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
);
257 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
258 * We don't protect the slave list iteration with a lock because:
259 * a. This operation is performed in IOCTL context,
260 * b. The operation is protected by the RTNL semaphore in the 8021q code,
261 * c. Holding a lock with BH disabled while directly calling a base driver
262 * entry point is generally a BAD idea.
264 * The design of synchronization/protection for this operation in the 8021q
265 * module is good for one or more VLAN devices over a single physical device
266 * and cannot be extended for a teaming solution like bonding, so there is a
267 * potential race condition here where a net device from the vlan group might
268 * be referenced (either by a base driver or the 8021q code) while it is being
269 * removed from the system. However, it turns out we're not making matters
270 * worse, and if it works for regular VLAN usage it will work here too.
274 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
275 * @bond_dev: bonding net device that got called
276 * @vid: vlan id being added
278 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
,
279 __be16 proto
, u16 vid
)
281 struct bonding
*bond
= netdev_priv(bond_dev
);
282 struct slave
*slave
, *rollback_slave
;
283 struct list_head
*iter
;
286 bond_for_each_slave(bond
, slave
, iter
) {
287 res
= vlan_vid_add(slave
->dev
, proto
, vid
);
295 /* unwind to the slave that failed */
296 bond_for_each_slave(bond
, rollback_slave
, iter
) {
297 if (rollback_slave
== slave
)
300 vlan_vid_del(rollback_slave
->dev
, proto
, vid
);
307 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
308 * @bond_dev: bonding net device that got called
309 * @vid: vlan id being removed
311 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
,
312 __be16 proto
, u16 vid
)
314 struct bonding
*bond
= netdev_priv(bond_dev
);
315 struct list_head
*iter
;
318 bond_for_each_slave(bond
, slave
, iter
)
319 vlan_vid_del(slave
->dev
, proto
, vid
);
321 if (bond_is_lb(bond
))
322 bond_alb_clear_vlan(bond
, vid
);
327 /*------------------------------- Link status -------------------------------*/
330 * Set the carrier state for the master according to the state of its
331 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
332 * do special 802.3ad magic.
334 * Returns zero if carrier state does not change, nonzero if it does.
336 static int bond_set_carrier(struct bonding
*bond
)
338 struct list_head
*iter
;
341 if (!bond_has_slaves(bond
))
344 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
345 return bond_3ad_set_carrier(bond
);
347 bond_for_each_slave(bond
, slave
, iter
) {
348 if (slave
->link
== BOND_LINK_UP
) {
349 if (!netif_carrier_ok(bond
->dev
)) {
350 netif_carrier_on(bond
->dev
);
358 if (netif_carrier_ok(bond
->dev
)) {
359 netif_carrier_off(bond
->dev
);
366 * Get link speed and duplex from the slave's base driver
367 * using ethtool. If for some reason the call fails or the
368 * values are invalid, set speed and duplex to -1,
371 static void bond_update_speed_duplex(struct slave
*slave
)
373 struct net_device
*slave_dev
= slave
->dev
;
374 struct ethtool_cmd ecmd
;
378 slave
->speed
= SPEED_UNKNOWN
;
379 slave
->duplex
= DUPLEX_UNKNOWN
;
381 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
385 slave_speed
= ethtool_cmd_speed(&ecmd
);
386 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
389 switch (ecmd
.duplex
) {
397 slave
->speed
= slave_speed
;
398 slave
->duplex
= ecmd
.duplex
;
403 const char *bond_slave_link_status(s8 link
)
420 * if <dev> supports MII link status reporting, check its link status.
422 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
423 * depending upon the setting of the use_carrier parameter.
425 * Return either BMSR_LSTATUS, meaning that the link is up (or we
426 * can't tell and just pretend it is), or 0, meaning that the link is
429 * If reporting is non-zero, instead of faking link up, return -1 if
430 * both ETHTOOL and MII ioctls fail (meaning the device does not
431 * support them). If use_carrier is set, return whatever it says.
432 * It'd be nice if there was a good way to tell if a driver supports
433 * netif_carrier, but there really isn't.
435 static int bond_check_dev_link(struct bonding
*bond
,
436 struct net_device
*slave_dev
, int reporting
)
438 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
439 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
441 struct mii_ioctl_data
*mii
;
443 if (!reporting
&& !netif_running(slave_dev
))
446 if (bond
->params
.use_carrier
)
447 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
449 /* Try to get link status using Ethtool first. */
450 if (slave_dev
->ethtool_ops
->get_link
)
451 return slave_dev
->ethtool_ops
->get_link(slave_dev
) ?
454 /* Ethtool can't be used, fallback to MII ioctls. */
455 ioctl
= slave_ops
->ndo_do_ioctl
;
457 /* TODO: set pointer to correct ioctl on a per team member */
458 /* bases to make this more efficient. that is, once */
459 /* we determine the correct ioctl, we will always */
460 /* call it and not the others for that team */
464 * We cannot assume that SIOCGMIIPHY will also read a
465 * register; not all network drivers (e.g., e100)
469 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
470 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
472 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
473 mii
->reg_num
= MII_BMSR
;
474 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
475 return mii
->val_out
& BMSR_LSTATUS
;
480 * If reporting, report that either there's no dev->do_ioctl,
481 * or both SIOCGMIIREG and get_link failed (meaning that we
482 * cannot report link status). If not reporting, pretend
485 return reporting
? -1 : BMSR_LSTATUS
;
488 /*----------------------------- Multicast list ------------------------------*/
491 * Push the promiscuity flag down to appropriate slaves
493 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
495 struct list_head
*iter
;
498 if (bond_uses_primary(bond
)) {
499 struct slave
*curr_active
= rtnl_dereference(bond
->curr_active_slave
);
502 err
= dev_set_promiscuity(curr_active
->dev
, inc
);
506 bond_for_each_slave(bond
, slave
, iter
) {
507 err
= dev_set_promiscuity(slave
->dev
, inc
);
516 * Push the allmulti flag down to all slaves
518 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
520 struct list_head
*iter
;
523 if (bond_uses_primary(bond
)) {
524 struct slave
*curr_active
= rtnl_dereference(bond
->curr_active_slave
);
527 err
= dev_set_allmulti(curr_active
->dev
, inc
);
531 bond_for_each_slave(bond
, slave
, iter
) {
532 err
= dev_set_allmulti(slave
->dev
, inc
);
541 * Retrieve the list of registered multicast addresses for the bonding
542 * device and retransmit an IGMP JOIN request to the current active
545 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
547 struct bonding
*bond
= container_of(work
, struct bonding
,
550 if (!rtnl_trylock()) {
551 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
554 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, bond
->dev
);
556 if (bond
->igmp_retrans
> 1) {
557 bond
->igmp_retrans
--;
558 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
563 /* Flush bond's hardware addresses from slave
565 static void bond_hw_addr_flush(struct net_device
*bond_dev
,
566 struct net_device
*slave_dev
)
568 struct bonding
*bond
= netdev_priv(bond_dev
);
570 dev_uc_unsync(slave_dev
, bond_dev
);
571 dev_mc_unsync(slave_dev
, bond_dev
);
573 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
574 /* del lacpdu mc addr from mc list */
575 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
577 dev_mc_del(slave_dev
, lacpdu_multicast
);
581 /*--------------------------- Active slave change ---------------------------*/
583 /* Update the hardware address list and promisc/allmulti for the new and
584 * old active slaves (if any). Modes that are not using primary keep all
585 * slaves up date at all times; only the modes that use primary need to call
586 * this function to swap these settings during a failover.
588 static void bond_hw_addr_swap(struct bonding
*bond
, struct slave
*new_active
,
589 struct slave
*old_active
)
594 if (bond
->dev
->flags
& IFF_PROMISC
)
595 dev_set_promiscuity(old_active
->dev
, -1);
597 if (bond
->dev
->flags
& IFF_ALLMULTI
)
598 dev_set_allmulti(old_active
->dev
, -1);
600 bond_hw_addr_flush(bond
->dev
, old_active
->dev
);
604 /* FIXME: Signal errors upstream. */
605 if (bond
->dev
->flags
& IFF_PROMISC
)
606 dev_set_promiscuity(new_active
->dev
, 1);
608 if (bond
->dev
->flags
& IFF_ALLMULTI
)
609 dev_set_allmulti(new_active
->dev
, 1);
611 netif_addr_lock_bh(bond
->dev
);
612 dev_uc_sync(new_active
->dev
, bond
->dev
);
613 dev_mc_sync(new_active
->dev
, bond
->dev
);
614 netif_addr_unlock_bh(bond
->dev
);
619 * bond_set_dev_addr - clone slave's address to bond
620 * @bond_dev: bond net device
621 * @slave_dev: slave net device
623 * Should be called with RTNL held.
625 static void bond_set_dev_addr(struct net_device
*bond_dev
,
626 struct net_device
*slave_dev
)
628 netdev_dbg(bond_dev
, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
629 bond_dev
, slave_dev
, slave_dev
->addr_len
);
630 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
631 bond_dev
->addr_assign_type
= NET_ADDR_STOLEN
;
632 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
636 * bond_do_fail_over_mac
638 * Perform special MAC address swapping for fail_over_mac settings
642 static void bond_do_fail_over_mac(struct bonding
*bond
,
643 struct slave
*new_active
,
644 struct slave
*old_active
)
646 u8 tmp_mac
[ETH_ALEN
];
647 struct sockaddr saddr
;
650 switch (bond
->params
.fail_over_mac
) {
651 case BOND_FOM_ACTIVE
:
653 bond_set_dev_addr(bond
->dev
, new_active
->dev
);
655 case BOND_FOM_FOLLOW
:
657 * if new_active && old_active, swap them
658 * if just old_active, do nothing (going to no active slave)
659 * if just new_active, set new_active to bond's MAC
665 ether_addr_copy(tmp_mac
, new_active
->dev
->dev_addr
);
666 ether_addr_copy(saddr
.sa_data
,
667 old_active
->dev
->dev_addr
);
668 saddr
.sa_family
= new_active
->dev
->type
;
670 ether_addr_copy(saddr
.sa_data
, bond
->dev
->dev_addr
);
671 saddr
.sa_family
= bond
->dev
->type
;
674 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
676 netdev_err(bond
->dev
, "Error %d setting MAC of slave %s\n",
677 -rv
, new_active
->dev
->name
);
684 ether_addr_copy(saddr
.sa_data
, tmp_mac
);
685 saddr
.sa_family
= old_active
->dev
->type
;
687 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
689 netdev_err(bond
->dev
, "Error %d setting MAC of slave %s\n",
690 -rv
, new_active
->dev
->name
);
694 netdev_err(bond
->dev
, "bond_do_fail_over_mac impossible: bad policy %d\n",
695 bond
->params
.fail_over_mac
);
701 static bool bond_should_change_active(struct bonding
*bond
)
703 struct slave
*prim
= rtnl_dereference(bond
->primary_slave
);
704 struct slave
*curr
= rtnl_dereference(bond
->curr_active_slave
);
706 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
708 if (bond
->force_primary
) {
709 bond
->force_primary
= false;
712 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
713 (prim
->speed
< curr
->speed
||
714 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
716 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
722 * find_best_interface - select the best available slave to be the active one
723 * @bond: our bonding struct
725 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
727 struct slave
*slave
, *bestslave
= NULL
, *primary
;
728 struct list_head
*iter
;
729 int mintime
= bond
->params
.updelay
;
731 primary
= rtnl_dereference(bond
->primary_slave
);
732 if (primary
&& primary
->link
== BOND_LINK_UP
&&
733 bond_should_change_active(bond
))
736 bond_for_each_slave(bond
, slave
, iter
) {
737 if (slave
->link
== BOND_LINK_UP
)
739 if (slave
->link
== BOND_LINK_BACK
&& bond_slave_is_up(slave
) &&
740 slave
->delay
< mintime
) {
741 mintime
= slave
->delay
;
749 static bool bond_should_notify_peers(struct bonding
*bond
)
754 slave
= rcu_dereference(bond
->curr_active_slave
);
757 netdev_dbg(bond
->dev
, "bond_should_notify_peers: slave %s\n",
758 slave
? slave
->dev
->name
: "NULL");
760 if (!slave
|| !bond
->send_peer_notif
||
761 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
768 * change_active_interface - change the active slave into the specified one
769 * @bond: our bonding struct
770 * @new: the new slave to make the active one
772 * Set the new slave to the bond's settings and unset them on the old
774 * Setting include flags, mc-list, promiscuity, allmulti, etc.
