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/bonding.h>
81 #include <net/bond_3ad.h>
82 #include <net/bond_alb.h>
84 #include "bonding_priv.h"
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
96 static int use_carrier = 1;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *arp_all_targets;
108 static char *fail_over_mac;
109 static int all_slaves_active;
110 static struct bond_params bonding_defaults;
111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 static int packets_per_slave = 1;
113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125 module_param(miimon, int, 0);
126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127 module_param(updelay, int, 0);
128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129 module_param(downdelay, int, 0);
130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 module_param(use_carrier, int, 0);
133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135 module_param(mode, charp, 0);
136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140 module_param(primary, charp, 0);
141 MODULE_PARM_DESC(primary, "Primary network device to use");
142 module_param(primary_reselect, charp, 0);
143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
148 "2 for only on active slave "
150 module_param(lacp_rate, charp, 0);
151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153 module_param(ad_select, charp, 0);
154 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
157 module_param(min_links, int, 0);
158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160 module_param(xmit_hash_policy, charp, 0);
161 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165 module_param(arp_interval, int, 0);
166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167 module_param_array(arp_ip_target, charp, NULL, 0);
168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169 module_param(arp_validate, charp, 0);
170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173 module_param(arp_all_targets, charp, 0);
174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac, charp, 0);
176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179 module_param(all_slaves_active, int, 0);
180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183 module_param(resend_igmp, int, 0);
184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 module_param(packets_per_slave, int, 0);
187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190 module_param(lp_interval, uint, 0);
191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
195 /*----------------------------- Global variables ----------------------------*/
197 #ifdef CONFIG_NET_POLL_CONTROLLER
198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
201 unsigned int bond_net_id __read_mostly;
203 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
205 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
206 .offset = offsetof(struct flow_keys, control),
209 .key_id = FLOW_DISSECTOR_KEY_BASIC,
210 .offset = offsetof(struct flow_keys, basic),
213 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
214 .offset = offsetof(struct flow_keys, addrs.v4addrs),
217 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
218 .offset = offsetof(struct flow_keys, addrs.v6addrs),
221 .key_id = FLOW_DISSECTOR_KEY_TIPC,
222 .offset = offsetof(struct flow_keys, addrs.tipckey),
225 .key_id = FLOW_DISSECTOR_KEY_PORTS,
226 .offset = offsetof(struct flow_keys, ports),
229 .key_id = FLOW_DISSECTOR_KEY_ICMP,
230 .offset = offsetof(struct flow_keys, icmp),
233 .key_id = FLOW_DISSECTOR_KEY_VLAN,
234 .offset = offsetof(struct flow_keys, vlan),
237 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
238 .offset = offsetof(struct flow_keys, tags),
241 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
242 .offset = offsetof(struct flow_keys, keyid),
246 static struct flow_dissector flow_keys_bonding __read_mostly;
248 /*-------------------------- Forward declarations ---------------------------*/
250 static int bond_init(struct net_device *bond_dev);
251 static void bond_uninit(struct net_device *bond_dev);
252 static void bond_get_stats(struct net_device *bond_dev,
253 struct rtnl_link_stats64 *stats);
254 static void bond_slave_arr_handler(struct work_struct *work);
255 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
257 static void bond_netdev_notify_work(struct work_struct *work);
259 /*---------------------------- General routines -----------------------------*/
261 const char *bond_mode_name(int mode)
263 static const char *names[] = {
264 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
265 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
266 [BOND_MODE_XOR] = "load balancing (xor)",
267 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
268 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
269 [BOND_MODE_TLB] = "transmit load balancing",
270 [BOND_MODE_ALB] = "adaptive load balancing",
273 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
279 /*---------------------------------- VLAN -----------------------------------*/
282 * bond_dev_queue_xmit - Prepare skb for xmit.
284 * @bond: bond device that got this skb for tx.
285 * @skb: hw accel VLAN tagged skb to transmit
286 * @slave_dev: slave that is supposed to xmit this skbuff
288 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
289 struct net_device *slave_dev)
291 skb->dev = slave_dev;
293 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
294 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
295 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
297 if (unlikely(netpoll_tx_running(bond->dev)))
298 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
303 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
304 * We don't protect the slave list iteration with a lock because:
305 * a. This operation is performed in IOCTL context,
306 * b. The operation is protected by the RTNL semaphore in the 8021q code,
307 * c. Holding a lock with BH disabled while directly calling a base driver
308 * entry point is generally a BAD idea.
310 * The design of synchronization/protection for this operation in the 8021q
311 * module is good for one or more VLAN devices over a single physical device
312 * and cannot be extended for a teaming solution like bonding, so there is a
313 * potential race condition here where a net device from the vlan group might
314 * be referenced (either by a base driver or the 8021q code) while it is being
315 * removed from the system. However, it turns out we're not making matters
316 * worse, and if it works for regular VLAN usage it will work here too.
320 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
321 * @bond_dev: bonding net device that got called
322 * @vid: vlan id being added
324 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
325 __be16 proto, u16 vid)
327 struct bonding *bond = netdev_priv(bond_dev);
328 struct slave *slave, *rollback_slave;
329 struct list_head *iter;
332 bond_for_each_slave(bond, slave, iter) {
333 res = vlan_vid_add(slave->dev, proto, vid);
341 /* unwind to the slave that failed */
342 bond_for_each_slave(bond, rollback_slave, iter) {
343 if (rollback_slave == slave)
346 vlan_vid_del(rollback_slave->dev, proto, vid);
353 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
354 * @bond_dev: bonding net device that got called
355 * @vid: vlan id being removed
357 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
358 __be16 proto, u16 vid)
360 struct bonding *bond = netdev_priv(bond_dev);
361 struct list_head *iter;
364 bond_for_each_slave(bond, slave, iter)
365 vlan_vid_del(slave->dev, proto, vid);
367 if (bond_is_lb(bond))
368 bond_alb_clear_vlan(bond, vid);
373 /*------------------------------- Link status -------------------------------*/
375 /* Set the carrier state for the master according to the state of its
376 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
377 * do special 802.3ad magic.
379 * Returns zero if carrier state does not change, nonzero if it does.
381 int bond_set_carrier(struct bonding *bond)
383 struct list_head *iter;
386 if (!bond_has_slaves(bond))
389 if (BOND_MODE(bond) == BOND_MODE_8023AD)
390 return bond_3ad_set_carrier(bond);
392 bond_for_each_slave(bond, slave, iter) {
393 if (slave->link == BOND_LINK_UP) {
394 if (!netif_carrier_ok(bond->dev)) {
395 netif_carrier_on(bond->dev);
403 if (netif_carrier_ok(bond->dev)) {
404 netif_carrier_off(bond->dev);
410 /* Get link speed and duplex from the slave's base driver
411 * using ethtool. If for some reason the call fails or the
412 * values are invalid, set speed and duplex to -1,
413 * and return. Return 1 if speed or duplex settings are
414 * UNKNOWN; 0 otherwise.
416 static int bond_update_speed_duplex(struct slave *slave)
418 struct net_device *slave_dev = slave->dev;
419 struct ethtool_link_ksettings ecmd;
422 slave->speed = SPEED_UNKNOWN;
423 slave->duplex = DUPLEX_UNKNOWN;
425 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
428 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
430 switch (ecmd.base.duplex) {
438 slave->speed = ecmd.base.speed;
439 slave->duplex = ecmd.base.duplex;
444 const char *bond_slave_link_status(s8 link)
460 /* if <dev> supports MII link status reporting, check its link status.
462 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
463 * depending upon the setting of the use_carrier parameter.
465 * Return either BMSR_LSTATUS, meaning that the link is up (or we
466 * can't tell and just pretend it is), or 0, meaning that the link is
469 * If reporting is non-zero, instead of faking link up, return -1 if
470 * both ETHTOOL and MII ioctls fail (meaning the device does not
471 * support them). If use_carrier is set, return whatever it says.
472 * It'd be nice if there was a good way to tell if a driver supports
473 * netif_carrier, but there really isn't.
475 static int bond_check_dev_link(struct bonding *bond,
476 struct net_device *slave_dev, int reporting)
478 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
479 int (*ioctl)(struct net_device *, struct ifreq *, int);
481 struct mii_ioctl_data *mii;
483 if (!reporting && !netif_running(slave_dev))
486 if (bond->params.use_carrier)
487 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
489 /* Try to get link status using Ethtool first. */
490 if (slave_dev->ethtool_ops->get_link)
491 return slave_dev->ethtool_ops->get_link(slave_dev) ?
494 /* Ethtool can't be used, fallback to MII ioctls. */
495 ioctl = slave_ops->ndo_do_ioctl;
497 /* TODO: set pointer to correct ioctl on a per team member
498 * bases to make this more efficient. that is, once
499 * we determine the correct ioctl, we will always
500 * call it and not the others for that team
504 /* We cannot assume that SIOCGMIIPHY will also read a
505 * register; not all network drivers (e.g., e100)
509 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
510 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
512 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
513 mii->reg_num = MII_BMSR;
514 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
515 return mii->val_out & BMSR_LSTATUS;
519 /* If reporting, report that either there's no dev->do_ioctl,
520 * or both SIOCGMIIREG and get_link failed (meaning that we
521 * cannot report link status). If not reporting, pretend
524 return reporting ? -1 : BMSR_LSTATUS;
527 /*----------------------------- Multicast list ------------------------------*/
529 /* Push the promiscuity flag down to appropriate slaves */
530 static int bond_set_promiscuity(struct bonding *bond, int inc)
532 struct list_head *iter;
535 if (bond_uses_primary(bond)) {
536 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
539 err = dev_set_promiscuity(curr_active->dev, inc);
543 bond_for_each_slave(bond, slave, iter) {
544 err = dev_set_promiscuity(slave->dev, inc);
552 /* Push the allmulti flag down to all slaves */
553 static int bond_set_allmulti(struct bonding *bond, int inc)
555 struct list_head *iter;
558 if (bond_uses_primary(bond)) {
559 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
562 err = dev_set_allmulti(curr_active->dev, inc);
566 bond_for_each_slave(bond, slave, iter) {
567 err = dev_set_allmulti(slave->dev, inc);
575 /* Retrieve the list of registered multicast addresses for the bonding
576 * device and retransmit an IGMP JOIN request to the current active
579 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
581 struct bonding *bond = container_of(work, struct bonding,
584 if (!rtnl_trylock()) {
585 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
588 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
590 if (bond->igmp_retrans > 1) {
591 bond->igmp_retrans--;
592 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
597 /* Flush bond's hardware addresses from slave */
598 static void bond_hw_addr_flush(struct net_device *bond_dev,
599 struct net_device *slave_dev)
601 struct bonding *bond = netdev_priv(bond_dev);
603 dev_uc_unsync(slave_dev, bond_dev);
604 dev_mc_unsync(slave_dev, bond_dev);
606 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
607 /* del lacpdu mc addr from mc list */
608 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
610 dev_mc_del(slave_dev, lacpdu_multicast);
614 /*--------------------------- Active slave change ---------------------------*/
616 /* Update the hardware address list and promisc/allmulti for the new and
617 * old active slaves (if any). Modes that are not using primary keep all
618 * slaves up date at all times; only the modes that use primary need to call
619 * this function to swap these settings during a failover.
621 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
622 struct slave *old_active)
625 if (bond->dev->flags & IFF_PROMISC)
626 dev_set_promiscuity(old_active->dev, -1);
628 if (bond->dev->flags & IFF_ALLMULTI)
629 dev_set_allmulti(old_active->dev, -1);
631 bond_hw_addr_flush(bond->dev, old_active->dev);
635 /* FIXME: Signal errors upstream. */
636 if (bond->dev->flags & IFF_PROMISC)
637 dev_set_promiscuity(new_active->dev, 1);
639 if (bond->dev->flags & IFF_ALLMULTI)
640 dev_set_allmulti(new_active->dev, 1);
642 netif_addr_lock_bh(bond->dev);
643 dev_uc_sync(new_active->dev, bond->dev);
644 dev_mc_sync(new_active->dev, bond->dev);
645 netif_addr_unlock_bh(bond->dev);
650 * bond_set_dev_addr - clone slave's address to bond
651 * @bond_dev: bond net device
652 * @slave_dev: slave net device
654 * Should be called with RTNL held.
656 static int bond_set_dev_addr(struct net_device *bond_dev,
657 struct net_device *slave_dev)
661 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
662 bond_dev, slave_dev, slave_dev->addr_len);
663 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
667 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
668 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
669 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
673 static struct slave *bond_get_old_active(struct bonding *bond,
674 struct slave *new_active)
677 struct list_head *iter;
679 bond_for_each_slave(bond, slave, iter) {
680 if (slave == new_active)
683 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
690 /* bond_do_fail_over_mac
692 * Perform special MAC address swapping for fail_over_mac settings
696 static void bond_do_fail_over_mac(struct bonding *bond,
697 struct slave *new_active,
698 struct slave *old_active)
700 u8 tmp_mac[MAX_ADDR_LEN];
701 struct sockaddr_storage ss;
704 switch (bond->params.fail_over_mac) {
705 case BOND_FOM_ACTIVE:
707 rv = bond_set_dev_addr(bond->dev, new_active->dev);
709 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
713 case BOND_FOM_FOLLOW:
714 /* if new_active && old_active, swap them
715 * if just old_active, do nothing (going to no active slave)
716 * if just new_active, set new_active to bond's MAC
722 old_active = bond_get_old_active(bond, new_active);
725 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
726 new_active->dev->addr_len);
727 bond_hw_addr_copy(ss.__data,
728 old_active->dev->dev_addr,
729 old_active->dev->addr_len);
730 ss.ss_family = new_active->dev->type;
732 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
733 bond->dev->addr_len);
734 ss.ss_family = bond->dev->type;
737 rv = dev_set_mac_address(new_active->dev,
738 (struct sockaddr *)&ss, NULL);
740 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
748 bond_hw_addr_copy(ss.__data, tmp_mac,
749 new_active->dev->addr_len);
750 ss.ss_family = old_active->dev->type;
752 rv = dev_set_mac_address(old_active->dev,
753 (struct sockaddr *)&ss, NULL);
755 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
760 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
761 bond->params.fail_over_mac);
767 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
769 struct slave *prim = rtnl_dereference(bond->primary_slave);
770 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
772 if (!prim || prim->link != BOND_LINK_UP) {
773 if (!curr || curr->link != BOND_LINK_UP)
778 if (bond->force_primary) {
779 bond->force_primary = false;
783 if (!curr || curr->link != BOND_LINK_UP)
786 /* At this point, prim and curr are both up */
787 switch (bond->params.primary_reselect) {
788 case BOND_PRI_RESELECT_ALWAYS:
790 case BOND_PRI_RESELECT_BETTER:
791 if (prim->speed < curr->speed)
793 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
796 case BOND_PRI_RESELECT_FAILURE:
799 netdev_err(bond->dev, "impossible primary_reselect %d\n",
800 bond->params.primary_reselect);
806 * bond_find_best_slave - select the best available slave to be the active one
807 * @bond: our bonding struct
809 static struct slave *bond_find_best_slave(struct bonding *bond)
811 struct slave *slave, *bestslave = NULL;
812 struct list_head *iter;
813 int mintime = bond->params.updelay;
815 slave = bond_choose_primary_or_current(bond);
819 bond_for_each_slave(bond, slave, iter) {
820 if (slave->link == BOND_LINK_UP)
822 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
823 slave->delay < mintime) {
824 mintime = slave->delay;
832 static bool bond_should_notify_peers(struct bonding *bond)
837 slave = rcu_dereference(bond->curr_active_slave);
840 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
841 slave ? slave->dev->name : "NULL");
843 if (!slave || !bond->send_peer_notif ||
844 bond->send_peer_notif %
845 max(1, bond->params.peer_notif_delay) != 0 ||
846 !netif_carrier_ok(bond->dev) ||
847 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
854 * change_active_interface - change the active slave into the specified one
855 * @bond: our bonding struct
856 * @new: the new slave to make the active one
858 * Set the new slave to the bond's settings and unset them on the old
860 * Setting include flags, mc-list, promiscuity, allmulti, etc.
