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_keys.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
89 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
90 static int num_peer_notif
= 1;
94 static int use_carrier
= 1;
97 static char *primary_reselect
;
98 static char *lacp_rate
;
100 static char *ad_select
;
101 static char *xmit_hash_policy
;
102 static int arp_interval
;
103 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
104 static char *arp_validate
;
105 static char *arp_all_targets
;
106 static char *fail_over_mac
;
107 static int all_slaves_active
;
108 static struct bond_params bonding_defaults
;
109 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
110 static int packets_per_slave
= 1;
111 static int lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
113 module_param(max_bonds
, int, 0);
114 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
115 module_param(tx_queues
, int, 0);
116 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
117 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
118 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
119 "failover event (alias of num_unsol_na)");
120 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
121 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
122 "failover event (alias of num_grat_arp)");
123 module_param(miimon
, int, 0);
124 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
125 module_param(updelay
, int, 0);
126 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
127 module_param(downdelay
, int, 0);
128 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
130 module_param(use_carrier
, int, 0);
131 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132 "0 for off, 1 for on (default)");
133 module_param(mode
, charp
, 0);
134 MODULE_PARM_DESC(mode
, "Mode of operation; 0 for balance-rr, "
135 "1 for active-backup, 2 for balance-xor, "
136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137 "6 for balance-alb");
138 module_param(primary
, charp
, 0);
139 MODULE_PARM_DESC(primary
, "Primary network device to use");
140 module_param(primary_reselect
, charp
, 0);
141 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
143 "0 for always (default), "
144 "1 for only if speed of primary is "
146 "2 for only on active slave "
148 module_param(lacp_rate
, charp
, 0);
149 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner; "
150 "0 for slow, 1 for fast");
151 module_param(ad_select
, charp
, 0);
152 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
153 "0 for stable (default), 1 for bandwidth, "
155 module_param(min_links
, int, 0);
156 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
158 module_param(xmit_hash_policy
, charp
, 0);
159 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
160 "0 for layer 2 (default), 1 for layer 3+4, "
161 "2 for layer 2+3, 3 for encap layer 2+3, "
162 "4 for encap layer 3+4");
163 module_param(arp_interval
, int, 0);
164 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
165 module_param_array(arp_ip_target
, charp
, NULL
, 0);
166 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
167 module_param(arp_validate
, charp
, 0);
168 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
169 "0 for none (default), 1 for active, "
170 "2 for backup, 3 for all");
171 module_param(arp_all_targets
, charp
, 0);
172 MODULE_PARM_DESC(arp_all_targets
, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
173 module_param(fail_over_mac
, charp
, 0);
174 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
175 "the same MAC; 0 for none (default), "
176 "1 for active, 2 for follow");
177 module_param(all_slaves_active
, int, 0);
178 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
179 "by setting active flag for all slaves; "
180 "0 for never (default), 1 for always.");
181 module_param(resend_igmp
, int, 0);
182 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
184 module_param(packets_per_slave
, int, 0);
185 MODULE_PARM_DESC(packets_per_slave
, "Packets to send per slave in balance-rr "
186 "mode; 0 for a random slave, 1 packet per "
187 "slave (default), >1 packets per slave.");
188 module_param(lp_interval
, uint
, 0);
189 MODULE_PARM_DESC(lp_interval
, "The number of seconds between instances where "
190 "the bonding driver sends learning packets to "
191 "each slaves peer switch. The default is 1.");
193 /*----------------------------- Global variables ----------------------------*/
195 #ifdef CONFIG_NET_POLL_CONTROLLER
196 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
199 int bond_net_id __read_mostly
;
201 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
202 static int arp_ip_count
;
203 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
204 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
205 static int lacp_fast
;
207 /*-------------------------- Forward declarations ---------------------------*/
209 static int bond_init(struct net_device
*bond_dev
);
210 static void bond_uninit(struct net_device
*bond_dev
);
212 /*---------------------------- General routines -----------------------------*/
214 const char *bond_mode_name(int mode
)
216 static const char *names
[] = {
217 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
218 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
219 [BOND_MODE_XOR
] = "load balancing (xor)",
220 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
221 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
222 [BOND_MODE_TLB
] = "transmit load balancing",
223 [BOND_MODE_ALB
] = "adaptive load balancing",
226 if (mode
< BOND_MODE_ROUNDROBIN
|| mode
> BOND_MODE_ALB
)
232 /*---------------------------------- VLAN -----------------------------------*/
235 * bond_dev_queue_xmit - Prepare skb for xmit.
237 * @bond: bond device that got this skb for tx.
238 * @skb: hw accel VLAN tagged skb to transmit
239 * @slave_dev: slave that is supposed to xmit this skbuff
241 void bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
242 struct net_device
*slave_dev
)
244 skb
->dev
= slave_dev
;
246 BUILD_BUG_ON(sizeof(skb
->queue_mapping
) !=
247 sizeof(qdisc_skb_cb(skb
)->slave_dev_queue_mapping
));
248 skb
->queue_mapping
= qdisc_skb_cb(skb
)->slave_dev_queue_mapping
;
250 if (unlikely(netpoll_tx_running(bond
->dev
)))
251 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
256 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
257 * We don't protect the slave list iteration with a lock because:
258 * a. This operation is performed in IOCTL context,
259 * b. The operation is protected by the RTNL semaphore in the 8021q code,
260 * c. Holding a lock with BH disabled while directly calling a base driver
261 * entry point is generally a BAD idea.
263 * The design of synchronization/protection for this operation in the 8021q
264 * module is good for one or more VLAN devices over a single physical device
265 * and cannot be extended for a teaming solution like bonding, so there is a
266 * potential race condition here where a net device from the vlan group might
267 * be referenced (either by a base driver or the 8021q code) while it is being
268 * removed from the system. However, it turns out we're not making matters
269 * worse, and if it works for regular VLAN usage it will work here too.
273 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
274 * @bond_dev: bonding net device that got called
275 * @vid: vlan id being added
277 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
,
278 __be16 proto
, u16 vid
)
280 struct bonding
*bond
= netdev_priv(bond_dev
);
281 struct slave
*slave
, *rollback_slave
;
282 struct list_head
*iter
;
285 bond_for_each_slave(bond
, slave
, iter
) {
286 res
= vlan_vid_add(slave
->dev
, proto
, vid
);
294 /* unwind to the slave that failed */
295 bond_for_each_slave(bond
, rollback_slave
, iter
) {
296 if (rollback_slave
== slave
)
299 vlan_vid_del(rollback_slave
->dev
, proto
, vid
);
306 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
307 * @bond_dev: bonding net device that got called
308 * @vid: vlan id being removed
310 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
,
311 __be16 proto
, u16 vid
)
313 struct bonding
*bond
= netdev_priv(bond_dev
);
314 struct list_head
*iter
;
317 bond_for_each_slave(bond
, slave
, iter
)
318 vlan_vid_del(slave
->dev
, proto
, vid
);
320 if (bond_is_lb(bond
))
321 bond_alb_clear_vlan(bond
, vid
);
326 /*------------------------------- Link status -------------------------------*/
328 /* Set the carrier state for the master according to the state of its
329 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
330 * do special 802.3ad magic.
332 * Returns zero if carrier state does not change, nonzero if it does.
334 static int bond_set_carrier(struct bonding
*bond
)
336 struct list_head
*iter
;
339 if (!bond_has_slaves(bond
))
342 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
343 return bond_3ad_set_carrier(bond
);
345 bond_for_each_slave(bond
, slave
, iter
) {
346 if (slave
->link
== BOND_LINK_UP
) {
347 if (!netif_carrier_ok(bond
->dev
)) {
348 netif_carrier_on(bond
->dev
);
356 if (netif_carrier_ok(bond
->dev
)) {
357 netif_carrier_off(bond
->dev
);
363 /* Get link speed and duplex from the slave's base driver
364 * using ethtool. If for some reason the call fails or the
365 * values are invalid, set speed and duplex to -1,
368 static void bond_update_speed_duplex(struct slave
*slave
)
370 struct net_device
*slave_dev
= slave
->dev
;
371 struct ethtool_cmd ecmd
;
375 slave
->speed
= SPEED_UNKNOWN
;
376 slave
->duplex
= DUPLEX_UNKNOWN
;
378 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
382 slave_speed
= ethtool_cmd_speed(&ecmd
);
383 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
386 switch (ecmd
.duplex
) {
394 slave
->speed
= slave_speed
;
395 slave
->duplex
= ecmd
.duplex
;
400 const char *bond_slave_link_status(s8 link
)
416 /* if <dev> supports MII link status reporting, check its link status.
418 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
419 * depending upon the setting of the use_carrier parameter.
421 * Return either BMSR_LSTATUS, meaning that the link is up (or we
422 * can't tell and just pretend it is), or 0, meaning that the link is
425 * If reporting is non-zero, instead of faking link up, return -1 if
426 * both ETHTOOL and MII ioctls fail (meaning the device does not
427 * support them). If use_carrier is set, return whatever it says.
428 * It'd be nice if there was a good way to tell if a driver supports
429 * netif_carrier, but there really isn't.
431 static int bond_check_dev_link(struct bonding
*bond
,
432 struct net_device
*slave_dev
, int reporting
)
434 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
435 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
437 struct mii_ioctl_data
*mii
;
439 if (!reporting
&& !netif_running(slave_dev
))
442 if (bond
->params
.use_carrier
)
443 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
445 /* Try to get link status using Ethtool first. */
446 if (slave_dev
->ethtool_ops
->get_link
)
447 return slave_dev
->ethtool_ops
->get_link(slave_dev
) ?
450 /* Ethtool can't be used, fallback to MII ioctls. */
451 ioctl
= slave_ops
->ndo_do_ioctl
;
453 /* TODO: set pointer to correct ioctl on a per team member
454 * bases to make this more efficient. that is, once
455 * we determine the correct ioctl, we will always
456 * call it and not the others for that team
460 /* We cannot assume that SIOCGMIIPHY will also read a
461 * register; not all network drivers (e.g., e100)
465 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
466 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
468 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
469 mii
->reg_num
= MII_BMSR
;
470 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
471 return mii
->val_out
& BMSR_LSTATUS
;
475 /* If reporting, report that either there's no dev->do_ioctl,
476 * or both SIOCGMIIREG and get_link failed (meaning that we
477 * cannot report link status). If not reporting, pretend
480 return reporting
? -1 : BMSR_LSTATUS
;
483 /*----------------------------- Multicast list ------------------------------*/
485 /* Push the promiscuity flag down to appropriate slaves */
486 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
488 struct list_head
*iter
;
491 if (bond_uses_primary(bond
)) {
492 struct slave
*curr_active
= rtnl_dereference(bond
->curr_active_slave
);
495 err
= dev_set_promiscuity(curr_active
->dev
, inc
);
499 bond_for_each_slave(bond
, slave
, iter
) {
500 err
= dev_set_promiscuity(slave
->dev
, inc
);
508 /* Push the allmulti flag down to all slaves */
509 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
511 struct list_head
*iter
;
514 if (bond_uses_primary(bond
)) {
515 struct slave
*curr_active
= rtnl_dereference(bond
->curr_active_slave
);
518 err
= dev_set_allmulti(curr_active
->dev
, inc
);
522 bond_for_each_slave(bond
, slave
, iter
) {
523 err
= dev_set_allmulti(slave
->dev
, inc
);
531 /* Retrieve the list of registered multicast addresses for the bonding
532 * device and retransmit an IGMP JOIN request to the current active
535 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
537 struct bonding
*bond
= container_of(work
, struct bonding
,
540 if (!rtnl_trylock()) {
541 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
544 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, bond
->dev
);
546 if (bond
->igmp_retrans
> 1) {
547 bond
->igmp_retrans
--;
548 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
553 /* Flush bond's hardware addresses from slave */
554 static void bond_hw_addr_flush(struct net_device
*bond_dev
,
555 struct net_device
*slave_dev
)
557 struct bonding
*bond
= netdev_priv(bond_dev
);
559 dev_uc_unsync(slave_dev
, bond_dev
);
560 dev_mc_unsync(slave_dev
, bond_dev
);
562 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
563 /* del lacpdu mc addr from mc list */
564 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
566 dev_mc_del(slave_dev
, lacpdu_multicast
);
570 /*--------------------------- Active slave change ---------------------------*/
572 /* Update the hardware address list and promisc/allmulti for the new and
573 * old active slaves (if any). Modes that are not using primary keep all
574 * slaves up date at all times; only the modes that use primary need to call
575 * this function to swap these settings during a failover.
577 static void bond_hw_addr_swap(struct bonding
*bond
, struct slave
*new_active
,
578 struct slave
*old_active
)
581 if (bond
->dev
->flags
& IFF_PROMISC
)
582 dev_set_promiscuity(old_active
->dev
, -1);
584 if (bond
->dev
->flags
& IFF_ALLMULTI
)
585 dev_set_allmulti(old_active
->dev
, -1);
587 bond_hw_addr_flush(bond
->dev
, old_active
->dev
);
591 /* FIXME: Signal errors upstream. */
592 if (bond
->dev
->flags
& IFF_PROMISC
)
593 dev_set_promiscuity(new_active
->dev
, 1);
595 if (bond
->dev
->flags
& IFF_ALLMULTI
)
596 dev_set_allmulti(new_active
->dev
, 1);
598 netif_addr_lock_bh(bond
->dev
);
599 dev_uc_sync(new_active
->dev
, bond
->dev
);
600 dev_mc_sync(new_active
->dev
, bond
->dev
);
601 netif_addr_unlock_bh(bond
->dev
);
606 * bond_set_dev_addr - clone slave's address to bond
607 * @bond_dev: bond net device
608 * @slave_dev: slave net device
610 * Should be called with RTNL held.