776 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
777 * because it is apparently the best available slave we have, even though its
778 * updelay hasn't timed out yet.
780 * Caller must hold RTNL.
782 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
784 struct slave
*old_active
;
788 old_active
= rtnl_dereference(bond
->curr_active_slave
);
790 if (old_active
== new_active
)
794 new_active
->last_link_up
= jiffies
;
796 if (new_active
->link
== BOND_LINK_BACK
) {
797 if (bond_uses_primary(bond
)) {
798 netdev_info(bond
->dev
, "making interface %s the new active one %d ms earlier\n",
799 new_active
->dev
->name
,
800 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
803 new_active
->delay
= 0;
804 new_active
->link
= BOND_LINK_UP
;
806 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
807 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
809 if (bond_is_lb(bond
))
810 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
812 if (bond_uses_primary(bond
)) {
813 netdev_info(bond
->dev
, "making interface %s the new active one\n",
814 new_active
->dev
->name
);
819 if (bond_uses_primary(bond
))
820 bond_hw_addr_swap(bond
, new_active
, old_active
);
822 if (bond_is_lb(bond
)) {
823 bond_alb_handle_active_change(bond
, new_active
);
825 bond_set_slave_inactive_flags(old_active
,
826 BOND_SLAVE_NOTIFY_NOW
);
828 bond_set_slave_active_flags(new_active
,
829 BOND_SLAVE_NOTIFY_NOW
);
831 rcu_assign_pointer(bond
->curr_active_slave
, new_active
);
834 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
) {
836 bond_set_slave_inactive_flags(old_active
,
837 BOND_SLAVE_NOTIFY_NOW
);
840 bool should_notify_peers
= false;
842 bond_set_slave_active_flags(new_active
,
843 BOND_SLAVE_NOTIFY_NOW
);
845 if (bond
->params
.fail_over_mac
)
846 bond_do_fail_over_mac(bond
, new_active
,
849 if (netif_running(bond
->dev
)) {
850 bond
->send_peer_notif
=
851 bond
->params
.num_peer_notif
;
852 should_notify_peers
=
853 bond_should_notify_peers(bond
);
856 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
857 if (should_notify_peers
)
858 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
863 /* resend IGMP joins since active slave has changed or
864 * all were sent on curr_active_slave.
865 * resend only if bond is brought up with the affected
866 * bonding modes and the retransmission is enabled */
867 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
868 ((bond_uses_primary(bond
) && new_active
) ||
869 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
)) {
870 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
871 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
876 * bond_select_active_slave - select a new active slave, if needed
877 * @bond: our bonding struct
879 * This functions should be called when one of the following occurs:
880 * - The old curr_active_slave has been released or lost its link.
881 * - The primary_slave has got its link back.
882 * - A slave has got its link back and there's no old curr_active_slave.
884 * Caller must hold RTNL.
886 void bond_select_active_slave(struct bonding
*bond
)
888 struct slave
*best_slave
;
891 best_slave
= bond_find_best_slave(bond
);
892 if (best_slave
!= rtnl_dereference(bond
->curr_active_slave
)) {
893 bond_change_active_slave(bond
, best_slave
);
894 rv
= bond_set_carrier(bond
);
898 if (netif_carrier_ok(bond
->dev
)) {
899 netdev_info(bond
->dev
, "first active interface up!\n");
901 netdev_info(bond
->dev
, "now running without any active interface!\n");
906 #ifdef CONFIG_NET_POLL_CONTROLLER
907 static inline int slave_enable_netpoll(struct slave
*slave
)
912 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
917 err
= __netpoll_setup(np
, slave
->dev
);
926 static inline void slave_disable_netpoll(struct slave
*slave
)
928 struct netpoll
*np
= slave
->np
;
934 __netpoll_free_async(np
);
937 static void bond_poll_controller(struct net_device
*bond_dev
)
941 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
943 struct bonding
*bond
= netdev_priv(bond_dev
);
944 struct list_head
*iter
;
947 bond_for_each_slave(bond
, slave
, iter
)
948 if (bond_slave_is_up(slave
))
949 slave_disable_netpoll(slave
);
952 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
954 struct bonding
*bond
= netdev_priv(dev
);
955 struct list_head
*iter
;
959 bond_for_each_slave(bond
, slave
, iter
) {
960 err
= slave_enable_netpoll(slave
);
962 bond_netpoll_cleanup(dev
);
969 static inline int slave_enable_netpoll(struct slave
*slave
)
973 static inline void slave_disable_netpoll(struct slave
*slave
)
976 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
981 /*---------------------------------- IOCTL ----------------------------------*/
983 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
984 netdev_features_t features
)
986 struct bonding
*bond
= netdev_priv(dev
);
987 struct list_head
*iter
;
988 netdev_features_t mask
;
992 features
&= ~NETIF_F_ONE_FOR_ALL
;
993 features
|= NETIF_F_ALL_FOR_ALL
;
995 bond_for_each_slave(bond
, slave
, iter
) {
996 features
= netdev_increment_features(features
,
997 slave
->dev
->features
,
1000 features
= netdev_add_tso_features(features
, mask
);
1005 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1006 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1007 NETIF_F_HIGHDMA | NETIF_F_LRO)
1009 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1010 NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL)
1012 static void bond_compute_features(struct bonding
*bond
)
1014 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1015 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1016 netdev_features_t enc_features
= BOND_ENC_FEATURES
;
1017 struct net_device
*bond_dev
= bond
->dev
;
1018 struct list_head
*iter
;
1019 struct slave
*slave
;
1020 unsigned short max_hard_header_len
= ETH_HLEN
;
1021 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1022 u16 gso_max_segs
= GSO_MAX_SEGS
;
1024 if (!bond_has_slaves(bond
))
1026 vlan_features
&= NETIF_F_ALL_FOR_ALL
;
1028 bond_for_each_slave(bond
, slave
, iter
) {
1029 vlan_features
= netdev_increment_features(vlan_features
,
1030 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1032 enc_features
= netdev_increment_features(enc_features
,
1033 slave
->dev
->hw_enc_features
,
1035 dst_release_flag
&= slave
->dev
->priv_flags
;
1036 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1037 max_hard_header_len
= slave
->dev
->hard_header_len
;
1039 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1040 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1044 bond_dev
->vlan_features
= vlan_features
;
1045 bond_dev
->hw_enc_features
= enc_features
;
1046 bond_dev
->hard_header_len
= max_hard_header_len
;
1047 bond_dev
->gso_max_segs
= gso_max_segs
;
1048 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1050 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1051 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1053 netdev_change_features(bond_dev
);
1056 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1057 struct net_device
*slave_dev
)
1059 bond_dev
->header_ops
= slave_dev
->header_ops
;
1061 bond_dev
->type
= slave_dev
->type
;
1062 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1063 bond_dev
->addr_len
= slave_dev
->addr_len
;
1065 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1066 slave_dev
->addr_len
);
1069 /* On bonding slaves other than the currently active slave, suppress
1070 * duplicates except for alb non-mcast/bcast.
1072 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1073 struct slave
*slave
,
1074 struct bonding
*bond
)
1076 if (bond_is_slave_inactive(slave
)) {
1077 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1078 skb
->pkt_type
!= PACKET_BROADCAST
&&
1079 skb
->pkt_type
!= PACKET_MULTICAST
)
1086 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1088 struct sk_buff
*skb
= *pskb
;
1089 struct slave
*slave
;
1090 struct bonding
*bond
;
1091 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1093 int ret
= RX_HANDLER_ANOTHER
;
1095 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1097 return RX_HANDLER_CONSUMED
;
1101 slave
= bond_slave_get_rcu(skb
->dev
);
1104 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1106 ret
= recv_probe(skb
, bond
, slave
);
1107 if (ret
== RX_HANDLER_CONSUMED
) {
1113 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1114 return RX_HANDLER_EXACT
;
1117 skb
->dev
= bond
->dev
;
1119 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1120 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1121 skb
->pkt_type
== PACKET_HOST
) {
1123 if (unlikely(skb_cow_head(skb
,
1124 skb
->data
- skb_mac_header(skb
)))) {
1126 return RX_HANDLER_CONSUMED
;
1128 ether_addr_copy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
);
1134 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1135 struct net_device
*slave_dev
,
1136 struct slave
*slave
)
1140 err
= netdev_master_upper_dev_link_private(slave_dev
, bond_dev
, slave
);
1143 slave_dev
->flags
|= IFF_SLAVE
;
1144 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1148 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1149 struct net_device
*slave_dev
)
1151 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1152 slave_dev
->flags
&= ~IFF_SLAVE
;
1153 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1156 static struct slave
*bond_alloc_slave(struct bonding
*bond
)
1158 struct slave
*slave
= NULL
;
1160 slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1164 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1165 SLAVE_AD_INFO(slave
) = kzalloc(sizeof(struct ad_slave_info
),
1167 if (!SLAVE_AD_INFO(slave
)) {
1175 static void bond_free_slave(struct slave
*slave
)
1177 struct bonding
*bond
= bond_get_bond_by_slave(slave
);
1179 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1180 kfree(SLAVE_AD_INFO(slave
));
1185 /* enslave device <slave> to bond device <master> */
1186 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1188 struct bonding
*bond
= netdev_priv(bond_dev
);
1189 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1190 struct slave
*new_slave
= NULL
, *prev_slave
;
1191 struct sockaddr addr
;
1195 if (!bond
->params
.use_carrier
&&
1196 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1197 slave_ops
->ndo_do_ioctl
== NULL
) {
1198 netdev_warn(bond_dev
, "no link monitoring support for %s\n",
1202 /* already enslaved */
1203 if (slave_dev
->flags
& IFF_SLAVE
) {
1204 netdev_dbg(bond_dev
, "Error: Device was already enslaved\n");
1208 if (bond_dev
== slave_dev
) {
1209 netdev_err(bond_dev
, "cannot enslave bond to itself.\n");
1213 /* vlan challenged mutual exclusion */
1214 /* no need to lock since we're protected by rtnl_lock */
1215 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1216 netdev_dbg(bond_dev
, "%s is NETIF_F_VLAN_CHALLENGED\n",
1218 if (vlan_uses_dev(bond_dev
)) {
1219 netdev_err(bond_dev
, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1220 slave_dev
->name
, bond_dev
->name
);
1223 netdev_warn(bond_dev
, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1224 slave_dev
->name
, slave_dev
->name
,
1228 netdev_dbg(bond_dev
, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1233 * Old ifenslave binaries are no longer supported. These can
1234 * be identified with moderate accuracy by the state of the slave:
1235 * the current ifenslave will set the interface down prior to
1236 * enslaving it; the old ifenslave will not.