862 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
863 * because it is apparently the best available slave we have, even though its
864 * updelay hasn't timed out yet.
866 * Caller must hold RTNL.
868 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
870 struct slave *old_active;
874 old_active = rtnl_dereference(bond->curr_active_slave);
876 if (old_active == new_active)
880 new_active->last_link_up = jiffies;
882 if (new_active->link == BOND_LINK_BACK) {
883 if (bond_uses_primary(bond)) {
884 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
885 (bond->params.updelay - new_active->delay) * bond->params.miimon);
888 new_active->delay = 0;
889 bond_set_slave_link_state(new_active, BOND_LINK_UP,
890 BOND_SLAVE_NOTIFY_NOW);
892 if (BOND_MODE(bond) == BOND_MODE_8023AD)
893 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
895 if (bond_is_lb(bond))
896 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
898 if (bond_uses_primary(bond)) {
899 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
904 if (bond_uses_primary(bond))
905 bond_hw_addr_swap(bond, new_active, old_active);
907 if (bond_is_lb(bond)) {
908 bond_alb_handle_active_change(bond, new_active);
910 bond_set_slave_inactive_flags(old_active,
911 BOND_SLAVE_NOTIFY_NOW);
913 bond_set_slave_active_flags(new_active,
914 BOND_SLAVE_NOTIFY_NOW);
916 rcu_assign_pointer(bond->curr_active_slave, new_active);
919 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
921 bond_set_slave_inactive_flags(old_active,
922 BOND_SLAVE_NOTIFY_NOW);
925 bool should_notify_peers = false;
927 bond_set_slave_active_flags(new_active,
928 BOND_SLAVE_NOTIFY_NOW);
930 if (bond->params.fail_over_mac)
931 bond_do_fail_over_mac(bond, new_active,
934 if (netif_running(bond->dev)) {
935 bond->send_peer_notif =
936 bond->params.num_peer_notif *
937 max(1, bond->params.peer_notif_delay);
938 should_notify_peers =
939 bond_should_notify_peers(bond);
942 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
943 if (should_notify_peers) {
944 bond->send_peer_notif--;
945 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
951 /* resend IGMP joins since active slave has changed or
952 * all were sent on curr_active_slave.
953 * resend only if bond is brought up with the affected
954 * bonding modes and the retransmission is enabled
956 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
957 ((bond_uses_primary(bond) && new_active) ||
958 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
959 bond->igmp_retrans = bond->params.resend_igmp;
960 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
965 * bond_select_active_slave - select a new active slave, if needed
966 * @bond: our bonding struct
968 * This functions should be called when one of the following occurs:
969 * - The old curr_active_slave has been released or lost its link.
970 * - The primary_slave has got its link back.
971 * - A slave has got its link back and there's no old curr_active_slave.
973 * Caller must hold RTNL.
975 void bond_select_active_slave(struct bonding *bond)
977 struct slave *best_slave;
982 best_slave = bond_find_best_slave(bond);
983 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
984 bond_change_active_slave(bond, best_slave);
985 rv = bond_set_carrier(bond);
989 if (netif_carrier_ok(bond->dev))
990 netdev_info(bond->dev, "active interface up!\n");
992 netdev_info(bond->dev, "now running without any active interface!\n");
996 #ifdef CONFIG_NET_POLL_CONTROLLER
997 static inline int slave_enable_netpoll(struct slave *slave)
1002 np = kzalloc(sizeof(*np), GFP_KERNEL);
1007 err = __netpoll_setup(np, slave->dev);
1016 static inline void slave_disable_netpoll(struct slave *slave)
1018 struct netpoll *np = slave->np;
1028 static void bond_poll_controller(struct net_device *bond_dev)
1030 struct bonding *bond = netdev_priv(bond_dev);
1031 struct slave *slave = NULL;
1032 struct list_head *iter;
1033 struct ad_info ad_info;
1035 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1036 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1039 bond_for_each_slave_rcu(bond, slave, iter) {
1040 if (!bond_slave_is_up(slave))
1043 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1044 struct aggregator *agg =
1045 SLAVE_AD_INFO(slave)->port.aggregator;
1048 agg->aggregator_identifier != ad_info.aggregator_id)
1052 netpoll_poll_dev(slave->dev);
1056 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1058 struct bonding *bond = netdev_priv(bond_dev);
1059 struct list_head *iter;
1060 struct slave *slave;
1062 bond_for_each_slave(bond, slave, iter)
1063 if (bond_slave_is_up(slave))
1064 slave_disable_netpoll(slave);
1067 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1069 struct bonding *bond = netdev_priv(dev);
1070 struct list_head *iter;
1071 struct slave *slave;
1074 bond_for_each_slave(bond, slave, iter) {
1075 err = slave_enable_netpoll(slave);
1077 bond_netpoll_cleanup(dev);
1084 static inline int slave_enable_netpoll(struct slave *slave)
1088 static inline void slave_disable_netpoll(struct slave *slave)
1091 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1096 /*---------------------------------- IOCTL ----------------------------------*/
1098 static netdev_features_t bond_fix_features(struct net_device *dev,
1099 netdev_features_t features)
1101 struct bonding *bond = netdev_priv(dev);
1102 struct list_head *iter;
1103 netdev_features_t mask;
1104 struct slave *slave;
1108 features &= ~NETIF_F_ONE_FOR_ALL;
1109 features |= NETIF_F_ALL_FOR_ALL;
1111 bond_for_each_slave(bond, slave, iter) {
1112 features = netdev_increment_features(features,
1113 slave->dev->features,
1116 features = netdev_add_tso_features(features, mask);
1121 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1122 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1123 NETIF_F_HIGHDMA | NETIF_F_LRO)
1125 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1126 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1128 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1131 static void bond_compute_features(struct bonding *bond)
1133 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1134 IFF_XMIT_DST_RELEASE_PERM;
1135 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1136 netdev_features_t enc_features = BOND_ENC_FEATURES;
1137 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1138 struct net_device *bond_dev = bond->dev;
1139 struct list_head *iter;
1140 struct slave *slave;
1141 unsigned short max_hard_header_len = ETH_HLEN;
1142 unsigned int gso_max_size = GSO_MAX_SIZE;
1143 u16 gso_max_segs = GSO_MAX_SEGS;
1145 if (!bond_has_slaves(bond))
1147 vlan_features &= NETIF_F_ALL_FOR_ALL;
1148 mpls_features &= NETIF_F_ALL_FOR_ALL;
1150 bond_for_each_slave(bond, slave, iter) {
1151 vlan_features = netdev_increment_features(vlan_features,
1152 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1154 enc_features = netdev_increment_features(enc_features,
1155 slave->dev->hw_enc_features,
1158 mpls_features = netdev_increment_features(mpls_features,
1159 slave->dev->mpls_features,
1160 BOND_MPLS_FEATURES);
1162 dst_release_flag &= slave->dev->priv_flags;
1163 if (slave->dev->hard_header_len > max_hard_header_len)
1164 max_hard_header_len = slave->dev->hard_header_len;
1166 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1167 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1169 bond_dev->hard_header_len = max_hard_header_len;
1172 bond_dev->vlan_features = vlan_features;
1173 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1174 NETIF_F_HW_VLAN_CTAG_TX |
1175 NETIF_F_HW_VLAN_STAG_TX |
1177 bond_dev->mpls_features = mpls_features;
1178 bond_dev->gso_max_segs = gso_max_segs;
1179 netif_set_gso_max_size(bond_dev, gso_max_size);
1181 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1182 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1183 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1184 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1186 netdev_change_features(bond_dev);
1189 static void bond_setup_by_slave(struct net_device *bond_dev,
1190 struct net_device *slave_dev)
1192 bond_dev->header_ops = slave_dev->header_ops;
1194 bond_dev->type = slave_dev->type;
1195 bond_dev->hard_header_len = slave_dev->hard_header_len;
1196 bond_dev->addr_len = slave_dev->addr_len;
1198 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1199 slave_dev->addr_len);
1202 /* On bonding slaves other than the currently active slave, suppress
1203 * duplicates except for alb non-mcast/bcast.
1205 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1206 struct slave *slave,
1207 struct bonding *bond)
1209 if (bond_is_slave_inactive(slave)) {
1210 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1211 skb->pkt_type != PACKET_BROADCAST &&
1212 skb->pkt_type != PACKET_MULTICAST)
1219 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1221 struct sk_buff *skb = *pskb;
1222 struct slave *slave;
1223 struct bonding *bond;
1224 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1226 int ret = RX_HANDLER_ANOTHER;
1228 skb = skb_share_check(skb, GFP_ATOMIC);
1230 return RX_HANDLER_CONSUMED;
1234 slave = bond_slave_get_rcu(skb->dev);
1237 recv_probe = READ_ONCE(bond->recv_probe);
1239 ret = recv_probe(skb, bond, slave);
1240 if (ret == RX_HANDLER_CONSUMED) {
1247 * For packets determined by bond_should_deliver_exact_match() call to
1248 * be suppressed we want to make an exception for link-local packets.
1249 * This is necessary for e.g. LLDP daemons to be able to monitor
1250 * inactive slave links without being forced to bind to them
1253 * At the same time, packets that are passed to the bonding master
1254 * (including link-local ones) can have their originating interface
1255 * determined via PACKET_ORIGDEV socket option.
1257 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1258 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1259 return RX_HANDLER_PASS;
1260 return RX_HANDLER_EXACT;
1263 skb->dev = bond->dev;
1265 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1266 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1267 skb->pkt_type == PACKET_HOST) {
1269 if (unlikely(skb_cow_head(skb,
1270 skb->data - skb_mac_header(skb)))) {
1272 return RX_HANDLER_CONSUMED;
1274 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1275 bond->dev->addr_len);
1281 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1283 switch (BOND_MODE(bond)) {
1284 case BOND_MODE_ROUNDROBIN:
1285 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1286 case BOND_MODE_ACTIVEBACKUP:
1287 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1288 case BOND_MODE_BROADCAST:
1289 return NETDEV_LAG_TX_TYPE_BROADCAST;
1291 case BOND_MODE_8023AD:
1292 return NETDEV_LAG_TX_TYPE_HASH;
1294 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1298 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1299 enum netdev_lag_tx_type type)
1301 if (type != NETDEV_LAG_TX_TYPE_HASH)
1302 return NETDEV_LAG_HASH_NONE;
1304 switch (bond->params.xmit_policy) {
1305 case BOND_XMIT_POLICY_LAYER2:
1306 return NETDEV_LAG_HASH_L2;
1307 case BOND_XMIT_POLICY_LAYER34:
1308 return NETDEV_LAG_HASH_L34;
1309 case BOND_XMIT_POLICY_LAYER23:
1310 return NETDEV_LAG_HASH_L23;
1311 case BOND_XMIT_POLICY_ENCAP23:
1312 return NETDEV_LAG_HASH_E23;
1313 case BOND_XMIT_POLICY_ENCAP34:
1314 return NETDEV_LAG_HASH_E34;
1316 return NETDEV_LAG_HASH_UNKNOWN;
1320 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1321 struct netlink_ext_ack *extack)
1323 struct netdev_lag_upper_info lag_upper_info;
1324 enum netdev_lag_tx_type type;
1326 type = bond_lag_tx_type(bond);
1327 lag_upper_info.tx_type = type;
1328 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1330 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1331 &lag_upper_info, extack);
1334 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1336 netdev_upper_dev_unlink(slave->dev, bond->dev);
1337 slave->dev->flags &= ~IFF_SLAVE;
1340 static struct slave *bond_alloc_slave(struct bonding *bond)
1342 struct slave *slave = NULL;
1344 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1348 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1349 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1351 if (!SLAVE_AD_INFO(slave)) {
1356 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1361 static void bond_free_slave(struct slave *slave)
1363 struct bonding *bond = bond_get_bond_by_slave(slave);
1365 cancel_delayed_work_sync(&slave->notify_work);
1366 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1367 kfree(SLAVE_AD_INFO(slave));
1372 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1374 info->bond_mode = BOND_MODE(bond);
1375 info->miimon = bond->params.miimon;
1376 info->num_slaves = bond->slave_cnt;
1379 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1381 strcpy(info->slave_name, slave->dev->name);
1382 info->link = slave->link;
1383 info->state = bond_slave_state(slave);
1384 info->link_failure_count = slave->link_failure_count;
1387 static void bond_netdev_notify_work(struct work_struct *_work)
1389 struct slave *slave = container_of(_work, struct slave,
1392 if (rtnl_trylock()) {
1393 struct netdev_bonding_info binfo;
1395 bond_fill_ifslave(slave, &binfo.slave);
1396 bond_fill_ifbond(slave->bond, &binfo.master);
1397 netdev_bonding_info_change(slave->dev, &binfo);
1400 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1404 void bond_queue_slave_event(struct slave *slave)
1406 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1409 void bond_lower_state_changed(struct slave *slave)
1411 struct netdev_lag_lower_state_info info;
1413 info.link_up = slave->link == BOND_LINK_UP ||
1414 slave->link == BOND_LINK_FAIL;
1415 info.tx_enabled = bond_is_active_slave(slave);
1416 netdev_lower_state_changed(slave->dev, &info);
1419 /* enslave device <slave> to bond device <master> */
1420 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1421 struct netlink_ext_ack *extack)
1423 struct bonding *bond = netdev_priv(bond_dev);
1424 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1425 struct slave *new_slave = NULL, *prev_slave;
1426 struct sockaddr_storage ss;
1430 if (!bond->params.use_carrier &&
1431 slave_dev->ethtool_ops->get_link == NULL &&
1432 slave_ops->ndo_do_ioctl == NULL) {
1433 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1436 /* already in-use? */
1437 if (netdev_is_rx_handler_busy(slave_dev)) {
1438 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1439 slave_err(bond_dev, slave_dev,
1440 "Error: Device is in use and cannot be enslaved\n");
1444 if (bond_dev == slave_dev) {
1445 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1446 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1450 /* vlan challenged mutual exclusion */
1451 /* no need to lock since we're protected by rtnl_lock */
1452 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1453 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1454 if (vlan_uses_dev(bond_dev)) {
1455 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1456 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1459 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1462 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1465 /* Old ifenslave binaries are no longer supported. These can
1466 * be identified with moderate accuracy by the state of the slave:
1467 * the current ifenslave will set the interface down prior to
1468 * enslaving it; the old ifenslave will not.
