slave;
1024 bond_for_each_slave(bond, slave, iter) {
1025 err = slave_enable_netpoll(slave);
1027 bond_netpoll_cleanup(dev);
1034 static inline int slave_enable_netpoll(struct slave *slave)
1038 static inline void slave_disable_netpoll(struct slave *slave)
1041 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1046 /*---------------------------------- IOCTL ----------------------------------*/
1048 static netdev_features_t bond_fix_features(struct net_device *dev,
1049 netdev_features_t features)
1051 struct bonding *bond = netdev_priv(dev);
1052 struct list_head *iter;
1053 netdev_features_t mask;
1054 struct slave *slave;
1058 features &= ~NETIF_F_ONE_FOR_ALL;
1059 features |= NETIF_F_ALL_FOR_ALL;
1061 bond_for_each_slave(bond, slave, iter) {
1062 features = netdev_increment_features(features,
1063 slave->dev->features,
1066 features = netdev_add_tso_features(features, mask);
1071 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1072 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1073 NETIF_F_HIGHDMA | NETIF_F_LRO)
1075 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1078 static void bond_compute_features(struct bonding *bond)
1080 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1081 IFF_XMIT_DST_RELEASE_PERM;
1082 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1083 netdev_features_t enc_features = BOND_ENC_FEATURES;
1084 struct net_device *bond_dev = bond->dev;
1085 struct list_head *iter;
1086 struct slave *slave;
1087 unsigned short max_hard_header_len = ETH_HLEN;
1088 unsigned int gso_max_size = GSO_MAX_SIZE;
1089 u16 gso_max_segs = GSO_MAX_SEGS;
1091 if (!bond_has_slaves(bond))
1093 vlan_features &= NETIF_F_ALL_FOR_ALL;
1095 bond_for_each_slave(bond, slave, iter) {
1096 vlan_features = netdev_increment_features(vlan_features,
1097 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1099 enc_features = netdev_increment_features(enc_features,
1100 slave->dev->hw_enc_features,
1102 dst_release_flag &= slave->dev->priv_flags;
1103 if (slave->dev->hard_header_len > max_hard_header_len)
1104 max_hard_header_len = slave->dev->hard_header_len;
1106 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1107 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1111 bond_dev->vlan_features = vlan_features;
1112 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1113 bond_dev->hard_header_len = max_hard_header_len;
1114 bond_dev->gso_max_segs = gso_max_segs;
1115 netif_set_gso_max_size(bond_dev, gso_max_size);
1117 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1118 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1119 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1120 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1122 netdev_change_features(bond_dev);
1125 static void bond_setup_by_slave(struct net_device *bond_dev,
1126 struct net_device *slave_dev)
1128 bond_dev->header_ops = slave_dev->header_ops;
1130 bond_dev->type = slave_dev->type;
1131 bond_dev->hard_header_len = slave_dev->hard_header_len;
1132 bond_dev->addr_len = slave_dev->addr_len;
1134 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1135 slave_dev->addr_len);
1138 /* On bonding slaves other than the currently active slave, suppress
1139 * duplicates except for alb non-mcast/bcast.
1141 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1142 struct slave *slave,
1143 struct bonding *bond)
1145 if (bond_is_slave_inactive(slave)) {
1146 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1147 skb->pkt_type != PACKET_BROADCAST &&
1148 skb->pkt_type != PACKET_MULTICAST)
1155 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1157 struct sk_buff *skb = *pskb;
1158 struct slave *slave;
1159 struct bonding *bond;
1160 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1162 int ret = RX_HANDLER_ANOTHER;
1164 skb = skb_share_check(skb, GFP_ATOMIC);
1166 return RX_HANDLER_CONSUMED;
1170 slave = bond_slave_get_rcu(skb->dev);
1173 recv_probe = ACCESS_ONCE(bond->recv_probe);
1175 ret = recv_probe(skb, bond, slave);
1176 if (ret == RX_HANDLER_CONSUMED) {
1182 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1183 return RX_HANDLER_EXACT;
1186 skb->dev = bond->dev;
1188 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1189 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1190 skb->pkt_type == PACKET_HOST) {
1192 if (unlikely(skb_cow_head(skb,
1193 skb->data - skb_mac_header(skb)))) {
1195 return RX_HANDLER_CONSUMED;
1197 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1203 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1204 struct net_device *slave_dev,
1205 struct slave *slave)
1209 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1212 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1216 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1217 struct net_device *slave_dev)
1219 netdev_upper_dev_unlink(slave_dev, bond_dev);
1220 slave_dev->flags &= ~IFF_SLAVE;
1221 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1224 static struct slave *bond_alloc_slave(struct bonding *bond)
1226 struct slave *slave = NULL;
1228 slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1232 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1233 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1235 if (!SLAVE_AD_INFO(slave)) {
1243 static void bond_free_slave(struct slave *slave)
1245 struct bonding *bond = bond_get_bond_by_slave(slave);
1247 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1248 kfree(SLAVE_AD_INFO(slave));
1253 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1255 info->bond_mode = BOND_MODE(bond);
1256 info->miimon = bond->params.miimon;
1257 info->num_slaves = bond->slave_cnt;
1260 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1262 strcpy(info->slave_name, slave->dev->name);
1263 info->link = slave->link;
1264 info->state = bond_slave_state(slave);
1265 info->link_failure_count = slave->link_failure_count;
1268 static void bond_netdev_notify(struct net_device *dev,
1269 struct netdev_bonding_info *info)
1272 netdev_bonding_info_change(dev, info);
1276 static void bond_netdev_notify_work(struct work_struct *_work)
1278 struct netdev_notify_work *w =
1279 container_of(_work, struct netdev_notify_work, work.work);
1281 bond_netdev_notify(w->dev, &w->bonding_info);
1286 void bond_queue_slave_event(struct slave *slave)
1288 struct bonding *bond = slave->bond;
1289 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1294 dev_hold(slave->dev);
1295 nnw->dev = slave->dev;
1296 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1297 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1298 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1300 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1303 /* enslave device <slave> to bond device <master> */
1304 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1306 struct bonding *bond = netdev_priv(bond_dev);
1307 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1308 struct slave *new_slave = NULL, *prev_slave;
1309 struct sockaddr addr;
1313 if (!bond->params.use_carrier &&
1314 slave_dev->ethtool_ops->get_link == NULL &&
1315 slave_ops->ndo_do_ioctl == NULL) {
1316 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1320 /* already enslaved */
1321 if (slave_dev->flags & IFF_SLAVE) {
1322 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1326 if (bond_dev == slave_dev) {
1327 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1331 /* vlan challenged mutual exclusion */
1332 /* no need to lock since we're protected by rtnl_lock */
1333 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1334 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1336 if (vlan_uses_dev(bond_dev)) {
1337 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1338 slave_dev->name, bond_dev->name);
1341 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1342 slave_dev->name, slave_dev->name,
1346 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1350 /* Old ifenslave binaries are no longer supported. These can
1351 * be identified with moderate accuracy by the state of the slave:
1352 * the current ifenslave will set the interface down prior to
1353 * enslaving it; the old ifenslave will not.
1355 if ((slave_dev->flags & IFF_UP)) {
1356 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1359 goto err_undo_flags;
1362 /* set bonding device ether type by slave - bonding netdevices are
1363 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1364 * there is a need to override some of the type dependent attribs/funcs.
1366 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1367 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1369 if (!bond_has_slaves(bond)) {
1370 if (bond_dev->type != slave_dev->type) {
1371 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1372 bond_dev->type, slave_dev->type);
1374 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1376 res = notifier_to_errno(res);
1378 netdev_err(bond_dev, "refused to change device type\n");
1380 goto err_undo_flags;
1383 /* Flush unicast and multicast addresses */
1384 dev_uc_flush(bond_dev);
1385 dev_mc_flush(bond_dev);
1387 if (slave_dev->type != ARPHRD_ETHER)
1388 bond_setup_by_slave(bond_dev, slave_dev);
1390 ether_setup(bond_dev);
1391 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1394 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1397 } else if (bond_dev->type != slave_dev->type) {
1398 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1399 slave_dev->name, slave_dev->type, bond_dev->type);
1401 goto err_undo_flags;
1404 if (slave_ops->ndo_set_mac_address == NULL) {
1405 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1406 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1407 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1408 if (!bond_has_slaves(bond)) {
1409 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1410 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1412 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1414 goto err_undo_flags;
1419 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1421 /* If this is the first slave, then we need to set the master's hardware
1422 * address to be the same as the slave's.
1424 if (!bond_has_slaves(bond) &&
1425 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1426 bond_set_dev_addr(bond->dev, slave_dev);
1428 new_slave = bond_alloc_slave(bond);
1431 goto err_undo_flags;
1434 new_slave->bond = bond;
1435 new_slave->dev = slave_dev;
1436 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1437 * is set via sysfs or module option if desired.
1439 new_slave->queue_id = 0;
1441 /* Save slave's original mtu and then set it to match the bond */
1442 new_slave->original_mtu = slave_dev->mtu;
1443 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1445 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1449 /* Save slave's original ("permanent") mac address for modes
1450 * that need it, and for restoring it upon release, and then
1451 * set it to the master's address
1453 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1455 if (!bond->params.fail_over_mac ||
1456 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1457 /* Set slave to master's mac address. The application already
1458 * set the master's mac address to that of the first slave
1460 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1461 addr.sa_family = slave_dev->type;
1462 res = dev_set_mac_address(slave_dev, &addr);
1464 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1465 goto err_restore_mtu;
1469 /* set slave flag before open to prevent IPv6 addrconf */
1470 slave_dev->flags |= IFF_SLAVE;
1472 /* open the slave since the application closed it */
1473 res = dev_open(slave_dev);
1475 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1476 goto err_restore_mac;
1479 slave_dev->priv_flags |= IFF_BONDING;
1480 /* initialize slave stats */
1481 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1483 if (bond_is_lb(bond)) {
1484 /* bond_alb_init_slave() must be called before all other stages since
1485 * it might fail and we do not want to have to undo everything
1487 res = bond_alb_init_slave(bond, new_slave);
1492 /* If the mode uses primary, then the following is handled by
1493 * bond_change_active_slave().
1495 if (!bond_uses_primary(bond)) {
1496 /* set promiscuity level to new slave */
1497 if (bond_dev->flags & IFF_PROMISC) {
1498 res = dev_set_promiscuity(slave_dev, 1);
1503 /* set allmulti level to new slave */
1504 if (bond_dev->flags & IFF_ALLMULTI) {
1505 res = dev_set_allmulti(slave_dev, 1);
1510 netif_addr_lock_bh(bond_dev);
1512 dev_mc_sync_multiple(slave_dev, bond_dev);
1513 dev_uc_sync_multiple(slave_dev, bond_dev);
1515 netif_addr_unlock_bh(bond_dev);
1518 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1519 /* add lacpdu mc addr to mc list */
1520 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1522 dev_mc_add(slave_dev, lacpdu_multicast);
1525 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1527 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1532 prev_slave = bond_last_slave(bond);
1534 new_slave->delay = 0;
1535 new_slave->link_failure_count = 0;
1537 bond_update_speed_duplex(new_slave);
1539 new_slave->last_rx = jiffies -
1540 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1541 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1542 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1544 if (bond->params.miimon && !bond->params.use_carrier) {
1545 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1547 if ((link_reporting == -1) && !bond->params.arp_interval) {
1548 /* miimon is set but a bonded network driver
1549 * does not support ETHTOOL/MII and
1550 * arp_interval is not set. Note: if
1551 * use_carrier is enabled, we will never go
1552 * here (because netif_carrier is always
1553 * supported); thus, we don't need to change
1554 * the messages for netif_carrier.