612 static void bond_set_dev_addr(struct net_device
*bond_dev
,
613 struct net_device
*slave_dev
)
615 netdev_dbg(bond_dev
, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
616 bond_dev
, slave_dev
, slave_dev
->addr_len
);
617 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
618 bond_dev
->addr_assign_type
= NET_ADDR_STOLEN
;
619 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
622 /* bond_do_fail_over_mac
624 * Perform special MAC address swapping for fail_over_mac settings
628 static void bond_do_fail_over_mac(struct bonding
*bond
,
629 struct slave
*new_active
,
630 struct slave
*old_active
)
632 u8 tmp_mac
[ETH_ALEN
];
633 struct sockaddr saddr
;
636 switch (bond
->params
.fail_over_mac
) {
637 case BOND_FOM_ACTIVE
:
639 bond_set_dev_addr(bond
->dev
, new_active
->dev
);
641 case BOND_FOM_FOLLOW
:
642 /* if new_active && old_active, swap them
643 * if just old_active, do nothing (going to no active slave)
644 * if just new_active, set new_active to bond's MAC
650 ether_addr_copy(tmp_mac
, new_active
->dev
->dev_addr
);
651 ether_addr_copy(saddr
.sa_data
,
652 old_active
->dev
->dev_addr
);
653 saddr
.sa_family
= new_active
->dev
->type
;
655 ether_addr_copy(saddr
.sa_data
, bond
->dev
->dev_addr
);
656 saddr
.sa_family
= bond
->dev
->type
;
659 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
661 netdev_err(bond
->dev
, "Error %d setting MAC of slave %s\n",
662 -rv
, new_active
->dev
->name
);
669 ether_addr_copy(saddr
.sa_data
, tmp_mac
);
670 saddr
.sa_family
= old_active
->dev
->type
;
672 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
674 netdev_err(bond
->dev
, "Error %d setting MAC of slave %s\n",
675 -rv
, new_active
->dev
->name
);
679 netdev_err(bond
->dev
, "bond_do_fail_over_mac impossible: bad policy %d\n",
680 bond
->params
.fail_over_mac
);
686 static bool bond_should_change_active(struct bonding
*bond
)
688 struct slave
*prim
= rtnl_dereference(bond
->primary_slave
);
689 struct slave
*curr
= rtnl_dereference(bond
->curr_active_slave
);
691 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
693 if (bond
->force_primary
) {
694 bond
->force_primary
= false;
697 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
698 (prim
->speed
< curr
->speed
||
699 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
701 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
707 * find_best_interface - select the best available slave to be the active one
708 * @bond: our bonding struct
710 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
712 struct slave
*slave
, *bestslave
= NULL
, *primary
;
713 struct list_head
*iter
;
714 int mintime
= bond
->params
.updelay
;
716 primary
= rtnl_dereference(bond
->primary_slave
);
717 if (primary
&& primary
->link
== BOND_LINK_UP
&&
718 bond_should_change_active(bond
))
721 bond_for_each_slave(bond
, slave
, iter
) {
722 if (slave
->link
== BOND_LINK_UP
)
724 if (slave
->link
== BOND_LINK_BACK
&& bond_slave_is_up(slave
) &&
725 slave
->delay
< mintime
) {
726 mintime
= slave
->delay
;
734 static bool bond_should_notify_peers(struct bonding
*bond
)
739 slave
= rcu_dereference(bond
->curr_active_slave
);
742 netdev_dbg(bond
->dev
, "bond_should_notify_peers: slave %s\n",
743 slave
? slave
->dev
->name
: "NULL");
745 if (!slave
|| !bond
->send_peer_notif
||
746 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
753 * change_active_interface - change the active slave into the specified one
754 * @bond: our bonding struct
755 * @new: the new slave to make the active one
757 * Set the new slave to the bond's settings and unset them on the old
759 * Setting include flags, mc-list, promiscuity, allmulti, etc.
761 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
762 * because it is apparently the best available slave we have, even though its
763 * updelay hasn't timed out yet.
765 * Caller must hold RTNL.
767 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
769 struct slave
*old_active
;
773 old_active
= rtnl_dereference(bond
->curr_active_slave
);
775 if (old_active
== new_active
)
779 new_active
->last_link_up
= jiffies
;
781 if (new_active
->link
== BOND_LINK_BACK
) {
782 if (bond_uses_primary(bond
)) {
783 netdev_info(bond
->dev
, "making interface %s the new active one %d ms earlier\n",
784 new_active
->dev
->name
,
785 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
788 new_active
->delay
= 0;
789 new_active
->link
= BOND_LINK_UP
;
791 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
792 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
794 if (bond_is_lb(bond
))
795 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
797 if (bond_uses_primary(bond
)) {
798 netdev_info(bond
->dev
, "making interface %s the new active one\n",
799 new_active
->dev
->name
);
804 if (bond_uses_primary(bond
))
805 bond_hw_addr_swap(bond
, new_active
, old_active
);
807 if (bond_is_lb(bond
)) {
808 bond_alb_handle_active_change(bond
, new_active
);
810 bond_set_slave_inactive_flags(old_active
,
811 BOND_SLAVE_NOTIFY_NOW
);
813 bond_set_slave_active_flags(new_active
,
814 BOND_SLAVE_NOTIFY_NOW
);
816 rcu_assign_pointer(bond
->curr_active_slave
, new_active
);
819 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
) {
821 bond_set_slave_inactive_flags(old_active
,
822 BOND_SLAVE_NOTIFY_NOW
);
825 bool should_notify_peers
= false;
827 bond_set_slave_active_flags(new_active
,
828 BOND_SLAVE_NOTIFY_NOW
);
830 if (bond
->params
.fail_over_mac
)
831 bond_do_fail_over_mac(bond
, new_active
,
834 if (netif_running(bond
->dev
)) {
835 bond
->send_peer_notif
=
836 bond
->params
.num_peer_notif
;
837 should_notify_peers
=
838 bond_should_notify_peers(bond
);
841 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
842 if (should_notify_peers
)
843 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
848 /* resend IGMP joins since active slave has changed or
849 * all were sent on curr_active_slave.
850 * resend only if bond is brought up with the affected
851 * bonding modes and the retransmission is enabled
853 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
854 ((bond_uses_primary(bond
) && new_active
) ||
855 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
)) {
856 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
857 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
862 * bond_select_active_slave - select a new active slave, if needed
863 * @bond: our bonding struct
865 * This functions should be called when one of the following occurs:
866 * - The old curr_active_slave has been released or lost its link.
867 * - The primary_slave has got its link back.
868 * - A slave has got its link back and there's no old curr_active_slave.
870 * Caller must hold RTNL.
872 void bond_select_active_slave(struct bonding
*bond
)
874 struct slave
*best_slave
;
879 best_slave
= bond_find_best_slave(bond
);
880 if (best_slave
!= rtnl_dereference(bond
->curr_active_slave
)) {
881 bond_change_active_slave(bond
, best_slave
);
882 rv
= bond_set_carrier(bond
);
886 if (netif_carrier_ok(bond
->dev
)) {
887 netdev_info(bond
->dev
, "first active interface up!\n");
889 netdev_info(bond
->dev
, "now running without any active interface!\n");
894 #ifdef CONFIG_NET_POLL_CONTROLLER
895 static inline int slave_enable_netpoll(struct slave
*slave
)
900 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
905 err
= __netpoll_setup(np
, slave
->dev
);
914 static inline void slave_disable_netpoll(struct slave
*slave
)
916 struct netpoll
*np
= slave
->np
;
922 __netpoll_free_async(np
);
925 static void bond_poll_controller(struct net_device
*bond_dev
)
929 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
931 struct bonding
*bond
= netdev_priv(bond_dev
);
932 struct list_head
*iter
;
935 bond_for_each_slave(bond
, slave
, iter
)
936 if (bond_slave_is_up(slave
))
937 slave_disable_netpoll(slave
);
940 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
942 struct bonding
*bond
= netdev_priv(dev
);
943 struct list_head
*iter
;
947 bond_for_each_slave(bond
, slave
, iter
) {
948 err
= slave_enable_netpoll(slave
);
950 bond_netpoll_cleanup(dev
);
957 static inline int slave_enable_netpoll(struct slave
*slave
)
961 static inline void slave_disable_netpoll(struct slave
*slave
)
964 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
969 /*---------------------------------- IOCTL ----------------------------------*/
971 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
972 netdev_features_t features
)
974 struct bonding
*bond
= netdev_priv(dev
);
975 struct list_head
*iter
;
976 netdev_features_t mask
;
980 features
&= ~NETIF_F_ONE_FOR_ALL
;
981 features
|= NETIF_F_ALL_FOR_ALL
;
983 bond_for_each_slave(bond
, slave
, iter
) {
984 features
= netdev_increment_features(features
,
985 slave
->dev
->features
,
988 features
= netdev_add_tso_features(features
, mask
);
993 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
994 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
995 NETIF_F_HIGHDMA | NETIF_F_LRO)
997 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
998 NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL)
1000 static void bond_compute_features(struct bonding
*bond
)
1002 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1003 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1004 netdev_features_t enc_features
= BOND_ENC_FEATURES
;
1005 struct net_device
*bond_dev
= bond
->dev
;
1006 struct list_head
*iter
;
1007 struct slave
*slave
;
1008 unsigned short max_hard_header_len
= ETH_HLEN
;
1009 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1010 u16 gso_max_segs
= GSO_MAX_SEGS
;
1012 if (!bond_has_slaves(bond
))
1014 vlan_features
&= NETIF_F_ALL_FOR_ALL
;
1016 bond_for_each_slave(bond
, slave
, iter
) {
1017 vlan_features
= netdev_increment_features(vlan_features
,
1018 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1020 enc_features
= netdev_increment_features(enc_features
,
1021 slave
->dev
->hw_enc_features
,
1023 dst_release_flag
&= slave
->dev
->priv_flags
;
1024 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1025 max_hard_header_len
= slave
->dev
->hard_header_len
;
1027 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1028 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1032 bond_dev
->vlan_features
= vlan_features
;
1033 bond_dev
->hw_enc_features
= enc_features
;
1034 bond_dev
->hard_header_len
= max_hard_header_len
;
1035 bond_dev
->gso_max_segs
= gso_max_segs
;
1036 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1038 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1039 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1041 netdev_change_features(bond_dev
);
1044 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1045 struct net_device
*slave_dev
)
1047 bond_dev
->header_ops
= slave_dev
->header_ops
;
1049 bond_dev
->type
= slave_dev
->type
;
1050 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1051 bond_dev
->addr_len
= slave_dev
->addr_len
;
1053 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1054 slave_dev
->addr_len
);
1057 /* On bonding slaves other than the currently active slave, suppress
1058 * duplicates except for alb non-mcast/bcast.
1060 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1061 struct slave
*slave
,
1062 struct bonding
*bond
)
1064 if (bond_is_slave_inactive(slave
)) {
1065 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1066 skb
->pkt_type
!= PACKET_BROADCAST
&&
1067 skb
->pkt_type
!= PACKET_MULTICAST
)
1074 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1076 struct sk_buff
*skb
= *pskb
;
1077 struct slave
*slave
;
1078 struct bonding
*bond
;
1079 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1081 int ret
= RX_HANDLER_ANOTHER
;
1083 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1085 return RX_HANDLER_CONSUMED
;
1089 slave
= bond_slave_get_rcu(skb
->dev
);
1092 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1094 ret
= recv_probe(skb
, bond
, slave
);
1095 if (ret
== RX_HANDLER_CONSUMED
) {
1101 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1102 return RX_HANDLER_EXACT
;
1105 skb
->dev
= bond
->dev
;
1107 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1108 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1109 skb
->pkt_type
== PACKET_HOST
) {
1111 if (unlikely(skb_cow_head(skb
,
1112 skb
->data
- skb_mac_header(skb
)))) {
1114 return RX_HANDLER_CONSUMED
;
1116 ether_addr_copy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
);
1122 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1123 struct net_device
*slave_dev
,
1124 struct slave
*slave
)
1128 err
= netdev_master_upper_dev_link_private(slave_dev
, bond_dev
, slave
);
1131 slave_dev
->flags
|= IFF_SLAVE
;
1132 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1136 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1137 struct net_device
*slave_dev
)
1139 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1140 slave_dev
->flags
&= ~IFF_SLAVE
;
1141 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1144 static struct slave
*bond_alloc_slave(struct bonding
*bond
)
1146 struct slave
*slave
= NULL
;
1148 slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1152 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1153 SLAVE_AD_INFO(slave
) = kzalloc(sizeof(struct ad_slave_info
),
1155 if (!SLAVE_AD_INFO(slave
)) {
1163 static void bond_free_slave(struct slave
*slave
)
1165 struct bonding
*bond
= bond_get_bond_by_slave(slave
);
1167 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1168 kfree(SLAVE_AD_INFO(slave
));
1173 /* enslave device <slave> to bond device <master> */
1174 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1176 struct bonding
*bond
= netdev_priv(bond_dev
);
1177 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1178 struct slave
*new_slave
= NULL
, *prev_slave
;
1179 struct sockaddr addr
;
1183 if (!bond
->params
.use_carrier
&&
1184 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1185 slave_ops
->ndo_do_ioctl
== NULL
) {
1186 netdev_warn(bond_dev
, "no link monitoring support for %s\n",
1190 /* already enslaved */
1191 if (slave_dev
->flags
& IFF_SLAVE
) {
1192 netdev_dbg(bond_dev
, "Error: Device was already enslaved\n");
1196 if (bond_dev
== slave_dev
) {
1197 netdev_err(bond_dev
, "cannot enslave bond to itself.\n");
1201 /* vlan challenged mutual exclusion */
1202 /* no need to lock since we're protected by rtnl_lock */
1203 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1204 netdev_dbg(bond_dev
, "%s is NETIF_F_VLAN_CHALLENGED\n",
1206 if (vlan_uses_dev(bond_dev
)) {
1207 netdev_err(bond_dev
, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1208 slave_dev
->name
, bond_dev
->name
);
1211 netdev_warn(bond_dev
, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1212 slave_dev
->name
, slave_dev
->name
,
1216 netdev_dbg(bond_dev
, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1220 /* Old ifenslave binaries are no longer supported. These can
1221 * be identified with moderate accuracy by the state of the slave:
1222 * the current ifenslave will set the interface down prior to
1223 * enslaving it; the old ifenslave will not.
1225 if ((slave_dev
->flags
& IFF_UP
)) {
1226 netdev_err(bond_dev
, "%s is up - this may be due to an out of date ifenslave\n",
1229 goto err_undo_flags
;
1232 /* set bonding device ether type by slave - bonding netdevices are
1233 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1234 * there is a need to override some of the type dependent attribs/funcs.