1238 if ((slave_dev
->flags
& IFF_UP
)) {
1239 netdev_err(bond_dev
, "%s is up - this may be due to an out of date ifenslave\n",
1242 goto err_undo_flags
;
1245 /* set bonding device ether type by slave - bonding netdevices are
1246 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1247 * there is a need to override some of the type dependent attribs/funcs.
1249 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1250 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1252 if (!bond_has_slaves(bond
)) {
1253 if (bond_dev
->type
!= slave_dev
->type
) {
1254 netdev_dbg(bond_dev
, "change device type from %d to %d\n",
1255 bond_dev
->type
, slave_dev
->type
);
1257 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1259 res
= notifier_to_errno(res
);
1261 netdev_err(bond_dev
, "refused to change device type\n");
1263 goto err_undo_flags
;
1266 /* Flush unicast and multicast addresses */
1267 dev_uc_flush(bond_dev
);
1268 dev_mc_flush(bond_dev
);
1270 if (slave_dev
->type
!= ARPHRD_ETHER
)
1271 bond_setup_by_slave(bond_dev
, slave_dev
);
1273 ether_setup(bond_dev
);
1274 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1277 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1280 } else if (bond_dev
->type
!= slave_dev
->type
) {
1281 netdev_err(bond_dev
, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1282 slave_dev
->name
, slave_dev
->type
, bond_dev
->type
);
1284 goto err_undo_flags
;
1287 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1288 netdev_warn(bond_dev
, "The slave device specified does not support setting the MAC address\n");
1289 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
&&
1290 bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1291 if (!bond_has_slaves(bond
)) {
1292 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1293 netdev_warn(bond_dev
, "Setting fail_over_mac to active for active-backup mode\n");
1295 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");
1297 goto err_undo_flags
;
1302 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1304 /* If this is the first slave, then we need to set the master's hardware
1305 * address to be the same as the slave's. */
1306 if (!bond_has_slaves(bond
) &&
1307 bond
->dev
->addr_assign_type
== NET_ADDR_RANDOM
)
1308 bond_set_dev_addr(bond
->dev
, slave_dev
);
1310 new_slave
= bond_alloc_slave(bond
);
1313 goto err_undo_flags
;
1316 new_slave
->bond
= bond
;
1317 new_slave
->dev
= slave_dev
;
1319 * Set the new_slave's queue_id to be zero. Queue ID mapping
1320 * is set via sysfs or module option if desired.
1322 new_slave
->queue_id
= 0;
1324 /* Save slave's original mtu and then set it to match the bond */
1325 new_slave
->original_mtu
= slave_dev
->mtu
;
1326 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1328 netdev_dbg(bond_dev
, "Error %d calling dev_set_mtu\n", res
);
1333 * Save slave's original ("permanent") mac address for modes
1334 * that need it, and for restoring it upon release, and then
1335 * set it to the master's address
1337 ether_addr_copy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
);
1339 if (!bond
->params
.fail_over_mac
||
1340 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1342 * Set slave to master's mac address. The application already
1343 * set the master's mac address to that of the first slave
1345 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1346 addr
.sa_family
= slave_dev
->type
;
1347 res
= dev_set_mac_address(slave_dev
, &addr
);
1349 netdev_dbg(bond_dev
, "Error %d calling set_mac_address\n", res
);
1350 goto err_restore_mtu
;
1354 /* open the slave since the application closed it */
1355 res
= dev_open(slave_dev
);
1357 netdev_dbg(bond_dev
, "Opening slave %s failed\n", slave_dev
->name
);
1358 goto err_restore_mac
;
1361 slave_dev
->priv_flags
|= IFF_BONDING
;
1363 if (bond_is_lb(bond
)) {
1364 /* bond_alb_init_slave() must be called before all other stages since
1365 * it might fail and we do not want to have to undo everything
1367 res
= bond_alb_init_slave(bond
, new_slave
);
1372 /* If the mode uses primary, then the following is handled by
1373 * bond_change_active_slave().
1375 if (!bond_uses_primary(bond
)) {
1376 /* set promiscuity level to new slave */
1377 if (bond_dev
->flags
& IFF_PROMISC
) {
1378 res
= dev_set_promiscuity(slave_dev
, 1);
1383 /* set allmulti level to new slave */
1384 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1385 res
= dev_set_allmulti(slave_dev
, 1);
1390 netif_addr_lock_bh(bond_dev
);
1392 dev_mc_sync_multiple(slave_dev
, bond_dev
);
1393 dev_uc_sync_multiple(slave_dev
, bond_dev
);
1395 netif_addr_unlock_bh(bond_dev
);
1398 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1399 /* add lacpdu mc addr to mc list */
1400 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1402 dev_mc_add(slave_dev
, lacpdu_multicast
);
1405 res
= vlan_vids_add_by_dev(slave_dev
, bond_dev
);
1407 netdev_err(bond_dev
, "Couldn't add bond vlan ids to %s\n",
1412 prev_slave
= bond_last_slave(bond
);
1414 new_slave
->delay
= 0;
1415 new_slave
->link_failure_count
= 0;
1417 bond_update_speed_duplex(new_slave
);
1419 new_slave
->last_rx
= jiffies
-
1420 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1421 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
1422 new_slave
->target_last_arp_rx
[i
] = new_slave
->last_rx
;
1424 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1425 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1427 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1429 * miimon is set but a bonded network driver
1430 * does not support ETHTOOL/MII and
1431 * arp_interval is not set. Note: if
1432 * use_carrier is enabled, we will never go
1433 * here (because netif_carrier is always
1434 * supported); thus, we don't need to change
1435 * the messages for netif_carrier.
1437 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",
1439 } else if (link_reporting
== -1) {
1440 /* unable get link status using mii/ethtool */
1441 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",
1446 /* check for initial state */
1447 if (bond
->params
.miimon
) {
1448 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1449 if (bond
->params
.updelay
) {
1450 new_slave
->link
= BOND_LINK_BACK
;
1451 new_slave
->delay
= bond
->params
.updelay
;
1453 new_slave
->link
= BOND_LINK_UP
;
1456 new_slave
->link
= BOND_LINK_DOWN
;
1458 } else if (bond
->params
.arp_interval
) {
1459 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1460 BOND_LINK_UP
: BOND_LINK_DOWN
);
1462 new_slave
->link
= BOND_LINK_UP
;
1465 if (new_slave
->link
!= BOND_LINK_DOWN
)
1466 new_slave
->last_link_up
= jiffies
;
1467 netdev_dbg(bond_dev
, "Initial state of slave_dev is BOND_LINK_%s\n",
1468 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1469 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1471 if (bond_uses_primary(bond
) && bond
->params
.primary
[0]) {
1472 /* if there is a primary slave, remember it */
1473 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1474 rcu_assign_pointer(bond
->primary_slave
, new_slave
);
1475 bond
->force_primary
= true;
1479 switch (BOND_MODE(bond
)) {
1480 case BOND_MODE_ACTIVEBACKUP
:
1481 bond_set_slave_inactive_flags(new_slave
,
1482 BOND_SLAVE_NOTIFY_NOW
);
1484 case BOND_MODE_8023AD
:
1485 /* in 802.3ad mode, the internal mechanism
1486 * will activate the slaves in the selected
1489 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1490 /* if this is the first slave */
1492 SLAVE_AD_INFO(new_slave
)->id
= 1;
1493 /* Initialize AD with the number of times that the AD timer is called in 1 second
1494 * can be called only after the mac address of the bond is set
1496 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1498 SLAVE_AD_INFO(new_slave
)->id
=
1499 SLAVE_AD_INFO(prev_slave
)->id
+ 1;
1502 bond_3ad_bind_slave(new_slave
);
1506 bond_set_active_slave(new_slave
);
1507 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1510 netdev_dbg(bond_dev
, "This slave is always active in trunk mode\n");
1512 /* always active in trunk mode */
1513 bond_set_active_slave(new_slave
);
1515 /* In trunking mode there is little meaning to curr_active_slave
1516 * anyway (it holds no special properties of the bond device),
1517 * so we can change it without calling change_active_interface()
1519 if (!rcu_access_pointer(bond
->curr_active_slave
) &&
1520 new_slave
->link
== BOND_LINK_UP
)
1521 rcu_assign_pointer(bond
->curr_active_slave
, new_slave
);
1524 } /* switch(bond_mode) */
1526 #ifdef CONFIG_NET_POLL_CONTROLLER
1527 slave_dev
->npinfo
= bond
->dev
->npinfo
;
1528 if (slave_dev
->npinfo
) {
1529 if (slave_enable_netpoll(new_slave
)) {
1530 netdev_info(bond_dev
, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1537 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1540 netdev_dbg(bond_dev
, "Error %d calling netdev_rx_handler_register\n", res
);
1544 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
, new_slave
);
1546 netdev_dbg(bond_dev
, "Error %d calling bond_master_upper_dev_link\n", res
);
1547 goto err_unregister
;
1550 res
= bond_sysfs_slave_add(new_slave
);
1552 netdev_dbg(bond_dev
, "Error %d calling bond_sysfs_slave_add\n", res
);
1553 goto err_upper_unlink
;
1557 bond_compute_features(bond
);
1558 bond_set_carrier(bond
);
1560 if (bond_uses_primary(bond
)) {
1562 bond_select_active_slave(bond
);
1563 unblock_netpoll_tx();
1566 netdev_info(bond_dev
, "Enslaving %s as %s interface with %s link\n",
1568 bond_is_active_slave(new_slave
) ? "an active" : "a backup",
1569 new_slave
->link
!= BOND_LINK_DOWN
? "an up" : "a down");
1571 /* enslave is successful */
1574 /* Undo stages on error */
1576 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1579 netdev_rx_handler_unregister(slave_dev
);
1582 if (!bond_uses_primary(bond
))
1583 bond_hw_addr_flush(bond_dev
, slave_dev
);
1585 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1586 if (rcu_access_pointer(bond
->primary_slave
) == new_slave
)
1587 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
1588 if (rcu_access_pointer(bond
->curr_active_slave
) == new_slave
) {
1590 bond_change_active_slave(bond
, NULL
);
1591 bond_select_active_slave(bond
);
1592 unblock_netpoll_tx();
1594 /* either primary_slave or curr_active_slave might've changed */
1596 slave_disable_netpoll(new_slave
);
1599 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1600 dev_close(slave_dev
);
1603 if (!bond
->params
.fail_over_mac
||
1604 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1605 /* XXX TODO - fom follow mode needs to change master's
1606 * MAC if this slave's MAC is in use by the bond, or at
1607 * least print a warning.