1470 if (slave_dev->flags & IFF_UP) {
1471 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1472 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1476 /* set bonding device ether type by slave - bonding netdevices are
1477 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1478 * there is a need to override some of the type dependent attribs/funcs.
1480 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1481 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1483 if (!bond_has_slaves(bond)) {
1484 if (bond_dev->type != slave_dev->type) {
1485 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1486 bond_dev->type, slave_dev->type);
1488 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1490 res = notifier_to_errno(res);
1492 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1496 /* Flush unicast and multicast addresses */
1497 dev_uc_flush(bond_dev);
1498 dev_mc_flush(bond_dev);
1500 if (slave_dev->type != ARPHRD_ETHER)
1501 bond_setup_by_slave(bond_dev, slave_dev);
1503 ether_setup(bond_dev);
1504 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1507 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1510 } else if (bond_dev->type != slave_dev->type) {
1511 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1512 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1513 slave_dev->type, bond_dev->type);
1517 if (slave_dev->type == ARPHRD_INFINIBAND &&
1518 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1519 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1520 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1523 goto err_undo_flags;
1526 if (!slave_ops->ndo_set_mac_address ||
1527 slave_dev->type == ARPHRD_INFINIBAND) {
1528 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1529 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1530 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1531 if (!bond_has_slaves(bond)) {
1532 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1533 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1535 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1536 slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1538 goto err_undo_flags;
1543 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1545 /* If this is the first slave, then we need to set the master's hardware
1546 * address to be the same as the slave's.
1548 if (!bond_has_slaves(bond) &&
1549 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1550 res = bond_set_dev_addr(bond->dev, slave_dev);
1552 goto err_undo_flags;
1555 new_slave = bond_alloc_slave(bond);
1558 goto err_undo_flags;
1561 new_slave->bond = bond;
1562 new_slave->dev = slave_dev;
1563 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1564 * is set via sysfs or module option if desired.
1566 new_slave->queue_id = 0;
1568 /* Save slave's original mtu and then set it to match the bond */
1569 new_slave->original_mtu = slave_dev->mtu;
1570 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1572 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1576 /* Save slave's original ("permanent") mac address for modes
1577 * that need it, and for restoring it upon release, and then
1578 * set it to the master's address
1580 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1581 slave_dev->addr_len);
1583 if (!bond->params.fail_over_mac ||
1584 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1585 /* Set slave to master's mac address. The application already
1586 * set the master's mac address to that of the first slave
1588 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1589 ss.ss_family = slave_dev->type;
1590 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1593 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1594 goto err_restore_mtu;
1598 /* set slave flag before open to prevent IPv6 addrconf */
1599 slave_dev->flags |= IFF_SLAVE;
1601 /* open the slave since the application closed it */
1602 res = dev_open(slave_dev, extack);
1604 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1605 goto err_restore_mac;
1608 slave_dev->priv_flags |= IFF_BONDING;
1609 /* initialize slave stats */
1610 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1612 if (bond_is_lb(bond)) {
1613 /* bond_alb_init_slave() must be called before all other stages since
1614 * it might fail and we do not want to have to undo everything
1616 res = bond_alb_init_slave(bond, new_slave);
1621 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1623 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1627 prev_slave = bond_last_slave(bond);
1629 new_slave->delay = 0;
1630 new_slave->link_failure_count = 0;
1632 if (bond_update_speed_duplex(new_slave) &&
1633 bond_needs_speed_duplex(bond))
1634 new_slave->link = BOND_LINK_DOWN;
1636 new_slave->last_rx = jiffies -
1637 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1638 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1639 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1641 if (bond->params.miimon && !bond->params.use_carrier) {
1642 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1644 if ((link_reporting == -1) && !bond->params.arp_interval) {
1645 /* miimon is set but a bonded network driver
1646 * does not support ETHTOOL/MII and
1647 * arp_interval is not set. Note: if
1648 * use_carrier is enabled, we will never go
1649 * here (because netif_carrier is always
1650 * supported); thus, we don't need to change
1651 * the messages for netif_carrier.
1653 slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n");
1654 } else if (link_reporting == -1) {
1655 /* unable get link status using mii/ethtool */
1656 slave_warn(bond_dev, slave_dev, "can't get link status from slave; 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");
1660 /* check for initial state */
1661 new_slave->link = BOND_LINK_NOCHANGE;
1662 if (bond->params.miimon) {
1663 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1664 if (bond->params.updelay) {
1665 bond_set_slave_link_state(new_slave,
1667 BOND_SLAVE_NOTIFY_NOW);
1668 new_slave->delay = bond->params.updelay;
1670 bond_set_slave_link_state(new_slave,
1672 BOND_SLAVE_NOTIFY_NOW);
1675 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1676 BOND_SLAVE_NOTIFY_NOW);
1678 } else if (bond->params.arp_interval) {
1679 bond_set_slave_link_state(new_slave,
1680 (netif_carrier_ok(slave_dev) ?
1681 BOND_LINK_UP : BOND_LINK_DOWN),
1682 BOND_SLAVE_NOTIFY_NOW);
1684 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1685 BOND_SLAVE_NOTIFY_NOW);
1688 if (new_slave->link != BOND_LINK_DOWN)
1689 new_slave->last_link_up = jiffies;
1690 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1691 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1692 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1694 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1695 /* if there is a primary slave, remember it */
1696 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1697 rcu_assign_pointer(bond->primary_slave, new_slave);
1698 bond->force_primary = true;
1702 switch (BOND_MODE(bond)) {
1703 case BOND_MODE_ACTIVEBACKUP:
1704 bond_set_slave_inactive_flags(new_slave,
1705 BOND_SLAVE_NOTIFY_NOW);
1707 case BOND_MODE_8023AD:
1708 /* in 802.3ad mode, the internal mechanism
1709 * will activate the slaves in the selected
1712 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1713 /* if this is the first slave */
1715 SLAVE_AD_INFO(new_slave)->id = 1;
1716 /* Initialize AD with the number of times that the AD timer is called in 1 second
1717 * can be called only after the mac address of the bond is set
1719 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1721 SLAVE_AD_INFO(new_slave)->id =
1722 SLAVE_AD_INFO(prev_slave)->id + 1;
1725 bond_3ad_bind_slave(new_slave);
1729 bond_set_active_slave(new_slave);
1730 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1733 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
1735 /* always active in trunk mode */
1736 bond_set_active_slave(new_slave);
1738 /* In trunking mode there is little meaning to curr_active_slave
1739 * anyway (it holds no special properties of the bond device),
1740 * so we can change it without calling change_active_interface()
1742 if (!rcu_access_pointer(bond->curr_active_slave) &&
1743 new_slave->link == BOND_LINK_UP)
1744 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1747 } /* switch(bond_mode) */
1749 #ifdef CONFIG_NET_POLL_CONTROLLER
1750 if (bond->dev->npinfo) {
1751 if (slave_enable_netpoll(new_slave)) {
1752 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1759 if (!(bond_dev->features & NETIF_F_LRO))
1760 dev_disable_lro(slave_dev);
1762 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1765 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
1769 res = bond_master_upper_dev_link(bond, new_slave, extack);
1771 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1772 goto err_unregister;
1775 res = bond_sysfs_slave_add(new_slave);
1777 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1778 goto err_upper_unlink;
1781 /* If the mode uses primary, then the following is handled by
1782 * bond_change_active_slave().
1784 if (!bond_uses_primary(bond)) {
1785 /* set promiscuity level to new slave */
1786 if (bond_dev->flags & IFF_PROMISC) {
1787 res = dev_set_promiscuity(slave_dev, 1);
1792 /* set allmulti level to new slave */
1793 if (bond_dev->flags & IFF_ALLMULTI) {
1794 res = dev_set_allmulti(slave_dev, 1);
1796 if (bond_dev->flags & IFF_PROMISC)
1797 dev_set_promiscuity(slave_dev, -1);
1802 netif_addr_lock_bh(bond_dev);
1803 dev_mc_sync_multiple(slave_dev, bond_dev);
1804 dev_uc_sync_multiple(slave_dev, bond_dev);
1805 netif_addr_unlock_bh(bond_dev);
1807 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1808 /* add lacpdu mc addr to mc list */
1809 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1811 dev_mc_add(slave_dev, lacpdu_multicast);
1816 bond_compute_features(bond);
1817 bond_set_carrier(bond);
1819 if (bond_uses_primary(bond)) {
1821 bond_select_active_slave(bond);
1822 unblock_netpoll_tx();
1825 if (bond_mode_can_use_xmit_hash(bond))
1826 bond_update_slave_arr(bond, NULL);
1829 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
1830 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1831 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1833 /* enslave is successful */
1834 bond_queue_slave_event(new_slave);
1837 /* Undo stages on error */
1839 bond_sysfs_slave_del(new_slave);
1842 bond_upper_dev_unlink(bond, new_slave);
1845 netdev_rx_handler_unregister(slave_dev);
1848 vlan_vids_del_by_dev(slave_dev, bond_dev);
1849 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1850 RCU_INIT_POINTER(bond->primary_slave, NULL);
1851 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1853 bond_change_active_slave(bond, NULL);
1854 bond_select_active_slave(bond);
1855 unblock_netpoll_tx();
1857 /* either primary_slave or curr_active_slave might've changed */
1859 slave_disable_netpoll(new_slave);
1862 if (!netif_is_bond_master(slave_dev))
1863 slave_dev->priv_flags &= ~IFF_BONDING;
1864 dev_close(slave_dev);
1867 slave_dev->flags &= ~IFF_SLAVE;
1868 if (!bond->params.fail_over_mac ||
1869 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1870 /* XXX TODO - fom follow mode needs to change master's
1871 * MAC if this slave's MAC is in use by the bond, or at
1872 * least print a warning.
1874 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1875 new_slave->dev->addr_len);
1876 ss.ss_family = slave_dev->type;
1877 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
1881 dev_set_mtu(slave_dev, new_slave->original_mtu);
1884 bond_free_slave(new_slave);
1887 /* Enslave of first slave has failed and we need to fix master's mac */
1888 if (!bond_has_slaves(bond)) {
1889 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1890 slave_dev->dev_addr))
1891 eth_hw_addr_random(bond_dev);
1892 if (bond_dev->type != ARPHRD_ETHER) {
1893 dev_close(bond_dev);
1894 ether_setup(bond_dev);
1895 bond_dev->flags |= IFF_MASTER;
1896 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1903 /* Try to release the slave device <slave> from the bond device <master>
1904 * It is legal to access curr_active_slave without a lock because all the function
1905 * is RTNL-locked. If "all" is true it means that the function is being called
1906 * while destroying a bond interface and all slaves are being released.
1908 * The rules for slave state should be:
1909 * for Active/Backup:
1910 * Active stays on all backups go down
1911 * for Bonded connections:
1912 * The first up interface should be left on and all others downed.
1914 static int __bond_release_one(struct net_device *bond_dev,
1915 struct net_device *slave_dev,
1916 bool all, bool unregister)
1918 struct bonding *bond = netdev_priv(bond_dev);
1919 struct slave *slave, *oldcurrent;
1920 struct sockaddr_storage ss;
1921 int old_flags = bond_dev->flags;
1922 netdev_features_t old_features = bond_dev->features;
1924 /* slave is not a slave or master is not master of this slave */
1925 if (!(slave_dev->flags & IFF_SLAVE) ||
1926 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1927 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
1933 slave = bond_get_slave_by_dev(bond, slave_dev);
1935 /* not a slave of this bond */
1936 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
1937 unblock_netpoll_tx();
1941 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1943 bond_sysfs_slave_del(slave);
1945 /* recompute stats just before removing the slave */
1946 bond_get_stats(bond->dev, &bond->bond_stats);
1948 bond_upper_dev_unlink(bond, slave);
1949 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1950 * for this slave anymore.
1952 netdev_rx_handler_unregister(slave_dev);
1954 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1955 bond_3ad_unbind_slave(slave);
1957 if (bond_mode_can_use_xmit_hash(bond))
1958 bond_update_slave_arr(bond, slave);
1960 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
1961 bond_is_active_slave(slave) ? "active" : "backup");
1963 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1965 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1967 if (!all && (!bond->params.fail_over_mac ||
1968 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1969 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1970 bond_has_slaves(bond))
1971 slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n",
1972 slave->perm_hwaddr);
1975 if (rtnl_dereference(bond->primary_slave) == slave)
1976 RCU_INIT_POINTER(bond->primary_slave, NULL);
1978 if (oldcurrent == slave)
1979 bond_change_active_slave(bond, NULL);
1981 if (bond_is_lb(bond)) {
1982 /* Must be called only after the slave has been
1983 * detached from the list and the curr_active_slave
1984 * has been cleared (if our_slave == old_current),
1985 * but before a new active slave is selected.