1236 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1237 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1239 if (!bond_has_slaves(bond
)) {
1240 if (bond_dev
->type
!= slave_dev
->type
) {
1241 netdev_dbg(bond_dev
, "change device type from %d to %d\n",
1242 bond_dev
->type
, slave_dev
->type
);
1244 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1246 res
= notifier_to_errno(res
);
1248 netdev_err(bond_dev
, "refused to change device type\n");
1250 goto err_undo_flags
;
1253 /* Flush unicast and multicast addresses */
1254 dev_uc_flush(bond_dev
);
1255 dev_mc_flush(bond_dev
);
1257 if (slave_dev
->type
!= ARPHRD_ETHER
)
1258 bond_setup_by_slave(bond_dev
, slave_dev
);
1260 ether_setup(bond_dev
);
1261 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1264 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1267 } else if (bond_dev
->type
!= slave_dev
->type
) {
1268 netdev_err(bond_dev
, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1269 slave_dev
->name
, slave_dev
->type
, bond_dev
->type
);
1271 goto err_undo_flags
;
1274 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1275 netdev_warn(bond_dev
, "The slave device specified does not support setting the MAC address\n");
1276 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
&&
1277 bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1278 if (!bond_has_slaves(bond
)) {
1279 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1280 netdev_warn(bond_dev
, "Setting fail_over_mac to active for active-backup mode\n");
1282 netdev_err(bond_dev
, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1284 goto err_undo_flags
;
1289 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1291 /* If this is the first slave, then we need to set the master's hardware
1292 * address to be the same as the slave's.
1294 if (!bond_has_slaves(bond
) &&
1295 bond
->dev
->addr_assign_type
== NET_ADDR_RANDOM
)
1296 bond_set_dev_addr(bond
->dev
, slave_dev
);
1298 new_slave
= bond_alloc_slave(bond
);
1301 goto err_undo_flags
;
1304 new_slave
->bond
= bond
;
1305 new_slave
->dev
= slave_dev
;
1306 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1307 * is set via sysfs or module option if desired.
1309 new_slave
->queue_id
= 0;
1311 /* Save slave's original mtu and then set it to match the bond */
1312 new_slave
->original_mtu
= slave_dev
->mtu
;
1313 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1315 netdev_dbg(bond_dev
, "Error %d calling dev_set_mtu\n", res
);
1319 /* Save slave's original ("permanent") mac address for modes
1320 * that need it, and for restoring it upon release, and then
1321 * set it to the master's address
1323 ether_addr_copy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
);
1325 if (!bond
->params
.fail_over_mac
||
1326 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1327 /* Set slave to master's mac address. The application already
1328 * set the master's mac address to that of the first slave
1330 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1331 addr
.sa_family
= slave_dev
->type
;
1332 res
= dev_set_mac_address(slave_dev
, &addr
);
1334 netdev_dbg(bond_dev
, "Error %d calling set_mac_address\n", res
);
1335 goto err_restore_mtu
;
1339 /* open the slave since the application closed it */
1340 res
= dev_open(slave_dev
);
1342 netdev_dbg(bond_dev
, "Opening slave %s failed\n", slave_dev
->name
);
1343 goto err_restore_mac
;
1346 slave_dev
->priv_flags
|= IFF_BONDING
;
1348 if (bond_is_lb(bond
)) {
1349 /* bond_alb_init_slave() must be called before all other stages since
1350 * it might fail and we do not want to have to undo everything
1352 res
= bond_alb_init_slave(bond
, new_slave
);
1357 /* If the mode uses primary, then the following is handled by
1358 * bond_change_active_slave().
1360 if (!bond_uses_primary(bond
)) {
1361 /* set promiscuity level to new slave */
1362 if (bond_dev
->flags
& IFF_PROMISC
) {
1363 res
= dev_set_promiscuity(slave_dev
, 1);
1368 /* set allmulti level to new slave */
1369 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1370 res
= dev_set_allmulti(slave_dev
, 1);
1375 netif_addr_lock_bh(bond_dev
);
1377 dev_mc_sync_multiple(slave_dev
, bond_dev
);
1378 dev_uc_sync_multiple(slave_dev
, bond_dev
);
1380 netif_addr_unlock_bh(bond_dev
);
1383 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1384 /* add lacpdu mc addr to mc list */
1385 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1387 dev_mc_add(slave_dev
, lacpdu_multicast
);
1390 res
= vlan_vids_add_by_dev(slave_dev
, bond_dev
);
1392 netdev_err(bond_dev
, "Couldn't add bond vlan ids to %s\n",
1397 prev_slave
= bond_last_slave(bond
);
1399 new_slave
->delay
= 0;
1400 new_slave
->link_failure_count
= 0;
1402 bond_update_speed_duplex(new_slave
);
1404 new_slave
->last_rx
= jiffies
-
1405 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1406 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
1407 new_slave
->target_last_arp_rx
[i
] = new_slave
->last_rx
;
1409 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1410 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1412 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1413 /* miimon is set but a bonded network driver
1414 * does not support ETHTOOL/MII and
1415 * arp_interval is not set. Note: if
1416 * use_carrier is enabled, we will never go
1417 * here (because netif_carrier is always
1418 * supported); thus, we don't need to change
1419 * the messages for netif_carrier.
1421 netdev_warn(bond_dev
, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1423 } else if (link_reporting
== -1) {
1424 /* unable get link status using mii/ethtool */
1425 netdev_warn(bond_dev
, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1430 /* check for initial state */
1431 if (bond
->params
.miimon
) {
1432 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1433 if (bond
->params
.updelay
) {
1434 new_slave
->link
= BOND_LINK_BACK
;
1435 new_slave
->delay
= bond
->params
.updelay
;
1437 new_slave
->link
= BOND_LINK_UP
;
1440 new_slave
->link
= BOND_LINK_DOWN
;
1442 } else if (bond
->params
.arp_interval
) {
1443 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1444 BOND_LINK_UP
: BOND_LINK_DOWN
);
1446 new_slave
->link
= BOND_LINK_UP
;
1449 if (new_slave
->link
!= BOND_LINK_DOWN
)
1450 new_slave
->last_link_up
= jiffies
;
1451 netdev_dbg(bond_dev
, "Initial state of slave_dev is BOND_LINK_%s\n",
1452 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1453 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1455 if (bond_uses_primary(bond
) && bond
->params
.primary
[0]) {
1456 /* if there is a primary slave, remember it */
1457 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1458 rcu_assign_pointer(bond
->primary_slave
, new_slave
);
1459 bond
->force_primary
= true;
1463 switch (BOND_MODE(bond
)) {
1464 case BOND_MODE_ACTIVEBACKUP
:
1465 bond_set_slave_inactive_flags(new_slave
,
1466 BOND_SLAVE_NOTIFY_NOW
);
1468 case BOND_MODE_8023AD
:
1469 /* in 802.3ad mode, the internal mechanism
1470 * will activate the slaves in the selected
1473 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1474 /* if this is the first slave */
1476 SLAVE_AD_INFO(new_slave
)->id
= 1;
1477 /* Initialize AD with the number of times that the AD timer is called in 1 second
1478 * can be called only after the mac address of the bond is set
1480 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1482 SLAVE_AD_INFO(new_slave
)->id
=
1483 SLAVE_AD_INFO(prev_slave
)->id
+ 1;
1486 bond_3ad_bind_slave(new_slave
);
1490 bond_set_active_slave(new_slave
);
1491 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1494 netdev_dbg(bond_dev
, "This slave is always active in trunk mode\n");
1496 /* always active in trunk mode */
1497 bond_set_active_slave(new_slave
);
1499 /* In trunking mode there is little meaning to curr_active_slave
1500 * anyway (it holds no special properties of the bond device),
1501 * so we can change it without calling change_active_interface()
1503 if (!rcu_access_pointer(bond
->curr_active_slave
) &&
1504 new_slave
->link
== BOND_LINK_UP
)
1505 rcu_assign_pointer(bond
->curr_active_slave
, new_slave
);
1508 } /* switch(bond_mode) */
1510 #ifdef CONFIG_NET_POLL_CONTROLLER
1511 slave_dev
->npinfo
= bond
->dev
->npinfo
;
1512 if (slave_dev
->npinfo
) {
1513 if (slave_enable_netpoll(new_slave
)) {
1514 netdev_info(bond_dev
, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1521 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1524 netdev_dbg(bond_dev
, "Error %d calling netdev_rx_handler_register\n", res
);
1528 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
, new_slave
);
1530 netdev_dbg(bond_dev
, "Error %d calling bond_master_upper_dev_link\n", res
);
1531 goto err_unregister
;
1534 res
= bond_sysfs_slave_add(new_slave
);
1536 netdev_dbg(bond_dev
, "Error %d calling bond_sysfs_slave_add\n", res
);
1537 goto err_upper_unlink
;
1541 bond_compute_features(bond
);
1542 bond_set_carrier(bond
);
1544 if (bond_uses_primary(bond
)) {
1546 bond_select_active_slave(bond
);
1547 unblock_netpoll_tx();
1550 netdev_info(bond_dev
, "Enslaving %s as %s interface with %s link\n",
1552 bond_is_active_slave(new_slave
) ? "an active" : "a backup",
1553 new_slave
->link
!= BOND_LINK_DOWN
? "an up" : "a down");
1555 /* enslave is successful */
1558 /* Undo stages on error */
1560 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1563 netdev_rx_handler_unregister(slave_dev
);
1566 if (!bond_uses_primary(bond
))
1567 bond_hw_addr_flush(bond_dev
, slave_dev
);
1569 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1570 if (rcu_access_pointer(bond
->primary_slave
) == new_slave
)
1571 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
1572 if (rcu_access_pointer(bond
->curr_active_slave
) == new_slave
) {
1574 bond_change_active_slave(bond
, NULL
);
1575 bond_select_active_slave(bond
);
1576 unblock_netpoll_tx();
1578 /* either primary_slave or curr_active_slave might've changed */
1580 slave_disable_netpoll(new_slave
);
1583 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1584 dev_close(slave_dev
);
1587 if (!bond
->params
.fail_over_mac
||
1588 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1589 /* XXX TODO - fom follow mode needs to change master's
1590 * MAC if this slave's MAC is in use by the bond, or at
1591 * least print a warning.
1593 ether_addr_copy(addr
.sa_data
, new_slave
->perm_hwaddr
);
1594 addr
.sa_family
= slave_dev
->type
;
1595 dev_set_mac_address(slave_dev
, &addr
);
1599 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1602 bond_free_slave(new_slave
);
1605 /* Enslave of first slave has failed and we need to fix master's mac */
1606 if (!bond_has_slaves(bond
) &&
1607 ether_addr_equal_64bits(bond_dev
->dev_addr
, slave_dev
->dev_addr
))
1608 eth_hw_addr_random(bond_dev
);
1613 /* Try to release the slave device <slave> from the bond device <master>
1614 * It is legal to access curr_active_slave without a lock because all the function
1615 * is RTNL-locked. If "all" is true it means that the function is being called
1616 * while destroying a bond interface and all slaves are being released.
1618 * The rules for slave state should be:
1619 * for Active/Backup:
1620 * Active stays on all backups go down
1621 * for Bonded connections:
1622 * The first up interface should be left on and all others downed.
1624 static int __bond_release_one(struct net_device
*bond_dev
,
1625 struct net_device
*slave_dev
,
1628 struct bonding
*bond
= netdev_priv(bond_dev
);
1629 struct slave
*slave
, *oldcurrent
;
1630 struct sockaddr addr
;
1631 int old_flags
= bond_dev
->flags
;
1632 netdev_features_t old_features
= bond_dev
->features
;
1634 /* slave is not a slave or master is not master of this slave */
1635 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1636 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1637 netdev_err(bond_dev
, "cannot release %s\n",
1644 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1646 /* not a slave of this bond */
1647 netdev_info(bond_dev
, "%s not enslaved\n",
1649 unblock_netpoll_tx();
1653 bond_sysfs_slave_del(slave
);
1655 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1656 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1657 * for this slave anymore.
1659 netdev_rx_handler_unregister(slave_dev
);
1661 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1662 bond_3ad_unbind_slave(slave
);
1664 netdev_info(bond_dev
, "Releasing %s interface %s\n",
1665 bond_is_active_slave(slave
) ? "active" : "backup",
1668 oldcurrent
= rcu_access_pointer(bond
->curr_active_slave
);
1670 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
1672 if (!all
&& (!bond
->params
.fail_over_mac
||
1673 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
)) {
1674 if (ether_addr_equal_64bits(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1675 bond_has_slaves(bond
))
1676 netdev_warn(bond_dev
, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1677 slave_dev
->name
, slave
->perm_hwaddr
,
1678 bond_dev
->name
, slave_dev
->name
);
1681 if (rtnl_dereference(bond
->primary_slave
) == slave
)
1682 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
1684 if (oldcurrent
== slave
)
1685 bond_change_active_slave(bond
, NULL
);
1687 if (bond_is_lb(bond
)) {
1688 /* Must be called only after the slave has been
1689 * detached from the list and the curr_active_slave
1690 * has been cleared (if our_slave == old_current),
1691 * but before a new active slave is selected.