1556 netdev_warn(bond_dev, "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",
1558 } else if (link_reporting == -1) {
1559 /* unable get link status using mii/ethtool */
1560 netdev_warn(bond_dev, "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",
1565 /* check for initial state */
1566 if (bond->params.miimon) {
1567 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1568 if (bond->params.updelay) {
1569 bond_set_slave_link_state(new_slave,
1571 new_slave->delay = bond->params.updelay;
1573 bond_set_slave_link_state(new_slave,
1577 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1579 } else if (bond->params.arp_interval) {
1580 bond_set_slave_link_state(new_slave,
1581 (netif_carrier_ok(slave_dev) ?
1582 BOND_LINK_UP : BOND_LINK_DOWN));
1584 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1587 if (new_slave->link != BOND_LINK_DOWN)
1588 new_slave->last_link_up = jiffies;
1589 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1590 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1591 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1593 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1594 /* if there is a primary slave, remember it */
1595 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1596 rcu_assign_pointer(bond->primary_slave, new_slave);
1597 bond->force_primary = true;
1601 switch (BOND_MODE(bond)) {
1602 case BOND_MODE_ACTIVEBACKUP:
1603 bond_set_slave_inactive_flags(new_slave,
1604 BOND_SLAVE_NOTIFY_NOW);
1606 case BOND_MODE_8023AD:
1607 /* in 802.3ad mode, the internal mechanism
1608 * will activate the slaves in the selected
1611 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1612 /* if this is the first slave */
1614 SLAVE_AD_INFO(new_slave)->id = 1;
1615 /* Initialize AD with the number of times that the AD timer is called in 1 second
1616 * can be called only after the mac address of the bond is set
1618 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1620 SLAVE_AD_INFO(new_slave)->id =
1621 SLAVE_AD_INFO(prev_slave)->id + 1;
1624 bond_3ad_bind_slave(new_slave);
1628 bond_set_active_slave(new_slave);
1629 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1632 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1634 /* always active in trunk mode */
1635 bond_set_active_slave(new_slave);
1637 /* In trunking mode there is little meaning to curr_active_slave
1638 * anyway (it holds no special properties of the bond device),
1639 * so we can change it without calling change_active_interface()
1641 if (!rcu_access_pointer(bond->curr_active_slave) &&
1642 new_slave->link == BOND_LINK_UP)
1643 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1646 } /* switch(bond_mode) */
1648 #ifdef CONFIG_NET_POLL_CONTROLLER
1649 slave_dev->npinfo = bond->dev->npinfo;
1650 if (slave_dev->npinfo) {
1651 if (slave_enable_netpoll(new_slave)) {
1652 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1659 if (!(bond_dev->features & NETIF_F_LRO))
1660 dev_disable_lro(slave_dev);
1662 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1665 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1669 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1671 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1672 goto err_unregister;
1675 res = bond_sysfs_slave_add(new_slave);
1677 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1678 goto err_upper_unlink;
1682 bond_compute_features(bond);
1683 bond_set_carrier(bond);
1685 if (bond_uses_primary(bond)) {
1687 bond_select_active_slave(bond);
1688 unblock_netpoll_tx();
1691 if (bond_mode_uses_xmit_hash(bond))
1692 bond_update_slave_arr(bond, NULL);
1694 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1696 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1697 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1699 /* enslave is successful */
1700 bond_queue_slave_event(new_slave);
1703 /* Undo stages on error */
1705 bond_upper_dev_unlink(bond_dev, slave_dev);
1708 netdev_rx_handler_unregister(slave_dev);
1711 if (!bond_uses_primary(bond))
1712 bond_hw_addr_flush(bond_dev, slave_dev);
1714 vlan_vids_del_by_dev(slave_dev, bond_dev);
1715 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1716 RCU_INIT_POINTER(bond->primary_slave, NULL);
1717 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1719 bond_change_active_slave(bond, NULL);
1720 bond_select_active_slave(bond);
1721 unblock_netpoll_tx();
1723 /* either primary_slave or curr_active_slave might've changed */
1725 slave_disable_netpoll(new_slave);
1728 slave_dev->priv_flags &= ~IFF_BONDING;
1729 dev_close(slave_dev);
1732 slave_dev->flags &= ~IFF_SLAVE;
1733 if (!bond->params.fail_over_mac ||
1734 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1735 /* XXX TODO - fom follow mode needs to change master's
1736 * MAC if this slave's MAC is in use by the bond, or at
1737 * least print a warning.
1739 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1740 addr.sa_family = slave_dev->type;
1741 dev_set_mac_address(slave_dev, &addr);
1745 dev_set_mtu(slave_dev, new_slave->original_mtu);
1748 bond_free_slave(new_slave);
1751 /* Enslave of first slave has failed and we need to fix master's mac */
1752 if (!bond_has_slaves(bond)) {
1753 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1754 slave_dev->dev_addr))
1755 eth_hw_addr_random(bond_dev);
1756 if (bond_dev->type != ARPHRD_ETHER) {
1757 dev_close(bond_dev);
1758 ether_setup(bond_dev);
1759 bond_dev->flags |= IFF_MASTER;
1760 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1767 /* Try to release the slave device <slave> from the bond device <master>
1768 * It is legal to access curr_active_slave without a lock because all the function
1769 * is RTNL-locked. If "all" is true it means that the function is being called
1770 * while destroying a bond interface and all slaves are being released.
1772 * The rules for slave state should be:
1773 * for Active/Backup:
1774 * Active stays on all backups go down
1775 * for Bonded connections:
1776 * The first up interface should be left on and all others downed.
1778 static int __bond_release_one(struct net_device *bond_dev,
1779 struct net_device *slave_dev,
1782 struct bonding *bond = netdev_priv(bond_dev);
1783 struct slave *slave, *oldcurrent;
1784 struct sockaddr addr;
1785 int old_flags = bond_dev->flags;
1786 netdev_features_t old_features = bond_dev->features;
1788 /* slave is not a slave or master is not master of this slave */
1789 if (!(slave_dev->flags & IFF_SLAVE) ||
1790 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1791 netdev_dbg(bond_dev, "cannot release %s\n",
1798 slave = bond_get_slave_by_dev(bond, slave_dev);
1800 /* not a slave of this bond */
1801 netdev_info(bond_dev, "%s not enslaved\n",
1803 unblock_netpoll_tx();
1807 bond_sysfs_slave_del(slave);
1809 /* recompute stats just before removing the slave */
1810 bond_get_stats(bond->dev, &bond->bond_stats);
1812 bond_upper_dev_unlink(bond_dev, slave_dev);
1813 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1814 * for this slave anymore.
1816 netdev_rx_handler_unregister(slave_dev);
1818 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1819 bond_3ad_unbind_slave(slave);
1821 if (bond_mode_uses_xmit_hash(bond))
1822 bond_update_slave_arr(bond, slave);
1824 netdev_info(bond_dev, "Releasing %s interface %s\n",
1825 bond_is_active_slave(slave) ? "active" : "backup",
1828 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1830 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1832 if (!all && (!bond->params.fail_over_mac ||
1833 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1834 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1835 bond_has_slaves(bond))
1836 netdev_warn(bond_dev, "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",
1837 slave_dev->name, slave->perm_hwaddr,
1838 bond_dev->name, slave_dev->name);
1841 if (rtnl_dereference(bond->primary_slave) == slave)
1842 RCU_INIT_POINTER(bond->primary_slave, NULL);
1844 if (oldcurrent == slave)
1845 bond_change_active_slave(bond, NULL);
1847 if (bond_is_lb(bond)) {
1848 /* Must be called only after the slave has been
1849 * detached from the list and the curr_active_slave
1850 * has been cleared (if our_slave == old_current),
1851 * but before a new active slave is selected.
1853 bond_alb_deinit_slave(bond, slave);
1857 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1858 } else if (oldcurrent == slave) {
1859 /* Note that we hold RTNL over this sequence, so there
1860 * is no concern that another slave add/remove event
1863 bond_select_active_slave(bond);
1866 if (!bond_has_slaves(bond)) {
1867 bond_set_carrier(bond);
1868 eth_hw_addr_random(bond_dev);
1871 unblock_netpoll_tx();
1875 if (!bond_has_slaves(bond)) {
1876 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1877 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1880 bond_compute_features(bond);
1881 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1882 (old_features & NETIF_F_VLAN_CHALLENGED))
1883 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1884 slave_dev->name, bond_dev->name);
1886 vlan_vids_del_by_dev(slave_dev, bond_dev);
1888 /* If the mode uses primary, then this case was handled above by
1889 * bond_change_active_slave(..., NULL)
1891 if (!bond_uses_primary(bond)) {
1892 /* unset promiscuity level from slave
1893 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1894 * of the IFF_PROMISC flag in the bond_dev, but we need the
1895 * value of that flag before that change, as that was the value
1896 * when this slave was attached, so we cache at the start of the
1897 * function and use it here. Same goes for ALLMULTI below
1899 if (old_flags & IFF_PROMISC)
1900 dev_set_promiscuity(slave_dev, -1);
1902 /* unset allmulti level from slave */
1903 if (old_flags & IFF_ALLMULTI)
1904 dev_set_allmulti(slave_dev, -1);
1906 bond_hw_addr_flush(bond_dev, slave_dev);
1909 slave_disable_netpoll(slave);
1911 /* close slave before restoring its mac address */
1912 dev_close(slave_dev);
1914 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1915 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1916 /* restore original ("permanent") mac address */
1917 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1918 addr.sa_family = slave_dev->type;
1919 dev_set_mac_address(slave_dev, &addr);
1922 dev_set_mtu(slave_dev, slave->original_mtu);
1924 slave_dev->priv_flags &= ~IFF_BONDING;
1926 bond_free_slave(slave);
1931 /* A wrapper used because of ndo_del_link */
1932 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1934 return __bond_release_one(bond_dev, slave_dev, false);
1937 /* First release a slave and then destroy the bond if no more slaves are left.
1938 * Must be under rtnl_lock when this function is called.