1987 bond_alb_deinit_slave(bond, slave);
1991 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1992 } else if (oldcurrent == slave) {
1993 /* Note that we hold RTNL over this sequence, so there
1994 * is no concern that another slave add/remove event
1997 bond_select_active_slave(bond);
2000 if (!bond_has_slaves(bond)) {
2001 bond_set_carrier(bond);
2002 eth_hw_addr_random(bond_dev);
2005 unblock_netpoll_tx();
2009 if (!bond_has_slaves(bond)) {
2010 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2011 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2014 bond_compute_features(bond);
2015 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2016 (old_features & NETIF_F_VLAN_CHALLENGED))
2017 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2019 vlan_vids_del_by_dev(slave_dev, bond_dev);
2021 /* If the mode uses primary, then this case was handled above by
2022 * bond_change_active_slave(..., NULL)
2024 if (!bond_uses_primary(bond)) {
2025 /* unset promiscuity level from slave
2026 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2027 * of the IFF_PROMISC flag in the bond_dev, but we need the
2028 * value of that flag before that change, as that was the value
2029 * when this slave was attached, so we cache at the start of the
2030 * function and use it here. Same goes for ALLMULTI below
2032 if (old_flags & IFF_PROMISC)
2033 dev_set_promiscuity(slave_dev, -1);
2035 /* unset allmulti level from slave */
2036 if (old_flags & IFF_ALLMULTI)
2037 dev_set_allmulti(slave_dev, -1);
2039 bond_hw_addr_flush(bond_dev, slave_dev);
2042 slave_disable_netpoll(slave);
2044 /* close slave before restoring its mac address */
2045 dev_close(slave_dev);
2047 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2048 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2049 /* restore original ("permanent") mac address */
2050 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2051 slave->dev->addr_len);
2052 ss.ss_family = slave_dev->type;
2053 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2057 __dev_set_mtu(slave_dev, slave->original_mtu);
2059 dev_set_mtu(slave_dev, slave->original_mtu);
2061 if (!netif_is_bond_master(slave_dev))
2062 slave_dev->priv_flags &= ~IFF_BONDING;
2064 bond_free_slave(slave);
2069 /* A wrapper used because of ndo_del_link */
2070 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2072 return __bond_release_one(bond_dev, slave_dev, false, false);
2075 /* First release a slave and then destroy the bond if no more slaves are left.
2076 * Must be under rtnl_lock when this function is called.
2078 static int bond_release_and_destroy(struct net_device *bond_dev,
2079 struct net_device *slave_dev)
2081 struct bonding *bond = netdev_priv(bond_dev);
2084 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2085 if (ret == 0 && !bond_has_slaves(bond)) {
2086 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2087 netdev_info(bond_dev, "Destroying bond\n");
2088 bond_remove_proc_entry(bond);
2089 unregister_netdevice(bond_dev);
2094 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2096 struct bonding *bond = netdev_priv(bond_dev);
2097 bond_fill_ifbond(bond, info);
2100 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2102 struct bonding *bond = netdev_priv(bond_dev);
2103 struct list_head *iter;
2104 int i = 0, res = -ENODEV;
2105 struct slave *slave;
2107 bond_for_each_slave(bond, slave, iter) {
2108 if (i++ == (int)info->slave_id) {
2110 bond_fill_ifslave(slave, info);
2118 /*-------------------------------- Monitoring -------------------------------*/
2120 /* called with rcu_read_lock() */
2121 static int bond_miimon_inspect(struct bonding *bond)
2123 int link_state, commit = 0;
2124 struct list_head *iter;
2125 struct slave *slave;
2126 bool ignore_updelay;
2128 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2130 bond_for_each_slave_rcu(bond, slave, iter) {
2131 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2133 link_state = bond_check_dev_link(bond, slave->dev, 0);
2135 switch (slave->link) {
2140 bond_propose_link_state(slave, BOND_LINK_FAIL);
2142 slave->delay = bond->params.downdelay;
2144 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2146 BOND_MODE_ACTIVEBACKUP) ?
2147 (bond_is_active_slave(slave) ?
2148 "active " : "backup ") : "",
2149 bond->params.downdelay * bond->params.miimon);
2152 case BOND_LINK_FAIL:
2154 /* recovered before downdelay expired */
2155 bond_propose_link_state(slave, BOND_LINK_UP);
2156 slave->last_link_up = jiffies;
2157 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2158 (bond->params.downdelay - slave->delay) *
2159 bond->params.miimon);
2164 if (slave->delay <= 0) {
2165 bond_propose_link_state(slave, BOND_LINK_DOWN);
2173 case BOND_LINK_DOWN:
2177 bond_propose_link_state(slave, BOND_LINK_BACK);
2179 slave->delay = bond->params.updelay;
2182 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2183 ignore_updelay ? 0 :
2184 bond->params.updelay *
2185 bond->params.miimon);
2188 case BOND_LINK_BACK:
2190 bond_propose_link_state(slave, BOND_LINK_DOWN);
2191 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2192 (bond->params.updelay - slave->delay) *
2193 bond->params.miimon);
2201 if (slave->delay <= 0) {
2202 bond_propose_link_state(slave, BOND_LINK_UP);
2204 ignore_updelay = false;
2216 static void bond_miimon_link_change(struct bonding *bond,
2217 struct slave *slave,
2220 switch (BOND_MODE(bond)) {
2221 case BOND_MODE_8023AD:
2222 bond_3ad_handle_link_change(slave, link);
2226 bond_alb_handle_link_change(bond, slave, link);
2229 bond_update_slave_arr(bond, NULL);
2234 static void bond_miimon_commit(struct bonding *bond)
2236 struct list_head *iter;
2237 struct slave *slave, *primary;
2239 bond_for_each_slave(bond, slave, iter) {
2240 switch (slave->link_new_state) {
2241 case BOND_LINK_NOCHANGE:
2242 /* For 802.3ad mode, check current slave speed and
2243 * duplex again in case its port was disabled after
2244 * invalid speed/duplex reporting but recovered before
2245 * link monitoring could make a decision on the actual
2248 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2249 slave->link == BOND_LINK_UP)
2250 bond_3ad_adapter_speed_duplex_changed(slave);
2254 if (bond_update_speed_duplex(slave) &&
2255 bond_needs_speed_duplex(bond)) {
2256 slave->link = BOND_LINK_DOWN;
2257 if (net_ratelimit())
2258 slave_warn(bond->dev, slave->dev,
2259 "failed to get link speed/duplex\n");
2262 bond_set_slave_link_state(slave, BOND_LINK_UP,
2263 BOND_SLAVE_NOTIFY_NOW);
2264 slave->last_link_up = jiffies;
2266 primary = rtnl_dereference(bond->primary_slave);
2267 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2268 /* prevent it from being the active one */
2269 bond_set_backup_slave(slave);
2270 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2271 /* make it immediately active */
2272 bond_set_active_slave(slave);
2273 } else if (slave != primary) {
2274 /* prevent it from being the active one */
2275 bond_set_backup_slave(slave);
2278 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2279 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2280 slave->duplex ? "full" : "half");
2282 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2284 if (!bond->curr_active_slave || slave == primary)
2289 case BOND_LINK_DOWN:
2290 if (slave->link_failure_count < UINT_MAX)
2291 slave->link_failure_count++;
2293 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2294 BOND_SLAVE_NOTIFY_NOW);
2296 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2297 BOND_MODE(bond) == BOND_MODE_8023AD)
2298 bond_set_slave_inactive_flags(slave,
2299 BOND_SLAVE_NOTIFY_NOW);
2301 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2303 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2305 if (slave == rcu_access_pointer(bond->curr_active_slave))
2311 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2312 slave->link_new_state);
2313 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2320 bond_select_active_slave(bond);
2321 unblock_netpoll_tx();
2324 bond_set_carrier(bond);
2329 * Really a wrapper that splits the mii monitor into two phases: an
2330 * inspection, then (if inspection indicates something needs to be done)
2331 * an acquisition of appropriate locks followed by a commit phase to
2332 * implement whatever link state changes are indicated.
2334 static void bond_mii_monitor(struct work_struct *work)
2336 struct bonding *bond = container_of(work, struct bonding,
2338 bool should_notify_peers = false;
2340 unsigned long delay;
2341 struct slave *slave;
2342 struct list_head *iter;
2344 delay = msecs_to_jiffies(bond->params.miimon);
2346 if (!bond_has_slaves(bond))
2350 should_notify_peers = bond_should_notify_peers(bond);
2351 commit = !!bond_miimon_inspect(bond);
2352 if (bond->send_peer_notif) {
2354 if (rtnl_trylock()) {
2355 bond->send_peer_notif--;
2363 /* Race avoidance with bond_close cancel of workqueue */
2364 if (!rtnl_trylock()) {
2366 should_notify_peers = false;
2370 bond_for_each_slave(bond, slave, iter) {
2371 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2373 bond_miimon_commit(bond);
2375 rtnl_unlock(); /* might sleep, hold no other locks */
2379 if (bond->params.miimon)
2380 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2382 if (should_notify_peers) {
2383 if (!rtnl_trylock())
2385 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2390 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2392 __be32 ip = *((__be32 *)data);
2394 return ip == bond_confirm_addr(upper, 0, ip);
2397 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2401 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2405 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2412 /* We go to the (large) trouble of VLAN tagging ARP frames because
2413 * switches in VLAN mode (especially if ports are configured as
2414 * "native" to a VLAN) might not pass non-tagged frames.
2416 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2417 __be32 src_ip, struct bond_vlan_tag *tags)
2419 struct sk_buff *skb;
2420 struct bond_vlan_tag *outer_tag = tags;
2421 struct net_device *slave_dev = slave->dev;
2422 struct net_device *bond_dev = slave->bond->dev;
2424 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2425 arp_op, &dest_ip, &src_ip);
2427 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2428 NULL, slave_dev->dev_addr, NULL);
2431 net_err_ratelimited("ARP packet allocation failed\n");
2435 if (!tags || tags->vlan_proto == VLAN_N_VID)
2440 /* Go through all the tags backwards and add them to the packet */
2441 while (tags->vlan_proto != VLAN_N_VID) {
2442 if (!tags->vlan_id) {
2447 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2448 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2449 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2452 net_err_ratelimited("failed to insert inner VLAN tag\n");
2458 /* Set the outer tag */
2459 if (outer_tag->vlan_id) {
2460 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2461 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2462 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2463 outer_tag->vlan_id);
2470 /* Validate the device path between the @start_dev and the @end_dev.
2471 * The path is valid if the @end_dev is reachable through device
2473 * When the path is validated, collect any vlan information in the
2476 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2477 struct net_device *end_dev,
2480 struct bond_vlan_tag *tags;
2481 struct net_device *upper;
2482 struct list_head *iter;
2484 if (start_dev == end_dev) {
2485 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2487 return ERR_PTR(-ENOMEM);
2488 tags[level].vlan_proto = VLAN_N_VID;
2492 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2493 tags = bond_verify_device_path(upper, end_dev, level + 1);
2494 if (IS_ERR_OR_NULL(tags)) {
2499 if (is_vlan_dev(upper)) {
2500 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2501 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2510 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2513 struct bond_vlan_tag *tags;
2514 __be32 *targets = bond->params.arp_targets, addr;
2517 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2518 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2519 __func__, &targets[i]);
2522 /* Find out through which dev should the packet go */
2523 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2526 /* there's no route to target - try to send arp
2527 * probe to generate any traffic (arp_validate=0)
2529 if (bond->params.arp_validate)
2530 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2533 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2538 /* bond device itself */
2539 if (rt->dst.dev == bond->dev)
2543 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2546 if (!IS_ERR_OR_NULL(tags))
2549 /* Not our device - skip */
2550 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2551 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2557 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2559 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2564 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2568 if (!sip || !bond_has_this_ip(bond, tip)) {
2569 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2570 __func__, &sip, &tip);
2574 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2576 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2580 slave->last_rx = jiffies;
2581 slave->target_last_arp_rx[i] = jiffies;
2584 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2585 struct slave *slave)
2587 struct arphdr *arp = (struct arphdr *)skb->data;
2588 struct slave *curr_active_slave, *curr_arp_slave;
2589 unsigned char *arp_ptr;
2591 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2594 if (!slave_do_arp_validate(bond, slave)) {
2595 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2596 !slave_do_arp_validate_only(bond))
2597 slave->last_rx = jiffies;
2598 return RX_HANDLER_ANOTHER;
2599 } else if (!is_arp) {
2600 return RX_HANDLER_ANOTHER;
2603 alen = arp_hdr_len(bond->dev);
2605 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2606 __func__, skb->dev->name);
2608 if (alen > skb_headlen(skb)) {
2609 arp = kmalloc(alen, GFP_ATOMIC);
2612 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2616 if (arp->ar_hln != bond->dev->addr_len ||
2617 skb->pkt_type == PACKET_OTHERHOST ||
2618 skb->pkt_type == PACKET_LOOPBACK ||
2619 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2620 arp->ar_pro != htons(ETH_P_IP) ||
2624 arp_ptr = (unsigned char *)(arp + 1);
2625 arp_ptr += bond->dev->addr_len;
2626 memcpy(&sip, arp_ptr, 4);
2627 arp_ptr += 4 + bond->dev->addr_len;
2628 memcpy(&tip, arp_ptr, 4);
2630 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2631 __func__, slave->dev->name, bond_slave_state(slave),
2632 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2635 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2636 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2638 /* We 'trust' the received ARP enough to validate it if:
2640 * (a) the slave receiving the ARP is active (which includes the
2641 * current ARP slave, if any), or
2643 * (b) the receiving slave isn't active, but there is a currently
2644 * active slave and it received valid arp reply(s) after it became
2645 * the currently active slave, or
2647 * (c) there is an ARP slave that sent an ARP during the prior ARP
2648 * interval, and we receive an ARP reply on any slave. We accept
2649 * these because switch FDB update delays may deliver the ARP
2650 * reply to a slave other than the sender of the ARP request.