1693 bond_alb_deinit_slave(bond
, slave
);
1697 RCU_INIT_POINTER(bond
->curr_active_slave
, NULL
);
1698 } else if (oldcurrent
== slave
) {
1699 /* Note that we hold RTNL over this sequence, so there
1700 * is no concern that another slave add/remove event
1703 bond_select_active_slave(bond
);
1706 if (!bond_has_slaves(bond
)) {
1707 bond_set_carrier(bond
);
1708 eth_hw_addr_random(bond_dev
);
1711 unblock_netpoll_tx();
1715 if (!bond_has_slaves(bond
)) {
1716 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
1717 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
1720 bond_compute_features(bond
);
1721 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1722 (old_features
& NETIF_F_VLAN_CHALLENGED
))
1723 netdev_info(bond_dev
, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1724 slave_dev
->name
, bond_dev
->name
);
1726 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1728 /* If the mode uses primary, then this case was handled above by
1729 * bond_change_active_slave(..., NULL)
1731 if (!bond_uses_primary(bond
)) {
1732 /* unset promiscuity level from slave
1733 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1734 * of the IFF_PROMISC flag in the bond_dev, but we need the
1735 * value of that flag before that change, as that was the value
1736 * when this slave was attached, so we cache at the start of the
1737 * function and use it here. Same goes for ALLMULTI below
1739 if (old_flags
& IFF_PROMISC
)
1740 dev_set_promiscuity(slave_dev
, -1);
1742 /* unset allmulti level from slave */
1743 if (old_flags
& IFF_ALLMULTI
)
1744 dev_set_allmulti(slave_dev
, -1);
1746 bond_hw_addr_flush(bond_dev
, slave_dev
);
1749 slave_disable_netpoll(slave
);
1751 /* close slave before restoring its mac address */
1752 dev_close(slave_dev
);
1754 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
||
1755 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1756 /* restore original ("permanent") mac address */
1757 ether_addr_copy(addr
.sa_data
, slave
->perm_hwaddr
);
1758 addr
.sa_family
= slave_dev
->type
;
1759 dev_set_mac_address(slave_dev
, &addr
);
1762 dev_set_mtu(slave_dev
, slave
->original_mtu
);
1764 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1766 bond_free_slave(slave
);
1771 /* A wrapper used because of ndo_del_link */
1772 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1774 return __bond_release_one(bond_dev
, slave_dev
, false);
1777 /* First release a slave and then destroy the bond if no more slaves are left.
1778 * Must be under rtnl_lock when this function is called.
1780 static int bond_release_and_destroy(struct net_device
*bond_dev
,
1781 struct net_device
*slave_dev
)
1783 struct bonding
*bond
= netdev_priv(bond_dev
);
1786 ret
= bond_release(bond_dev
, slave_dev
);
1787 if (ret
== 0 && !bond_has_slaves(bond
)) {
1788 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1789 netdev_info(bond_dev
, "Destroying bond %s\n",
1791 unregister_netdevice(bond_dev
);
1796 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1798 struct bonding
*bond
= netdev_priv(bond_dev
);
1800 info
->bond_mode
= BOND_MODE(bond
);
1801 info
->miimon
= bond
->params
.miimon
;
1802 info
->num_slaves
= bond
->slave_cnt
;
1807 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
1809 struct bonding
*bond
= netdev_priv(bond_dev
);
1810 struct list_head
*iter
;
1811 int i
= 0, res
= -ENODEV
;
1812 struct slave
*slave
;
1814 bond_for_each_slave(bond
, slave
, iter
) {
1815 if (i
++ == (int)info
->slave_id
) {
1817 strcpy(info
->slave_name
, slave
->dev
->name
);
1818 info
->link
= slave
->link
;
1819 info
->state
= bond_slave_state(slave
);
1820 info
->link_failure_count
= slave
->link_failure_count
;
1828 /*-------------------------------- Monitoring -------------------------------*/
1830 /* called with rcu_read_lock() */
1831 static int bond_miimon_inspect(struct bonding
*bond
)
1833 int link_state
, commit
= 0;
1834 struct list_head
*iter
;
1835 struct slave
*slave
;
1836 bool ignore_updelay
;
1838 ignore_updelay
= !rcu_dereference(bond
->curr_active_slave
);
1840 bond_for_each_slave_rcu(bond
, slave
, iter
) {
1841 slave
->new_link
= BOND_LINK_NOCHANGE
;
1843 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
1845 switch (slave
->link
) {
1850 slave
->link
= BOND_LINK_FAIL
;
1851 slave
->delay
= bond
->params
.downdelay
;
1853 netdev_info(bond
->dev
, "link status down for %sinterface %s, disabling it in %d ms\n",
1855 BOND_MODE_ACTIVEBACKUP
) ?
1856 (bond_is_active_slave(slave
) ?
1857 "active " : "backup ") : "",
1859 bond
->params
.downdelay
* bond
->params
.miimon
);
1862 case BOND_LINK_FAIL
:
1864 /* recovered before downdelay expired */
1865 slave
->link
= BOND_LINK_UP
;
1866 slave
->last_link_up
= jiffies
;
1867 netdev_info(bond
->dev
, "link status up again after %d ms for interface %s\n",
1868 (bond
->params
.downdelay
- slave
->delay
) *
1869 bond
->params
.miimon
,
1874 if (slave
->delay
<= 0) {
1875 slave
->new_link
= BOND_LINK_DOWN
;
1883 case BOND_LINK_DOWN
:
1887 slave
->link
= BOND_LINK_BACK
;
1888 slave
->delay
= bond
->params
.updelay
;
1891 netdev_info(bond
->dev
, "link status up for interface %s, enabling it in %d ms\n",
1893 ignore_updelay
? 0 :
1894 bond
->params
.updelay
*
1895 bond
->params
.miimon
);
1898 case BOND_LINK_BACK
:
1900 slave
->link
= BOND_LINK_DOWN
;
1901 netdev_info(bond
->dev
, "link status down again after %d ms for interface %s\n",
1902 (bond
->params
.updelay
- slave
->delay
) *
1903 bond
->params
.miimon
,
1912 if (slave
->delay
<= 0) {
1913 slave
->new_link
= BOND_LINK_UP
;
1915 ignore_updelay
= false;
1927 static void bond_miimon_commit(struct bonding
*bond
)
1929 struct list_head
*iter
;
1930 struct slave
*slave
, *primary
;
1932 bond_for_each_slave(bond
, slave
, iter
) {
1933 switch (slave
->new_link
) {
1934 case BOND_LINK_NOCHANGE
:
1938 slave
->link
= BOND_LINK_UP
;
1939 slave
->last_link_up
= jiffies
;
1941 primary
= rtnl_dereference(bond
->primary_slave
);
1942 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1943 /* prevent it from being the active one */
1944 bond_set_backup_slave(slave
);
1945 } else if (BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1946 /* make it immediately active */
1947 bond_set_active_slave(slave
);
1948 } else if (slave
!= primary
) {
1949 /* prevent it from being the active one */
1950 bond_set_backup_slave(slave
);
1953 netdev_info(bond
->dev
, "link status definitely up for interface %s, %u Mbps %s duplex\n",
1955 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
1956 slave
->duplex
? "full" : "half");
1958 /* notify ad that the link status has changed */
1959 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1960 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
1962 if (bond_is_lb(bond
))
1963 bond_alb_handle_link_change(bond
, slave
,
1966 if (!bond
->curr_active_slave
|| slave
== primary
)
1971 case BOND_LINK_DOWN
:
1972 if (slave
->link_failure_count
< UINT_MAX
)
1973 slave
->link_failure_count
++;
1975 slave
->link
= BOND_LINK_DOWN
;
1977 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
1978 BOND_MODE(bond
) == BOND_MODE_8023AD
)
1979 bond_set_slave_inactive_flags(slave
,
1980 BOND_SLAVE_NOTIFY_NOW
);
1982 netdev_info(bond
->dev
, "link status definitely down for interface %s, disabling it\n",
1985 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1986 bond_3ad_handle_link_change(slave
,
1989 if (bond_is_lb(bond
))
1990 bond_alb_handle_link_change(bond
, slave
,
1993 if (slave
== rcu_access_pointer(bond
->curr_active_slave
))
1999 netdev_err(bond
->dev
, "invalid new link %d on slave %s\n",
2000 slave
->new_link
, slave
->dev
->name
);
2001 slave
->new_link
= BOND_LINK_NOCHANGE
;
2008 bond_select_active_slave(bond
);
2009 unblock_netpoll_tx();
2012 bond_set_carrier(bond
);
2017 * Really a wrapper that splits the mii monitor into two phases: an
2018 * inspection, then (if inspection indicates something needs to be done)
2019 * an acquisition of appropriate locks followed by a commit phase to
2020 * implement whatever link state changes are indicated.
2022 static void bond_mii_monitor(struct work_struct
*work
)
2024 struct bonding
*bond
= container_of(work
, struct bonding
,
2026 bool should_notify_peers
= false;
2027 unsigned long delay
;
2029 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2031 if (!bond_has_slaves(bond
))
2036 should_notify_peers
= bond_should_notify_peers(bond
);
2038 if (bond_miimon_inspect(bond
)) {
2041 /* Race avoidance with bond_close cancel of workqueue */
2042 if (!rtnl_trylock()) {
2044 should_notify_peers
= false;
2048 bond_miimon_commit(bond
);
2050 rtnl_unlock(); /* might sleep, hold no other locks */
2055 if (bond
->params
.miimon
)
2056 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2058 if (should_notify_peers
) {
2059 if (!rtnl_trylock())
2061 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2066 static bool bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2068 struct net_device
*upper
;
2069 struct list_head
*iter
;
2072 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2076 netdev_for_each_all_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2077 if (ip
== bond_confirm_addr(upper
, 0, ip
)) {
2087 /* We go to the (large) trouble of VLAN tagging ARP frames because
2088 * switches in VLAN mode (especially if ports are configured as
2089 * "native" to a VLAN) might not pass non-tagged frames.
2091 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
,
2092 __be32 dest_ip
, __be32 src_ip
,
2093 struct bond_vlan_tag
*tags
)
2095 struct sk_buff
*skb
;
2096 struct bond_vlan_tag
*outer_tag
= tags
;
2098 netdev_dbg(slave_dev
, "arp %d on slave %s: dst %pI4 src %pI4\n",
2099 arp_op
, slave_dev
->name
, &dest_ip
, &src_ip
);
2101 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2102 NULL
, slave_dev
->dev_addr
, NULL
);
2105 net_err_ratelimited("ARP packet allocation failed\n");
2109 if (!tags
|| tags
->vlan_proto
== VLAN_N_VID
)
2114 /* Go through all the tags backwards and add them to the packet */
2115 while (tags
->vlan_proto
!= VLAN_N_VID
) {
2116 if (!tags
->vlan_id
) {
2121 netdev_dbg(slave_dev
, "inner tag: proto %X vid %X\n",
2122 ntohs(outer_tag
->vlan_proto
), tags
->vlan_id
);
2123 skb
= __vlan_put_tag(skb
, tags
->vlan_proto
,
2126 net_err_ratelimited("failed to insert inner VLAN tag\n");
2132 /* Set the outer tag */
2133 if (outer_tag
->vlan_id
) {
2134 netdev_dbg(slave_dev
, "outer tag: proto %X vid %X\n",
2135 ntohs(outer_tag
->vlan_proto
), outer_tag
->vlan_id
);
2136 skb
= vlan_put_tag(skb
, outer_tag
->vlan_proto
,
2137 outer_tag
->vlan_id
);
2139 net_err_ratelimited("failed to insert outer VLAN tag\n");
2148 /* Validate the device path between the @start_dev and the @end_dev.
2149 * The path is valid if the @end_dev is reachable through device
2151 * When the path is validated, collect any vlan information in the
2154 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
2155 struct net_device
*end_dev
,
2158 struct bond_vlan_tag
*tags
;
2159 struct net_device
*upper
;
2160 struct list_head
*iter
;
2162 if (start_dev
== end_dev
) {
2163 tags
= kzalloc(sizeof(*tags
) * (level
+ 1), GFP_ATOMIC
);
2165 return ERR_PTR(-ENOMEM
);
2166 tags
[level
].vlan_proto
= VLAN_N_VID
;
2170 netdev_for_each_upper_dev_rcu(start_dev
, upper
, iter
) {
2171 tags
= bond_verify_device_path(upper
, end_dev
, level
+ 1);
2172 if (IS_ERR_OR_NULL(tags
)) {
2177 if (is_vlan_dev(upper
)) {
2178 tags
[level
].vlan_proto
= vlan_dev_vlan_proto(upper
);
2179 tags
[level
].vlan_id
= vlan_dev_vlan_id(upper
);
2188 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2191 struct bond_vlan_tag
*tags
;
2192 __be32
*targets
= bond
->params
.arp_targets
, addr
;
2195 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
&& targets
[i
]; i
++) {
2196 netdev_dbg(bond
->dev
, "basa: target %pI4\n", &targets
[i
]);
2199 /* Find out through which dev should the packet go */
2200 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2203 /* there's no route to target - try to send arp
2204 * probe to generate any traffic (arp_validate=0)
2206 if (bond
->params
.arp_validate
)
2207 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2210 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2215 /* bond device itself */
2216 if (rt
->dst
.dev
== bond
->dev
)
2220 tags
= bond_verify_device_path(bond
->dev
, rt
->dst
.dev
, 0);
2223 if (!IS_ERR_OR_NULL(tags
))
2226 /* Not our device - skip */
2227 netdev_dbg(bond
->dev
, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2228 &targets
[i
], rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2234 addr
= bond_confirm_addr(rt
->dst
.dev
, targets
[i
], 0);
2236 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2242 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2246 if (!sip
|| !bond_has_this_ip(bond
, tip
)) {
2247 netdev_dbg(bond
->dev
, "bva: sip %pI4 tip %pI4 not found\n",
2252 i
= bond_get_targets_ip(bond
->params
.arp_targets
, sip
);
2254 netdev_dbg(bond
->dev
, "bva: sip %pI4 not found in targets\n",
2258 slave
->last_rx
= jiffies
;
2259 slave
->target_last_arp_rx
[i
] = jiffies
;
2262 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2263 struct slave
*slave
)
2265 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2266 struct slave
*curr_active_slave
;
2267 unsigned char *arp_ptr
;
2269 int alen
, is_arp
= skb
->protocol
== __cpu_to_be16(ETH_P_ARP
);
2271 if (!slave_do_arp_validate(bond
, slave
)) {
2272 if ((slave_do_arp_validate_only(bond
) && is_arp
) ||
2273 !slave_do_arp_validate_only(bond
))
2274 slave
->last_rx
= jiffies
;
2275 return RX_HANDLER_ANOTHER
;
2276 } else if (!is_arp
) {
2277 return RX_HANDLER_ANOTHER
;
2280 alen
= arp_hdr_len(bond
->dev
);
2282 netdev_dbg(bond
->dev
, "bond_arp_rcv: skb->dev %s\n",
2285 if (alen
> skb_headlen(skb
)) {
2286 arp
= kmalloc(alen
, GFP_ATOMIC
);
2289 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2293 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2294 skb
->pkt_type
== PACKET_OTHERHOST
||
2295 skb
->pkt_type
== PACKET_LOOPBACK
||
2296 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2297 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2301 arp_ptr
= (unsigned char *)(arp
+ 1);
2302 arp_ptr
+= bond
->dev
->addr_len
;
2303 memcpy(&sip
, arp_ptr
, 4);
2304 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2305 memcpy(&tip
, arp_ptr
, 4);
2307 netdev_dbg(bond
->dev
, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2308 slave
->dev
->name
, bond_slave_state(slave
),
2309 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2312 curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2314 /* Backup slaves won't see the ARP reply, but do come through
2315 * here for each ARP probe (so we swap the sip/tip to validate
2316 * the probe). In a "redundant switch, common router" type of
2317 * configuration, the ARP probe will (hopefully) travel from
2318 * the active, through one switch, the router, then the other
2319 * switch before reaching the backup.