1609 ether_addr_copy(addr
.sa_data
, new_slave
->perm_hwaddr
);
1610 addr
.sa_family
= slave_dev
->type
;
1611 dev_set_mac_address(slave_dev
, &addr
);
1615 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1618 bond_free_slave(new_slave
);
1621 /* Enslave of first slave has failed and we need to fix master's mac */
1622 if (!bond_has_slaves(bond
) &&
1623 ether_addr_equal_64bits(bond_dev
->dev_addr
, slave_dev
->dev_addr
))
1624 eth_hw_addr_random(bond_dev
);
1630 * Try to release the slave device <slave> from the bond device <master>
1631 * It is legal to access curr_active_slave without a lock because all the function
1632 * is write-locked. If "all" is true it means that the function is being called
1633 * while destroying a bond interface and all slaves are being released.
1635 * The rules for slave state should be:
1636 * for Active/Backup:
1637 * Active stays on all backups go down
1638 * for Bonded connections:
1639 * The first up interface should be left on and all others downed.
1641 static int __bond_release_one(struct net_device
*bond_dev
,
1642 struct net_device
*slave_dev
,
1645 struct bonding
*bond
= netdev_priv(bond_dev
);
1646 struct slave
*slave
, *oldcurrent
;
1647 struct sockaddr addr
;
1648 int old_flags
= bond_dev
->flags
;
1649 netdev_features_t old_features
= bond_dev
->features
;
1651 /* slave is not a slave or master is not master of this slave */
1652 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1653 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1654 netdev_err(bond_dev
, "cannot release %s\n",
1661 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1663 /* not a slave of this bond */
1664 netdev_info(bond_dev
, "%s not enslaved\n",
1666 unblock_netpoll_tx();
1670 bond_sysfs_slave_del(slave
);
1672 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1673 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1674 * for this slave anymore.
1676 netdev_rx_handler_unregister(slave_dev
);
1678 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1679 bond_3ad_unbind_slave(slave
);
1681 netdev_info(bond_dev
, "Releasing %s interface %s\n",
1682 bond_is_active_slave(slave
) ? "active" : "backup",
1685 oldcurrent
= rcu_access_pointer(bond
->curr_active_slave
);
1687 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
1689 if (!all
&& (!bond
->params
.fail_over_mac
||
1690 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
)) {
1691 if (ether_addr_equal_64bits(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1692 bond_has_slaves(bond
))
1693 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",
1694 slave_dev
->name
, slave
->perm_hwaddr
,
1695 bond_dev
->name
, slave_dev
->name
);
1698 if (rtnl_dereference(bond
->primary_slave
) == slave
)
1699 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
1701 if (oldcurrent
== slave
)
1702 bond_change_active_slave(bond
, NULL
);
1704 if (bond_is_lb(bond
)) {
1705 /* Must be called only after the slave has been
1706 * detached from the list and the curr_active_slave
1707 * has been cleared (if our_slave == old_current),
1708 * but before a new active slave is selected.
1710 bond_alb_deinit_slave(bond
, slave
);
1714 RCU_INIT_POINTER(bond
->curr_active_slave
, NULL
);
1715 } else if (oldcurrent
== slave
) {
1717 * Note that we hold RTNL over this sequence, so there
1718 * is no concern that another slave add/remove event
1721 bond_select_active_slave(bond
);
1724 if (!bond_has_slaves(bond
)) {
1725 bond_set_carrier(bond
);
1726 eth_hw_addr_random(bond_dev
);
1729 unblock_netpoll_tx();
1733 if (!bond_has_slaves(bond
)) {
1734 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
1735 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
1738 bond_compute_features(bond
);
1739 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1740 (old_features
& NETIF_F_VLAN_CHALLENGED
))
1741 netdev_info(bond_dev
, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1742 slave_dev
->name
, bond_dev
->name
);
1744 /* must do this from outside any spinlocks */
1745 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1747 /* If the mode uses primary, then this cases was handled above by
1748 * bond_change_active_slave(..., NULL)
1750 if (!bond_uses_primary(bond
)) {
1751 /* unset promiscuity level from slave
1752 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1753 * of the IFF_PROMISC flag in the bond_dev, but we need the
1754 * value of that flag before that change, as that was the value
1755 * when this slave was attached, so we cache at the start of the
1756 * function and use it here. Same goes for ALLMULTI below
1758 if (old_flags
& IFF_PROMISC
)
1759 dev_set_promiscuity(slave_dev
, -1);
1761 /* unset allmulti level from slave */
1762 if (old_flags
& IFF_ALLMULTI
)
1763 dev_set_allmulti(slave_dev
, -1);
1765 bond_hw_addr_flush(bond_dev
, slave_dev
);
1768 slave_disable_netpoll(slave
);
1770 /* close slave before restoring its mac address */
1771 dev_close(slave_dev
);
1773 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
||
1774 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1775 /* restore original ("permanent") mac address */
1776 ether_addr_copy(addr
.sa_data
, slave
->perm_hwaddr
);
1777 addr
.sa_family
= slave_dev
->type
;
1778 dev_set_mac_address(slave_dev
, &addr
);
1781 dev_set_mtu(slave_dev
, slave
->original_mtu
);
1783 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1785 bond_free_slave(slave
);
1787 return 0; /* deletion OK */
1790 /* A wrapper used because of ndo_del_link */
1791 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1793 return __bond_release_one(bond_dev
, slave_dev
, false);
1797 * First release a slave and then destroy the bond if no more slaves are left.
1798 * Must be under rtnl_lock when this function is called.
1800 static int bond_release_and_destroy(struct net_device
*bond_dev
,
1801 struct net_device
*slave_dev
)
1803 struct bonding
*bond
= netdev_priv(bond_dev
);
1806 ret
= bond_release(bond_dev
, slave_dev
);
1807 if (ret
== 0 && !bond_has_slaves(bond
)) {
1808 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1809 netdev_info(bond_dev
, "Destroying bond %s\n",
1811 unregister_netdevice(bond_dev
);
1816 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1818 struct bonding
*bond
= netdev_priv(bond_dev
);
1820 info
->bond_mode
= BOND_MODE(bond
);
1821 info
->miimon
= bond
->params
.miimon
;
1823 info
->num_slaves
= bond
->slave_cnt
;
1828 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
1830 struct bonding
*bond
= netdev_priv(bond_dev
);
1831 struct list_head
*iter
;
1832 int i
= 0, res
= -ENODEV
;
1833 struct slave
*slave
;
1835 bond_for_each_slave(bond
, slave
, iter
) {
1836 if (i
++ == (int)info
->slave_id
) {
1838 strcpy(info
->slave_name
, slave
->dev
->name
);
1839 info
->link
= slave
->link
;
1840 info
->state
= bond_slave_state(slave
);
1841 info
->link_failure_count
= slave
->link_failure_count
;
1849 /*-------------------------------- Monitoring -------------------------------*/
1851 /* called with rcu_read_lock() */
1852 static int bond_miimon_inspect(struct bonding
*bond
)
1854 int link_state
, commit
= 0;
1855 struct list_head
*iter
;
1856 struct slave
*slave
;
1857 bool ignore_updelay
;
1859 ignore_updelay
= !rcu_dereference(bond
->curr_active_slave
);
1861 bond_for_each_slave_rcu(bond
, slave
, iter
) {
1862 slave
->new_link
= BOND_LINK_NOCHANGE
;
1864 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
1866 switch (slave
->link
) {
1871 slave
->link
= BOND_LINK_FAIL
;
1872 slave
->delay
= bond
->params
.downdelay
;
1874 netdev_info(bond
->dev
, "link status down for %sinterface %s, disabling it in %d ms\n",
1876 BOND_MODE_ACTIVEBACKUP
) ?
1877 (bond_is_active_slave(slave
) ?