2652 * Note: for (b), backup slaves are receiving the broadcast ARP
2653 * request, not a reply. This request passes from the sending
2654 * slave through the L2 switch(es) to the receiving slave. Since
2655 * this is checking the request, sip/tip are swapped for
2658 * This is done to avoid endless looping when we can't reach the
2659 * arp_ip_target and fool ourselves with our own arp requests.
2661 if (bond_is_active_slave(slave))
2662 bond_validate_arp(bond, slave, sip, tip);
2663 else if (curr_active_slave &&
2664 time_after(slave_last_rx(bond, curr_active_slave),
2665 curr_active_slave->last_link_up))
2666 bond_validate_arp(bond, slave, tip, sip);
2667 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2668 bond_time_in_interval(bond,
2669 dev_trans_start(curr_arp_slave->dev), 1))
2670 bond_validate_arp(bond, slave, sip, tip);
2673 if (arp != (struct arphdr *)skb->data)
2675 return RX_HANDLER_ANOTHER;
2678 /* function to verify if we're in the arp_interval timeslice, returns true if
2679 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2680 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2682 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2685 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2687 return time_in_range(jiffies,
2688 last_act - delta_in_ticks,
2689 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2692 /* This function is called regularly to monitor each slave's link
2693 * ensuring that traffic is being sent and received when arp monitoring
2694 * is used in load-balancing mode. if the adapter has been dormant, then an
2695 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2696 * arp monitoring in active backup mode.
2698 static void bond_loadbalance_arp_mon(struct bonding *bond)
2700 struct slave *slave, *oldcurrent;
2701 struct list_head *iter;
2702 int do_failover = 0, slave_state_changed = 0;
2704 if (!bond_has_slaves(bond))
2709 oldcurrent = rcu_dereference(bond->curr_active_slave);
2710 /* see if any of the previous devices are up now (i.e. they have
2711 * xmt and rcv traffic). the curr_active_slave does not come into
2712 * the picture unless it is null. also, slave->last_link_up is not
2713 * needed here because we send an arp on each slave and give a slave
2714 * as long as it needs to get the tx/rx within the delta.
2715 * TODO: what about up/down delay in arp mode? it wasn't here before
2718 bond_for_each_slave_rcu(bond, slave, iter) {
2719 unsigned long trans_start = dev_trans_start(slave->dev);
2721 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2723 if (slave->link != BOND_LINK_UP) {
2724 if (bond_time_in_interval(bond, trans_start, 1) &&
2725 bond_time_in_interval(bond, slave->last_rx, 1)) {
2727 bond_propose_link_state(slave, BOND_LINK_UP);
2728 slave_state_changed = 1;
2730 /* primary_slave has no meaning in round-robin
2731 * mode. the window of a slave being up and
2732 * curr_active_slave being null after enslaving
2736 slave_info(bond->dev, slave->dev, "link status definitely up\n");
2739 slave_info(bond->dev, slave->dev, "interface is now up\n");
2743 /* slave->link == BOND_LINK_UP */
2745 /* not all switches will respond to an arp request
2746 * when the source ip is 0, so don't take the link down
2747 * if we don't know our ip yet
2749 if (!bond_time_in_interval(bond, trans_start, 2) ||
2750 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2752 bond_propose_link_state(slave, BOND_LINK_DOWN);
2753 slave_state_changed = 1;
2755 if (slave->link_failure_count < UINT_MAX)
2756 slave->link_failure_count++;
2758 slave_info(bond->dev, slave->dev, "interface is now down\n");
2760 if (slave == oldcurrent)
2765 /* note: if switch is in round-robin mode, all links
2766 * must tx arp to ensure all links rx an arp - otherwise
2767 * links may oscillate or not come up at all; if switch is
2768 * in something like xor mode, there is nothing we can
2769 * do - all replies will be rx'ed on same link causing slaves
2770 * to be unstable during low/no traffic periods
2772 if (bond_slave_is_up(slave))
2773 bond_arp_send_all(bond, slave);
2778 if (do_failover || slave_state_changed) {
2779 if (!rtnl_trylock())
2782 bond_for_each_slave(bond, slave, iter) {
2783 if (slave->link_new_state != BOND_LINK_NOCHANGE)
2784 slave->link = slave->link_new_state;
2787 if (slave_state_changed) {
2788 bond_slave_state_change(bond);
2789 if (BOND_MODE(bond) == BOND_MODE_XOR)
2790 bond_update_slave_arr(bond, NULL);
2794 bond_select_active_slave(bond);
2795 unblock_netpoll_tx();
2801 if (bond->params.arp_interval)
2802 queue_delayed_work(bond->wq, &bond->arp_work,
2803 msecs_to_jiffies(bond->params.arp_interval));
2806 /* Called to inspect slaves for active-backup mode ARP monitor link state
2807 * changes. Sets proposed link state in slaves to specify what action
2808 * should take place for the slave. Returns 0 if no changes are found, >0
2809 * if changes to link states must be committed.
2811 * Called with rcu_read_lock held.
2813 static int bond_ab_arp_inspect(struct bonding *bond)
2815 unsigned long trans_start, last_rx;
2816 struct list_head *iter;
2817 struct slave *slave;
2820 bond_for_each_slave_rcu(bond, slave, iter) {
2821 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2822 last_rx = slave_last_rx(bond, slave);
2824 if (slave->link != BOND_LINK_UP) {
2825 if (bond_time_in_interval(bond, last_rx, 1)) {
2826 bond_propose_link_state(slave, BOND_LINK_UP);
2832 /* Give slaves 2*delta after being enslaved or made
2833 * active. This avoids bouncing, as the last receive
2834 * times need a full ARP monitor cycle to be updated.
2836 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2839 /* Backup slave is down if:
2840 * - No current_arp_slave AND
2841 * - more than 3*delta since last receive AND
2842 * - the bond has an IP address
2844 * Note: a non-null current_arp_slave indicates
2845 * the curr_active_slave went down and we are
2846 * searching for a new one; under this condition
2847 * we only take the curr_active_slave down - this
2848 * gives each slave a chance to tx/rx traffic
2849 * before being taken out
2851 if (!bond_is_active_slave(slave) &&
2852 !rcu_access_pointer(bond->current_arp_slave) &&
2853 !bond_time_in_interval(bond, last_rx, 3)) {
2854 bond_propose_link_state(slave, BOND_LINK_DOWN);
2858 /* Active slave is down if:
2859 * - more than 2*delta since transmitting OR
2860 * - (more than 2*delta since receive AND
2861 * the bond has an IP address)
2863 trans_start = dev_trans_start(slave->dev);
2864 if (bond_is_active_slave(slave) &&
2865 (!bond_time_in_interval(bond, trans_start, 2) ||
2866 !bond_time_in_interval(bond, last_rx, 2))) {
2867 bond_propose_link_state(slave, BOND_LINK_DOWN);
2875 /* Called to commit link state changes noted by inspection step of
2876 * active-backup mode ARP monitor.
2878 * Called with RTNL hold.
2880 static void bond_ab_arp_commit(struct bonding *bond)
2882 unsigned long trans_start;
2883 struct list_head *iter;
2884 struct slave *slave;
2886 bond_for_each_slave(bond, slave, iter) {
2887 switch (slave->link_new_state) {
2888 case BOND_LINK_NOCHANGE:
2892 trans_start = dev_trans_start(slave->dev);
2893 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2894 (!rtnl_dereference(bond->curr_active_slave) &&
2895 bond_time_in_interval(bond, trans_start, 1))) {
2896 struct slave *current_arp_slave;
2898 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2899 bond_set_slave_link_state(slave, BOND_LINK_UP,
2900 BOND_SLAVE_NOTIFY_NOW);
2901 if (current_arp_slave) {
2902 bond_set_slave_inactive_flags(
2904 BOND_SLAVE_NOTIFY_NOW);
2905 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2908 slave_info(bond->dev, slave->dev, "link status definitely up\n");
2910 if (!rtnl_dereference(bond->curr_active_slave) ||
2911 slave == rtnl_dereference(bond->primary_slave))
2918 case BOND_LINK_DOWN:
2919 if (slave->link_failure_count < UINT_MAX)
2920 slave->link_failure_count++;
2922 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2923 BOND_SLAVE_NOTIFY_NOW);
2924 bond_set_slave_inactive_flags(slave,
2925 BOND_SLAVE_NOTIFY_NOW);
2927 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2929 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2930 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2937 slave_err(bond->dev, slave->dev,
2938 "impossible: link_new_state %d on slave\n",
2939 slave->link_new_state);
2945 bond_select_active_slave(bond);
2946 unblock_netpoll_tx();
2949 bond_set_carrier(bond);
2952 /* Send ARP probes for active-backup mode ARP monitor.
2954 * Called with rcu_read_lock held.
2956 static bool bond_ab_arp_probe(struct bonding *bond)
2958 struct slave *slave, *before = NULL, *new_slave = NULL,
2959 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2960 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2961 struct list_head *iter;
2963 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2965 if (curr_arp_slave && curr_active_slave)
2966 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2967 curr_arp_slave->dev->name,
2968 curr_active_slave->dev->name);
2970 if (curr_active_slave) {
2971 bond_arp_send_all(bond, curr_active_slave);
2972 return should_notify_rtnl;
2975 /* if we don't have a curr_active_slave, search for the next available
2976 * backup slave from the current_arp_slave and make it the candidate
2977 * for becoming the curr_active_slave
2980 if (!curr_arp_slave) {
2981 curr_arp_slave = bond_first_slave_rcu(bond);
2982 if (!curr_arp_slave)
2983 return should_notify_rtnl;
2986 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2988 bond_for_each_slave_rcu(bond, slave, iter) {
2989 if (!found && !before && bond_slave_is_up(slave))
2992 if (found && !new_slave && bond_slave_is_up(slave))
2994 /* if the link state is up at this point, we
2995 * mark it down - this can happen if we have
2996 * simultaneous link failures and
2997 * reselect_active_interface doesn't make this
2998 * one the current slave so it is still marked
2999 * up when it is actually down
3001 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3002 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3003 BOND_SLAVE_NOTIFY_LATER);
3004 if (slave->link_failure_count < UINT_MAX)
3005 slave->link_failure_count++;
3007 bond_set_slave_inactive_flags(slave,
3008 BOND_SLAVE_NOTIFY_LATER);
3010 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3012 if (slave == curr_arp_slave)
3016 if (!new_slave && before)
3022 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3023 BOND_SLAVE_NOTIFY_LATER);
3024 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3025 bond_arp_send_all(bond, new_slave);
3026 new_slave->last_link_up = jiffies;
3027 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3030 bond_for_each_slave_rcu(bond, slave, iter) {
3031 if (slave->should_notify || slave->should_notify_link) {
3032 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3036 return should_notify_rtnl;
3039 static void bond_activebackup_arp_mon(struct bonding *bond)
3041 bool should_notify_peers = false;
3042 bool should_notify_rtnl = false;
3045 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3047 if (!bond_has_slaves(bond))
3052 should_notify_peers = bond_should_notify_peers(bond);
3054 if (bond_ab_arp_inspect(bond)) {
3057 /* Race avoidance with bond_close flush of workqueue */
3058 if (!rtnl_trylock()) {
3060 should_notify_peers = false;
3064 bond_ab_arp_commit(bond);
3070 should_notify_rtnl = bond_ab_arp_probe(bond);
3074 if (bond->params.arp_interval)
3075 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3077 if (should_notify_peers || should_notify_rtnl) {
3078 if (!rtnl_trylock())
3081 if (should_notify_peers)
3082 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3084 if (should_notify_rtnl) {
3085 bond_slave_state_notify(bond);
3086 bond_slave_link_notify(bond);
3093 static void bond_arp_monitor(struct work_struct *work)
3095 struct bonding *bond = container_of(work, struct bonding,
3098 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3099 bond_activebackup_arp_mon(bond);
3101 bond_loadbalance_arp_mon(bond);
3104 /*-------------------------- netdev event handling --------------------------*/
3106 /* Change device name */
3107 static int bond_event_changename(struct bonding *bond)
3109 bond_remove_proc_entry(bond);
3110 bond_create_proc_entry(bond);
3112 bond_debug_reregister(bond);
3117 static int bond_master_netdev_event(unsigned long event,
3118 struct net_device *bond_dev)
3120 struct bonding *event_bond = netdev_priv(bond_dev);
3122 netdev_dbg(bond_dev, "%s called\n", __func__);
3125 case NETDEV_CHANGENAME:
3126 return bond_event_changename(event_bond);
3127 case NETDEV_UNREGISTER:
3128 bond_remove_proc_entry(event_bond);
3130 case NETDEV_REGISTER:
3131 bond_create_proc_entry(event_bond);
3140 static int bond_slave_netdev_event(unsigned long event,
3141 struct net_device *slave_dev)
3143 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3144 struct bonding *bond;
3145 struct net_device *bond_dev;
3147 /* A netdev event can be generated while enslaving a device
3148 * before netdev_rx_handler_register is called in which case
3149 * slave will be NULL
3152 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3156 bond_dev = slave->bond->dev;
3158 primary = rtnl_dereference(bond->primary_slave);
3160 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3163 case NETDEV_UNREGISTER:
3164 if (bond_dev->type != ARPHRD_ETHER)
3165 bond_release_and_destroy(bond_dev, slave_dev);
3167 __bond_release_one(bond_dev, slave_dev, false, true);
3171 /* For 802.3ad mode only:
3172 * Getting invalid Speed/Duplex values here will put slave
3173 * in weird state. Mark it as link-fail if the link was
3174 * previously up or link-down if it hasn't yet come up, and
3175 * let link-monitoring (miimon) set it right when correct
3176 * speeds/duplex are available.
3178 if (bond_update_speed_duplex(slave) &&
3179 BOND_MODE(bond) == BOND_MODE_8023AD) {
3180 if (slave->last_link_up)
3181 slave->link = BOND_LINK_FAIL;
3183 slave->link = BOND_LINK_DOWN;
3186 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3187 bond_3ad_adapter_speed_duplex_changed(slave);
3190 /* Refresh slave-array if applicable!