2321 * We 'trust' the arp requests if there is an active slave and
2322 * it received valid arp reply(s) after it became active. This
2323 * is done to avoid endless looping when we can't reach the
2324 * arp_ip_target and fool ourselves with our own arp requests.
2327 if (bond_is_active_slave(slave
))
2328 bond_validate_arp(bond
, slave
, sip
, tip
);
2329 else if (curr_active_slave
&&
2330 time_after(slave_last_rx(bond
, curr_active_slave
),
2331 curr_active_slave
->last_link_up
))
2332 bond_validate_arp(bond
, slave
, tip
, sip
);
2335 if (arp
!= (struct arphdr
*)skb
->data
)
2337 return RX_HANDLER_ANOTHER
;
2340 /* function to verify if we're in the arp_interval timeslice, returns true if
2341 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2342 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2344 static bool bond_time_in_interval(struct bonding
*bond
, unsigned long last_act
,
2347 int delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2349 return time_in_range(jiffies
,
2350 last_act
- delta_in_ticks
,
2351 last_act
+ mod
* delta_in_ticks
+ delta_in_ticks
/2);
2354 /* This function is called regularly to monitor each slave's link
2355 * ensuring that traffic is being sent and received when arp monitoring
2356 * is used in load-balancing mode. if the adapter has been dormant, then an
2357 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2358 * arp monitoring in active backup mode.
2360 static void bond_loadbalance_arp_mon(struct work_struct
*work
)
2362 struct bonding
*bond
= container_of(work
, struct bonding
,
2364 struct slave
*slave
, *oldcurrent
;
2365 struct list_head
*iter
;
2366 int do_failover
= 0, slave_state_changed
= 0;
2368 if (!bond_has_slaves(bond
))
2373 oldcurrent
= rcu_dereference(bond
->curr_active_slave
);
2374 /* see if any of the previous devices are up now (i.e. they have
2375 * xmt and rcv traffic). the curr_active_slave does not come into
2376 * the picture unless it is null. also, slave->last_link_up is not
2377 * needed here because we send an arp on each slave and give a slave
2378 * as long as it needs to get the tx/rx within the delta.
2379 * TODO: what about up/down delay in arp mode? it wasn't here before
2382 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2383 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2385 if (slave
->link
!= BOND_LINK_UP
) {
2386 if (bond_time_in_interval(bond
, trans_start
, 1) &&
2387 bond_time_in_interval(bond
, slave
->last_rx
, 1)) {
2389 slave
->link
= BOND_LINK_UP
;
2390 slave_state_changed
= 1;
2392 /* primary_slave has no meaning in round-robin
2393 * mode. the window of a slave being up and
2394 * curr_active_slave being null after enslaving
2398 netdev_info(bond
->dev
, "link status definitely up for interface %s\n",
2402 netdev_info(bond
->dev
, "interface %s is now up\n",
2407 /* slave->link == BOND_LINK_UP */
2409 /* not all switches will respond to an arp request
2410 * when the source ip is 0, so don't take the link down
2411 * if we don't know our ip yet
2413 if (!bond_time_in_interval(bond
, trans_start
, 2) ||
2414 !bond_time_in_interval(bond
, slave
->last_rx
, 2)) {
2416 slave
->link
= BOND_LINK_DOWN
;
2417 slave_state_changed
= 1;
2419 if (slave
->link_failure_count
< UINT_MAX
)
2420 slave
->link_failure_count
++;
2422 netdev_info(bond
->dev
, "interface %s is now down\n",
2425 if (slave
== oldcurrent
)
2430 /* note: if switch is in round-robin mode, all links
2431 * must tx arp to ensure all links rx an arp - otherwise
2432 * links may oscillate or not come up at all; if switch is
2433 * in something like xor mode, there is nothing we can
2434 * do - all replies will be rx'ed on same link causing slaves
2435 * to be unstable during low/no traffic periods
2437 if (bond_slave_is_up(slave
))
2438 bond_arp_send_all(bond
, slave
);
2443 if (do_failover
|| slave_state_changed
) {
2444 if (!rtnl_trylock())
2447 if (slave_state_changed
) {
2448 bond_slave_state_change(bond
);
2449 } else if (do_failover
) {
2451 bond_select_active_slave(bond
);
2452 unblock_netpoll_tx();
2458 if (bond
->params
.arp_interval
)
2459 queue_delayed_work(bond
->wq
, &bond
->arp_work
,
2460 msecs_to_jiffies(bond
->params
.arp_interval
));
2463 /* Called to inspect slaves for active-backup mode ARP monitor link state
2464 * changes. Sets new_link in slaves to specify what action should take
2465 * place for the slave. Returns 0 if no changes are found, >0 if changes
2466 * to link states must be committed.
2468 * Called with rcu_read_lock held.
2470 static int bond_ab_arp_inspect(struct bonding
*bond
)
2472 unsigned long trans_start
, last_rx
;
2473 struct list_head
*iter
;
2474 struct slave
*slave
;
2477 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2478 slave
->new_link
= BOND_LINK_NOCHANGE
;
2479 last_rx
= slave_last_rx(bond
, slave
);
2481 if (slave
->link
!= BOND_LINK_UP
) {
2482 if (bond_time_in_interval(bond
, last_rx
, 1)) {
2483 slave
->new_link
= BOND_LINK_UP
;
2489 /* Give slaves 2*delta after being enslaved or made
2490 * active. This avoids bouncing, as the last receive
2491 * times need a full ARP monitor cycle to be updated.
2493 if (bond_time_in_interval(bond
, slave
->last_link_up
, 2))
2496 /* Backup slave is down if:
2497 * - No current_arp_slave AND
2498 * - more than 3*delta since last receive AND
2499 * - the bond has an IP address
2501 * Note: a non-null current_arp_slave indicates
2502 * the curr_active_slave went down and we are
2503 * searching for a new one; under this condition
2504 * we only take the curr_active_slave down - this
2505 * gives each slave a chance to tx/rx traffic
2506 * before being taken out
2508 if (!bond_is_active_slave(slave
) &&
2509 !rcu_access_pointer(bond
->current_arp_slave
) &&
2510 !bond_time_in_interval(bond
, last_rx
, 3)) {
2511 slave
->new_link
= BOND_LINK_DOWN
;
2515 /* Active slave is down if:
2516 * - more than 2*delta since transmitting OR
2517 * - (more than 2*delta since receive AND
2518 * the bond has an IP address)
2520 trans_start
= dev_trans_start(slave
->dev
);
2521 if (bond_is_active_slave(slave
) &&
2522 (!bond_time_in_interval(bond
, trans_start
, 2) ||
2523 !bond_time_in_interval(bond
, last_rx
, 2))) {
2524 slave
->new_link
= BOND_LINK_DOWN
;
2532 /* Called to commit link state changes noted by inspection step of
2533 * active-backup mode ARP monitor.
2535 * Called with RTNL hold.
2537 static void bond_ab_arp_commit(struct bonding
*bond
)
2539 unsigned long trans_start
;
2540 struct list_head
*iter
;
2541 struct slave
*slave
;
2543 bond_for_each_slave(bond
, slave
, iter
) {
2544 switch (slave
->new_link
) {
2545 case BOND_LINK_NOCHANGE
:
2549 trans_start
= dev_trans_start(slave
->dev
);
2550 if (rtnl_dereference(bond
->curr_active_slave
) != slave
||
2551 (!rtnl_dereference(bond
->curr_active_slave
) &&
2552 bond_time_in_interval(bond
, trans_start
, 1))) {
2553 struct slave
*current_arp_slave
;
2555 current_arp_slave
= rtnl_dereference(bond
->current_arp_slave
);
2556 slave
->link
= BOND_LINK_UP
;
2557 if (current_arp_slave
) {
2558 bond_set_slave_inactive_flags(
2560 BOND_SLAVE_NOTIFY_NOW
);
2561 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
2564 netdev_info(bond
->dev
, "link status definitely up for interface %s\n",
2567 if (!rtnl_dereference(bond
->curr_active_slave
) ||
2568 slave
== rtnl_dereference(bond
->primary_slave
))
2575 case BOND_LINK_DOWN
:
2576 if (slave
->link_failure_count
< UINT_MAX
)
2577 slave
->link_failure_count
++;
2579 slave
->link
= BOND_LINK_DOWN
;
2580 bond_set_slave_inactive_flags(slave
,
2581 BOND_SLAVE_NOTIFY_NOW
);
2583 netdev_info(bond
->dev
, "link status definitely down for interface %s, disabling it\n",
2586 if (slave
== rtnl_dereference(bond
->curr_active_slave
)) {
2587 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
2594 netdev_err(bond
->dev
, "impossible: new_link %d on slave %s\n",
2595 slave
->new_link
, slave
->dev
->name
);
2601 bond_select_active_slave(bond
);
2602 unblock_netpoll_tx();
2605 bond_set_carrier(bond
);
2608 /* Send ARP probes for active-backup mode ARP monitor.
2610 * Called with rcu_read_lock held.
2612 static bool bond_ab_arp_probe(struct bonding
*bond
)
2614 struct slave
*slave
, *before
= NULL
, *new_slave
= NULL
,
2615 *curr_arp_slave
= rcu_dereference(bond
->current_arp_slave
),
2616 *curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2617 struct list_head
*iter
;
2619 bool should_notify_rtnl
= BOND_SLAVE_NOTIFY_LATER
;
2621 if (curr_arp_slave
&& curr_active_slave
)
2622 netdev_info(bond
->dev
, "PROBE: c_arp %s && cas %s BAD\n",
2623 curr_arp_slave
->dev
->name
,
2624 curr_active_slave
->dev
->name
);
2626 if (curr_active_slave
) {
2627 bond_arp_send_all(bond
, curr_active_slave
);
2628 return should_notify_rtnl
;
2631 /* if we don't have a curr_active_slave, search for the next available
2632 * backup slave from the current_arp_slave and make it the candidate
2633 * for becoming the curr_active_slave
2636 if (!curr_arp_slave
) {
2637 curr_arp_slave
= bond_first_slave_rcu(bond
);
2638 if (!curr_arp_slave
)
2639 return should_notify_rtnl
;
2642 bond_set_slave_inactive_flags(curr_arp_slave
, BOND_SLAVE_NOTIFY_LATER
);
2644 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2645 if (!found
&& !before
&& bond_slave_is_up(slave
))
2648 if (found
&& !new_slave
&& bond_slave_is_up(slave
))
2650 /* if the link state is up at this point, we
2651 * mark it down - this can happen if we have
2652 * simultaneous link failures and
2653 * reselect_active_interface doesn't make this
2654 * one the current slave so it is still marked
2655 * up when it is actually down
2657 if (!bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
2658 slave
->link
= BOND_LINK_DOWN
;
2659 if (slave
->link_failure_count
< UINT_MAX
)
2660 slave
->link_failure_count
++;
2662 bond_set_slave_inactive_flags(slave
,
2663 BOND_SLAVE_NOTIFY_LATER
);
2665 netdev_info(bond
->dev
, "backup interface %s is now down\n",
2668 if (slave
== curr_arp_slave
)
2672 if (!new_slave
&& before
)
2678 new_slave
->link
= BOND_LINK_BACK
;
2679 bond_set_slave_active_flags(new_slave
, BOND_SLAVE_NOTIFY_LATER
);
2680 bond_arp_send_all(bond
, new_slave
);
2681 new_slave
->last_link_up
= jiffies
;
2682 rcu_assign_pointer(bond
->current_arp_slave
, new_slave
);
2685 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2686 if (slave
->should_notify
) {
2687 should_notify_rtnl
= BOND_SLAVE_NOTIFY_NOW
;
2691 return should_notify_rtnl
;
2694 static void bond_activebackup_arp_mon(struct work_struct
*work
)
2696 struct bonding
*bond
= container_of(work
, struct bonding
,
2698 bool should_notify_peers
= false;
2699 bool should_notify_rtnl
= false;
2702 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2704 if (!bond_has_slaves(bond
))
2709 should_notify_peers
= bond_should_notify_peers(bond
);
2711 if (bond_ab_arp_inspect(bond
)) {
2714 /* Race avoidance with bond_close flush of workqueue */
2715 if (!rtnl_trylock()) {
2717 should_notify_peers
= false;
2721 bond_ab_arp_commit(bond
);
2727 should_notify_rtnl
= bond_ab_arp_probe(bond
);
2731 if (bond
->params
.arp_interval
)
2732 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2734 if (should_notify_peers
|| should_notify_rtnl
) {
2735 if (!rtnl_trylock())
2738 if (should_notify_peers
)
2739 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
2741 if (should_notify_rtnl
)
2742 bond_slave_state_notify(bond
);
2748 /*-------------------------- netdev event handling --------------------------*/
2750 /* Change device name */
2751 static int bond_event_changename(struct bonding
*bond
)
2753 bond_remove_proc_entry(bond
);
2754 bond_create_proc_entry(bond
);
2756 bond_debug_reregister(bond
);
2761 static int bond_master_netdev_event(unsigned long event
,
2762 struct net_device
*bond_dev
)
2764 struct bonding
*event_bond
= netdev_priv(bond_dev
);
2767 case NETDEV_CHANGENAME
:
2768 return bond_event_changename(event_bond
);
2769 case NETDEV_UNREGISTER
:
2770 bond_remove_proc_entry(event_bond
);
2772 case NETDEV_REGISTER
:
2773 bond_create_proc_entry(event_bond
);
2775 case NETDEV_NOTIFY_PEERS
:
2776 if (event_bond
->send_peer_notif
)
2777 event_bond
->send_peer_notif
--;
2786 static int bond_slave_netdev_event(unsigned long event
,
2787 struct net_device
*slave_dev
)
2789 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
), *primary
;
2790 struct bonding
*bond
;
2791 struct net_device
*bond_dev
;
2795 /* A netdev event can be generated while enslaving a device
2796 * before netdev_rx_handler_register is called in which case
2797 * slave will be NULL
2801 bond_dev
= slave
->bond
->dev
;
2803 primary
= rtnl_dereference(bond
->primary_slave
);
2806 case NETDEV_UNREGISTER
:
2807 if (bond_dev
->type
!= ARPHRD_ETHER
)
2808 bond_release_and_destroy(bond_dev
, slave_dev
);
2810 bond_release(bond_dev
, slave_dev
);
2814 old_speed
= slave
->speed
;
2815 old_duplex
= slave
->duplex
;
2817 bond_update_speed_duplex(slave
);
2819 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
2820 if (old_speed
!= slave
->speed
)
2821 bond_3ad_adapter_speed_changed(slave
);
2822 if (old_duplex
!= slave
->duplex
)
2823 bond_3ad_adapter_duplex_changed(slave
);
2828 case NETDEV_CHANGEMTU
:
2829 /* TODO: Should slaves be allowed to
2830 * independently alter their MTU? For
2831 * an active-backup bond, slaves need
2832 * not be the same type of device, so
2833 * MTUs may vary. For other modes,
2834 * slaves arguably should have the
2835 * same MTUs. To do this, we'd need to
2836 * take over the slave's change_mtu
2837 * function for the duration of their
2841 case NETDEV_CHANGENAME
:
2842 /* we don't care if we don't have primary set */
2843 if (!bond_uses_primary(bond
) ||
2844 !bond
->params
.primary
[0])
2847 if (slave
== primary
) {
2848 /* slave's name changed - he's no longer primary */
2849 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
2850 } else if (!strcmp(slave_dev
->name
, bond
->params
.primary
)) {
2851 /* we have a new primary slave */
2852 rcu_assign_pointer(bond
->primary_slave
, slave
);
2853 } else { /* we didn't change primary - exit */
2857 netdev_info(bond
->dev
, "Primary slave changed to %s, reselecting active slave\n",
2858 primary
? slave_dev
->name
: "none");
2861 bond_select_active_slave(bond
);
2862 unblock_netpoll_tx();
2864 case NETDEV_FEAT_CHANGE
:
2865 bond_compute_features(bond
);
2867 case NETDEV_RESEND_IGMP
:
2868 /* Propagate to master device */
2869 call_netdevice_notifiers(event
, slave
->bond
->dev
);
2878 /* bond_netdev_event: handle netdev notifier chain events.