1940 static int bond_release_and_destroy(struct net_device *bond_dev,
1941 struct net_device *slave_dev)
1943 struct bonding *bond = netdev_priv(bond_dev);
1946 ret = bond_release(bond_dev, slave_dev);
1947 if (ret == 0 && !bond_has_slaves(bond)) {
1948 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1949 netdev_info(bond_dev, "Destroying bond %s\n",
1951 bond_remove_proc_entry(bond);
1952 unregister_netdevice(bond_dev);
1957 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1959 struct bonding *bond = netdev_priv(bond_dev);
1960 bond_fill_ifbond(bond, info);
1964 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1966 struct bonding *bond = netdev_priv(bond_dev);
1967 struct list_head *iter;
1968 int i = 0, res = -ENODEV;
1969 struct slave *slave;
1971 bond_for_each_slave(bond, slave, iter) {
1972 if (i++ == (int)info->slave_id) {
1974 bond_fill_ifslave(slave, info);
1982 /*-------------------------------- Monitoring -------------------------------*/
1984 /* called with rcu_read_lock() */
1985 static int bond_miimon_inspect(struct bonding *bond)
1987 int link_state, commit = 0;
1988 struct list_head *iter;
1989 struct slave *slave;
1990 bool ignore_updelay;
1992 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1994 bond_for_each_slave_rcu(bond, slave, iter) {
1995 slave->new_link = BOND_LINK_NOCHANGE;
1997 link_state = bond_check_dev_link(bond, slave->dev, 0);
1999 switch (slave->link) {
2004 bond_set_slave_link_state(slave, BOND_LINK_FAIL);
2005 slave->delay = bond->params.downdelay;
2007 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2009 BOND_MODE_ACTIVEBACKUP) ?
2010 (bond_is_active_slave(slave) ?
2011 "active " : "backup ") : "",
2013 bond->params.downdelay * bond->params.miimon);
2016 case BOND_LINK_FAIL:
2018 /* recovered before downdelay expired */
2019 bond_set_slave_link_state(slave, BOND_LINK_UP);
2020 slave->last_link_up = jiffies;
2021 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2022 (bond->params.downdelay - slave->delay) *
2023 bond->params.miimon,
2028 if (slave->delay <= 0) {
2029 slave->new_link = BOND_LINK_DOWN;
2037 case BOND_LINK_DOWN:
2041 bond_set_slave_link_state(slave, BOND_LINK_BACK);
2042 slave->delay = bond->params.updelay;
2045 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2047 ignore_updelay ? 0 :
2048 bond->params.updelay *
2049 bond->params.miimon);
2052 case BOND_LINK_BACK:
2054 bond_set_slave_link_state(slave,
2056 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2057 (bond->params.updelay - slave->delay) *
2058 bond->params.miimon,
2067 if (slave->delay <= 0) {
2068 slave->new_link = BOND_LINK_UP;
2070 ignore_updelay = false;
2082 static void bond_miimon_commit(struct bonding *bond)
2084 struct list_head *iter;
2085 struct slave *slave, *primary;
2087 bond_for_each_slave(bond, slave, iter) {
2088 switch (slave->new_link) {
2089 case BOND_LINK_NOCHANGE:
2093 bond_set_slave_link_state(slave, BOND_LINK_UP);
2094 slave->last_link_up = jiffies;
2096 primary = rtnl_dereference(bond->primary_slave);
2097 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2098 /* prevent it from being the active one */
2099 bond_set_backup_slave(slave);
2100 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2101 /* make it immediately active */
2102 bond_set_active_slave(slave);
2103 } else if (slave != primary) {
2104 /* prevent it from being the active one */
2105 bond_set_backup_slave(slave);
2108 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2110 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2111 slave->duplex ? "full" : "half");
2113 /* notify ad that the link status has changed */
2114 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2115 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2117 if (bond_is_lb(bond))
2118 bond_alb_handle_link_change(bond, slave,
2121 if (BOND_MODE(bond) == BOND_MODE_XOR)
2122 bond_update_slave_arr(bond, NULL);
2124 if (!bond->curr_active_slave || slave == primary)
2129 case BOND_LINK_DOWN:
2130 if (slave->link_failure_count < UINT_MAX)
2131 slave->link_failure_count++;
2133 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2135 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2136 BOND_MODE(bond) == BOND_MODE_8023AD)
2137 bond_set_slave_inactive_flags(slave,
2138 BOND_SLAVE_NOTIFY_NOW);
2140 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2143 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2144 bond_3ad_handle_link_change(slave,
2147 if (bond_is_lb(bond))
2148 bond_alb_handle_link_change(bond, slave,
2151 if (BOND_MODE(bond) == BOND_MODE_XOR)
2152 bond_update_slave_arr(bond, NULL);
2154 if (slave == rcu_access_pointer(bond->curr_active_slave))
2160 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2161 slave->new_link, slave->dev->name);
2162 slave->new_link = BOND_LINK_NOCHANGE;
2169 bond_select_active_slave(bond);
2170 unblock_netpoll_tx();
2173 bond_set_carrier(bond);
2178 * Really a wrapper that splits the mii monitor into two phases: an
2179 * inspection, then (if inspection indicates something needs to be done)
2180 * an acquisition of appropriate locks followed by a commit phase to
2181 * implement whatever link state changes are indicated.
2183 static void bond_mii_monitor(struct work_struct *work)
2185 struct bonding *bond = container_of(work, struct bonding,
2187 bool should_notify_peers = false;
2188 unsigned long delay;
2190 delay = msecs_to_jiffies(bond->params.miimon);
2192 if (!bond_has_slaves(bond))
2197 should_notify_peers = bond_should_notify_peers(bond);
2199 if (bond_miimon_inspect(bond)) {
2202 /* Race avoidance with bond_close cancel of workqueue */
2203 if (!rtnl_trylock()) {
2205 should_notify_peers = false;
2209 bond_miimon_commit(bond);
2211 rtnl_unlock(); /* might sleep, hold no other locks */
2216 if (bond->params.miimon)
2217 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2219 if (should_notify_peers) {
2220 if (!rtnl_trylock())
2222 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2227 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2229 struct net_device *upper;
2230 struct list_head *iter;
2233 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2237 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2238 if (ip == bond_confirm_addr(upper, 0, ip)) {
2248 /* We go to the (large) trouble of VLAN tagging ARP frames because
2249 * switches in VLAN mode (especially if ports are configured as
2250 * "native" to a VLAN) might not pass non-tagged frames.
2252 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2253 __be32 dest_ip, __be32 src_ip,
2254 struct bond_vlan_tag *tags)
2256 struct sk_buff *skb;
2257 struct bond_vlan_tag *outer_tag = tags;
2259 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2260 arp_op, slave_dev->name, &dest_ip, &src_ip);
2262 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2263 NULL, slave_dev->dev_addr, NULL);
2266 net_err_ratelimited("ARP packet allocation failed\n");
2270 if (!tags || tags->vlan_proto == VLAN_N_VID)
2275 /* Go through all the tags backwards and add them to the packet */
2276 while (tags->vlan_proto != VLAN_N_VID) {
2277 if (!tags->vlan_id) {
2282 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2283 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2284 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2287 net_err_ratelimited("failed to insert inner VLAN tag\n");
2293 /* Set the outer tag */
2294 if (outer_tag->vlan_id) {
2295 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2296 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2297 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2298 outer_tag->vlan_id);
2305 /* Validate the device path between the @start_dev and the @end_dev.
2306 * The path is valid if the @end_dev is reachable through device
2308 * When the path is validated, collect any vlan information in the
2311 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2312 struct net_device *end_dev,
2315 struct bond_vlan_tag *tags;
2316 struct net_device *upper;
2317 struct list_head *iter;
2319 if (start_dev == end_dev) {
2320 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2322 return ERR_PTR(-ENOMEM);
2323 tags[level].vlan_proto = VLAN_N_VID;
2327 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2328 tags = bond_verify_device_path(upper, end_dev, level + 1);
2329 if (IS_ERR_OR_NULL(tags)) {
2334 if (is_vlan_dev(upper)) {
2335 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2336 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2345 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2348 struct bond_vlan_tag *tags;
2349 __be32 *targets = bond->params.arp_targets, addr;
2352 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2353 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2356 /* Find out through which dev should the packet go */
2357 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2360 /* there's no route to target - try to send arp
2361 * probe to generate any traffic (arp_validate=0)
2363 if (bond->params.arp_validate)
2364 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2367 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2372 /* bond device itself */
2373 if (rt->dst.dev == bond->dev)
2377 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2380 if (!IS_ERR_OR_NULL(tags))
2383 /* Not our device - skip */
2384 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2385 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2391 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2393 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2399 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2403 if (!sip || !bond_has_this_ip(bond, tip)) {
2404 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2409 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2411 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2415 slave->last_rx = jiffies;
2416 slave->target_last_arp_rx[i] = jiffies;
2419 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2420 struct slave *slave)
2422 struct arphdr *arp = (struct arphdr *)skb->data;
2423 struct slave *curr_active_slave, *curr_arp_slave;
2424 unsigned char *arp_ptr;
2426 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2428 if (!slave_do_arp_validate(bond, slave)) {
2429 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2430 !slave_do_arp_validate_only(bond))
2431 slave->last_rx = jiffies;
2432 return RX_HANDLER_ANOTHER;
2433 } else if (!is_arp) {
2434 return RX_HANDLER_ANOTHER;
2437 alen = arp_hdr_len(bond->dev);
2439 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2442 if (alen > skb_headlen(skb)) {
2443 arp = kmalloc(alen, GFP_ATOMIC);
2446 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2450 if (arp->ar_hln != bond->dev->addr_len ||
2451 skb->pkt_type == PACKET_OTHERHOST ||
2452 skb->pkt_type == PACKET_LOOPBACK ||
2453 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2454 arp->ar_pro != htons(ETH_P_IP) ||
2458 arp_ptr = (unsigned char *)(arp + 1);
2459 arp_ptr += bond->dev->addr_len;
2460 memcpy(&sip, arp_ptr, 4);
2461 arp_ptr += 4 + bond->dev->addr_len;
2462 memcpy(&tip, arp_ptr, 4);
2464 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2465 slave->dev->name, bond_slave_state(slave),
2466 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2469 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2470 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2472 /* We 'trust' the received ARP enough to validate it if:
2474 * (a) the slave receiving the ARP is active (which includes the
2475 * current ARP slave, if any), or
2477 * (b) the receiving slave isn't active, but there is a currently
2478 * active slave and it received valid arp reply(s) after it became
2479 * the currently active slave, or
2481 * (c) there is an ARP slave that sent an ARP during the prior ARP
2482 * interval, and we receive an ARP reply on any slave. We accept
2483 * these because switch FDB update delays may deliver the ARP
2484 * reply to a slave other than the sender of the ARP request.
2486 * Note: for (b), backup slaves are receiving the broadcast ARP
2487 * request, not a reply. This request passes from the sending
2488 * slave through the L2 switch(es) to the receiving slave. Since
2489 * this is checking the request, sip/tip are swapped for
2492 * This is done to avoid endless looping when we can't reach the
2493 * arp_ip_target and fool ourselves with our own arp requests.