3191 * If the setup does not use miimon or arpmon (mode-specific!),
3192 * then these events will not cause the slave-array to be
3193 * refreshed. This will cause xmit to use a slave that is not
3194 * usable. Avoid such situation by refeshing the array at these
3195 * events. If these (miimon/arpmon) parameters are configured
3196 * then array gets refreshed twice and that should be fine!
3198 if (bond_mode_can_use_xmit_hash(bond))
3199 bond_update_slave_arr(bond, NULL);
3201 case NETDEV_CHANGEMTU:
3202 /* TODO: Should slaves be allowed to
3203 * independently alter their MTU? For
3204 * an active-backup bond, slaves need
3205 * not be the same type of device, so
3206 * MTUs may vary. For other modes,
3207 * slaves arguably should have the
3208 * same MTUs. To do this, we'd need to
3209 * take over the slave's change_mtu
3210 * function for the duration of their
3214 case NETDEV_CHANGENAME:
3215 /* we don't care if we don't have primary set */
3216 if (!bond_uses_primary(bond) ||
3217 !bond->params.primary[0])
3220 if (slave == primary) {
3221 /* slave's name changed - he's no longer primary */
3222 RCU_INIT_POINTER(bond->primary_slave, NULL);
3223 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3224 /* we have a new primary slave */
3225 rcu_assign_pointer(bond->primary_slave, slave);
3226 } else { /* we didn't change primary - exit */
3230 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3231 primary ? slave_dev->name : "none");
3234 bond_select_active_slave(bond);
3235 unblock_netpoll_tx();
3237 case NETDEV_FEAT_CHANGE:
3238 bond_compute_features(bond);
3240 case NETDEV_RESEND_IGMP:
3241 /* Propagate to master device */
3242 call_netdevice_notifiers(event, slave->bond->dev);
3251 /* bond_netdev_event: handle netdev notifier chain events.
3253 * This function receives events for the netdev chain. The caller (an
3254 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3255 * locks for us to safely manipulate the slave devices (RTNL lock,
3258 static int bond_netdev_event(struct notifier_block *this,
3259 unsigned long event, void *ptr)
3261 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3263 netdev_dbg(event_dev, "%s received %s\n",
3264 __func__, netdev_cmd_to_name(event));
3266 if (!(event_dev->priv_flags & IFF_BONDING))
3269 if (event_dev->flags & IFF_MASTER) {
3272 ret = bond_master_netdev_event(event, event_dev);
3273 if (ret != NOTIFY_DONE)
3277 if (event_dev->flags & IFF_SLAVE)
3278 return bond_slave_netdev_event(event, event_dev);
3283 static struct notifier_block bond_netdev_notifier = {
3284 .notifier_call = bond_netdev_event,
3287 /*---------------------------- Hashing Policies -----------------------------*/
3289 /* L2 hash helper */
3290 static inline u32 bond_eth_hash(struct sk_buff *skb)
3292 struct ethhdr *ep, hdr_tmp;
3294 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3296 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3300 /* Extract the appropriate headers based on bond's xmit policy */
3301 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3302 struct flow_keys *fk)
3304 const struct ipv6hdr *iph6;
3305 const struct iphdr *iph;
3306 int noff, proto = -1;
3308 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) {
3309 memset(fk, 0, sizeof(*fk));
3310 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3311 fk, NULL, 0, 0, 0, 0);
3314 fk->ports.ports = 0;
3315 memset(&fk->icmp, 0, sizeof(fk->icmp));
3316 noff = skb_network_offset(skb);
3317 if (skb->protocol == htons(ETH_P_IP)) {
3318 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3321 iph_to_flow_copy_v4addrs(fk, iph);
3322 noff += iph->ihl << 2;
3323 if (!ip_is_fragment(iph))
3324 proto = iph->protocol;
3325 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3326 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3328 iph6 = ipv6_hdr(skb);
3329 iph_to_flow_copy_v6addrs(fk, iph6);
3330 noff += sizeof(*iph6);
3331 proto = iph6->nexthdr;
3335 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0) {
3336 if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6)
3337 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data,
3338 skb_transport_offset(skb),
3341 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3348 * bond_xmit_hash - generate a hash value based on the xmit policy
3349 * @bond: bonding device
3350 * @skb: buffer to use for headers
3352 * This function will extract the necessary headers from the skb buffer and use
3353 * them to generate a hash based on the xmit_policy set in the bonding device
3355 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3357 struct flow_keys flow;
3360 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3364 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3365 !bond_flow_dissect(bond, skb, &flow))
3366 return bond_eth_hash(skb);
3368 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3369 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3370 hash = bond_eth_hash(skb);
3373 memcpy(&hash, &flow.icmp, sizeof(hash));
3375 memcpy(&hash, &flow.ports.ports, sizeof(hash));
3377 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3378 (__force u32)flow_get_u32_src(&flow);
3379 hash ^= (hash >> 16);
3380 hash ^= (hash >> 8);
3385 /*-------------------------- Device entry points ----------------------------*/
3387 void bond_work_init_all(struct bonding *bond)
3389 INIT_DELAYED_WORK(&bond->mcast_work,
3390 bond_resend_igmp_join_requests_delayed);
3391 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3392 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3393 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3394 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3395 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3398 static void bond_work_cancel_all(struct bonding *bond)
3400 cancel_delayed_work_sync(&bond->mii_work);
3401 cancel_delayed_work_sync(&bond->arp_work);
3402 cancel_delayed_work_sync(&bond->alb_work);
3403 cancel_delayed_work_sync(&bond->ad_work);
3404 cancel_delayed_work_sync(&bond->mcast_work);
3405 cancel_delayed_work_sync(&bond->slave_arr_work);
3408 static int bond_open(struct net_device *bond_dev)
3410 struct bonding *bond = netdev_priv(bond_dev);
3411 struct list_head *iter;
3412 struct slave *slave;
3414 /* reset slave->backup and slave->inactive */
3415 if (bond_has_slaves(bond)) {
3416 bond_for_each_slave(bond, slave, iter) {
3417 if (bond_uses_primary(bond) &&
3418 slave != rcu_access_pointer(bond->curr_active_slave)) {
3419 bond_set_slave_inactive_flags(slave,
3420 BOND_SLAVE_NOTIFY_NOW);
3421 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3422 bond_set_slave_active_flags(slave,
3423 BOND_SLAVE_NOTIFY_NOW);
3428 if (bond_is_lb(bond)) {
3429 /* bond_alb_initialize must be called before the timer
3432 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3434 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3435 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3438 if (bond->params.miimon) /* link check interval, in milliseconds. */
3439 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3441 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3442 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3443 bond->recv_probe = bond_arp_rcv;
3446 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3447 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3448 /* register to receive LACPDUs */
3449 bond->recv_probe = bond_3ad_lacpdu_recv;
3450 bond_3ad_initiate_agg_selection(bond, 1);
3453 if (bond_mode_can_use_xmit_hash(bond))
3454 bond_update_slave_arr(bond, NULL);
3459 static int bond_close(struct net_device *bond_dev)
3461 struct bonding *bond = netdev_priv(bond_dev);
3463 bond_work_cancel_all(bond);
3464 bond->send_peer_notif = 0;
3465 if (bond_is_lb(bond))
3466 bond_alb_deinitialize(bond);
3467 bond->recv_probe = NULL;
3472 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3473 * that some drivers can provide 32bit values only.
3475 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3476 const struct rtnl_link_stats64 *_new,
3477 const struct rtnl_link_stats64 *_old)
3479 const u64 *new = (const u64 *)_new;
3480 const u64 *old = (const u64 *)_old;
3481 u64 *res = (u64 *)_res;
3484 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3487 s64 delta = nv - ov;
3489 /* detects if this particular field is 32bit only */
3490 if (((nv | ov) >> 32) == 0)
3491 delta = (s64)(s32)((u32)nv - (u32)ov);
3493 /* filter anomalies, some drivers reset their stats
3494 * at down/up events.
3501 static void bond_get_stats(struct net_device *bond_dev,
3502 struct rtnl_link_stats64 *stats)
3504 struct bonding *bond = netdev_priv(bond_dev);
3505 struct rtnl_link_stats64 temp;
3506 struct list_head *iter;
3507 struct slave *slave;
3509 spin_lock(&bond->stats_lock);
3510 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3513 bond_for_each_slave_rcu(bond, slave, iter) {
3514 const struct rtnl_link_stats64 *new =
3515 dev_get_stats(slave->dev, &temp);
3517 bond_fold_stats(stats, new, &slave->slave_stats);
3519 /* save off the slave stats for the next run */
3520 memcpy(&slave->slave_stats, new, sizeof(*new));
3524 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3525 spin_unlock(&bond->stats_lock);
3528 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3530 struct bonding *bond = netdev_priv(bond_dev);
3531 struct net_device *slave_dev = NULL;
3532 struct ifbond k_binfo;
3533 struct ifbond __user *u_binfo = NULL;
3534 struct ifslave k_sinfo;
3535 struct ifslave __user *u_sinfo = NULL;
3536 struct mii_ioctl_data *mii = NULL;
3537 struct bond_opt_value newval;
3541 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3552 /* We do this again just in case we were called by SIOCGMIIREG
3553 * instead of SIOCGMIIPHY.
3559 if (mii->reg_num == 1) {
3561 if (netif_carrier_ok(bond->dev))
3562 mii->val_out = BMSR_LSTATUS;
3566 case BOND_INFO_QUERY_OLD:
3567 case SIOCBONDINFOQUERY:
3568 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3570 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3573 bond_info_query(bond_dev, &k_binfo);
3574 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3578 case BOND_SLAVE_INFO_QUERY_OLD:
3579 case SIOCBONDSLAVEINFOQUERY:
3580 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3582 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3585 res = bond_slave_info_query(bond_dev, &k_sinfo);
3587 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3595 net = dev_net(bond_dev);
3597 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3600 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3602 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
3608 case BOND_ENSLAVE_OLD:
3609 case SIOCBONDENSLAVE:
3610 res = bond_enslave(bond_dev, slave_dev, NULL);
3612 case BOND_RELEASE_OLD:
3613 case SIOCBONDRELEASE:
3614 res = bond_release(bond_dev, slave_dev);
3616 case BOND_SETHWADDR_OLD:
3617 case SIOCBONDSETHWADDR:
3618 res = bond_set_dev_addr(bond_dev, slave_dev);
3620 case BOND_CHANGE_ACTIVE_OLD:
3621 case SIOCBONDCHANGEACTIVE:
3622 bond_opt_initstr(&newval, slave_dev->name);
3623 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3633 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3635 struct bonding *bond = netdev_priv(bond_dev);
3637 if (change & IFF_PROMISC)
3638 bond_set_promiscuity(bond,
3639 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3641 if (change & IFF_ALLMULTI)
3642 bond_set_allmulti(bond,
3643 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3646 static void bond_set_rx_mode(struct net_device *bond_dev)
3648 struct bonding *bond = netdev_priv(bond_dev);
3649 struct list_head *iter;
3650 struct slave *slave;
3653 if (bond_uses_primary(bond)) {
3654 slave = rcu_dereference(bond->curr_active_slave);
3656 dev_uc_sync(slave->dev, bond_dev);
3657 dev_mc_sync(slave->dev, bond_dev);
3660 bond_for_each_slave_rcu(bond, slave, iter) {
3661 dev_uc_sync_multiple(slave->dev, bond_dev);
3662 dev_mc_sync_multiple(slave->dev, bond_dev);
3668 static int bond_neigh_init(struct neighbour *n)
3670 struct bonding *bond = netdev_priv(n->dev);
3671 const struct net_device_ops *slave_ops;
3672 struct neigh_parms parms;
3673 struct slave *slave;
3676 slave = bond_first_slave(bond);
3679 slave_ops = slave->dev->netdev_ops;
3680 if (!slave_ops->ndo_neigh_setup)
3683 parms.neigh_setup = NULL;
3684 parms.neigh_cleanup = NULL;
3685 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3689 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3690 * after the last slave has been detached. Assumes that all slaves
3691 * utilize the same neigh_cleanup (true at this writing as only user
3694 n->parms->neigh_cleanup = parms.neigh_cleanup;
3696 if (!parms.neigh_setup)
3699 return parms.neigh_setup(n);
3702 /* The bonding ndo_neigh_setup is called at init time beofre any
3703 * slave exists. So we must declare proxy setup function which will
3704 * be used at run time to resolve the actual slave neigh param setup.
3706 * It's also called by master devices (such as vlans) to setup their
3707 * underlying devices. In that case - do nothing, we're already set up from
3710 static int bond_neigh_setup(struct net_device *dev,
3711 struct neigh_parms *parms)
3713 /* modify only our neigh_parms */
3714 if (parms->dev == dev)
3715 parms->neigh_setup = bond_neigh_init;
3720 /* Change the MTU of all of a master's slaves to match the master */
3721 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3723 struct bonding *bond = netdev_priv(bond_dev);
3724 struct slave *slave, *rollback_slave;
3725 struct list_head *iter;
3728 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3730 bond_for_each_slave(bond, slave, iter) {
3731 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
3732 slave, slave->dev->netdev_ops->ndo_change_mtu);
3734 res = dev_set_mtu(slave->dev, new_mtu);
3737 /* If we failed to set the slave's mtu to the new value
3738 * we must abort the operation even in ACTIVE_BACKUP
3739 * mode, because if we allow the backup slaves to have
3740 * different mtu values than the active slave we'll
3741 * need to change their mtu when doing a failover. That
3742 * means changing their mtu from timer context, which
3743 * is probably not a good idea.
3745 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
3751 bond_dev->mtu = new_mtu;
3756 /* unwind from head to the slave that failed */
3757 bond_for_each_slave(bond, rollback_slave, iter) {
3760 if (rollback_slave == slave)
3763 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3765 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
3772 /* Change HW address
3774 * Note that many devices must be down to change the HW address, and
3775 * downing the master releases all slaves. We can make bonds full of
3776 * bonding devices to test this, however.
3778 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3780 struct bonding *bond = netdev_priv(bond_dev);
3781 struct slave *slave, *rollback_slave;
3782 struct sockaddr_storage *ss = addr, tmp_ss;
3783 struct list_head *iter;
3786 if (BOND_MODE(bond) == BOND_MODE_ALB)
3787 return bond_alb_set_mac_address(bond_dev, addr);
3790 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
3792 /* If fail_over_mac is enabled, do nothing and return success.