2880 * This function receives events for the netdev chain. The caller (an
2881 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2882 * locks for us to safely manipulate the slave devices (RTNL lock,
2885 static int bond_netdev_event(struct notifier_block
*this,
2886 unsigned long event
, void *ptr
)
2888 struct net_device
*event_dev
= netdev_notifier_info_to_dev(ptr
);
2890 netdev_dbg(event_dev
, "event: %lx\n", event
);
2892 if (!(event_dev
->priv_flags
& IFF_BONDING
))
2895 if (event_dev
->flags
& IFF_MASTER
) {
2896 netdev_dbg(event_dev
, "IFF_MASTER\n");
2897 return bond_master_netdev_event(event
, event_dev
);
2900 if (event_dev
->flags
& IFF_SLAVE
) {
2901 netdev_dbg(event_dev
, "IFF_SLAVE\n");
2902 return bond_slave_netdev_event(event
, event_dev
);
2908 static struct notifier_block bond_netdev_notifier
= {
2909 .notifier_call
= bond_netdev_event
,
2912 /*---------------------------- Hashing Policies -----------------------------*/
2914 /* L2 hash helper */
2915 static inline u32
bond_eth_hash(struct sk_buff
*skb
)
2917 struct ethhdr
*ep
, hdr_tmp
;
2919 ep
= skb_header_pointer(skb
, 0, sizeof(hdr_tmp
), &hdr_tmp
);
2921 return ep
->h_dest
[5] ^ ep
->h_source
[5] ^ ep
->h_proto
;
2925 /* Extract the appropriate headers based on bond's xmit policy */
2926 static bool bond_flow_dissect(struct bonding
*bond
, struct sk_buff
*skb
,
2927 struct flow_keys
*fk
)
2929 const struct ipv6hdr
*iph6
;
2930 const struct iphdr
*iph
;
2931 int noff
, proto
= -1;
2933 if (bond
->params
.xmit_policy
> BOND_XMIT_POLICY_LAYER23
)
2934 return skb_flow_dissect(skb
, fk
);
2937 noff
= skb_network_offset(skb
);
2938 if (skb
->protocol
== htons(ETH_P_IP
)) {
2939 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph
))))
2942 fk
->src
= iph
->saddr
;
2943 fk
->dst
= iph
->daddr
;
2944 noff
+= iph
->ihl
<< 2;
2945 if (!ip_is_fragment(iph
))
2946 proto
= iph
->protocol
;
2947 } else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
2948 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph6
))))
2950 iph6
= ipv6_hdr(skb
);
2951 fk
->src
= (__force __be32
)ipv6_addr_hash(&iph6
->saddr
);
2952 fk
->dst
= (__force __be32
)ipv6_addr_hash(&iph6
->daddr
);
2953 noff
+= sizeof(*iph6
);
2954 proto
= iph6
->nexthdr
;
2958 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
&& proto
>= 0)
2959 fk
->ports
= skb_flow_get_ports(skb
, noff
, proto
);
2965 * bond_xmit_hash - generate a hash value based on the xmit policy
2966 * @bond: bonding device
2967 * @skb: buffer to use for headers
2969 * This function will extract the necessary headers from the skb buffer and use
2970 * them to generate a hash based on the xmit_policy set in the bonding device
2972 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
)
2974 struct flow_keys flow
;
2977 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER2
||
2978 !bond_flow_dissect(bond
, skb
, &flow
))
2979 return bond_eth_hash(skb
);
2981 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER23
||
2982 bond
->params
.xmit_policy
== BOND_XMIT_POLICY_ENCAP23
)
2983 hash
= bond_eth_hash(skb
);
2985 hash
= (__force u32
)flow
.ports
;
2986 hash
^= (__force u32
)flow
.dst
^ (__force u32
)flow
.src
;
2987 hash
^= (hash
>> 16);
2988 hash
^= (hash
>> 8);
2993 /*-------------------------- Device entry points ----------------------------*/
2995 static void bond_work_init_all(struct bonding
*bond
)
2997 INIT_DELAYED_WORK(&bond
->mcast_work
,
2998 bond_resend_igmp_join_requests_delayed
);
2999 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3000 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3001 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3002 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3004 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3005 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3008 static void bond_work_cancel_all(struct bonding
*bond
)
3010 cancel_delayed_work_sync(&bond
->mii_work
);
3011 cancel_delayed_work_sync(&bond
->arp_work
);
3012 cancel_delayed_work_sync(&bond
->alb_work
);
3013 cancel_delayed_work_sync(&bond
->ad_work
);
3014 cancel_delayed_work_sync(&bond
->mcast_work
);
3017 static int bond_open(struct net_device
*bond_dev
)
3019 struct bonding
*bond
= netdev_priv(bond_dev
);
3020 struct list_head
*iter
;
3021 struct slave
*slave
;
3023 /* reset slave->backup and slave->inactive */
3024 if (bond_has_slaves(bond
)) {
3025 bond_for_each_slave(bond
, slave
, iter
) {
3026 if (bond_uses_primary(bond
) &&
3027 slave
!= rcu_access_pointer(bond
->curr_active_slave
)) {
3028 bond_set_slave_inactive_flags(slave
,
3029 BOND_SLAVE_NOTIFY_NOW
);
3031 bond_set_slave_active_flags(slave
,
3032 BOND_SLAVE_NOTIFY_NOW
);
3037 bond_work_init_all(bond
);
3039 if (bond_is_lb(bond
)) {
3040 /* bond_alb_initialize must be called before the timer
3043 if (bond_alb_initialize(bond
, (BOND_MODE(bond
) == BOND_MODE_ALB
)))
3045 if (bond
->params
.tlb_dynamic_lb
)
3046 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3049 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3050 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3052 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3053 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3054 bond
->recv_probe
= bond_arp_rcv
;
3057 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
3058 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3059 /* register to receive LACPDUs */
3060 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3061 bond_3ad_initiate_agg_selection(bond
, 1);
3067 static int bond_close(struct net_device
*bond_dev
)
3069 struct bonding
*bond
= netdev_priv(bond_dev
);
3071 bond_work_cancel_all(bond
);
3072 bond
->send_peer_notif
= 0;
3073 if (bond_is_lb(bond
))
3074 bond_alb_deinitialize(bond
);
3075 bond
->recv_probe
= NULL
;
3080 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3081 struct rtnl_link_stats64
*stats
)
3083 struct bonding
*bond
= netdev_priv(bond_dev
);
3084 struct rtnl_link_stats64 temp
;
3085 struct list_head
*iter
;
3086 struct slave
*slave
;
3088 memset(stats
, 0, sizeof(*stats
));
3090 bond_for_each_slave(bond
, slave
, iter
) {
3091 const struct rtnl_link_stats64
*sstats
=
3092 dev_get_stats(slave
->dev
, &temp
);
3094 stats
->rx_packets
+= sstats
->rx_packets
;
3095 stats
->rx_bytes
+= sstats
->rx_bytes
;
3096 stats
->rx_errors
+= sstats
->rx_errors
;
3097 stats
->rx_dropped
+= sstats
->rx_dropped
;
3099 stats
->tx_packets
+= sstats
->tx_packets
;
3100 stats
->tx_bytes
+= sstats
->tx_bytes
;
3101 stats
->tx_errors
+= sstats
->tx_errors
;
3102 stats
->tx_dropped
+= sstats
->tx_dropped
;
3104 stats
->multicast
+= sstats
->multicast
;
3105 stats
->collisions
+= sstats
->collisions
;
3107 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3108 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3109 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3110 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3111 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3112 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3114 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3115 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3116 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3117 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3118 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3124 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3126 struct bonding
*bond
= netdev_priv(bond_dev
);
3127 struct net_device
*slave_dev
= NULL
;
3128 struct ifbond k_binfo
;
3129 struct ifbond __user
*u_binfo
= NULL
;
3130 struct ifslave k_sinfo
;
3131 struct ifslave __user
*u_sinfo
= NULL
;
3132 struct mii_ioctl_data
*mii
= NULL
;
3133 struct bond_opt_value newval
;
3137 netdev_dbg(bond_dev
, "bond_ioctl: cmd=%d\n", cmd
);
3148 /* We do this again just in case we were called by SIOCGMIIREG
3149 * instead of SIOCGMIIPHY.
3155 if (mii
->reg_num
== 1) {
3157 if (netif_carrier_ok(bond
->dev
))
3158 mii
->val_out
= BMSR_LSTATUS
;
3162 case BOND_INFO_QUERY_OLD
:
3163 case SIOCBONDINFOQUERY
:
3164 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3166 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3169 res
= bond_info_query(bond_dev
, &k_binfo
);
3171 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3175 case BOND_SLAVE_INFO_QUERY_OLD
:
3176 case SIOCBONDSLAVEINFOQUERY
:
3177 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3179 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3182 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3184 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3192 net
= dev_net(bond_dev
);
3194 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3197 slave_dev
= __dev_get_by_name(net
, ifr
->ifr_slave
);
3199 netdev_dbg(bond_dev
, "slave_dev=%p:\n", slave_dev
);
3204 netdev_dbg(bond_dev
, "slave_dev->name=%s:\n", slave_dev
->name
);
3206 case BOND_ENSLAVE_OLD
:
3207 case SIOCBONDENSLAVE
:
3208 res
= bond_enslave(bond_dev
, slave_dev
);
3210 case BOND_RELEASE_OLD
:
3211 case SIOCBONDRELEASE
:
3212 res
= bond_release(bond_dev
, slave_dev
);
3214 case BOND_SETHWADDR_OLD
:
3215 case SIOCBONDSETHWADDR
:
3216 bond_set_dev_addr(bond_dev
, slave_dev
);
3219 case BOND_CHANGE_ACTIVE_OLD
:
3220 case SIOCBONDCHANGEACTIVE
:
3221 bond_opt_initstr(&newval
, slave_dev
->name
);
3222 res
= __bond_opt_set(bond
, BOND_OPT_ACTIVE_SLAVE
, &newval
);
3231 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3233 struct bonding
*bond
= netdev_priv(bond_dev
);
3235 if (change
& IFF_PROMISC
)
3236 bond_set_promiscuity(bond
,
3237 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3239 if (change
& IFF_ALLMULTI
)
3240 bond_set_allmulti(bond
,
3241 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3244 static void bond_set_rx_mode(struct net_device
*bond_dev
)
3246 struct bonding
*bond
= netdev_priv(bond_dev
);
3247 struct list_head
*iter
;
3248 struct slave
*slave
;
3251 if (bond_uses_primary(bond
)) {
3252 slave
= rcu_dereference(bond
->curr_active_slave
);
3254 dev_uc_sync(slave
->dev
, bond_dev
);
3255 dev_mc_sync(slave
->dev
, bond_dev
);
3258 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3259 dev_uc_sync_multiple(slave
->dev
, bond_dev
);
3260 dev_mc_sync_multiple(slave
->dev
, bond_dev
);
3266 static int bond_neigh_init(struct neighbour
*n
)
3268 struct bonding
*bond
= netdev_priv(n
->dev
);
3269 const struct net_device_ops
*slave_ops
;
3270 struct neigh_parms parms
;
3271 struct slave
*slave
;
3274 slave
= bond_first_slave(bond
);
3277 slave_ops
= slave
->dev
->netdev_ops
;
3278 if (!slave_ops
->ndo_neigh_setup
)
3281 parms
.neigh_setup
= NULL
;
3282 parms
.neigh_cleanup
= NULL
;
3283 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3287 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3288 * after the last slave has been detached. Assumes that all slaves
3289 * utilize the same neigh_cleanup (true at this writing as only user
3292 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3294 if (!parms
.neigh_setup
)
3297 return parms
.neigh_setup(n
);
3300 /* The bonding ndo_neigh_setup is called at init time beofre any
3301 * slave exists. So we must declare proxy setup function which will
3302 * be used at run time to resolve the actual slave neigh param setup.