1878 "active " : "backup ") : "",
1880 bond
->params
.downdelay
* bond
->params
.miimon
);
1883 case BOND_LINK_FAIL
:
1886 * recovered before downdelay expired
1888 slave
->link
= BOND_LINK_UP
;
1889 slave
->last_link_up
= jiffies
;
1890 netdev_info(bond
->dev
, "link status up again after %d ms for interface %s\n",
1891 (bond
->params
.downdelay
- slave
->delay
) *
1892 bond
->params
.miimon
,
1897 if (slave
->delay
<= 0) {
1898 slave
->new_link
= BOND_LINK_DOWN
;
1906 case BOND_LINK_DOWN
:
1910 slave
->link
= BOND_LINK_BACK
;
1911 slave
->delay
= bond
->params
.updelay
;
1914 netdev_info(bond
->dev
, "link status up for interface %s, enabling it in %d ms\n",
1916 ignore_updelay
? 0 :
1917 bond
->params
.updelay
*
1918 bond
->params
.miimon
);
1921 case BOND_LINK_BACK
:
1923 slave
->link
= BOND_LINK_DOWN
;
1924 netdev_info(bond
->dev
, "link status down again after %d ms for interface %s\n",
1925 (bond
->params
.updelay
- slave
->delay
) *
1926 bond
->params
.miimon
,
1935 if (slave
->delay
<= 0) {
1936 slave
->new_link
= BOND_LINK_UP
;
1938 ignore_updelay
= false;
1950 static void bond_miimon_commit(struct bonding
*bond
)
1952 struct list_head
*iter
;
1953 struct slave
*slave
, *primary
;
1955 bond_for_each_slave(bond
, slave
, iter
) {
1956 switch (slave
->new_link
) {
1957 case BOND_LINK_NOCHANGE
:
1961 slave
->link
= BOND_LINK_UP
;
1962 slave
->last_link_up
= jiffies
;
1964 primary
= rtnl_dereference(bond
->primary_slave
);
1965 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1966 /* prevent it from being the active one */
1967 bond_set_backup_slave(slave
);
1968 } else if (BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1969 /* make it immediately active */
1970 bond_set_active_slave(slave
);
1971 } else if (slave
!= primary
) {
1972 /* prevent it from being the active one */
1973 bond_set_backup_slave(slave
);
1976 netdev_info(bond
->dev
, "link status definitely up for interface %s, %u Mbps %s duplex\n",
1978 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
1979 slave
->duplex
? "full" : "half");
1981 /* notify ad that the link status has changed */
1982 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1983 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
1985 if (bond_is_lb(bond
))
1986 bond_alb_handle_link_change(bond
, slave
,
1989 if (!bond
->curr_active_slave
|| slave
== primary
)
1994 case BOND_LINK_DOWN
:
1995 if (slave
->link_failure_count
< UINT_MAX
)
1996 slave
->link_failure_count
++;
1998 slave
->link
= BOND_LINK_DOWN
;
2000 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
2001 BOND_MODE(bond
) == BOND_MODE_8023AD
)
2002 bond_set_slave_inactive_flags(slave
,
2003 BOND_SLAVE_NOTIFY_NOW
);
2005 netdev_info(bond
->dev
, "link status definitely down for interface %s, disabling it\n",
2008 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
2009 bond_3ad_handle_link_change(slave
,
2012 if (bond_is_lb(bond
))
2013 bond_alb_handle_link_change(bond
, slave
,
2016 if (slave
== rcu_access_pointer(bond
->curr_active_slave
))
2022 netdev_err(bond
->dev
, "invalid new link %d on slave %s\n",
2023 slave
->new_link
, slave
->dev
->name
);
2024 slave
->new_link
= BOND_LINK_NOCHANGE
;
2032 bond_select_active_slave(bond
);
2033 unblock_netpoll_tx();
2036 bond_set_carrier(bond
);
2042 * Really a wrapper that splits the mii monitor into two phases: an
2043 * inspection, then (if inspection indicates something needs to be done)
2044 * an acquisition of appropriate locks followed by a commit phase to
2045 * implement whatever link state changes are indicated.
2047 static void bond_mii_monitor(struct work_struct
*work
)
2049 struct bonding
*bond
= container_of(work
, struct bonding
,
2051 bool should_notify_peers
= false;
2052 unsigned long delay
;
2054 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2056 if (!bond_has_slaves(bond
))
2061 should_notify_peers
= bond_should_notify_peers(bond
);
2063 if (bond_miimon_inspect(bond
)) {
2066 /* Race avoidance with bond_close cancel of workqueue */
2067 if (!rtnl_trylock()) {
2069 should_notify_peers
= false;
2073 bond_miimon_commit(bond
);
2075 rtnl_unlock(); /* might sleep, hold no other locks */
2080 if (bond
->params
.miimon
)
2081 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2083 if (should_notify_peers
) {
2084 if (!rtnl_trylock())
2086 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2091 static bool bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2093 struct net_device
*upper
;
2094 struct list_head
*iter
;
2097 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2101 netdev_for_each_all_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2102 if (ip
== bond_confirm_addr(upper
, 0, ip
)) {
2113 * We go to the (large) trouble of VLAN tagging ARP frames because
2114 * switches in VLAN mode (especially if ports are configured as
2115 * "native" to a VLAN) might not pass non-tagged frames.
2117 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
,
2118 __be32 dest_ip
, __be32 src_ip
,
2119 struct bond_vlan_tag
*tags
)
2121 struct sk_buff
*skb
;
2122 struct bond_vlan_tag
*outer_tag
= tags
;
2124 netdev_dbg(slave_dev
, "arp %d on slave %s: dst %pI4 src %pI4\n",
2125 arp_op
, slave_dev
->name
, &dest_ip
, &src_ip
);
2127 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2128 NULL
, slave_dev
->dev_addr
, NULL
);
2131 net_err_ratelimited("ARP packet allocation failed\n");
2135 if (!tags
|| tags
->vlan_proto
== VLAN_N_VID
)
2140 /* Go through all the tags backwards and add them to the packet */
2141 while (tags
->vlan_proto
!= VLAN_N_VID
) {
2142 if (!tags
->vlan_id
) {
2147 netdev_dbg(slave_dev
, "inner tag: proto %X vid %X\n",
2148 ntohs(outer_tag
->vlan_proto
), tags
->vlan_id
);
2149 skb
= __vlan_put_tag(skb
, tags
->vlan_proto
,
2152 net_err_ratelimited("failed to insert inner VLAN tag\n");
2158 /* Set the outer tag */
2159 if (outer_tag
->vlan_id
) {
2160 netdev_dbg(slave_dev
, "outer tag: proto %X vid %X\n",
2161 ntohs(outer_tag
->vlan_proto
), outer_tag
->vlan_id
);
2162 skb
= vlan_put_tag(skb
, outer_tag
->vlan_proto
,
2163 outer_tag
->vlan_id
);
2165 net_err_ratelimited("failed to insert outer VLAN tag\n");
2174 /* Validate the device path between the @start_dev and the @end_dev.
2175 * The path is valid if the @end_dev is reachable through device
2177 * When the path is validated, collect any vlan information in the
2180 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
2181 struct net_device
*end_dev
,
2184 struct bond_vlan_tag
*tags
;
2185 struct net_device
*upper
;
2186 struct list_head
*iter
;
2188 if (start_dev
== end_dev
) {
2189 tags
= kzalloc(sizeof(*tags
) * (level
+ 1), GFP_ATOMIC
);
2191 return ERR_PTR(-ENOMEM
);
2192 tags
[level
].vlan_proto
= VLAN_N_VID
;
2196 netdev_for_each_upper_dev_rcu(start_dev
, upper
, iter
) {
2197 tags
= bond_verify_device_path(upper
, end_dev
, level
+ 1);
2198 if (IS_ERR_OR_NULL(tags
)) {
2203 if (is_vlan_dev(upper
)) {
2204 tags
[level
].vlan_proto
= vlan_dev_vlan_proto(upper
);
2205 tags
[level
].vlan_id
= vlan_dev_vlan_id(upper
);
2214 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2217 struct bond_vlan_tag
*tags
;
2218 __be32
*targets
= bond
->params
.arp_targets
, addr
;
2221 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
&& targets
[i
]; i
++) {
2222 netdev_dbg(bond
->dev
, "basa: target %pI4\n", &targets
[i
]);
2225 /* Find out through which dev should the packet go */
2226 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2229 /* there's no route to target - try to send arp
2230 * probe to generate any traffic (arp_validate=0)
2232 if (bond
->params
.arp_validate
)
2233 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2236 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2241 /* bond device itself */
2242 if (rt
->dst
.dev
== bond
->dev
)
2246 tags
= bond_verify_device_path(bond
->dev
, rt
->dst
.dev
, 0);
2249 if (!IS_ERR_OR_NULL(tags
))
2252 /* Not our device - skip */
2253 netdev_dbg(bond
->dev
, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2254 &targets
[i
], rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2260 addr
= bond_confirm_addr(rt
->dst
.dev
, targets
[i
], 0);
2262 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2268 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2272 if (!sip
|| !bond_has_this_ip(bond
, tip
)) {
2273 netdev_dbg(bond
->dev
, "bva: sip %pI4 tip %pI4 not found\n",
2278 i
= bond_get_targets_ip(bond
->params
.arp_targets
, sip
);
2280 netdev_dbg(bond
->dev
, "bva: sip %pI4 not found in targets\n",
2284 slave
->last_rx
= jiffies
;
2285 slave
->target_last_arp_rx
[i
] = jiffies
;
2288 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2289 struct slave
*slave
)
2291 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2292 struct slave
*curr_active_slave
;
2293 unsigned char *arp_ptr
;
2295 int alen
, is_arp
= skb
->protocol
== __cpu_to_be16(ETH_P_ARP
);
2297 if (!slave_do_arp_validate(bond
, slave
)) {
2298 if ((slave_do_arp_validate_only(bond
) && is_arp
) ||
2299 !slave_do_arp_validate_only(bond
))
2300 slave
->last_rx
= jiffies
;
2301 return RX_HANDLER_ANOTHER
;
2302 } else if (!is_arp
) {
2303 return RX_HANDLER_ANOTHER
;
2306 alen
= arp_hdr_len(bond
->dev
);
2308 netdev_dbg(bond
->dev
, "bond_arp_rcv: skb->dev %s\n",
2311 if (alen
> skb_headlen(skb
)) {
2312 arp
= kmalloc(alen
, GFP_ATOMIC
);
2315 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2319 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2320 skb
->pkt_type
== PACKET_OTHERHOST
||
2321 skb
->pkt_type
== PACKET_LOOPBACK
||
2322 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2323 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2327 arp_ptr
= (unsigned char *)(arp
+ 1);
2328 arp_ptr
+= bond
->dev
->addr_len
;
2329 memcpy(&sip
, arp_ptr
, 4);
2330 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2331 memcpy(&tip
, arp_ptr
, 4);
2333 netdev_dbg(bond
->dev
, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2334 slave
->dev
->name
, bond_slave_state(slave
),
2335 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2338 curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2341 * Backup slaves won't see the ARP reply, but do come through
2342 * here for each ARP probe (so we swap the sip/tip to validate
2343 * the probe). In a "redundant switch, common router" type of
2344 * configuration, the ARP probe will (hopefully) travel from
2345 * the active, through one switch, the router, then the other
2346 * switch before reaching the backup.
2348 * We 'trust' the arp requests if there is an active slave and
2349 * it received valid arp reply(s) after it became active. This
2350 * is done to avoid endless looping when we can't reach the
2351 * arp_ip_target and fool ourselves with our own arp requests.