2495 if (bond_is_active_slave(slave))
2496 bond_validate_arp(bond, slave, sip, tip);
2497 else if (curr_active_slave &&
2498 time_after(slave_last_rx(bond, curr_active_slave),
2499 curr_active_slave->last_link_up))
2500 bond_validate_arp(bond, slave, tip, sip);
2501 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2502 bond_time_in_interval(bond,
2503 dev_trans_start(curr_arp_slave->dev), 1))
2504 bond_validate_arp(bond, slave, sip, tip);
2507 if (arp != (struct arphdr *)skb->data)
2509 return RX_HANDLER_ANOTHER;
2512 /* function to verify if we're in the arp_interval timeslice, returns true if
2513 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2514 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2516 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2519 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2521 return time_in_range(jiffies,
2522 last_act - delta_in_ticks,
2523 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2526 /* This function is called regularly to monitor each slave's link
2527 * ensuring that traffic is being sent and received when arp monitoring
2528 * is used in load-balancing mode. if the adapter has been dormant, then an
2529 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2530 * arp monitoring in active backup mode.
2532 static void bond_loadbalance_arp_mon(struct work_struct *work)
2534 struct bonding *bond = container_of(work, struct bonding,
2536 struct slave *slave, *oldcurrent;
2537 struct list_head *iter;
2538 int do_failover = 0, slave_state_changed = 0;
2540 if (!bond_has_slaves(bond))
2545 oldcurrent = rcu_dereference(bond->curr_active_slave);
2546 /* see if any of the previous devices are up now (i.e. they have
2547 * xmt and rcv traffic). the curr_active_slave does not come into
2548 * the picture unless it is null. also, slave->last_link_up is not
2549 * needed here because we send an arp on each slave and give a slave
2550 * as long as it needs to get the tx/rx within the delta.
2551 * TODO: what about up/down delay in arp mode? it wasn't here before
2554 bond_for_each_slave_rcu(bond, slave, iter) {
2555 unsigned long trans_start = dev_trans_start(slave->dev);
2557 if (slave->link != BOND_LINK_UP) {
2558 if (bond_time_in_interval(bond, trans_start, 1) &&
2559 bond_time_in_interval(bond, slave->last_rx, 1)) {
2561 slave->link = BOND_LINK_UP;
2562 slave_state_changed = 1;
2564 /* primary_slave has no meaning in round-robin
2565 * mode. the window of a slave being up and
2566 * curr_active_slave being null after enslaving
2570 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2574 netdev_info(bond->dev, "interface %s is now up\n",
2579 /* slave->link == BOND_LINK_UP */
2581 /* not all switches will respond to an arp request
2582 * when the source ip is 0, so don't take the link down
2583 * if we don't know our ip yet
2585 if (!bond_time_in_interval(bond, trans_start, 2) ||
2586 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2588 slave->link = BOND_LINK_DOWN;
2589 slave_state_changed = 1;
2591 if (slave->link_failure_count < UINT_MAX)
2592 slave->link_failure_count++;
2594 netdev_info(bond->dev, "interface %s is now down\n",
2597 if (slave == oldcurrent)
2602 /* note: if switch is in round-robin mode, all links
2603 * must tx arp to ensure all links rx an arp - otherwise
2604 * links may oscillate or not come up at all; if switch is
2605 * in something like xor mode, there is nothing we can
2606 * do - all replies will be rx'ed on same link causing slaves
2607 * to be unstable during low/no traffic periods
2609 if (bond_slave_is_up(slave))
2610 bond_arp_send_all(bond, slave);
2615 if (do_failover || slave_state_changed) {
2616 if (!rtnl_trylock())
2619 if (slave_state_changed) {
2620 bond_slave_state_change(bond);
2621 if (BOND_MODE(bond) == BOND_MODE_XOR)
2622 bond_update_slave_arr(bond, NULL);
2626 bond_select_active_slave(bond);
2627 unblock_netpoll_tx();
2633 if (bond->params.arp_interval)
2634 queue_delayed_work(bond->wq, &bond->arp_work,
2635 msecs_to_jiffies(bond->params.arp_interval));
2638 /* Called to inspect slaves for active-backup mode ARP monitor link state
2639 * changes. Sets new_link in slaves to specify what action should take
2640 * place for the slave. Returns 0 if no changes are found, >0 if changes
2641 * to link states must be committed.
2643 * Called with rcu_read_lock held.
2645 static int bond_ab_arp_inspect(struct bonding *bond)
2647 unsigned long trans_start, last_rx;
2648 struct list_head *iter;
2649 struct slave *slave;
2652 bond_for_each_slave_rcu(bond, slave, iter) {
2653 slave->new_link = BOND_LINK_NOCHANGE;
2654 last_rx = slave_last_rx(bond, slave);
2656 if (slave->link != BOND_LINK_UP) {
2657 if (bond_time_in_interval(bond, last_rx, 1)) {
2658 slave->new_link = BOND_LINK_UP;
2664 /* Give slaves 2*delta after being enslaved or made
2665 * active. This avoids bouncing, as the last receive
2666 * times need a full ARP monitor cycle to be updated.
2668 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2671 /* Backup slave is down if:
2672 * - No current_arp_slave AND
2673 * - more than 3*delta since last receive AND
2674 * - the bond has an IP address
2676 * Note: a non-null current_arp_slave indicates
2677 * the curr_active_slave went down and we are
2678 * searching for a new one; under this condition
2679 * we only take the curr_active_slave down - this
2680 * gives each slave a chance to tx/rx traffic
2681 * before being taken out
2683 if (!bond_is_active_slave(slave) &&
2684 !rcu_access_pointer(bond->current_arp_slave) &&
2685 !bond_time_in_interval(bond, last_rx, 3)) {
2686 slave->new_link = BOND_LINK_DOWN;
2690 /* Active slave is down if:
2691 * - more than 2*delta since transmitting OR
2692 * - (more than 2*delta since receive AND
2693 * the bond has an IP address)
2695 trans_start = dev_trans_start(slave->dev);
2696 if (bond_is_active_slave(slave) &&
2697 (!bond_time_in_interval(bond, trans_start, 2) ||
2698 !bond_time_in_interval(bond, last_rx, 2))) {
2699 slave->new_link = BOND_LINK_DOWN;
2707 /* Called to commit link state changes noted by inspection step of
2708 * active-backup mode ARP monitor.
2710 * Called with RTNL hold.
2712 static void bond_ab_arp_commit(struct bonding *bond)
2714 unsigned long trans_start;
2715 struct list_head *iter;
2716 struct slave *slave;
2718 bond_for_each_slave(bond, slave, iter) {
2719 switch (slave->new_link) {
2720 case BOND_LINK_NOCHANGE:
2724 trans_start = dev_trans_start(slave->dev);
2725 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2726 (!rtnl_dereference(bond->curr_active_slave) &&
2727 bond_time_in_interval(bond, trans_start, 1))) {
2728 struct slave *current_arp_slave;
2730 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2731 bond_set_slave_link_state(slave, BOND_LINK_UP);
2732 if (current_arp_slave) {
2733 bond_set_slave_inactive_flags(
2735 BOND_SLAVE_NOTIFY_NOW);
2736 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2739 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2742 if (!rtnl_dereference(bond->curr_active_slave) ||
2743 slave == rtnl_dereference(bond->primary_slave))
2750 case BOND_LINK_DOWN:
2751 if (slave->link_failure_count < UINT_MAX)
2752 slave->link_failure_count++;
2754 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2755 bond_set_slave_inactive_flags(slave,
2756 BOND_SLAVE_NOTIFY_NOW);
2758 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2761 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2762 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2769 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2770 slave->new_link, slave->dev->name);
2776 bond_select_active_slave(bond);
2777 unblock_netpoll_tx();
2780 bond_set_carrier(bond);
2783 /* Send ARP probes for active-backup mode ARP monitor.
2785 * Called with rcu_read_lock held.
2787 static bool bond_ab_arp_probe(struct bonding *bond)
2789 struct slave *slave, *before = NULL, *new_slave = NULL,
2790 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2791 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2792 struct list_head *iter;
2794 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2796 if (curr_arp_slave && curr_active_slave)
2797 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2798 curr_arp_slave->dev->name,
2799 curr_active_slave->dev->name);
2801 if (curr_active_slave) {
2802 bond_arp_send_all(bond, curr_active_slave);
2803 return should_notify_rtnl;
2806 /* if we don't have a curr_active_slave, search for the next available
2807 * backup slave from the current_arp_slave and make it the candidate
2808 * for becoming the curr_active_slave
2811 if (!curr_arp_slave) {
2812 curr_arp_slave = bond_first_slave_rcu(bond);
2813 if (!curr_arp_slave)
2814 return should_notify_rtnl;
2817 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2819 bond_for_each_slave_rcu(bond, slave, iter) {
2820 if (!found && !before && bond_slave_is_up(slave))
2823 if (found && !new_slave && bond_slave_is_up(slave))
2825 /* if the link state is up at this point, we
2826 * mark it down - this can happen if we have
2827 * simultaneous link failures and
2828 * reselect_active_interface doesn't make this
2829 * one the current slave so it is still marked
2830 * up when it is actually down
2832 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2833 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2834 if (slave->link_failure_count < UINT_MAX)
2835 slave->link_failure_count++;
2837 bond_set_slave_inactive_flags(slave,
2838 BOND_SLAVE_NOTIFY_LATER);
2840 netdev_info(bond->dev, "backup interface %s is now down\n",
2843 if (slave == curr_arp_slave)
2847 if (!new_slave && before)
2853 bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2854 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2855 bond_arp_send_all(bond, new_slave);
2856 new_slave->last_link_up = jiffies;
2857 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2860 bond_for_each_slave_rcu(bond, slave, iter) {
2861 if (slave->should_notify) {
2862 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2866 return should_notify_rtnl;
2869 static void bond_activebackup_arp_mon(struct work_struct *work)
2871 struct bonding *bond = container_of(work, struct bonding,
2873 bool should_notify_peers = false;
2874 bool should_notify_rtnl = false;
2877 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2879 if (!bond_has_slaves(bond))
2884 should_notify_peers = bond_should_notify_peers(bond);
2886 if (bond_ab_arp_inspect(bond)) {
2889 /* Race avoidance with bond_close flush of workqueue */
2890 if (!rtnl_trylock()) {
2892 should_notify_peers = false;
2896 bond_ab_arp_commit(bond);
2902 should_notify_rtnl = bond_ab_arp_probe(bond);
2906 if (bond->params.arp_interval)
2907 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2909 if (should_notify_peers || should_notify_rtnl) {
2910 if (!rtnl_trylock())
2913 if (should_notify_peers)
2914 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2916 if (should_notify_rtnl)
2917 bond_slave_state_notify(bond);
2923 /*-------------------------- netdev event handling --------------------------*/
2925 /* Change device name */
2926 static int bond_event_changename(struct bonding *bond)
2928 bond_remove_proc_entry(bond);
2929 bond_create_proc_entry(bond);
2931 bond_debug_reregister(bond);
2936 static int bond_master_netdev_event(unsigned long event,
2937 struct net_device *bond_dev)
2939 struct bonding *event_bond = netdev_priv(bond_dev);
2942 case NETDEV_CHANGENAME:
2943 return bond_event_changename(event_bond);
2944 case NETDEV_UNREGISTER:
2945 bond_remove_proc_entry(event_bond);
2947 case NETDEV_REGISTER:
2948 bond_create_proc_entry(event_bond);
2950 case NETDEV_NOTIFY_PEERS:
2951 if (event_bond->send_peer_notif)
2952 event_bond->send_peer_notif--;
2961 static int bond_slave_netdev_event(unsigned long event,
2962 struct net_device *slave_dev)
2964 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2965 struct bonding *bond;
2966 struct net_device *bond_dev;
2968 /* A netdev event can be generated while enslaving a device
2969 * before netdev_rx_handler_register is called in which case
2970 * slave will be NULL
2974 bond_dev = slave->bond->dev;
2976 primary = rtnl_dereference(bond->primary_slave);
2979 case NETDEV_UNREGISTER:
2980 if (bond_dev->type != ARPHRD_ETHER)
2981 bond_release_and_destroy(bond_dev, slave_dev);
2983 bond_release(bond_dev, slave_dev);
2987 bond_update_speed_duplex(slave);
2988 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2989 bond_3ad_adapter_speed_duplex_changed(slave);
2992 /* Refresh slave-array if applicable!