3793 * Returning an error causes ifenslave to fail.
3795 if (bond->params.fail_over_mac &&
3796 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3799 if (!is_valid_ether_addr(ss->__data))
3800 return -EADDRNOTAVAIL;
3802 bond_for_each_slave(bond, slave, iter) {
3803 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
3805 res = dev_set_mac_address(slave->dev, addr, NULL);
3807 /* TODO: consider downing the slave
3809 * User should expect communications
3810 * breakage anyway until ARP finish
3813 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
3820 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3824 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3825 tmp_ss.ss_family = bond_dev->type;
3827 /* unwind from head to the slave that failed */
3828 bond_for_each_slave(bond, rollback_slave, iter) {
3831 if (rollback_slave == slave)
3834 tmp_res = dev_set_mac_address(rollback_slave->dev,
3835 (struct sockaddr *)&tmp_ss, NULL);
3837 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
3846 * bond_xmit_slave_id - transmit skb through slave with slave_id
3847 * @bond: bonding device that is transmitting
3848 * @skb: buffer to transmit
3849 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3851 * This function tries to transmit through slave with slave_id but in case
3852 * it fails, it tries to find the first available slave for transmission.
3853 * The skb is consumed in all cases, thus the function is void.
3855 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3857 struct list_head *iter;
3858 struct slave *slave;
3861 /* Here we start from the slave with slave_id */
3862 bond_for_each_slave_rcu(bond, slave, iter) {
3864 if (bond_slave_can_tx(slave)) {
3865 bond_dev_queue_xmit(bond, skb, slave->dev);
3871 /* Here we start from the first slave up to slave_id */
3873 bond_for_each_slave_rcu(bond, slave, iter) {
3876 if (bond_slave_can_tx(slave)) {
3877 bond_dev_queue_xmit(bond, skb, slave->dev);
3881 /* no slave that can tx has been found */
3882 bond_tx_drop(bond->dev, skb);
3886 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3887 * @bond: bonding device to use
3889 * Based on the value of the bonding device's packets_per_slave parameter
3890 * this function generates a slave id, which is usually used as the next
3891 * slave to transmit through.
3893 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3896 struct reciprocal_value reciprocal_packets_per_slave;
3897 int packets_per_slave = bond->params.packets_per_slave;
3899 switch (packets_per_slave) {
3901 slave_id = prandom_u32();
3904 slave_id = bond->rr_tx_counter;
3907 reciprocal_packets_per_slave =
3908 bond->params.reciprocal_packets_per_slave;
3909 slave_id = reciprocal_divide(bond->rr_tx_counter,
3910 reciprocal_packets_per_slave);
3913 bond->rr_tx_counter++;
3918 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3919 struct net_device *bond_dev)
3921 struct bonding *bond = netdev_priv(bond_dev);
3922 struct slave *slave;
3926 /* Start with the curr_active_slave that joined the bond as the
3927 * default for sending IGMP traffic. For failover purposes one
3928 * needs to maintain some consistency for the interface that will
3929 * send the join/membership reports. The curr_active_slave found
3930 * will send all of this type of traffic.
3932 if (skb->protocol == htons(ETH_P_IP)) {
3933 int noff = skb_network_offset(skb);
3936 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3940 if (iph->protocol == IPPROTO_IGMP) {
3941 slave = rcu_dereference(bond->curr_active_slave);
3943 bond_dev_queue_xmit(bond, skb, slave->dev);
3945 bond_xmit_slave_id(bond, skb, 0);
3946 return NETDEV_TX_OK;
3951 slave_cnt = READ_ONCE(bond->slave_cnt);
3952 if (likely(slave_cnt)) {
3953 slave_id = bond_rr_gen_slave_id(bond);
3954 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3956 bond_tx_drop(bond_dev, skb);
3958 return NETDEV_TX_OK;
3961 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3962 * the bond has a usable interface.
3964 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3965 struct net_device *bond_dev)
3967 struct bonding *bond = netdev_priv(bond_dev);
3968 struct slave *slave;
3970 slave = rcu_dereference(bond->curr_active_slave);
3972 bond_dev_queue_xmit(bond, skb, slave->dev);
3974 bond_tx_drop(bond_dev, skb);
3976 return NETDEV_TX_OK;
3979 /* Use this to update slave_array when (a) it's not appropriate to update
3980 * slave_array right away (note that update_slave_array() may sleep)
3981 * and / or (b) RTNL is not held.
3983 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3985 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3988 /* Slave array work handler. Holds only RTNL */
3989 static void bond_slave_arr_handler(struct work_struct *work)
3991 struct bonding *bond = container_of(work, struct bonding,
3992 slave_arr_work.work);
3995 if (!rtnl_trylock())
3998 ret = bond_update_slave_arr(bond, NULL);
4001 pr_warn_ratelimited("Failed to update slave array from WT\n");
4007 bond_slave_arr_work_rearm(bond, 1);
4010 /* Build the usable slaves array in control path for modes that use xmit-hash
4011 * to determine the slave interface -
4012 * (a) BOND_MODE_8023AD
4014 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4016 * The caller is expected to hold RTNL only and NO other lock!
4018 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4020 struct slave *slave;
4021 struct list_head *iter;
4022 struct bond_up_slave *new_arr, *old_arr;
4026 #ifdef CONFIG_LOCKDEP
4027 WARN_ON(lockdep_is_held(&bond->mode_lock));
4030 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
4034 pr_err("Failed to build slave-array.\n");
4037 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4038 struct ad_info ad_info;
4040 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4041 pr_debug("bond_3ad_get_active_agg_info failed\n");
4042 kfree_rcu(new_arr, rcu);
4043 /* No active aggragator means it's not safe to use
4044 * the previous array.
4046 old_arr = rtnl_dereference(bond->slave_arr);
4048 RCU_INIT_POINTER(bond->slave_arr, NULL);
4049 kfree_rcu(old_arr, rcu);
4053 agg_id = ad_info.aggregator_id;
4055 bond_for_each_slave(bond, slave, iter) {
4056 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4057 struct aggregator *agg;
4059 agg = SLAVE_AD_INFO(slave)->port.aggregator;
4060 if (!agg || agg->aggregator_identifier != agg_id)
4063 if (!bond_slave_can_tx(slave))
4065 if (skipslave == slave)
4068 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4071 new_arr->arr[new_arr->count++] = slave;
4074 old_arr = rtnl_dereference(bond->slave_arr);
4075 rcu_assign_pointer(bond->slave_arr, new_arr);
4077 kfree_rcu(old_arr, rcu);
4079 if (ret != 0 && skipslave) {
4082 /* Rare situation where caller has asked to skip a specific
4083 * slave but allocation failed (most likely!). BTW this is
4084 * only possible when the call is initiated from
4085 * __bond_release_one(). In this situation; overwrite the
4086 * skipslave entry in the array with the last entry from the
4087 * array to avoid a situation where the xmit path may choose
4088 * this to-be-skipped slave to send a packet out.
4090 old_arr = rtnl_dereference(bond->slave_arr);
4091 for (idx = 0; old_arr != NULL && idx < old_arr->count; idx++) {
4092 if (skipslave == old_arr->arr[idx]) {
4094 old_arr->arr[old_arr->count-1];
4103 /* Use this Xmit function for 3AD as well as XOR modes. The current
4104 * usable slave array is formed in the control path. The xmit function
4105 * just calculates hash and sends the packet out.
4107 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4108 struct net_device *dev)
4110 struct bonding *bond = netdev_priv(dev);
4111 struct slave *slave;
4112 struct bond_up_slave *slaves;
4115 slaves = rcu_dereference(bond->slave_arr);
4116 count = slaves ? READ_ONCE(slaves->count) : 0;
4117 if (likely(count)) {
4118 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4119 bond_dev_queue_xmit(bond, skb, slave->dev);
4121 bond_tx_drop(dev, skb);
4124 return NETDEV_TX_OK;
4127 /* in broadcast mode, we send everything to all usable interfaces. */
4128 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4129 struct net_device *bond_dev)
4131 struct bonding *bond = netdev_priv(bond_dev);
4132 struct slave *slave = NULL;
4133 struct list_head *iter;
4135 bond_for_each_slave_rcu(bond, slave, iter) {
4136 if (bond_is_last_slave(bond, slave))
4138 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4139 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4142 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4143 bond_dev->name, __func__);
4146 bond_dev_queue_xmit(bond, skb2, slave->dev);
4149 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4150 bond_dev_queue_xmit(bond, skb, slave->dev);
4152 bond_tx_drop(bond_dev, skb);
4154 return NETDEV_TX_OK;
4157 /*------------------------- Device initialization ---------------------------*/
4159 /* Lookup the slave that corresponds to a qid */
4160 static inline int bond_slave_override(struct bonding *bond,
4161 struct sk_buff *skb)
4163 struct slave *slave = NULL;
4164 struct list_head *iter;
4166 if (!skb_rx_queue_recorded(skb))
4169 /* Find out if any slaves have the same mapping as this skb. */
4170 bond_for_each_slave_rcu(bond, slave, iter) {
4171 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4172 if (bond_slave_is_up(slave) &&
4173 slave->link == BOND_LINK_UP) {
4174 bond_dev_queue_xmit(bond, skb, slave->dev);
4177 /* If the slave isn't UP, use default transmit policy. */
4186 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4187 struct net_device *sb_dev)
4189 /* This helper function exists to help dev_pick_tx get the correct
4190 * destination queue. Using a helper function skips a call to
4191 * skb_tx_hash and will put the skbs in the queue we expect on their
4192 * way down to the bonding driver.
4194 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4196 /* Save the original txq to restore before passing to the driver */
4197 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4199 if (unlikely(txq >= dev->real_num_tx_queues)) {
4201 txq -= dev->real_num_tx_queues;
4202 } while (txq >= dev->real_num_tx_queues);
4207 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4209 struct bonding *bond = netdev_priv(dev);
4211 if (bond_should_override_tx_queue(bond) &&
4212 !bond_slave_override(bond, skb))
4213 return NETDEV_TX_OK;
4215 switch (BOND_MODE(bond)) {
4216 case BOND_MODE_ROUNDROBIN:
4217 return bond_xmit_roundrobin(skb, dev);
4218 case BOND_MODE_ACTIVEBACKUP:
4219 return bond_xmit_activebackup(skb, dev);
4220 case BOND_MODE_8023AD:
4222 return bond_3ad_xor_xmit(skb, dev);
4223 case BOND_MODE_BROADCAST:
4224 return bond_xmit_broadcast(skb, dev);
4226 return bond_alb_xmit(skb, dev);
4228 return bond_tlb_xmit(skb, dev);
4230 /* Should never happen, mode already checked */
4231 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4233 bond_tx_drop(dev, skb);
4234 return NETDEV_TX_OK;
4238 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4240 struct bonding *bond = netdev_priv(dev);
4241 netdev_tx_t ret = NETDEV_TX_OK;
4243 /* If we risk deadlock from transmitting this in the
4244 * netpoll path, tell netpoll to queue the frame for later tx
4246 if (unlikely(is_netpoll_tx_blocked(dev)))
4247 return NETDEV_TX_BUSY;
4250 if (bond_has_slaves(bond))
4251 ret = __bond_start_xmit(skb, dev);
4253 bond_tx_drop(dev, skb);
4259 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4260 struct ethtool_link_ksettings *cmd)
4262 struct bonding *bond = netdev_priv(bond_dev);
4263 unsigned long speed = 0;
4264 struct list_head *iter;
4265 struct slave *slave;
4267 cmd->base.duplex = DUPLEX_UNKNOWN;
4268 cmd->base.port = PORT_OTHER;
4270 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4271 * do not need to check mode. Though link speed might not represent
4272 * the true receive or transmit bandwidth (not all modes are symmetric)
4273 * this is an accurate maximum.
4275 bond_for_each_slave(bond, slave, iter) {
4276 if (bond_slave_can_tx(slave)) {
4277 if (slave->speed != SPEED_UNKNOWN)
4278 speed += slave->speed;
4279 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4280 slave->duplex != DUPLEX_UNKNOWN)
4281 cmd->base.duplex = slave->duplex;
4284 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4289 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4290 struct ethtool_drvinfo *drvinfo)
4292 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4293 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4294 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4298 static const struct ethtool_ops bond_ethtool_ops = {
4299 .get_drvinfo = bond_ethtool_get_drvinfo,
4300 .get_link = ethtool_op_get_link,
4301 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4304 static const struct net_device_ops bond_netdev_ops = {
4305 .ndo_init = bond_init,
4306 .ndo_uninit = bond_uninit,
4307 .ndo_open = bond_open,
4308 .ndo_stop = bond_close,
4309 .ndo_start_xmit = bond_start_xmit,
4310 .ndo_select_queue = bond_select_queue,
4311 .ndo_get_stats64 = bond_get_stats,
4312 .ndo_do_ioctl = bond_do_ioctl,
4313 .ndo_change_rx_flags = bond_change_rx_flags,
4314 .ndo_set_rx_mode = bond_set_rx_mode,
4315 .ndo_change_mtu = bond_change_mtu,
4316 .ndo_set_mac_address = bond_set_mac_address,
4317 .ndo_neigh_setup = bond_neigh_setup,
4318 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4319 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4320 #ifdef CONFIG_NET_POLL_CONTROLLER
4321 .ndo_netpoll_setup = bond_netpoll_setup,
4322 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4323 .ndo_poll_controller = bond_poll_controller,
4325 .ndo_add_slave = bond_enslave,
4326 .ndo_del_slave = bond_release,
4327 .ndo_fix_features = bond_fix_features,
4328 .ndo_features_check = passthru_features_check,
4331 static const struct device_type bond_type = {
4335 static void bond_destructor(struct net_device *bond_dev)
4337 struct bonding *bond = netdev_priv(bond_dev);
4339 destroy_workqueue(bond->wq);
4342 void bond_setup(struct net_device *bond_dev)
4344 struct bonding *bond = netdev_priv(bond_dev);
4346 spin_lock_init(&bond->mode_lock);
4347 bond->params = bonding_defaults;
4349 /* Initialize pointers */
4350 bond->dev = bond_dev;
4352 /* Initialize the device entry points */
4353 ether_setup(bond_dev);
4354 bond_dev->max_mtu = ETH_MAX_MTU;
4355 bond_dev->netdev_ops = &bond_netdev_ops;
4356 bond_dev->ethtool_ops = &bond_ethtool_ops;
4358 bond_dev->needs_free_netdev = true;
4359 bond_dev->priv_destructor = bond_destructor;
4361 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4363 /* Initialize the device options */
4364 bond_dev->flags |= IFF_MASTER;
4365 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4366 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4368 /* don't acquire bond device's netif_tx_lock when transmitting */
4369 bond_dev->features |= NETIF_F_LLTX;
4371 /* By default, we declare the bond to be fully
4372 * VLAN hardware accelerated capable. Special
4373 * care is taken in the various xmit functions
4374 * when there are slaves that are not hw accel
4378 /* Don't allow bond devices to change network namespaces. */
4379 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4381 bond_dev->hw_features = BOND_VLAN_FEATURES |
4382 NETIF_F_HW_VLAN_CTAG_RX |
4383 NETIF_F_HW_VLAN_CTAG_FILTER;
4385 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4386 bond_dev->features |= bond_dev->hw_features;
4387 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4390 /* Destroy a bonding device.