2354 if (bond_is_active_slave(slave
))
2355 bond_validate_arp(bond
, slave
, sip
, tip
);
2356 else if (curr_active_slave
&&
2357 time_after(slave_last_rx(bond
, curr_active_slave
),
2358 curr_active_slave
->last_link_up
))
2359 bond_validate_arp(bond
, slave
, tip
, sip
);
2362 if (arp
!= (struct arphdr
*)skb
->data
)
2364 return RX_HANDLER_ANOTHER
;
2367 /* function to verify if we're in the arp_interval timeslice, returns true if
2368 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2369 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2371 static bool bond_time_in_interval(struct bonding
*bond
, unsigned long last_act
,
2374 int delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2376 return time_in_range(jiffies
,
2377 last_act
- delta_in_ticks
,
2378 last_act
+ mod
* delta_in_ticks
+ delta_in_ticks
/2);
2382 * this function is called regularly to monitor each slave's link
2383 * ensuring that traffic is being sent and received when arp monitoring
2384 * is used in load-balancing mode. if the adapter has been dormant, then an
2385 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2386 * arp monitoring in active backup mode.
2388 static void bond_loadbalance_arp_mon(struct work_struct
*work
)
2390 struct bonding
*bond
= container_of(work
, struct bonding
,
2392 struct slave
*slave
, *oldcurrent
;
2393 struct list_head
*iter
;
2394 int do_failover
= 0, slave_state_changed
= 0;
2396 if (!bond_has_slaves(bond
))
2401 oldcurrent
= rcu_dereference(bond
->curr_active_slave
);
2402 /* see if any of the previous devices are up now (i.e. they have
2403 * xmt and rcv traffic). the curr_active_slave does not come into
2404 * the picture unless it is null. also, slave->last_link_up is not
2405 * needed here because we send an arp on each slave and give a slave
2406 * as long as it needs to get the tx/rx within the delta.
2407 * TODO: what about up/down delay in arp mode? it wasn't here before
2410 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2411 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2413 if (slave
->link
!= BOND_LINK_UP
) {
2414 if (bond_time_in_interval(bond
, trans_start
, 1) &&
2415 bond_time_in_interval(bond
, slave
->last_rx
, 1)) {
2417 slave
->link
= BOND_LINK_UP
;
2418 slave_state_changed
= 1;
2420 /* primary_slave has no meaning in round-robin
2421 * mode. the window of a slave being up and
2422 * curr_active_slave being null after enslaving
2426 netdev_info(bond
->dev
, "link status definitely up for interface %s\n",
2430 netdev_info(bond
->dev
, "interface %s is now up\n",
2435 /* slave->link == BOND_LINK_UP */
2437 /* not all switches will respond to an arp request
2438 * when the source ip is 0, so don't take the link down
2439 * if we don't know our ip yet
2441 if (!bond_time_in_interval(bond
, trans_start
, 2) ||
2442 !bond_time_in_interval(bond
, slave
->last_rx
, 2)) {
2444 slave
->link
= BOND_LINK_DOWN
;
2445 slave_state_changed
= 1;
2447 if (slave
->link_failure_count
< UINT_MAX
)
2448 slave
->link_failure_count
++;
2450 netdev_info(bond
->dev
, "interface %s is now down\n",
2453 if (slave
== oldcurrent
)
2458 /* note: if switch is in round-robin mode, all links
2459 * must tx arp to ensure all links rx an arp - otherwise
2460 * links may oscillate or not come up at all; if switch is
2461 * in something like xor mode, there is nothing we can
2462 * do - all replies will be rx'ed on same link causing slaves
2463 * to be unstable during low/no traffic periods
2465 if (bond_slave_is_up(slave
))
2466 bond_arp_send_all(bond
, slave
);
2471 if (do_failover
|| slave_state_changed
) {
2472 if (!rtnl_trylock())
2475 if (slave_state_changed
) {
2476 bond_slave_state_change(bond
);
2477 } else if (do_failover
) {
2479 bond_select_active_slave(bond
);
2480 unblock_netpoll_tx();
2486 if (bond
->params
.arp_interval
)
2487 queue_delayed_work(bond
->wq
, &bond
->arp_work
,
2488 msecs_to_jiffies(bond
->params
.arp_interval
));
2492 * Called to inspect slaves for active-backup mode ARP monitor link state
2493 * changes. Sets new_link in slaves to specify what action should take
2494 * place for the slave. Returns 0 if no changes are found, >0 if changes
2495 * to link states must be committed.
2497 * Called with rcu_read_lock hold.
2499 static int bond_ab_arp_inspect(struct bonding
*bond
)
2501 unsigned long trans_start
, last_rx
;
2502 struct list_head
*iter
;
2503 struct slave
*slave
;
2506 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2507 slave
->new_link
= BOND_LINK_NOCHANGE
;
2508 last_rx
= slave_last_rx(bond
, slave
);
2510 if (slave
->link
!= BOND_LINK_UP
) {
2511 if (bond_time_in_interval(bond
, last_rx
, 1)) {
2512 slave
->new_link
= BOND_LINK_UP
;
2519 * Give slaves 2*delta after being enslaved or made
2520 * active. This avoids bouncing, as the last receive
2521 * times need a full ARP monitor cycle to be updated.
2523 if (bond_time_in_interval(bond
, slave
->last_link_up
, 2))
2527 * Backup slave is down if:
2528 * - No current_arp_slave AND
2529 * - more than 3*delta since last receive AND
2530 * - the bond has an IP address
2532 * Note: a non-null current_arp_slave indicates
2533 * the curr_active_slave went down and we are
2534 * searching for a new one; under this condition
2535 * we only take the curr_active_slave down - this
2536 * gives each slave a chance to tx/rx traffic
2537 * before being taken out
2539 if (!bond_is_active_slave(slave
) &&
2540 !rcu_access_pointer(bond
->current_arp_slave
) &&
2541 !bond_time_in_interval(bond
, last_rx
, 3)) {
2542 slave
->new_link
= BOND_LINK_DOWN
;
2547 * Active slave is down if:
2548 * - more than 2*delta since transmitting OR
2549 * - (more than 2*delta since receive AND
2550 * the bond has an IP address)
2552 trans_start
= dev_trans_start(slave
->dev
);
2553 if (bond_is_active_slave(slave
) &&
2554 (!bond_time_in_interval(bond
, trans_start
, 2) ||
2555 !bond_time_in_interval(bond
, last_rx
, 2))) {
2556 slave
->new_link
= BOND_LINK_DOWN
;
2565 * Called to commit link state changes noted by inspection step of
2566 * active-backup mode ARP monitor.
2568 * Called with RTNL hold.
2570 static void bond_ab_arp_commit(struct bonding
*bond
)
2572 unsigned long trans_start
;
2573 struct list_head
*iter
;
2574 struct slave
*slave
;
2576 bond_for_each_slave(bond
, slave
, iter
) {
2577 switch (slave
->new_link
) {
2578 case BOND_LINK_NOCHANGE
:
2582 trans_start
= dev_trans_start(slave
->dev
);
2583 if (rtnl_dereference(bond
->curr_active_slave
) != slave
||
2584 (!rtnl_dereference(bond
->curr_active_slave
) &&
2585 bond_time_in_interval(bond
, trans_start
, 1))) {
2586 struct slave
*current_arp_slave
;
2588 current_arp_slave
= rtnl_dereference(bond
->current_arp_slave
);
2589 slave
->link
= BOND_LINK_UP
;
2590 if (current_arp_slave
) {
2591 bond_set_slave_inactive_flags(
2593 BOND_SLAVE_NOTIFY_NOW
);
2594 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
2597 netdev_info(bond
->dev
, "link status definitely up for interface %s\n",
2600 if (!rtnl_dereference(bond
->curr_active_slave
) ||
2601 slave
== rtnl_dereference(bond
->primary_slave
))
2608 case BOND_LINK_DOWN
:
2609 if (slave
->link_failure_count
< UINT_MAX
)
2610 slave
->link_failure_count
++;
2612 slave
->link
= BOND_LINK_DOWN
;
2613 bond_set_slave_inactive_flags(slave
,
2614 BOND_SLAVE_NOTIFY_NOW
);
2616 netdev_info(bond
->dev
, "link status definitely down for interface %s, disabling it\n",
2619 if (slave
== rtnl_dereference(bond
->curr_active_slave
)) {
2620 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
2627 netdev_err(bond
->dev
, "impossible: new_link %d on slave %s\n",
2628 slave
->new_link
, slave
->dev
->name
);
2635 bond_select_active_slave(bond
);
2636 unblock_netpoll_tx();
2639 bond_set_carrier(bond
);
2643 * Send ARP probes for active-backup mode ARP monitor.
2645 * Called with rcu_read_lock hold.