3304 * It's also called by master devices (such as vlans) to setup their
3305 * underlying devices. In that case - do nothing, we're already set up from
3308 static int bond_neigh_setup(struct net_device
*dev
,
3309 struct neigh_parms
*parms
)
3311 /* modify only our neigh_parms */
3312 if (parms
->dev
== dev
)
3313 parms
->neigh_setup
= bond_neigh_init
;
3318 /* Change the MTU of all of a master's slaves to match the master */
3319 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3321 struct bonding
*bond
= netdev_priv(bond_dev
);
3322 struct slave
*slave
, *rollback_slave
;
3323 struct list_head
*iter
;
3326 netdev_dbg(bond_dev
, "bond=%p, new_mtu=%d\n", bond
, new_mtu
);
3328 bond_for_each_slave(bond
, slave
, iter
) {
3329 netdev_dbg(bond_dev
, "s %p c_m %p\n",
3330 slave
, slave
->dev
->netdev_ops
->ndo_change_mtu
);
3332 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3335 /* If we failed to set the slave's mtu to the new value
3336 * we must abort the operation even in ACTIVE_BACKUP
3337 * mode, because if we allow the backup slaves to have
3338 * different mtu values than the active slave we'll
3339 * need to change their mtu when doing a failover. That
3340 * means changing their mtu from timer context, which
3341 * is probably not a good idea.
3343 netdev_dbg(bond_dev
, "err %d %s\n", res
,
3349 bond_dev
->mtu
= new_mtu
;
3354 /* unwind from head to the slave that failed */
3355 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3358 if (rollback_slave
== slave
)
3361 tmp_res
= dev_set_mtu(rollback_slave
->dev
, bond_dev
->mtu
);
3363 netdev_dbg(bond_dev
, "unwind err %d dev %s\n",
3364 tmp_res
, rollback_slave
->dev
->name
);
3371 /* Change HW address
3373 * Note that many devices must be down to change the HW address, and
3374 * downing the master releases all slaves. We can make bonds full of
3375 * bonding devices to test this, however.
3377 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3379 struct bonding
*bond
= netdev_priv(bond_dev
);
3380 struct slave
*slave
, *rollback_slave
;
3381 struct sockaddr
*sa
= addr
, tmp_sa
;
3382 struct list_head
*iter
;
3385 if (BOND_MODE(bond
) == BOND_MODE_ALB
)
3386 return bond_alb_set_mac_address(bond_dev
, addr
);
3389 netdev_dbg(bond_dev
, "bond=%p\n", bond
);
3391 /* If fail_over_mac is enabled, do nothing and return success.
3392 * Returning an error causes ifenslave to fail.
3394 if (bond
->params
.fail_over_mac
&&
3395 BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3398 if (!is_valid_ether_addr(sa
->sa_data
))
3399 return -EADDRNOTAVAIL
;
3401 bond_for_each_slave(bond
, slave
, iter
) {
3402 netdev_dbg(bond_dev
, "slave %p %s\n", slave
, slave
->dev
->name
);
3403 res
= dev_set_mac_address(slave
->dev
, addr
);
3405 /* TODO: consider downing the slave
3407 * User should expect communications
3408 * breakage anyway until ARP finish
3411 netdev_dbg(bond_dev
, "err %d %s\n", res
, slave
->dev
->name
);
3417 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3421 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3422 tmp_sa
.sa_family
= bond_dev
->type
;
3424 /* unwind from head to the slave that failed */
3425 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3428 if (rollback_slave
== slave
)
3431 tmp_res
= dev_set_mac_address(rollback_slave
->dev
, &tmp_sa
);
3433 netdev_dbg(bond_dev
, "unwind err %d dev %s\n",
3434 tmp_res
, rollback_slave
->dev
->name
);
3442 * bond_xmit_slave_id - transmit skb through slave with slave_id
3443 * @bond: bonding device that is transmitting
3444 * @skb: buffer to transmit
3445 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3447 * This function tries to transmit through slave with slave_id but in case
3448 * it fails, it tries to find the first available slave for transmission.
3449 * The skb is consumed in all cases, thus the function is void.
3451 static void bond_xmit_slave_id(struct bonding
*bond
, struct sk_buff
*skb
, int slave_id
)
3453 struct list_head
*iter
;
3454 struct slave
*slave
;
3457 /* Here we start from the slave with slave_id */
3458 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3460 if (bond_slave_can_tx(slave
)) {
3461 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3467 /* Here we start from the first slave up to slave_id */
3469 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3472 if (bond_slave_can_tx(slave
)) {
3473 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3477 /* no slave that can tx has been found */
3478 dev_kfree_skb_any(skb
);
3482 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3483 * @bond: bonding device to use
3485 * Based on the value of the bonding device's packets_per_slave parameter
3486 * this function generates a slave id, which is usually used as the next
3487 * slave to transmit through.
3489 static u32
bond_rr_gen_slave_id(struct bonding
*bond
)
3492 struct reciprocal_value reciprocal_packets_per_slave
;
3493 int packets_per_slave
= bond
->params
.packets_per_slave
;
3495 switch (packets_per_slave
) {
3497 slave_id
= prandom_u32();
3500 slave_id
= bond
->rr_tx_counter
;
3503 reciprocal_packets_per_slave
=
3504 bond
->params
.reciprocal_packets_per_slave
;
3505 slave_id
= reciprocal_divide(bond
->rr_tx_counter
,
3506 reciprocal_packets_per_slave
);
3509 bond
->rr_tx_counter
++;
3514 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3516 struct bonding
*bond
= netdev_priv(bond_dev
);
3517 struct iphdr
*iph
= ip_hdr(skb
);
3518 struct slave
*slave
;
3521 /* Start with the curr_active_slave that joined the bond as the
3522 * default for sending IGMP traffic. For failover purposes one
3523 * needs to maintain some consistency for the interface that will
3524 * send the join/membership reports. The curr_active_slave found
3525 * will send all of this type of traffic.
3527 if (iph
->protocol
== IPPROTO_IGMP
&& skb
->protocol
== htons(ETH_P_IP
)) {
3528 slave
= rcu_dereference(bond
->curr_active_slave
);
3530 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3532 bond_xmit_slave_id(bond
, skb
, 0);
3534 slave_id
= bond_rr_gen_slave_id(bond
);
3535 bond_xmit_slave_id(bond
, skb
, slave_id
% bond
->slave_cnt
);
3538 return NETDEV_TX_OK
;
3541 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3542 * the bond has a usable interface.
3544 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3546 struct bonding
*bond
= netdev_priv(bond_dev
);
3547 struct slave
*slave
;
3549 slave
= rcu_dereference(bond
->curr_active_slave
);
3551 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3553 dev_kfree_skb_any(skb
);
3555 return NETDEV_TX_OK
;
3558 /* In bond_xmit_xor() , we determine the output device by using a pre-
3559 * determined xmit_hash_policy(), If the selected device is not enabled,
3560 * find the next active slave.
3562 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3564 struct bonding
*bond
= netdev_priv(bond_dev
);
3566 bond_xmit_slave_id(bond
, skb
, bond_xmit_hash(bond
, skb
) % bond
->slave_cnt
);
3568 return NETDEV_TX_OK
;
3571 /* in broadcast mode, we send everything to all usable interfaces. */
3572 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3574 struct bonding
*bond
= netdev_priv(bond_dev
);
3575 struct slave
*slave
= NULL
;
3576 struct list_head
*iter
;
3578 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3579 if (bond_is_last_slave(bond
, slave
))
3581 if (bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
3582 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
3585 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3586 bond_dev
->name
, __func__
);
3589 bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3592 if (slave
&& bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
)
3593 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3595 dev_kfree_skb_any(skb
);
3597 return NETDEV_TX_OK
;
3600 /*------------------------- Device initialization ---------------------------*/
3602 /* Lookup the slave that corresponds to a qid */
3603 static inline int bond_slave_override(struct bonding
*bond
,
3604 struct sk_buff
*skb
)
3606 struct slave
*slave
= NULL
;
3607 struct list_head
*iter
;
3609 if (!skb
->queue_mapping
)
3612 /* Find out if any slaves have the same mapping as this skb. */
3613 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3614 if (slave
->queue_id
== skb
->queue_mapping
) {
3615 if (bond_slave_can_tx(slave
)) {
3616 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3619 /* If the slave isn't UP, use default transmit policy. */
3628 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
3629 void *accel_priv
, select_queue_fallback_t fallback
)
3631 /* This helper function exists to help dev_pick_tx get the correct
3632 * destination queue. Using a helper function skips a call to
3633 * skb_tx_hash and will put the skbs in the queue we expect on their
3634 * way down to the bonding driver.
3636 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
3638 /* Save the original txq to restore before passing to the driver */
3639 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
3641 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
3643 txq
-= dev
->real_num_tx_queues
;
3644 } while (txq
>= dev
->real_num_tx_queues
);
3649 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3651 struct bonding
*bond
= netdev_priv(dev
);
3653 if (bond_should_override_tx_queue(bond
) &&
3654 !bond_slave_override(bond
, skb
))
3655 return NETDEV_TX_OK
;
3657 switch (BOND_MODE(bond
)) {
3658 case BOND_MODE_ROUNDROBIN
:
3659 return bond_xmit_roundrobin(skb
, dev
);
3660 case BOND_MODE_ACTIVEBACKUP
:
3661 return bond_xmit_activebackup(skb
, dev
);
3663 return bond_xmit_xor(skb
, dev
);
3664 case BOND_MODE_BROADCAST
:
3665 return bond_xmit_broadcast(skb
, dev
);
3666 case BOND_MODE_8023AD
:
3667 return bond_3ad_xmit_xor(skb
, dev
);
3669 return bond_alb_xmit(skb
, dev
);
3671 return bond_tlb_xmit(skb
, dev
);
3673 /* Should never happen, mode already checked */
3674 netdev_err(dev
, "Unknown bonding mode %d\n", BOND_MODE(bond
));
3676 dev_kfree_skb_any(skb
);
3677 return NETDEV_TX_OK
;
3681 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3683 struct bonding
*bond
= netdev_priv(dev
);
3684 netdev_tx_t ret
= NETDEV_TX_OK
;
3686 /* If we risk deadlock from transmitting this in the
3687 * netpoll path, tell netpoll to queue the frame for later tx
3689 if (unlikely(is_netpoll_tx_blocked(dev
)))
3690 return NETDEV_TX_BUSY
;
3693 if (bond_has_slaves(bond
))
3694 ret
= __bond_start_xmit(skb
, dev
);
3696 dev_kfree_skb_any(skb
);
3702 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
3703 struct ethtool_cmd
*ecmd
)
3705 struct bonding
*bond
= netdev_priv(bond_dev
);
3706 unsigned long speed
= 0;
3707 struct list_head
*iter
;
3708 struct slave
*slave
;
3710 ecmd
->duplex
= DUPLEX_UNKNOWN
;
3711 ecmd
->port
= PORT_OTHER
;
3713 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3714 * do not need to check mode. Though link speed might not represent
3715 * the true receive or transmit bandwidth (not all modes are symmetric)
3716 * this is an accurate maximum.
3718 bond_for_each_slave(bond
, slave
, iter
) {
3719 if (bond_slave_can_tx(slave
)) {
3720 if (slave
->speed
!= SPEED_UNKNOWN
)
3721 speed
+= slave
->speed
;
3722 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
3723 slave
->duplex
!= DUPLEX_UNKNOWN
)
3724 ecmd
->duplex
= slave
->duplex
;
3727 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
3732 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
3733 struct ethtool_drvinfo
*drvinfo
)
3735 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
3736 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
3737 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
3741 static const struct ethtool_ops bond_ethtool_ops
= {
3742 .get_drvinfo
= bond_ethtool_get_drvinfo
,
3743 .get_settings
= bond_ethtool_get_settings
,
3744 .get_link
= ethtool_op_get_link
,
3747 static const struct net_device_ops bond_netdev_ops
= {
3748 .ndo_init
= bond_init
,
3749 .ndo_uninit
= bond_uninit
,
3750 .ndo_open
= bond_open
,
3751 .ndo_stop
= bond_close
,
3752 .ndo_start_xmit
= bond_start_xmit
,
3753 .ndo_select_queue
= bond_select_queue
,
3754 .ndo_get_stats64
= bond_get_stats
,
3755 .ndo_do_ioctl
= bond_do_ioctl
,
3756 .ndo_change_rx_flags
= bond_change_rx_flags
,
3757 .ndo_set_rx_mode
= bond_set_rx_mode
,
3758 .ndo_change_mtu
= bond_change_mtu
,
3759 .ndo_set_mac_address
= bond_set_mac_address
,
3760 .ndo_neigh_setup
= bond_neigh_setup
,
3761 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
3762 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
3763 #ifdef CONFIG_NET_POLL_CONTROLLER
3764 .ndo_netpoll_setup
= bond_netpoll_setup
,
3765 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
3766 .ndo_poll_controller
= bond_poll_controller
,
3768 .ndo_add_slave
= bond_enslave
,
3769 .ndo_del_slave
= bond_release
,
3770 .ndo_fix_features
= bond_fix_features
,
3773 static const struct device_type bond_type
= {
3777 static void bond_destructor(struct net_device
*bond_dev
)
3779 struct bonding
*bond
= netdev_priv(bond_dev
);
3781 destroy_workqueue(bond
->wq
);
3782 free_netdev(bond_dev
);
3785 void bond_setup(struct net_device
*bond_dev
)
3787 struct bonding
*bond
= netdev_priv(bond_dev
);
3789 spin_lock_init(&bond
->mode_lock
);
3790 bond
->params
= bonding_defaults
;
3792 /* Initialize pointers */
3793 bond
->dev
= bond_dev
;
3795 /* Initialize the device entry points */
3796 ether_setup(bond_dev
);
3797 bond_dev
->netdev_ops
= &bond_netdev_ops
;
3798 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
3800 bond_dev
->destructor
= bond_destructor
;
3802 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
3804 /* Initialize the device options */
3805 bond_dev
->tx_queue_len
= 0;
3806 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
3807 bond_dev
->priv_flags
|= IFF_BONDING
| IFF_UNICAST_FLT
;
3808 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
3810 /* don't acquire bond device's netif_tx_lock when transmitting */
3811 bond_dev
->features
|= NETIF_F_LLTX
;
3813 /* By default, we declare the bond to be fully
3814 * VLAN hardware accelerated capable. Special
3815 * care is taken in the various xmit functions
3816 * when there are slaves that are not hw accel
3820 /* Don't allow bond devices to change network namespaces. */
3821 bond_dev
->features
|= NETIF_F_NETNS_LOCAL
;
3823 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
3824 NETIF_F_HW_VLAN_CTAG_TX
|
3825 NETIF_F_HW_VLAN_CTAG_RX
|
3826 NETIF_F_HW_VLAN_CTAG_FILTER
;
3828 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
3829 bond_dev
->hw_features
|= NETIF_F_GSO_UDP_TUNNEL
;
3830 bond_dev
->features
|= bond_dev
->hw_features
;
3833 /* Destroy a bonding device.