2993 * If the setup does not use miimon or arpmon (mode-specific!),
2994 * then these events will not cause the slave-array to be
2995 * refreshed. This will cause xmit to use a slave that is not
2996 * usable. Avoid such situation by refeshing the array at these
2997 * events. If these (miimon/arpmon) parameters are configured
2998 * then array gets refreshed twice and that should be fine!
3000 if (bond_mode_uses_xmit_hash(bond))
3001 bond_update_slave_arr(bond, NULL);
3003 case NETDEV_CHANGEMTU:
3004 /* TODO: Should slaves be allowed to
3005 * independently alter their MTU? For
3006 * an active-backup bond, slaves need
3007 * not be the same type of device, so
3008 * MTUs may vary. For other modes,
3009 * slaves arguably should have the
3010 * same MTUs. To do this, we'd need to
3011 * take over the slave's change_mtu
3012 * function for the duration of their
3016 case NETDEV_CHANGENAME:
3017 /* we don't care if we don't have primary set */
3018 if (!bond_uses_primary(bond) ||
3019 !bond->params.primary[0])
3022 if (slave == primary) {
3023 /* slave's name changed - he's no longer primary */
3024 RCU_INIT_POINTER(bond->primary_slave, NULL);
3025 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3026 /* we have a new primary slave */
3027 rcu_assign_pointer(bond->primary_slave, slave);
3028 } else { /* we didn't change primary - exit */
3032 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3033 primary ? slave_dev->name : "none");
3036 bond_select_active_slave(bond);
3037 unblock_netpoll_tx();
3039 case NETDEV_FEAT_CHANGE:
3040 bond_compute_features(bond);
3042 case NETDEV_RESEND_IGMP:
3043 /* Propagate to master device */
3044 call_netdevice_notifiers(event, slave->bond->dev);
3053 /* bond_netdev_event: handle netdev notifier chain events.
3055 * This function receives events for the netdev chain. The caller (an
3056 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3057 * locks for us to safely manipulate the slave devices (RTNL lock,
3060 static int bond_netdev_event(struct notifier_block *this,
3061 unsigned long event, void *ptr)
3063 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3065 netdev_dbg(event_dev, "event: %lx\n", event);
3067 if (!(event_dev->priv_flags & IFF_BONDING))
3070 if (event_dev->flags & IFF_MASTER) {
3071 netdev_dbg(event_dev, "IFF_MASTER\n");
3072 return bond_master_netdev_event(event, event_dev);
3075 if (event_dev->flags & IFF_SLAVE) {
3076 netdev_dbg(event_dev, "IFF_SLAVE\n");
3077 return bond_slave_netdev_event(event, event_dev);
3083 static struct notifier_block bond_netdev_notifier = {
3084 .notifier_call = bond_netdev_event,
3087 /*---------------------------- Hashing Policies -----------------------------*/
3089 /* L2 hash helper */
3090 static inline u32 bond_eth_hash(struct sk_buff *skb)
3092 struct ethhdr *ep, hdr_tmp;
3094 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3096 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3100 /* Extract the appropriate headers based on bond's xmit policy */
3101 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3102 struct flow_keys *fk)
3104 const struct ipv6hdr *iph6;
3105 const struct iphdr *iph;
3106 int noff, proto = -1;
3108 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3109 return skb_flow_dissect_flow_keys(skb, fk, 0);
3111 fk->ports.ports = 0;
3112 noff = skb_network_offset(skb);
3113 if (skb->protocol == htons(ETH_P_IP)) {
3114 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3117 iph_to_flow_copy_v4addrs(fk, iph);
3118 noff += iph->ihl << 2;
3119 if (!ip_is_fragment(iph))
3120 proto = iph->protocol;
3121 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3122 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3124 iph6 = ipv6_hdr(skb);
3125 iph_to_flow_copy_v6addrs(fk, iph6);
3126 noff += sizeof(*iph6);
3127 proto = iph6->nexthdr;
3131 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3132 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3138 * bond_xmit_hash - generate a hash value based on the xmit policy
3139 * @bond: bonding device
3140 * @skb: buffer to use for headers
3142 * This function will extract the necessary headers from the skb buffer and use
3143 * them to generate a hash based on the xmit_policy set in the bonding device
3145 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3147 struct flow_keys flow;
3150 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3154 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3155 !bond_flow_dissect(bond, skb, &flow))
3156 return bond_eth_hash(skb);
3158 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3159 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3160 hash = bond_eth_hash(skb);
3162 hash = (__force u32)flow.ports.ports;
3163 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3164 (__force u32)flow_get_u32_src(&flow);
3165 hash ^= (hash >> 16);
3166 hash ^= (hash >> 8);
3171 /*-------------------------- Device entry points ----------------------------*/
3173 static void bond_work_init_all(struct bonding *bond)
3175 INIT_DELAYED_WORK(&bond->mcast_work,
3176 bond_resend_igmp_join_requests_delayed);
3177 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3178 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3179 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3180 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3182 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3183 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3184 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3187 static void bond_work_cancel_all(struct bonding *bond)
3189 cancel_delayed_work_sync(&bond->mii_work);
3190 cancel_delayed_work_sync(&bond->arp_work);
3191 cancel_delayed_work_sync(&bond->alb_work);
3192 cancel_delayed_work_sync(&bond->ad_work);
3193 cancel_delayed_work_sync(&bond->mcast_work);
3194 cancel_delayed_work_sync(&bond->slave_arr_work);
3197 static int bond_open(struct net_device *bond_dev)
3199 struct bonding *bond = netdev_priv(bond_dev);
3200 struct list_head *iter;
3201 struct slave *slave;
3203 /* reset slave->backup and slave->inactive */
3204 if (bond_has_slaves(bond)) {
3205 bond_for_each_slave(bond, slave, iter) {
3206 if (bond_uses_primary(bond) &&
3207 slave != rcu_access_pointer(bond->curr_active_slave)) {
3208 bond_set_slave_inactive_flags(slave,
3209 BOND_SLAVE_NOTIFY_NOW);
3210 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3211 bond_set_slave_active_flags(slave,
3212 BOND_SLAVE_NOTIFY_NOW);
3217 bond_work_init_all(bond);
3219 if (bond_is_lb(bond)) {
3220 /* bond_alb_initialize must be called before the timer
3223 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3225 if (bond->params.tlb_dynamic_lb)
3226 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3229 if (bond->params.miimon) /* link check interval, in milliseconds. */
3230 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3232 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3233 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3234 bond->recv_probe = bond_arp_rcv;
3237 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3238 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3239 /* register to receive LACPDUs */
3240 bond->recv_probe = bond_3ad_lacpdu_recv;
3241 bond_3ad_initiate_agg_selection(bond, 1);
3244 if (bond_mode_uses_xmit_hash(bond))
3245 bond_update_slave_arr(bond, NULL);
3250 static int bond_close(struct net_device *bond_dev)
3252 struct bonding *bond = netdev_priv(bond_dev);
3254 bond_work_cancel_all(bond);
3255 bond->send_peer_notif = 0;
3256 if (bond_is_lb(bond))
3257 bond_alb_deinitialize(bond);
3258 bond->recv_probe = NULL;
3263 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3264 struct rtnl_link_stats64 *stats)
3266 struct bonding *bond = netdev_priv(bond_dev);
3267 struct rtnl_link_stats64 temp;
3268 struct list_head *iter;
3269 struct slave *slave;
3271 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3273 bond_for_each_slave(bond, slave, iter) {
3274 const struct rtnl_link_stats64 *sstats =
3275 dev_get_stats(slave->dev, &temp);
3276 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3278 stats->rx_packets += sstats->rx_packets - pstats->rx_packets;
3279 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3280 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3281 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3283 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3284 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3285 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3286 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3288 stats->multicast += sstats->multicast - pstats->multicast;
3289 stats->collisions += sstats->collisions - pstats->collisions;
3291 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3292 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3293 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3294 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3295 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3296 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3298 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3299 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3300 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3301 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3302 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3304 /* save off the slave stats for the next run */
3305 memcpy(pstats, sstats, sizeof(*sstats));
3307 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3312 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3314 struct bonding *bond = netdev_priv(bond_dev);
3315 struct net_device *slave_dev = NULL;
3316 struct ifbond k_binfo;
3317 struct ifbond __user *u_binfo = NULL;
3318 struct ifslave k_sinfo;
3319 struct ifslave __user *u_sinfo = NULL;
3320 struct mii_ioctl_data *mii = NULL;
3321 struct bond_opt_value newval;
3325 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3336 /* We do this again just in case we were called by SIOCGMIIREG
3337 * instead of SIOCGMIIPHY.