2647 static bool bond_ab_arp_probe(struct bonding
*bond
)
2649 struct slave
*slave
, *before
= NULL
, *new_slave
= NULL
,
2650 *curr_arp_slave
= rcu_dereference(bond
->current_arp_slave
),
2651 *curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2652 struct list_head
*iter
;
2654 bool should_notify_rtnl
= BOND_SLAVE_NOTIFY_LATER
;
2656 if (curr_arp_slave
&& curr_active_slave
)
2657 netdev_info(bond
->dev
, "PROBE: c_arp %s && cas %s BAD\n",
2658 curr_arp_slave
->dev
->name
,
2659 curr_active_slave
->dev
->name
);
2661 if (curr_active_slave
) {
2662 bond_arp_send_all(bond
, curr_active_slave
);
2663 return should_notify_rtnl
;
2666 /* if we don't have a curr_active_slave, search for the next available
2667 * backup slave from the current_arp_slave and make it the candidate
2668 * for becoming the curr_active_slave
2671 if (!curr_arp_slave
) {
2672 curr_arp_slave
= bond_first_slave_rcu(bond
);
2673 if (!curr_arp_slave
)
2674 return should_notify_rtnl
;
2677 bond_set_slave_inactive_flags(curr_arp_slave
, BOND_SLAVE_NOTIFY_LATER
);
2679 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2680 if (!found
&& !before
&& bond_slave_is_up(slave
))
2683 if (found
&& !new_slave
&& bond_slave_is_up(slave
))
2685 /* if the link state is up at this point, we
2686 * mark it down - this can happen if we have
2687 * simultaneous link failures and
2688 * reselect_active_interface doesn't make this
2689 * one the current slave so it is still marked
2690 * up when it is actually down
2692 if (!bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
2693 slave
->link
= BOND_LINK_DOWN
;
2694 if (slave
->link_failure_count
< UINT_MAX
)
2695 slave
->link_failure_count
++;
2697 bond_set_slave_inactive_flags(slave
,
2698 BOND_SLAVE_NOTIFY_LATER
);
2700 netdev_info(bond
->dev
, "backup interface %s is now down\n",
2703 if (slave
== curr_arp_slave
)
2707 if (!new_slave
&& before
)
2713 new_slave
->link
= BOND_LINK_BACK
;
2714 bond_set_slave_active_flags(new_slave
, BOND_SLAVE_NOTIFY_LATER
);
2715 bond_arp_send_all(bond
, new_slave
);
2716 new_slave
->last_link_up
= jiffies
;
2717 rcu_assign_pointer(bond
->current_arp_slave
, new_slave
);
2720 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2721 if (slave
->should_notify
) {
2722 should_notify_rtnl
= BOND_SLAVE_NOTIFY_NOW
;
2726 return should_notify_rtnl
;
2729 static void bond_activebackup_arp_mon(struct work_struct
*work
)
2731 struct bonding
*bond
= container_of(work
, struct bonding
,
2733 bool should_notify_peers
= false;
2734 bool should_notify_rtnl
= false;
2737 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2739 if (!bond_has_slaves(bond
))
2744 should_notify_peers
= bond_should_notify_peers(bond
);
2746 if (bond_ab_arp_inspect(bond
)) {
2749 /* Race avoidance with bond_close flush of workqueue */
2750 if (!rtnl_trylock()) {
2752 should_notify_peers
= false;
2756 bond_ab_arp_commit(bond
);
2762 should_notify_rtnl
= bond_ab_arp_probe(bond
);
2766 if (bond
->params
.arp_interval
)
2767 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2769 if (should_notify_peers
|| should_notify_rtnl
) {
2770 if (!rtnl_trylock())
2773 if (should_notify_peers
)
2774 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
2776 if (should_notify_rtnl
)
2777 bond_slave_state_notify(bond
);
2783 /*-------------------------- netdev event handling --------------------------*/
2786 * Change device name
2788 static int bond_event_changename(struct bonding
*bond
)
2790 bond_remove_proc_entry(bond
);
2791 bond_create_proc_entry(bond
);
2793 bond_debug_reregister(bond
);
2798 static int bond_master_netdev_event(unsigned long event
,
2799 struct net_device
*bond_dev
)
2801 struct bonding
*event_bond
= netdev_priv(bond_dev
);
2804 case NETDEV_CHANGENAME
:
2805 return bond_event_changename(event_bond
);
2806 case NETDEV_UNREGISTER
:
2807 bond_remove_proc_entry(event_bond
);
2809 case NETDEV_REGISTER
:
2810 bond_create_proc_entry(event_bond
);
2812 case NETDEV_NOTIFY_PEERS
:
2813 if (event_bond
->send_peer_notif
)
2814 event_bond
->send_peer_notif
--;
2823 static int bond_slave_netdev_event(unsigned long event
,
2824 struct net_device
*slave_dev
)
2826 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
), *primary
;
2827 struct bonding
*bond
;
2828 struct net_device
*bond_dev
;
2832 /* A netdev event can be generated while enslaving a device
2833 * before netdev_rx_handler_register is called in which case
2834 * slave will be NULL
2838 bond_dev
= slave
->bond
->dev
;
2840 primary
= rtnl_dereference(bond
->primary_slave
);
2843 case NETDEV_UNREGISTER
:
2844 if (bond_dev
->type
!= ARPHRD_ETHER
)
2845 bond_release_and_destroy(bond_dev
, slave_dev
);
2847 bond_release(bond_dev
, slave_dev
);
2851 old_speed
= slave
->speed
;
2852 old_duplex
= slave
->duplex
;
2854 bond_update_speed_duplex(slave
);
2856 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
2857 if (old_speed
!= slave
->speed
)
2858 bond_3ad_adapter_speed_changed(slave
);
2859 if (old_duplex
!= slave
->duplex
)
2860 bond_3ad_adapter_duplex_changed(slave
);
2865 * ... Or is it this?
2868 case NETDEV_CHANGEMTU
:
2870 * TODO: Should slaves be allowed to
2871 * independently alter their MTU? For
2872 * an active-backup bond, slaves need
2873 * not be the same type of device, so
2874 * MTUs may vary. For other modes,
2875 * slaves arguably should have the
2876 * same MTUs. To do this, we'd need to
2877 * take over the slave's change_mtu
2878 * function for the duration of their
2882 case NETDEV_CHANGENAME
:
2883 /* we don't care if we don't have primary set */
2884 if (!bond_uses_primary(bond
) ||
2885 !bond
->params
.primary
[0])
2888 if (slave
== primary
) {
2889 /* slave's name changed - he's no longer primary */
2890 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
2891 } else if (!strcmp(slave_dev
->name
, bond
->params
.primary
)) {
2892 /* we have a new primary slave */
2893 rcu_assign_pointer(bond
->primary_slave
, slave
);
2894 } else { /* we didn't change primary - exit */
2898 netdev_info(bond
->dev
, "Primary slave changed to %s, reselecting active slave\n",
2899 primary
? slave_dev
->name
: "none");
2902 bond_select_active_slave(bond
);
2903 unblock_netpoll_tx();
2905 case NETDEV_FEAT_CHANGE
:
2906 bond_compute_features(bond
);
2908 case NETDEV_RESEND_IGMP
:
2909 /* Propagate to master device */
2910 call_netdevice_notifiers(event
, slave
->bond
->dev
);
2920 * bond_netdev_event: handle netdev notifier chain events.
2922 * This function receives events for the netdev chain. The caller (an
2923 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2924 * locks for us to safely manipulate the slave devices (RTNL lock,
2927 static int bond_netdev_event(struct notifier_block
*this,
2928 unsigned long event
, void *ptr
)
2930 struct net_device
*event_dev
= netdev_notifier_info_to_dev(ptr
);
2932 netdev_dbg(event_dev
, "event: %lx\n", event
);
2934 if (!(event_dev
->priv_flags
& IFF_BONDING
))
2937 if (event_dev
->flags
& IFF_MASTER
) {
2938 netdev_dbg(event_dev
, "IFF_MASTER\n");
2939 return bond_master_netdev_event(event
, event_dev
);
2942 if (event_dev
->flags
& IFF_SLAVE
) {
2943 netdev_dbg(event_dev
, "IFF_SLAVE\n");
2944 return bond_slave_netdev_event(event
, event_dev
);
2950 static struct notifier_block bond_netdev_notifier
= {
2951 .notifier_call
= bond_netdev_event
,
2954 /*---------------------------- Hashing Policies -----------------------------*/
2956 /* L2 hash helper */
2957 static inline u32
bond_eth_hash(struct sk_buff
*skb
)
2959 struct ethhdr
*ep
, hdr_tmp
;
2961 ep
= skb_header_pointer(skb
, 0, sizeof(hdr_tmp
), &hdr_tmp
);
2963 return ep
->h_dest
[5] ^ ep
->h_source
[5] ^ ep
->h_proto
;
2967 /* Extract the appropriate headers based on bond's xmit policy */
2968 static bool bond_flow_dissect(struct bonding
*bond
, struct sk_buff
*skb
,
2969 struct flow_keys
*fk
)
2971 const struct ipv6hdr
*iph6
;
2972 const struct iphdr
*iph
;
2973 int noff
, proto
= -1;
2975 if (bond
->params
.xmit_policy
> BOND_XMIT_POLICY_LAYER23
)
2976 return skb_flow_dissect(skb
, fk
);
2979 noff
= skb_network_offset(skb
);
2980 if (skb
->protocol
== htons(ETH_P_IP
)) {
2981 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph
))))
2984 fk
->src
= iph
->saddr
;
2985 fk
->dst
= iph
->daddr
;
2986 noff
+= iph
->ihl
<< 2;
2987 if (!ip_is_fragment(iph
))
2988 proto
= iph
->protocol
;
2989 } else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
2990 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph6
))))
2992 iph6
= ipv6_hdr(skb
);
2993 fk
->src
= (__force __be32
)ipv6_addr_hash(&iph6
->saddr
);
2994 fk
->dst
= (__force __be32
)ipv6_addr_hash(&iph6
->daddr
);
2995 noff
+= sizeof(*iph6
);
2996 proto
= iph6
->nexthdr
;
3000 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
&& proto
>= 0)
3001 fk
->ports
= skb_flow_get_ports(skb
, noff
, proto
);
3007 * bond_xmit_hash - generate a hash value based on the xmit policy
3008 * @bond: bonding device
3009 * @skb: buffer to use for headers
3011 * This function will extract the necessary headers from the skb buffer and use
3012 * them to generate a hash based on the xmit_policy set in the bonding device
3014 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
)
3016 struct flow_keys flow
;
3019 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER2
||
3020 !bond_flow_dissect(bond
, skb
, &flow
))
3021 return bond_eth_hash(skb
);
3023 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER23
||
3024 bond
->params
.xmit_policy
== BOND_XMIT_POLICY_ENCAP23
)
3025 hash
= bond_eth_hash(skb
);
3027 hash
= (__force u32
)flow
.ports
;
3028 hash
^= (__force u32
)flow
.dst
^ (__force u32
)flow
.src
;
3029 hash
^= (hash
>> 16);
3030 hash
^= (hash
>> 8);
3035 /*-------------------------- Device entry points ----------------------------*/
3037 static void bond_work_init_all(struct bonding
*bond
)
3039 INIT_DELAYED_WORK(&bond
->mcast_work
,
3040 bond_resend_igmp_join_requests_delayed
);
3041 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3042 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3043 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3044 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3046 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3047 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3050 static void bond_work_cancel_all(struct bonding
*bond
)
3052 cancel_delayed_work_sync(&bond
->mii_work
);
3053 cancel_delayed_work_sync(&bond
->arp_work
);
3054 cancel_delayed_work_sync(&bond
->alb_work
);
3055 cancel_delayed_work_sync(&bond
->ad_work
);
3056 cancel_delayed_work_sync(&bond
->mcast_work
);
3059 static int bond_open(struct net_device
*bond_dev
)
3061 struct bonding
*bond
= netdev_priv(bond_dev
);
3062 struct list_head
*iter
;
3063 struct slave
*slave
;
3065 /* reset slave->backup and slave->inactive */
3066 if (bond_has_slaves(bond
)) {
3067 bond_for_each_slave(bond
, slave
, iter
) {
3068 if (bond_uses_primary(bond
) &&
3069 slave
!= rcu_access_pointer(bond
->curr_active_slave
)) {
3070 bond_set_slave_inactive_flags(slave
,
3071 BOND_SLAVE_NOTIFY_NOW
);
3073 bond_set_slave_active_flags(slave
,
3074 BOND_SLAVE_NOTIFY_NOW
);
3079 bond_work_init_all(bond
);
3081 if (bond_is_lb(bond
)) {
3082 /* bond_alb_initialize must be called before the timer
3085 if (bond_alb_initialize(bond
, (BOND_MODE(bond
) == BOND_MODE_ALB
)))
3087 if (bond
->params
.tlb_dynamic_lb
)
3088 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3091 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3092 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3094 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3095 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3096 bond
->recv_probe
= bond_arp_rcv
;
3099 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
3100 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3101 /* register to receive LACPDUs */
3102 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3103 bond_3ad_initiate_agg_selection(bond
, 1);
3109 static int bond_close(struct net_device
*bond_dev
)
3111 struct bonding
*bond
= netdev_priv(bond_dev
);
3113 bond_work_cancel_all(bond
);
3114 bond
->send_peer_notif
= 0;
3115 if (bond_is_lb(bond
))
3116 bond_alb_deinitialize(bond
);
3117 bond
->recv_probe
= NULL
;
3122 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3123 struct rtnl_link_stats64
*stats
)
3125 struct bonding
*bond
= netdev_priv(bond_dev
);
3126 struct rtnl_link_stats64 temp
;
3127 struct list_head
*iter
;
3128 struct slave
*slave
;
3130 memset(stats
, 0, sizeof(*stats
));
3132 bond_for_each_slave(bond
, slave
, iter
) {
3133 const struct rtnl_link_stats64
*sstats
=
3134 dev_get_stats(slave
->dev
, &temp
);
3136 stats
->rx_packets
+= sstats
->rx_packets
;
3137 stats
->rx_bytes
+= sstats
->rx_bytes
;
3138 stats
->rx_errors
+= sstats
->rx_errors
;
3139 stats
->rx_dropped
+= sstats
->rx_dropped
;
3141 stats
->tx_packets
+= sstats
->tx_packets
;
3142 stats
->tx_bytes
+= sstats
->tx_bytes
;
3143 stats
->tx_errors
+= sstats
->tx_errors
;
3144 stats
->tx_dropped
+= sstats
->tx_dropped
;
3146 stats
->multicast
+= sstats
->multicast
;
3147 stats
->collisions
+= sstats
->collisions
;
3149 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3150 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3151 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3152 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3153 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3154 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3156 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3157 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3158 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3159 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3160 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3166 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3168 struct bonding
*bond
= netdev_priv(bond_dev
);
3169 struct net_device
*slave_dev
= NULL
;
3170 struct ifbond k_binfo
;
3171 struct ifbond __user
*u_binfo
= NULL
;
3172 struct ifslave k_sinfo
;
3173 struct ifslave __user
*u_sinfo
= NULL
;
3174 struct mii_ioctl_data
*mii
= NULL
;
3175 struct bond_opt_value newval
;
3179 netdev_dbg(bond_dev
, "bond_ioctl: cmd=%d\n", cmd
);
3191 * We do this again just in case we were called by SIOCGMIIREG
3192 * instead of SIOCGMIIPHY.