3834 * Must be under rtnl_lock when this function is called.
3836 static void bond_uninit(struct net_device
*bond_dev
)
3838 struct bonding
*bond
= netdev_priv(bond_dev
);
3839 struct list_head
*iter
;
3840 struct slave
*slave
;
3842 bond_netpoll_cleanup(bond_dev
);
3844 /* Release the bonded slaves */
3845 bond_for_each_slave(bond
, slave
, iter
)
3846 __bond_release_one(bond_dev
, slave
->dev
, true);
3847 netdev_info(bond_dev
, "Released all slaves\n");
3849 list_del(&bond
->bond_list
);
3851 bond_debug_unregister(bond
);
3854 /*------------------------- Module initialization ---------------------------*/
3856 static int bond_check_params(struct bond_params
*params
)
3858 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
3859 struct bond_opt_value newval
;
3860 const struct bond_opt_value
*valptr
;
3861 int arp_all_targets_value
;
3863 /* Convert string parameters. */
3865 bond_opt_initstr(&newval
, mode
);
3866 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_MODE
), &newval
);
3868 pr_err("Error: Invalid bonding mode \"%s\"\n", mode
);
3871 bond_mode
= valptr
->value
;
3874 if (xmit_hash_policy
) {
3875 if ((bond_mode
!= BOND_MODE_XOR
) &&
3876 (bond_mode
!= BOND_MODE_8023AD
) &&
3877 (bond_mode
!= BOND_MODE_TLB
)) {
3878 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3879 bond_mode_name(bond_mode
));
3881 bond_opt_initstr(&newval
, xmit_hash_policy
);
3882 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH
),
3885 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
3889 xmit_hashtype
= valptr
->value
;
3894 if (bond_mode
!= BOND_MODE_8023AD
) {
3895 pr_info("lacp_rate param is irrelevant in mode %s\n",
3896 bond_mode_name(bond_mode
));
3898 bond_opt_initstr(&newval
, lacp_rate
);
3899 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE
),
3902 pr_err("Error: Invalid lacp rate \"%s\"\n",
3906 lacp_fast
= valptr
->value
;
3911 bond_opt_initstr(&newval
, ad_select
);
3912 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT
),
3915 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select
);
3918 params
->ad_select
= valptr
->value
;
3919 if (bond_mode
!= BOND_MODE_8023AD
)
3920 pr_warn("ad_select param only affects 802.3ad mode\n");
3922 params
->ad_select
= BOND_AD_STABLE
;
3925 if (max_bonds
< 0) {
3926 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
3927 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
3928 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
3932 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
3938 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
3943 if (downdelay
< 0) {
3944 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
3945 downdelay
, INT_MAX
);
3949 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
3950 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
3955 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
3956 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
3961 /* reset values for 802.3ad/TLB/ALB */
3962 if (!bond_mode_uses_arp(bond_mode
)) {
3964 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");
3965 pr_warn("Forcing miimon to 100msec\n");
3966 miimon
= BOND_DEFAULT_MIIMON
;
3970 if (tx_queues
< 1 || tx_queues
> 255) {
3971 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
3972 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
3973 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
3976 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
3977 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
3979 all_slaves_active
= 0;
3982 if (resend_igmp
< 0 || resend_igmp
> 255) {
3983 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
3984 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
3985 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
3988 bond_opt_initval(&newval
, packets_per_slave
);
3989 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE
), &newval
)) {
3990 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
3991 packets_per_slave
, USHRT_MAX
);
3992 packets_per_slave
= 1;
3995 if (bond_mode
== BOND_MODE_ALB
) {
3996 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",
4001 if (updelay
|| downdelay
) {
4002 /* just warn the user the up/down delay will have
4003 * no effect since miimon is zero...
4005 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",
4006 updelay
, downdelay
);
4009 /* don't allow arp monitoring */
4011 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4012 miimon
, arp_interval
);
4016 if ((updelay
% miimon
) != 0) {
4017 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4018 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4023 if ((downdelay
% miimon
) != 0) {
4024 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4026 (downdelay
/ miimon
) * miimon
);
4029 downdelay
/= miimon
;
4032 if (arp_interval
< 0) {
4033 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4034 arp_interval
, INT_MAX
);
4038 for (arp_ip_count
= 0, i
= 0;
4039 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4042 /* not a complete check, but good enough to catch mistakes */
4043 if (!in4_pton(arp_ip_target
[i
], -1, (u8
*)&ip
, -1, NULL
) ||
4044 !bond_is_ip_target_ok(ip
)) {
4045 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4049 if (bond_get_targets_ip(arp_target
, ip
) == -1)
4050 arp_target
[arp_ip_count
++] = ip
;
4052 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4057 if (arp_interval
&& !arp_ip_count
) {
4058 /* don't allow arping if no arp_ip_target given... */
4059 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4065 if (!arp_interval
) {
4066 pr_err("arp_validate requires arp_interval\n");
4070 bond_opt_initstr(&newval
, arp_validate
);
4071 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE
),
4074 pr_err("Error: invalid arp_validate \"%s\"\n",
4078 arp_validate_value
= valptr
->value
;
4080 arp_validate_value
= 0;
4083 arp_all_targets_value
= 0;
4084 if (arp_all_targets
) {
4085 bond_opt_initstr(&newval
, arp_all_targets
);
4086 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS
),
4089 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4091 arp_all_targets_value
= 0;
4093 arp_all_targets_value
= valptr
->value
;
4098 pr_info("MII link monitoring set to %d ms\n", miimon
);
4099 } else if (arp_interval
) {
4100 valptr
= bond_opt_get_val(BOND_OPT_ARP_VALIDATE
,
4101 arp_validate_value
);
4102 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4103 arp_interval
, valptr
->string
, arp_ip_count
);
4105 for (i
= 0; i
< arp_ip_count
; i
++)
4106 pr_cont(" %s", arp_ip_target
[i
]);
4110 } else if (max_bonds
) {
4111 /* miimon and arp_interval not set, we need one so things
4112 * work as expected, see bonding.txt for details
4114 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");
4117 if (primary
&& !bond_mode_uses_primary(bond_mode
)) {
4118 /* currently, using a primary only makes sense
4119 * in active backup, TLB or ALB modes
4121 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4122 primary
, bond_mode_name(bond_mode
));
4126 if (primary
&& primary_reselect
) {
4127 bond_opt_initstr(&newval
, primary_reselect
);
4128 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT
),
4131 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4135 primary_reselect_value
= valptr
->value
;
4137 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4140 if (fail_over_mac
) {
4141 bond_opt_initstr(&newval
, fail_over_mac
);
4142 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC
),
4145 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4149 fail_over_mac_value
= valptr
->value
;
4150 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4151 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4153 fail_over_mac_value
= BOND_FOM_NONE
;
4156 if (lp_interval
== 0) {
4157 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4158 INT_MAX
, BOND_ALB_DEFAULT_LP_INTERVAL
);
4159 lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
4162 /* fill params struct with the proper values */
4163 params
->mode
= bond_mode
;
4164 params
->xmit_policy
= xmit_hashtype
;
4165 params
->miimon
= miimon
;
4166 params
->num_peer_notif
= num_peer_notif
;
4167 params
->arp_interval
= arp_interval
;
4168 params
->arp_validate
= arp_validate_value
;
4169 params
->arp_all_targets
= arp_all_targets_value
;
4170 params
->updelay
= updelay
;
4171 params
->downdelay
= downdelay
;
4172 params
->use_carrier
= use_carrier
;
4173 params
->lacp_fast
= lacp_fast
;
4174 params
->primary
[0] = 0;
4175 params
->primary_reselect
= primary_reselect_value
;
4176 params
->fail_over_mac
= fail_over_mac_value
;
4177 params
->tx_queues
= tx_queues
;
4178 params
->all_slaves_active
= all_slaves_active
;
4179 params
->resend_igmp
= resend_igmp
;
4180 params
->min_links
= min_links
;
4181 params
->lp_interval
= lp_interval
;
4182 params
->packets_per_slave
= packets_per_slave
;
4183 params
->tlb_dynamic_lb
= 1; /* Default value */
4184 if (packets_per_slave
> 0) {
4185 params
->reciprocal_packets_per_slave
=
4186 reciprocal_value(packets_per_slave
);
4188 /* reciprocal_packets_per_slave is unused if
4189 * packets_per_slave is 0 or 1, just initialize it
4191 params
->reciprocal_packets_per_slave
=
4192 (struct reciprocal_value
) { 0 };
4196 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4197 params
->primary
[IFNAMSIZ
- 1] = 0;
4200 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4205 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4206 static struct lock_class_key bonding_netdev_addr_lock_key
;
4207 static struct lock_class_key bonding_tx_busylock_key
;
4209 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4210 struct netdev_queue
*txq
,
4213 lockdep_set_class(&txq
->_xmit_lock
,
4214 &bonding_netdev_xmit_lock_key
);
4217 static void bond_set_lockdep_class(struct net_device
*dev
)
4219 lockdep_set_class(&dev
->addr_list_lock
,
4220 &bonding_netdev_addr_lock_key
);
4221 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4222 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4225 /* Called from registration process */
4226 static int bond_init(struct net_device
*bond_dev
)
4228 struct bonding
*bond
= netdev_priv(bond_dev
);
4229 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4231 netdev_dbg(bond_dev
, "Begin bond_init\n");
4233 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4237 bond_set_lockdep_class(bond_dev
);
4239 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4241 bond_prepare_sysfs_group(bond
);
4243 bond_debug_register(bond
);
4245 /* Ensure valid dev_addr */
4246 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4247 bond_dev
->addr_assign_type
== NET_ADDR_PERM
)
4248 eth_hw_addr_random(bond_dev
);
4253 unsigned int bond_get_num_tx_queues(void)
4258 /* Create a new bond based on the specified name and bonding parameters.
4259 * If name is NULL, obtain a suitable "bond%d" name for us.
4260 * Caller must NOT hold rtnl_lock; we need to release it here before we
4261 * set up our sysfs entries.
4263 int bond_create(struct net
*net
, const char *name
)
4265 struct net_device
*bond_dev
;
4270 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4271 name
? name
: "bond%d", NET_NAME_UNKNOWN
,
4272 bond_setup
, tx_queues
);
4274 pr_err("%s: eek! can't alloc netdev!\n", name
);
4279 dev_net_set(bond_dev
, net
);
4280 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4282 res
= register_netdevice(bond_dev
);
4284 netif_carrier_off(bond_dev
);
4288 bond_destructor(bond_dev
);
4292 static int __net_init
bond_net_init(struct net
*net
)
4294 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4297 INIT_LIST_HEAD(&bn
->dev_list
);
4299 bond_create_proc_dir(bn
);
4300 bond_create_sysfs(bn
);
4305 static void __net_exit
bond_net_exit(struct net
*net
)
4307 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4308 struct bonding
*bond
, *tmp_bond
;
4311 bond_destroy_sysfs(bn
);
4313 /* Kill off any bonds created after unregistering bond rtnl ops */
4315 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4316 unregister_netdevice_queue(bond
->dev
, &list
);
4317 unregister_netdevice_many(&list
);
4320 bond_destroy_proc_dir(bn
);
4323 static struct pernet_operations bond_net_ops
= {
4324 .init
= bond_net_init
,
4325 .exit
= bond_net_exit
,
4327 .size
= sizeof(struct bond_net
),
4330 static int __init
bonding_init(void)
4335 pr_info("%s", bond_version
);
4337 res
= bond_check_params(&bonding_defaults
);
4341 res
= register_pernet_subsys(&bond_net_ops
);
4345 res
= bond_netlink_init();
4349 bond_create_debugfs();
4351 for (i
= 0; i
< max_bonds
; i
++) {
4352 res
= bond_create(&init_net
, NULL
);
4357 register_netdevice_notifier(&bond_netdev_notifier
);
4361 bond_destroy_debugfs();
4362 bond_netlink_fini();
4364 unregister_pernet_subsys(&bond_net_ops
);
4369 static void __exit
bonding_exit(void)
4371 unregister_netdevice_notifier(&bond_netdev_notifier
);
4373 bond_destroy_debugfs();
4375 bond_netlink_fini();
4376 unregister_pernet_subsys(&bond_net_ops
);
4378 #ifdef CONFIG_NET_POLL_CONTROLLER
4379 /* Make sure we don't have an imbalance on our netpoll blocking */
4380 WARN_ON(atomic_read(&netpoll_block_tx
));
4384 module_init(bonding_init
);
4385 module_exit(bonding_exit
);
4386 MODULE_LICENSE("GPL");
4387 MODULE_VERSION(DRV_VERSION
);
4388 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4389 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");