4391 * Must be under rtnl_lock when this function is called.
4393 static void bond_uninit(struct net_device *bond_dev)
4395 struct bonding *bond = netdev_priv(bond_dev);
4396 struct list_head *iter;
4397 struct slave *slave;
4398 struct bond_up_slave *arr;
4400 bond_netpoll_cleanup(bond_dev);
4402 /* Release the bonded slaves */
4403 bond_for_each_slave(bond, slave, iter)
4404 __bond_release_one(bond_dev, slave->dev, true, true);
4405 netdev_info(bond_dev, "Released all slaves\n");
4407 arr = rtnl_dereference(bond->slave_arr);
4409 RCU_INIT_POINTER(bond->slave_arr, NULL);
4410 kfree_rcu(arr, rcu);
4413 list_del(&bond->bond_list);
4415 lockdep_unregister_key(&bond->stats_lock_key);
4416 bond_debug_unregister(bond);
4419 /*------------------------- Module initialization ---------------------------*/
4421 static int bond_check_params(struct bond_params *params)
4423 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4424 struct bond_opt_value newval;
4425 const struct bond_opt_value *valptr;
4426 int arp_all_targets_value = 0;
4427 u16 ad_actor_sys_prio = 0;
4428 u16 ad_user_port_key = 0;
4429 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4431 int bond_mode = BOND_MODE_ROUNDROBIN;
4432 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4436 /* Convert string parameters. */
4438 bond_opt_initstr(&newval, mode);
4439 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4441 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4444 bond_mode = valptr->value;
4447 if (xmit_hash_policy) {
4448 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4449 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4450 bond_mode == BOND_MODE_BROADCAST) {
4451 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4452 bond_mode_name(bond_mode));
4454 bond_opt_initstr(&newval, xmit_hash_policy);
4455 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4458 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4462 xmit_hashtype = valptr->value;
4467 if (bond_mode != BOND_MODE_8023AD) {
4468 pr_info("lacp_rate param is irrelevant in mode %s\n",
4469 bond_mode_name(bond_mode));
4471 bond_opt_initstr(&newval, lacp_rate);
4472 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4475 pr_err("Error: Invalid lacp rate \"%s\"\n",
4479 lacp_fast = valptr->value;
4484 bond_opt_initstr(&newval, ad_select);
4485 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4488 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4491 params->ad_select = valptr->value;
4492 if (bond_mode != BOND_MODE_8023AD)
4493 pr_warn("ad_select param only affects 802.3ad mode\n");
4495 params->ad_select = BOND_AD_STABLE;
4498 if (max_bonds < 0) {
4499 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4500 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4501 max_bonds = BOND_DEFAULT_MAX_BONDS;
4505 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4511 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4516 if (downdelay < 0) {
4517 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4518 downdelay, INT_MAX);
4522 if ((use_carrier != 0) && (use_carrier != 1)) {
4523 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4528 if (num_peer_notif < 0 || num_peer_notif > 255) {
4529 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4534 /* reset values for 802.3ad/TLB/ALB */
4535 if (!bond_mode_uses_arp(bond_mode)) {
4537 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");
4538 pr_warn("Forcing miimon to 100msec\n");
4539 miimon = BOND_DEFAULT_MIIMON;
4543 if (tx_queues < 1 || tx_queues > 255) {
4544 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4545 tx_queues, BOND_DEFAULT_TX_QUEUES);
4546 tx_queues = BOND_DEFAULT_TX_QUEUES;
4549 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4550 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4552 all_slaves_active = 0;
4555 if (resend_igmp < 0 || resend_igmp > 255) {
4556 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4557 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4558 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4561 bond_opt_initval(&newval, packets_per_slave);
4562 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4563 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4564 packets_per_slave, USHRT_MAX);
4565 packets_per_slave = 1;
4568 if (bond_mode == BOND_MODE_ALB) {
4569 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",
4574 if (updelay || downdelay) {
4575 /* just warn the user the up/down delay will have
4576 * no effect since miimon is zero...
4578 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",
4579 updelay, downdelay);
4582 /* don't allow arp monitoring */
4584 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4585 miimon, arp_interval);
4589 if ((updelay % miimon) != 0) {
4590 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4591 updelay, miimon, (updelay / miimon) * miimon);
4596 if ((downdelay % miimon) != 0) {
4597 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4599 (downdelay / miimon) * miimon);
4602 downdelay /= miimon;
4605 if (arp_interval < 0) {
4606 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4607 arp_interval, INT_MAX);
4611 for (arp_ip_count = 0, i = 0;
4612 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4615 /* not a complete check, but good enough to catch mistakes */
4616 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4617 !bond_is_ip_target_ok(ip)) {
4618 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4622 if (bond_get_targets_ip(arp_target, ip) == -1)
4623 arp_target[arp_ip_count++] = ip;
4625 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4630 if (arp_interval && !arp_ip_count) {
4631 /* don't allow arping if no arp_ip_target given... */
4632 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4638 if (!arp_interval) {
4639 pr_err("arp_validate requires arp_interval\n");
4643 bond_opt_initstr(&newval, arp_validate);
4644 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4647 pr_err("Error: invalid arp_validate \"%s\"\n",
4651 arp_validate_value = valptr->value;
4653 arp_validate_value = 0;
4656 if (arp_all_targets) {
4657 bond_opt_initstr(&newval, arp_all_targets);
4658 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4661 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4663 arp_all_targets_value = 0;
4665 arp_all_targets_value = valptr->value;
4670 pr_info("MII link monitoring set to %d ms\n", miimon);
4671 } else if (arp_interval) {
4672 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4673 arp_validate_value);
4674 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4675 arp_interval, valptr->string, arp_ip_count);
4677 for (i = 0; i < arp_ip_count; i++)
4678 pr_cont(" %s", arp_ip_target[i]);
4682 } else if (max_bonds) {
4683 /* miimon and arp_interval not set, we need one so things
4684 * work as expected, see bonding.txt for details
4686 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");
4689 if (primary && !bond_mode_uses_primary(bond_mode)) {
4690 /* currently, using a primary only makes sense
4691 * in active backup, TLB or ALB modes
4693 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4694 primary, bond_mode_name(bond_mode));
4698 if (primary && primary_reselect) {
4699 bond_opt_initstr(&newval, primary_reselect);
4700 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4703 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4707 primary_reselect_value = valptr->value;
4709 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4712 if (fail_over_mac) {
4713 bond_opt_initstr(&newval, fail_over_mac);
4714 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4717 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4721 fail_over_mac_value = valptr->value;
4722 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4723 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4725 fail_over_mac_value = BOND_FOM_NONE;
4728 bond_opt_initstr(&newval, "default");
4729 valptr = bond_opt_parse(
4730 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4733 pr_err("Error: No ad_actor_sys_prio default value");
4736 ad_actor_sys_prio = valptr->value;
4738 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4741 pr_err("Error: No ad_user_port_key default value");
4744 ad_user_port_key = valptr->value;
4746 bond_opt_initstr(&newval, "default");
4747 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4749 pr_err("Error: No tlb_dynamic_lb default value");
4752 tlb_dynamic_lb = valptr->value;
4754 if (lp_interval == 0) {
4755 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4756 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4757 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4760 /* fill params struct with the proper values */
4761 params->mode = bond_mode;
4762 params->xmit_policy = xmit_hashtype;
4763 params->miimon = miimon;
4764 params->num_peer_notif = num_peer_notif;
4765 params->arp_interval = arp_interval;
4766 params->arp_validate = arp_validate_value;
4767 params->arp_all_targets = arp_all_targets_value;
4768 params->updelay = updelay;
4769 params->downdelay = downdelay;
4770 params->peer_notif_delay = 0;
4771 params->use_carrier = use_carrier;
4772 params->lacp_fast = lacp_fast;
4773 params->primary[0] = 0;
4774 params->primary_reselect = primary_reselect_value;
4775 params->fail_over_mac = fail_over_mac_value;
4776 params->tx_queues = tx_queues;
4777 params->all_slaves_active = all_slaves_active;
4778 params->resend_igmp = resend_igmp;
4779 params->min_links = min_links;
4780 params->lp_interval = lp_interval;
4781 params->packets_per_slave = packets_per_slave;
4782 params->tlb_dynamic_lb = tlb_dynamic_lb;
4783 params->ad_actor_sys_prio = ad_actor_sys_prio;
4784 eth_zero_addr(params->ad_actor_system);
4785 params->ad_user_port_key = ad_user_port_key;
4786 if (packets_per_slave > 0) {
4787 params->reciprocal_packets_per_slave =
4788 reciprocal_value(packets_per_slave);
4790 /* reciprocal_packets_per_slave is unused if
4791 * packets_per_slave is 0 or 1, just initialize it
4793 params->reciprocal_packets_per_slave =
4794 (struct reciprocal_value) { 0 };
4798 strncpy(params->primary, primary, IFNAMSIZ);
4799 params->primary[IFNAMSIZ - 1] = 0;
4802 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4807 /* Called from registration process */
4808 static int bond_init(struct net_device *bond_dev)
4810 struct bonding *bond = netdev_priv(bond_dev);
4811 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4813 netdev_dbg(bond_dev, "Begin bond_init\n");
4815 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4819 spin_lock_init(&bond->stats_lock);
4820 lockdep_register_key(&bond->stats_lock_key);
4821 lockdep_set_class(&bond->stats_lock, &bond->stats_lock_key);
4823 list_add_tail(&bond->bond_list, &bn->dev_list);
4825 bond_prepare_sysfs_group(bond);
4827 bond_debug_register(bond);
4829 /* Ensure valid dev_addr */
4830 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4831 bond_dev->addr_assign_type == NET_ADDR_PERM)
4832 eth_hw_addr_random(bond_dev);
4837 unsigned int bond_get_num_tx_queues(void)
4842 /* Create a new bond based on the specified name and bonding parameters.
4843 * If name is NULL, obtain a suitable "bond%d" name for us.
4844 * Caller must NOT hold rtnl_lock; we need to release it here before we
4845 * set up our sysfs entries.
4847 int bond_create(struct net *net, const char *name)
4849 struct net_device *bond_dev;
4850 struct bonding *bond;
4851 struct alb_bond_info *bond_info;
4856 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4857 name ? name : "bond%d", NET_NAME_UNKNOWN,
4858 bond_setup, tx_queues);
4860 pr_err("%s: eek! can't alloc netdev!\n", name);
4866 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4867 * It is set to 0 by default which is wrong.
4869 bond = netdev_priv(bond_dev);
4870 bond_info = &(BOND_ALB_INFO(bond));
4871 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4873 dev_net_set(bond_dev, net);
4874 bond_dev->rtnl_link_ops = &bond_link_ops;
4876 res = register_netdevice(bond_dev);
4878 netif_carrier_off(bond_dev);
4880 bond_work_init_all(bond);
4884 free_netdev(bond_dev);
4888 static int __net_init bond_net_init(struct net *net)
4890 struct bond_net *bn = net_generic(net, bond_net_id);
4893 INIT_LIST_HEAD(&bn->dev_list);
4895 bond_create_proc_dir(bn);
4896 bond_create_sysfs(bn);
4901 static void __net_exit bond_net_exit(struct net *net)
4903 struct bond_net *bn = net_generic(net, bond_net_id);
4904 struct bonding *bond, *tmp_bond;
4907 bond_destroy_sysfs(bn);
4909 /* Kill off any bonds created after unregistering bond rtnl ops */
4911 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4912 unregister_netdevice_queue(bond->dev, &list);
4913 unregister_netdevice_many(&list);
4916 bond_destroy_proc_dir(bn);
4919 static struct pernet_operations bond_net_ops = {
4920 .init = bond_net_init,
4921 .exit = bond_net_exit,
4923 .size = sizeof(struct bond_net),
4926 static int __init bonding_init(void)
4931 pr_info("%s", bond_version);
4933 res = bond_check_params(&bonding_defaults);
4937 res = register_pernet_subsys(&bond_net_ops);
4941 res = bond_netlink_init();
4945 bond_create_debugfs();
4947 for (i = 0; i < max_bonds; i++) {
4948 res = bond_create(&init_net, NULL);
4953 skb_flow_dissector_init(&flow_keys_bonding,
4954 flow_keys_bonding_keys,
4955 ARRAY_SIZE(flow_keys_bonding_keys));
4957 register_netdevice_notifier(&bond_netdev_notifier);
4961 bond_destroy_debugfs();
4962 bond_netlink_fini();
4964 unregister_pernet_subsys(&bond_net_ops);
4969 static void __exit bonding_exit(void)
4971 unregister_netdevice_notifier(&bond_netdev_notifier);
4973 bond_destroy_debugfs();
4975 bond_netlink_fini();
4976 unregister_pernet_subsys(&bond_net_ops);
4978 #ifdef CONFIG_NET_POLL_CONTROLLER
4979 /* Make sure we don't have an imbalance on our netpoll blocking */
4980 WARN_ON(atomic_read(&netpoll_block_tx));
4984 module_init(bonding_init);
4985 module_exit(bonding_exit);
4986 MODULE_LICENSE("GPL");
4987 MODULE_VERSION(DRV_VERSION);
4988 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4989 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");