3343 if (mii->reg_num == 1) {
3345 if (netif_carrier_ok(bond->dev))
3346 mii->val_out = BMSR_LSTATUS;
3350 case BOND_INFO_QUERY_OLD:
3351 case SIOCBONDINFOQUERY:
3352 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3354 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3357 res = bond_info_query(bond_dev, &k_binfo);
3359 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3363 case BOND_SLAVE_INFO_QUERY_OLD:
3364 case SIOCBONDSLAVEINFOQUERY:
3365 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3367 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3370 res = bond_slave_info_query(bond_dev, &k_sinfo);
3372 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3380 net = dev_net(bond_dev);
3382 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3385 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3387 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3392 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3394 case BOND_ENSLAVE_OLD:
3395 case SIOCBONDENSLAVE:
3396 res = bond_enslave(bond_dev, slave_dev);
3398 case BOND_RELEASE_OLD:
3399 case SIOCBONDRELEASE:
3400 res = bond_release(bond_dev, slave_dev);
3402 case BOND_SETHWADDR_OLD:
3403 case SIOCBONDSETHWADDR:
3404 bond_set_dev_addr(bond_dev, slave_dev);
3407 case BOND_CHANGE_ACTIVE_OLD:
3408 case SIOCBONDCHANGEACTIVE:
3409 bond_opt_initstr(&newval, slave_dev->name);
3410 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3419 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3421 struct bonding *bond = netdev_priv(bond_dev);
3423 if (change & IFF_PROMISC)
3424 bond_set_promiscuity(bond,
3425 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3427 if (change & IFF_ALLMULTI)
3428 bond_set_allmulti(bond,
3429 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3432 static void bond_set_rx_mode(struct net_device *bond_dev)
3434 struct bonding *bond = netdev_priv(bond_dev);
3435 struct list_head *iter;
3436 struct slave *slave;
3439 if (bond_uses_primary(bond)) {
3440 slave = rcu_dereference(bond->curr_active_slave);
3442 dev_uc_sync(slave->dev, bond_dev);
3443 dev_mc_sync(slave->dev, bond_dev);
3446 bond_for_each_slave_rcu(bond, slave, iter) {
3447 dev_uc_sync_multiple(slave->dev, bond_dev);
3448 dev_mc_sync_multiple(slave->dev, bond_dev);
3454 static int bond_neigh_init(struct neighbour *n)
3456 struct bonding *bond = netdev_priv(n->dev);
3457 const struct net_device_ops *slave_ops;
3458 struct neigh_parms parms;
3459 struct slave *slave;
3462 slave = bond_first_slave(bond);
3465 slave_ops = slave->dev->netdev_ops;
3466 if (!slave_ops->ndo_neigh_setup)
3469 parms.neigh_setup = NULL;
3470 parms.neigh_cleanup = NULL;
3471 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3475 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3476 * after the last slave has been detached. Assumes that all slaves
3477 * utilize the same neigh_cleanup (true at this writing as only user
3480 n->parms->neigh_cleanup = parms.neigh_cleanup;
3482 if (!parms.neigh_setup)
3485 return parms.neigh_setup(n);
3488 /* The bonding ndo_neigh_setup is called at init time beofre any
3489 * slave exists. So we must declare proxy setup function which will
3490 * be used at run time to resolve the actual slave neigh param setup.
3492 * It's also called by master devices (such as vlans) to setup their
3493 * underlying devices. In that case - do nothing, we're already set up from
3496 static int bond_neigh_setup(struct net_device *dev,
3497 struct neigh_parms *parms)
3499 /* modify only our neigh_parms */
3500 if (parms->dev == dev)
3501 parms->neigh_setup = bond_neigh_init;
3506 /* Change the MTU of all of a master's slaves to match the master */
3507 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3509 struct bonding *bond = netdev_priv(bond_dev);
3510 struct slave *slave, *rollback_slave;
3511 struct list_head *iter;
3514 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3516 bond_for_each_slave(bond, slave, iter) {
3517 netdev_dbg(bond_dev, "s %p c_m %p\n",
3518 slave, slave->dev->netdev_ops->ndo_change_mtu);
3520 res = dev_set_mtu(slave->dev, new_mtu);
3523 /* If we failed to set the slave's mtu to the new value
3524 * we must abort the operation even in ACTIVE_BACKUP
3525 * mode, because if we allow the backup slaves to have
3526 * different mtu values than the active slave we'll
3527 * need to change their mtu when doing a failover. That
3528 * means changing their mtu from timer context, which
3529 * is probably not a good idea.
3531 netdev_dbg(bond_dev, "err %d %s\n", res,
3537 bond_dev->mtu = new_mtu;
3542 /* unwind from head to the slave that failed */
3543 bond_for_each_slave(bond, rollback_slave, iter) {
3546 if (rollback_slave == slave)
3549 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3551 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3552 tmp_res, rollback_slave->dev->name);
3559 /* Change HW address
3561 * Note that many devices must be down to change the HW address, and
3562 * downing the master releases all slaves. We can make bonds full of
3563 * bonding devices to test this, however.
3565 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3567 struct bonding *bond = netdev_priv(bond_dev);
3568 struct slave *slave, *rollback_slave;
3569 struct sockaddr *sa = addr, tmp_sa;
3570 struct list_head *iter;
3573 if (BOND_MODE(bond) == BOND_MODE_ALB)
3574 return bond_alb_set_mac_address(bond_dev, addr);
3577 netdev_dbg(bond_dev, "bond=%p\n", bond);
3579 /* If fail_over_mac is enabled, do nothing and return success.
3580 * Returning an error causes ifenslave to fail.
3582 if (bond->params.fail_over_mac &&
3583 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3586 if (!is_valid_ether_addr(sa->sa_data))
3587 return -EADDRNOTAVAIL;
3589 bond_for_each_slave(bond, slave, iter) {
3590 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3591 res = dev_set_mac_address(slave->dev, addr);
3593 /* TODO: consider downing the slave
3595 * User should expect communications
3596 * breakage anyway until ARP finish
3599 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3605 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3609 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3610 tmp_sa.sa_family = bond_dev->type;
3612 /* unwind from head to the slave that failed */
3613 bond_for_each_slave(bond, rollback_slave, iter) {
3616 if (rollback_slave == slave)
3619 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3621 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3622 tmp_res, rollback_slave->dev->name);
3630 * bond_xmit_slave_id - transmit skb through slave with slave_id
3631 * @bond: bonding device that is transmitting
3632 * @skb: buffer to transmit
3633 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3635 * This function tries to transmit through slave with slave_id but in case
3636 * it fails, it tries to find the first available slave for transmission.
3637 * The skb is consumed in all cases, thus the function is void.
3639 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3641 struct list_head *iter;
3642 struct slave *slave;
3645 /* Here we start from the slave with slave_id */
3646 bond_for_each_slave_rcu(bond, slave, iter) {
3648 if (bond_slave_can_tx(slave)) {
3649 bond_dev_queue_xmit(bond, skb, slave->dev);
3655 /* Here we start from the first slave up to slave_id */
3657 bond_for_each_slave_rcu(bond, slave, iter) {
3660 if (bond_slave_can_tx(slave)) {
3661 bond_dev_queue_xmit(bond, skb, slave->dev);
3665 /* no slave that can tx has been found */
3666 bond_tx_drop(bond->dev, skb);
3670 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3671 * @bond: bonding device to use
3673 * Based on the value of the bonding device's packets_per_slave parameter
3674 * this function generates a slave id, which is usually used as the next
3675 * slave to transmit through.
3677 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3680 struct reciprocal_value reciprocal_packets_per_slave;
3681 int packets_per_slave = bond->params.packets_per_slave;
3683 switch (packets_per_slave) {
3685 slave_id = prandom_u32();
3688 slave_id = bond->rr_tx_counter;
3691 reciprocal_packets_per_slave =
3692 bond->params.reciprocal_packets_per_slave;
3693 slave_id = reciprocal_divide(bond->rr_tx_counter,
3694 reciprocal_packets_per_slave);
3697 bond->rr_tx_counter++;
3702 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3704 struct bonding *bond = netdev_priv(bond_dev);
3705 struct iphdr *iph = ip_hdr(skb);
3706 struct slave *slave;
3709 /* Start with the curr_active_slave that joined the bond as the
3710 * default for sending IGMP traffic. For failover purposes one
3711 * needs to maintain some consistency for the interface that will
3712 * send the join/membership reports. The curr_active_slave found
3713 * will send all of this type of traffic.
3715 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3716 slave = rcu_dereference(bond->curr_active_slave);
3718 bond_dev_queue_xmit(bond, skb, slave->dev);
3720 bond_xmit_slave_id(bond, skb, 0);
3722 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3724 if (likely(slave_cnt)) {
3725 slave_id = bond_rr_gen_slave_id(bond);
3726 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3728 bond_tx_drop(bond_dev, skb);
3732 return NETDEV_TX_OK;
3735 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3736 * the bond has a usable interface.
3738 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3740 struct bonding *bond = netdev_priv(bond_dev);
3741 struct slave *slave;
3743 slave = rcu_dereference(bond->curr_active_slave);
3745 bond_dev_queue_xmit(bond, skb, slave->dev);
3747 bond_tx_drop(bond_dev, skb);
3749 return NETDEV_TX_OK;
3752 /* Use this to update slave_array when (a) it's not appropriate to update
3753 * slave_array right away (note that update_slave_array() may sleep)
3754 * and / or (b) RTNL is not held.
3756 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3758 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3761 /* Slave array work handler. Holds only RTNL */
3762 static void bond_slave_arr_handler(struct work_struct *work)
3764 struct bonding *bond = container_of(work, struct bonding,
3765 slave_arr_work.work);
3768 if (!rtnl_trylock())
3771 ret = bond_update_slave_arr(bond, NULL);
3774 pr_warn_ratelimited("Failed to update slave array from WT\n");
3780 bond_slave_arr_work_rearm(bond, 1);
3783 /* Build the usable slaves array in control path for modes that use xmit-hash
3784 * to determine the slave interface -
3785 * (a) BOND_MODE_8023AD
3787 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3789 * The caller is expected to hold RTNL only and NO other lock!
3791 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3793 struct slave *slave;
3794 struct list_head *iter;
3795 struct bond_up_slave *new_arr, *old_arr;
3799 #ifdef CONFIG_LOCKDEP
3800 WARN_ON(lockdep_is_held(&bond->mode_lock));
3803 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3807 pr_err("Failed to build slave-array.\n");
3810 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3811 struct ad_info ad_info;
3813 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3814 pr_debug("bond_3ad_get_active_agg_info failed\n");
3815 kfree_rcu(new_arr, rcu);
3816 /* No active aggragator means it's not safe to use
3817 * the previous array.
3819 old_arr = rtnl_dereference(bond->slave_arr);
3821 RCU_INIT_POINTER(bond->slave_arr, NULL);
3822 kfree_rcu(old_arr, rcu);
3826 agg_id = ad_info.aggregator_id;
3828 bond_for_each_slave(bond, slave, iter) {
3829 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3830 struct aggregator *agg;
3832 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3833 if (!agg || agg->aggregator_identifier != agg_id)
3836 if (!bond_slave_can_tx(slave))
3838 if (skipslave == slave)
3840 new_arr->arr[new_arr->count++] = slave;
3843 old_arr = rtnl_dereference(bond->slave_arr);
3844 rcu_assign_pointer(bond->slave_arr, new_arr);
3846 kfree_rcu(old_arr, rcu);
3848 if (ret != 0 && skipslave) {
3851 /* Rare situation where caller has asked to skip a specific
3852 * slave but allocation failed (most likely!). BTW this is
3853 * only possible when the call is initiated from
3854 * __bond_release_one(). In this situation; overwrite the
3855 * skipslave entry in the array with the last entry from the
3856 * array to avoid a situation where the xmit path may choose
3857 * this to-be-skipped slave to send a packet out.
3859 old_arr = rtnl_dereference(bond->slave_arr);
3860 for (idx = 0; idx < old_arr->count; idx++) {
3861 if (skipslave == old_arr->arr[idx]) {
3863 old_arr->arr[old_arr->count-1];
3872 /* Use this Xmit function for 3AD as well as XOR modes. The current
3873 * usable slave array is formed in the control path. The xmit function
3874 * just calculates hash and sends the packet out.
3876 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3878 struct bonding *bond = netdev_priv(dev);
3879 struct slave *slave;
3880 struct bond_up_slave *slaves;
3883 slaves = rcu_dereference(bond->slave_arr);
3884 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3885 if (likely(count)) {
3886 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3887 bond_dev_queue_xmit(bond, skb, slave->dev);
3889 bond_tx_drop(dev, skb);
3892 return NETDEV_TX_OK;
3895 /* in broadcast mode, we send everything to all usable interfaces. */
3896 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3898 struct bonding *bond = netdev_priv(bond_dev);
3899 struct slave *slave = NULL;
3900 struct list_head *iter;
3902 bond_for_each_slave_rcu(bond, slave, iter) {
3903 if (bond_is_last_slave(bond, slave))
3905 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3906 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3909 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3910 bond_dev->name, __func__);
3913 bond_dev_queue_xmit(bond, skb2, slave->dev);
3916 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3917 bond_dev_queue_xmit(bond, skb, slave->dev);
3919 bond_tx_drop(bond_dev, skb);
3921 return NETDEV_TX_OK;
3924 /*------------------------- Device initialization ---------------------------*/
3926 /* Lookup the slave that corresponds to a qid */
3927 static inline int bond_slave_override(struct bonding *bond,
3928 struct sk_buff *skb)
3930 struct slave *slave = NULL;
3931 struct list_head *iter;
3933 if (!skb->queue_mapping)
3936 /* Find out if any slaves have the same mapping as this skb. */
3937 bond_for_each_slave_rcu(bond, slave, iter) {
3938 if (slave->queue_id == skb->queue_mapping) {
3939 if (bond_slave_is_up(slave) &&
3940 slave->link == BOND_LINK_UP) {
3941 bond_dev_queue_xmit(bond, skb, slave->dev);
3944 /* If the slave isn't UP, use default transmit policy. */
3953 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3954 void *accel_priv, select_queue_fallback_t fallback)
3956 /* This helper function exists to help dev_pick_tx get the correct
3957 * destination queue. Using a helper function skips a call to
3958 * skb_tx_hash and will put the skbs in the queue we expect on their
3959 * way down to the bonding driver.
3961 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3963 /* Save the original txq to restore before passing to the driver */
3964 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3966 if (unlikely(txq >= dev->real_num_tx_queues)) {
3968 txq -= dev->real_num_tx_queues;
3969 } while (txq >= dev->real_num_tx_queues);
3974 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3976 struct bonding *bond = netdev_priv(dev);
3978 if (bond_should_override_tx_queue(bond) &&
3979 !bond_slave_override(bond, skb))
3980 return NETDEV_TX_OK;
3982 switch (BOND_MODE(bond)) {
3983 case BOND_MODE_ROUNDROBIN:
3984 return bond_xmit_roundrobin(skb, dev);
3985 case BOND_MODE_ACTIVEBACKUP:
3986 return bond_xmit_activebackup(skb, dev);
3987 case BOND_MODE_8023AD:
3989 return bond_3ad_xor_xmit(skb, dev);
3990 case BOND_MODE_BROADCAST:
3991 return bond_xmit_broadcast(skb, dev);
3993 return bond_alb_xmit(skb, dev);
3995 return bond_tlb_xmit(skb, dev);
3997 /* Should never happen, mode already checked */
3998 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4000 bond_tx_drop(dev, skb);
4001 return NETDEV_TX_OK;
4005 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4007 struct bonding *bond = netdev_priv(dev);
4008 netdev_tx_t ret = NETDEV_TX_OK;
4010 /* If we risk deadlock from transmitting this in the
4011 * netpoll path, tell netpoll to queue the frame for later tx
4013 if (unlikely(is_netpoll_tx_blocked(dev)))
4014 return NETDEV_TX_BUSY;
4017 if (bond_has_slaves(bond))
4018 ret = __bond_start_xmit(skb, dev);
4020 bond_tx_drop(dev, skb);
4026 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4027 struct ethtool_cmd *ecmd)
4029 struct bonding *bond = netdev_priv(bond_dev);
4030 unsigned long speed = 0;
4031 struct list_head *iter;
4032 struct slave *slave;
4034 ecmd->duplex = DUPLEX_UNKNOWN;
4035 ecmd->port = PORT_OTHER;
4037 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4038 * do not need to check mode. Though link speed might not represent
4039 * the true receive or transmit bandwidth (not all modes are symmetric)
4040 * this is an accurate maximum.
4042 bond_for_each_slave(bond, slave, iter) {
4043 if (bond_slave_can_tx(slave)) {
4044 if (slave->speed != SPEED_UNKNOWN)
4045 speed += slave->speed;
4046 if (ecmd->duplex == DUPLEX_UNKNOWN &&
4047 slave->duplex != DUPLEX_UNKNOWN)
4048 ecmd->duplex = slave->duplex;
4051 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4056 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4057 struct ethtool_drvinfo *drvinfo)
4059 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4060 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4061 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4065 static const struct ethtool_ops bond_ethtool_ops = {
4066 .get_drvinfo = bond_ethtool_get_drvinfo,
4067 .get_settings = bond_ethtool_get_settings,
4068 .get_link = ethtool_op_get_link,
4071 static const struct net_device_ops bond_netdev_ops = {
4072 .ndo_init = bond_init,
4073 .ndo_uninit = bond_uninit,
4074 .ndo_open = bond_open,
4075 .ndo_stop = bond_close,
4076 .ndo_start_xmit = bond_start_xmit,
4077 .ndo_select_queue = bond_select_queue,
4078 .ndo_get_stats64 = bond_get_stats,
4079 .ndo_do_ioctl = bond_do_ioctl,
4080 .ndo_change_rx_flags = bond_change_rx_flags,
4081 .ndo_set_rx_mode = bond_set_rx_mode,
4082 .ndo_change_mtu = bond_change_mtu,
4083 .ndo_set_mac_address = bond_set_mac_address,
4084 .ndo_neigh_setup = bond_neigh_setup,
4085 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4086 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4087 #ifdef CONFIG_NET_POLL_CONTROLLER
4088 .ndo_netpoll_setup = bond_netpoll_setup,
4089 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4090 .ndo_poll_controller = bond_poll_controller,
4092 .ndo_add_slave = bond_enslave,
4093 .ndo_del_slave = bond_release,
4094 .ndo_fix_features = bond_fix_features,
4095 .ndo_bridge_setlink = switchdev_port_bridge_setlink,
4096 .ndo_bridge_getlink = switchdev_port_bridge_getlink,
4097 .ndo_bridge_dellink = switchdev_port_bridge_dellink,
4098 .ndo_fdb_add = switchdev_port_fdb_add,
4099 .ndo_fdb_del = switchdev_port_fdb_del,
4100 .ndo_fdb_dump = switchdev_port_fdb_dump,
4101 .ndo_features_check = passthru_features_check,
4104 static const struct device_type bond_type = {
4108 static void bond_destructor(struct net_device *bond_dev)
4110 struct bonding *bond = netdev_priv(bond_dev);
4112 destroy_workqueue(bond->wq);
4113 free_netdev(bond_dev);
4116 void bond_setup(struct net_device *bond_dev)
4118 struct bonding *bond = netdev_priv(bond_dev);
4120 spin_lock_init(&bond->mode_lock);
4121 bond->params = bonding_defaults;
4123 /* Initialize pointers */
4124 bond->dev = bond_dev;
4126 /* Initialize the device entry points */
4127 ether_setup(bond_dev);
4128 bond_dev->netdev_ops = &bond_netdev_ops;
4129 bond_dev->ethtool_ops = &bond_ethtool_ops;
4131 bond_dev->destructor = bond_destructor;
4133 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4135 /* Initialize the device options */
4136 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4137 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4138 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4140 /* don't acquire bond device's netif_tx_lock when transmitting */
4141 bond_dev->features |= NETIF_F_LLTX;
4143 /* By default, we declare the bond to be fully
4144 * VLAN hardware accelerated capable. Special
4145 * care is taken in the various xmit functions
4146 * when there are slaves that are not hw accel
4150 /* Don't allow bond devices to change network namespaces. */
4151 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4153 bond_dev->hw_features = BOND_VLAN_FEATURES |
4154 NETIF_F_HW_VLAN_CTAG_TX |
4155 NETIF_F_HW_VLAN_CTAG_RX |
4156 NETIF_F_HW_VLAN_CTAG_FILTER;
4158 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4159 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4160 bond_dev->features |= bond_dev->hw_features;
4163 /* Destroy a bonding device.
4164 * Must be under rtnl_lock when this function is called.
4166 static void bond_uninit(struct net_device *bond_dev)
4168 struct bonding *bond = netdev_priv(bond_dev);
4169 struct list_head *iter;
4170 struct slave *slave;
4171 struct bond_up_slave *arr;
4173 bond_netpoll_cleanup(bond_dev);
4175 /* Release the bonded slaves */
4176 bond_for_each_slave(bond, slave, iter)
4177 __bond_release_one(bond_dev, slave->dev, true);
4178 netdev_info(bond_dev, "Released all slaves\n");
4180 arr = rtnl_dereference(bond->slave_arr);
4182 RCU_INIT_POINTER(bond->slave_arr, NULL);
4183 kfree_rcu(arr, rcu);
4186 list_del(&bond->bond_list);
4188 bond_debug_unregister(bond);
4191 /*------------------------- Module initialization ---------------------------*/
4193 static int bond_check_params(struct bond_params *params)
4195 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4196 struct bond_opt_value newval;
4197 const struct bond_opt_value *valptr;
4198 int arp_all_targets_value;
4199 u16 ad_actor_sys_prio = 0;
4200 u16 ad_user_port_key = 0;
4202 /* Convert string parameters. */
4204 bond_opt_initstr(&newval, mode);
4205 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4207 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4210 bond_mode = valptr->value;
4213 if (xmit_hash_policy) {
4214 if ((bond_mode != BOND_MODE_XOR) &&
4215 (bond_mode != BOND_MODE_8023AD) &&
4216 (bond_mode != BOND_MODE_TLB)) {
4217 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4218 bond_mode_name(bond_mode));
4220 bond_opt_initstr(&newval, xmit_hash_policy);
4221 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4224 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4228 xmit_hashtype = valptr->value;
4233 if (bond_mode != BOND_MODE_8023AD) {
4234 pr_info("lacp_rate param is irrelevant in mode %s\n",
4235 bond_mode_name(bond_mode));
4237 bond_opt_initstr(&newval, lacp_rate);
4238 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4241 pr_err("Error: Invalid lacp rate \"%s\"\n",
4245 lacp_fast = valptr->value;
4250 bond_opt_initstr(&newval, ad_select);
4251 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4254 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4257 params->ad_select = valptr->value;
4258 if (bond_mode != BOND_MODE_8023AD)
4259 pr_warn("ad_select param only affects 802.3ad mode\n");
4261 params->ad_select = BOND_AD_STABLE;
4264 if (max_bonds < 0) {
4265 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4266 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4267 max_bonds = BOND_DEFAULT_MAX_BONDS;
4271 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4277 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4282 if (downdelay < 0) {
4283 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4284 downdelay, INT_MAX);
4288 if ((use_carrier != 0) && (use_carrier != 1)) {
4289 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4294 if (num_peer_notif < 0 || num_peer_notif > 255) {
4295 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4300 /* reset values for 802.3ad/TLB/ALB */
4301 if (!bond_mode_uses_arp(bond_mode)) {
4303 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4304 pr_warn("Forcing miimon to 100msec\n");
4305 miimon = BOND_DEFAULT_MIIMON;
4309 if (tx_queues < 1 || tx_queues > 255) {
4310 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4311 tx_queues, BOND_DEFAULT_TX_QUEUES);
4312 tx_queues = BOND_DEFAULT_TX_QUEUES;
4315 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4316 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4318 all_slaves_active = 0;
4321 if (resend_igmp < 0 || resend_igmp > 255) {
4322 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4323 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4324 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4327 bond_opt_initval(&newval, packets_per_slave);
4328 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4329 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4330 packets_per_slave, USHRT_MAX);
4331 packets_per_slave = 1;
4334 if (bond_mode == BOND_MODE_ALB) {
4335 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",
4340 if (updelay || downdelay) {
4341 /* just warn the user the up/down delay will have
4342 * no effect since miimon is zero...
4344 pr_warn("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",
4345 updelay, downdelay);
4348 /* don't allow arp monitoring */
4350 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4351 miimon, arp_interval);
4355 if ((updelay % miimon) != 0) {
4356 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4357 updelay, miimon, (updelay / miimon) * miimon);
4362 if ((downdelay % miimon) != 0) {
4363 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4365 (downdelay / miimon) * miimon);
4368 downdelay /= miimon;
4371 if (arp_interval < 0) {
4372 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4373 arp_interval, INT_MAX);
4377 for (arp_ip_count = 0, i = 0;
4378 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4381 /* not a complete check, but good enough to catch mistakes */
4382 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4383 !bond_is_ip_target_ok(ip)) {
4384 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4388 if (bond_get_targets_ip(arp_target, ip) == -1)
4389 arp_target[arp_ip_count++] = ip;
4391 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4396 if (arp_interval && !arp_ip_count) {
4397 /* don't allow arping if no arp_ip_target given... */
4398 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4404 if (!arp_interval) {
4405 pr_err("arp_validate requires arp_interval\n");
4409 bond_opt_initstr(&newval, arp_validate);
4410 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4413 pr_err("Error: invalid arp_validate \"%s\"\n",
4417 arp_validate_value = valptr->value;
4419 arp_validate_value = 0;
4422 arp_all_targets_value = 0;
4423 if (arp_all_targets) {
4424 bond_opt_initstr(&newval, arp_all_targets);
4425 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4428 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4430 arp_all_targets_value = 0;
4432 arp_all_targets_value = valptr->value;
4437 pr_info("MII link monitoring set to %d ms\n", miimon);
4438 } else if (arp_interval) {
4439 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4440 arp_validate_value);
4441 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4442 arp_interval, valptr->string, arp_ip_count);
4444 for (i = 0; i < arp_ip_count; i++)
4445 pr_cont(" %s", arp_ip_target[i]);
4449 } else if (max_bonds) {
4450 /* miimon and arp_interval not set, we need one so things
4451 * work as expected, see bonding.txt for details
4453 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");
4456 if (primary && !bond_mode_uses_primary(bond_mode)) {
4457 /* currently, using a primary only makes sense
4458 * in active backup, TLB or ALB modes
4460 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4461 primary, bond_mode_name(bond_mode));
4465 if (primary && primary_reselect) {
4466 bond_opt_initstr(&newval, primary_reselect);
4467 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4470 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4474 primary_reselect_value = valptr->value;
4476 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4479 if (fail_over_mac) {
4480 bond_opt_initstr(&newval, fail_over_mac);
4481 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4484 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4488 fail_over_mac_value = valptr->value;
4489 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4490 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4492 fail_over_mac_value = BOND_FOM_NONE;
4495 bond_opt_initstr(&newval, "default");
4496 valptr = bond_opt_parse(
4497 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4500 pr_err("Error: No ad_actor_sys_prio default value");
4503 ad_actor_sys_prio = valptr->value;
4505 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4508 pr_err("Error: No ad_user_port_key default value");
4511 ad_user_port_key = valptr->value;
4513 if (lp_interval == 0) {
4514 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4515 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4516 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4519 /* fill params struct with the proper values */
4520 params->mode = bond_mode;
4521 params->xmit_policy = xmit_hashtype;
4522 params->miimon = miimon;
4523 params->num_peer_notif = num_peer_notif;
4524 params->arp_interval = arp_interval;
4525 params->arp_validate = arp_validate_value;
4526 params->arp_all_targets = arp_all_targets_value;
4527 params->updelay = updelay;
4528 params->downdelay = downdelay;
4529 params->use_carrier = use_carrier;
4530 params->lacp_fast = lacp_fast;
4531 params->primary[0] = 0;
4532 params->primary_reselect = primary_reselect_value;
4533 params->fail_over_mac = fail_over_mac_value;
4534 params->tx_queues = tx_queues;
4535 params->all_slaves_active = all_slaves_active;
4536 params->resend_igmp = resend_igmp;
4537 params->min_links = min_links;
4538 params->lp_interval = lp_interval;
4539 params->packets_per_slave = packets_per_slave;
4540 params->tlb_dynamic_lb = 1; /* Default value */
4541 params->ad_actor_sys_prio = ad_actor_sys_prio;
4542 eth_zero_addr(params->ad_actor_system);
4543 params->ad_user_port_key = ad_user_port_key;
4544 if (packets_per_slave > 0) {
4545 params->reciprocal_packets_per_slave =
4546 reciprocal_value(packets_per_slave);
4548 /* reciprocal_packets_per_slave is unused if
4549 * packets_per_slave is 0 or 1, just initialize it
4551 params->reciprocal_packets_per_slave =
4552 (struct reciprocal_value) { 0 };
4556 strncpy(params->primary, primary, IFNAMSIZ);
4557 params->primary[IFNAMSIZ - 1] = 0;
4560 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4565 static struct lock_class_key bonding_netdev_xmit_lock_key;
4566 static struct lock_class_key bonding_netdev_addr_lock_key;
4567 static struct lock_class_key bonding_tx_busylock_key;
4569 static void bond_set_lockdep_class_one(struct net_device *dev,
4570 struct netdev_queue *txq,
4573 lockdep_set_class(&txq->_xmit_lock,
4574 &bonding_netdev_xmit_lock_key);
4577 static void bond_set_lockdep_class(struct net_device *dev)
4579 lockdep_set_class(&dev->addr_list_lock,
4580 &bonding_netdev_addr_lock_key);
4581 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4582 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4585 /* Called from registration process */
4586 static int bond_init(struct net_device *bond_dev)
4588 struct bonding *bond = netdev_priv(bond_dev);
4589 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4591 netdev_dbg(bond_dev, "Begin bond_init\n");
4593 bond->wq = create_singlethread_workqueue(bond_dev->name);
4597 bond_set_lockdep_class(bond_dev);
4599 list_add_tail(&bond->bond_list, &bn->dev_list);
4601 bond_prepare_sysfs_group(bond);
4603 bond_debug_register(bond);
4605 /* Ensure valid dev_addr */
4606 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4607 bond_dev->addr_assign_type == NET_ADDR_PERM)
4608 eth_hw_addr_random(bond_dev);
4613 unsigned int bond_get_num_tx_queues(void)
4618 /* Create a new bond based on the specified name and bonding parameters.
4619 * If name is NULL, obtain a suitable "bond%d" name for us.
4620 * Caller must NOT hold rtnl_lock; we need to release it here before we
4621 * set up our sysfs entries.
4623 int bond_create(struct net *net, const char *name)
4625 struct net_device *bond_dev;
4626 struct bonding *bond;
4627 struct alb_bond_info *bond_info;
4632 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4633 name ? name : "bond%d", NET_NAME_UNKNOWN,
4634 bond_setup, tx_queues);
4636 pr_err("%s: eek! can't alloc netdev!\n", name);
4642 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4643 * It is set to 0 by default which is wrong.
4645 bond = netdev_priv(bond_dev);
4646 bond_info = &(BOND_ALB_INFO(bond));
4647 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4649 dev_net_set(bond_dev, net);
4650 bond_dev->rtnl_link_ops = &bond_link_ops;
4652 res = register_netdevice(bond_dev);
4654 netif_carrier_off(bond_dev);
4658 bond_destructor(bond_dev);
4662 static int __net_init bond_net_init(struct net *net)
4664 struct bond_net *bn = net_generic(net, bond_net_id);
4667 INIT_LIST_HEAD(&bn->dev_list);
4669 bond_create_proc_dir(bn);
4670 bond_create_sysfs(bn);
4675 static void __net_exit bond_net_exit(struct net *net)
4677 struct bond_net *bn = net_generic(net, bond_net_id);
4678 struct bonding *bond, *tmp_bond;
4681 bond_destroy_sysfs(bn);
4683 /* Kill off any bonds created after unregistering bond rtnl ops */
4685 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4686 unregister_netdevice_queue(bond->dev, &list);
4687 unregister_netdevice_many(&list);
4690 bond_destroy_proc_dir(bn);
4693 static struct pernet_operations bond_net_ops = {
4694 .init = bond_net_init,
4695 .exit = bond_net_exit,
4697 .size = sizeof(struct bond_net),
4700 static int __init bonding_init(void)
4705 pr_info("%s", bond_version);
4707 res = bond_check_params(&bonding_defaults);
4711 res = register_pernet_subsys(&bond_net_ops);
4715 res = bond_netlink_init();
4719 bond_create_debugfs();
4721 for (i = 0; i < max_bonds; i++) {
4722 res = bond_create(&init_net, NULL);
4727 register_netdevice_notifier(&bond_netdev_notifier);
4731 bond_destroy_debugfs();
4732 bond_netlink_fini();
4734 unregister_pernet_subsys(&bond_net_ops);
4739 static void __exit bonding_exit(void)
4741 unregister_netdevice_notifier(&bond_netdev_notifier);
4743 bond_destroy_debugfs();
4745 bond_netlink_fini();
4746 unregister_pernet_subsys(&bond_net_ops);
4748 #ifdef CONFIG_NET_POLL_CONTROLLER
4749 /* Make sure we don't have an imbalance on our netpoll blocking */
4750 WARN_ON(atomic_read(&netpoll_block_tx));
4754 module_init(bonding_init);
4755 module_exit(bonding_exit);
4756 MODULE_LICENSE("GPL");
4757 MODULE_VERSION(DRV_VERSION);
4758 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4759 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");