3198 if (mii
->reg_num
== 1) {
3200 if (netif_carrier_ok(bond
->dev
))
3201 mii
->val_out
= BMSR_LSTATUS
;
3205 case BOND_INFO_QUERY_OLD
:
3206 case SIOCBONDINFOQUERY
:
3207 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3209 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3212 res
= bond_info_query(bond_dev
, &k_binfo
);
3214 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3218 case BOND_SLAVE_INFO_QUERY_OLD
:
3219 case SIOCBONDSLAVEINFOQUERY
:
3220 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3222 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3225 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3227 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3236 net
= dev_net(bond_dev
);
3238 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3241 slave_dev
= __dev_get_by_name(net
, ifr
->ifr_slave
);
3243 netdev_dbg(bond_dev
, "slave_dev=%p:\n", slave_dev
);
3248 netdev_dbg(bond_dev
, "slave_dev->name=%s:\n", slave_dev
->name
);
3250 case BOND_ENSLAVE_OLD
:
3251 case SIOCBONDENSLAVE
:
3252 res
= bond_enslave(bond_dev
, slave_dev
);
3254 case BOND_RELEASE_OLD
:
3255 case SIOCBONDRELEASE
:
3256 res
= bond_release(bond_dev
, slave_dev
);
3258 case BOND_SETHWADDR_OLD
:
3259 case SIOCBONDSETHWADDR
:
3260 bond_set_dev_addr(bond_dev
, slave_dev
);
3263 case BOND_CHANGE_ACTIVE_OLD
:
3264 case SIOCBONDCHANGEACTIVE
:
3265 bond_opt_initstr(&newval
, slave_dev
->name
);
3266 res
= __bond_opt_set(bond
, BOND_OPT_ACTIVE_SLAVE
, &newval
);
3275 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3277 struct bonding
*bond
= netdev_priv(bond_dev
);
3279 if (change
& IFF_PROMISC
)
3280 bond_set_promiscuity(bond
,
3281 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3283 if (change
& IFF_ALLMULTI
)
3284 bond_set_allmulti(bond
,
3285 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3288 static void bond_set_rx_mode(struct net_device
*bond_dev
)
3290 struct bonding
*bond
= netdev_priv(bond_dev
);
3291 struct list_head
*iter
;
3292 struct slave
*slave
;
3296 if (bond_uses_primary(bond
)) {
3297 slave
= rcu_dereference(bond
->curr_active_slave
);
3299 dev_uc_sync(slave
->dev
, bond_dev
);
3300 dev_mc_sync(slave
->dev
, bond_dev
);
3303 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3304 dev_uc_sync_multiple(slave
->dev
, bond_dev
);
3305 dev_mc_sync_multiple(slave
->dev
, bond_dev
);
3311 static int bond_neigh_init(struct neighbour
*n
)
3313 struct bonding
*bond
= netdev_priv(n
->dev
);
3314 const struct net_device_ops
*slave_ops
;
3315 struct neigh_parms parms
;
3316 struct slave
*slave
;
3319 slave
= bond_first_slave(bond
);
3322 slave_ops
= slave
->dev
->netdev_ops
;
3323 if (!slave_ops
->ndo_neigh_setup
)
3326 parms
.neigh_setup
= NULL
;
3327 parms
.neigh_cleanup
= NULL
;
3328 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3333 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3334 * after the last slave has been detached. Assumes that all slaves
3335 * utilize the same neigh_cleanup (true at this writing as only user
3338 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3340 if (!parms
.neigh_setup
)
3343 return parms
.neigh_setup(n
);
3347 * The bonding ndo_neigh_setup is called at init time beofre any
3348 * slave exists. So we must declare proxy setup function which will
3349 * be used at run time to resolve the actual slave neigh param setup.
3351 * It's also called by master devices (such as vlans) to setup their
3352 * underlying devices. In that case - do nothing, we're already set up from
3355 static int bond_neigh_setup(struct net_device
*dev
,
3356 struct neigh_parms
*parms
)
3358 /* modify only our neigh_parms */
3359 if (parms
->dev
== dev
)
3360 parms
->neigh_setup
= bond_neigh_init
;
3366 * Change the MTU of all of a master's slaves to match the master
3368 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3370 struct bonding
*bond
= netdev_priv(bond_dev
);
3371 struct slave
*slave
, *rollback_slave
;
3372 struct list_head
*iter
;
3375 netdev_dbg(bond_dev
, "bond=%p, new_mtu=%d\n", bond
, new_mtu
);
3377 bond_for_each_slave(bond
, slave
, iter
) {
3378 netdev_dbg(bond_dev
, "s %p c_m %p\n",
3379 slave
, slave
->dev
->netdev_ops
->ndo_change_mtu
);
3381 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3384 /* If we failed to set the slave's mtu to the new value
3385 * we must abort the operation even in ACTIVE_BACKUP
3386 * mode, because if we allow the backup slaves to have
3387 * different mtu values than the active slave we'll
3388 * need to change their mtu when doing a failover. That
3389 * means changing their mtu from timer context, which
3390 * is probably not a good idea.
3392 netdev_dbg(bond_dev
, "err %d %s\n", res
,
3398 bond_dev
->mtu
= new_mtu
;
3403 /* unwind from head to the slave that failed */
3404 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3407 if (rollback_slave
== slave
)
3410 tmp_res
= dev_set_mtu(rollback_slave
->dev
, bond_dev
->mtu
);
3412 netdev_dbg(bond_dev
, "unwind err %d dev %s\n",
3413 tmp_res
, rollback_slave
->dev
->name
);
3423 * Note that many devices must be down to change the HW address, and
3424 * downing the master releases all slaves. We can make bonds full of
3425 * bonding devices to test this, however.
3427 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3429 struct bonding
*bond
= netdev_priv(bond_dev
);
3430 struct slave
*slave
, *rollback_slave
;
3431 struct sockaddr
*sa
= addr
, tmp_sa
;
3432 struct list_head
*iter
;
3435 if (BOND_MODE(bond
) == BOND_MODE_ALB
)
3436 return bond_alb_set_mac_address(bond_dev
, addr
);
3439 netdev_dbg(bond_dev
, "bond=%p\n", bond
);
3441 /* If fail_over_mac is enabled, do nothing and return success.
3442 * Returning an error causes ifenslave to fail.
3444 if (bond
->params
.fail_over_mac
&&
3445 BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3448 if (!is_valid_ether_addr(sa
->sa_data
))
3449 return -EADDRNOTAVAIL
;
3451 bond_for_each_slave(bond
, slave
, iter
) {
3452 netdev_dbg(bond_dev
, "slave %p %s\n", slave
, slave
->dev
->name
);
3453 res
= dev_set_mac_address(slave
->dev
, addr
);
3455 /* TODO: consider downing the slave
3457 * User should expect communications
3458 * breakage anyway until ARP finish
3461 netdev_dbg(bond_dev
, "err %d %s\n", res
, slave
->dev
->name
);
3467 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3471 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3472 tmp_sa
.sa_family
= bond_dev
->type
;
3474 /* unwind from head to the slave that failed */
3475 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3478 if (rollback_slave
== slave
)
3481 tmp_res
= dev_set_mac_address(rollback_slave
->dev
, &tmp_sa
);
3483 netdev_dbg(bond_dev
, "unwind err %d dev %s\n",
3484 tmp_res
, rollback_slave
->dev
->name
);
3492 * bond_xmit_slave_id - transmit skb through slave with slave_id
3493 * @bond: bonding device that is transmitting
3494 * @skb: buffer to transmit
3495 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3497 * This function tries to transmit through slave with slave_id but in case
3498 * it fails, it tries to find the first available slave for transmission.
3499 * The skb is consumed in all cases, thus the function is void.
3501 static void bond_xmit_slave_id(struct bonding
*bond
, struct sk_buff
*skb
, int slave_id
)
3503 struct list_head
*iter
;
3504 struct slave
*slave
;
3507 /* Here we start from the slave with slave_id */
3508 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3510 if (bond_slave_can_tx(slave
)) {
3511 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3517 /* Here we start from the first slave up to slave_id */
3519 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3522 if (bond_slave_can_tx(slave
)) {
3523 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3527 /* no slave that can tx has been found */
3528 dev_kfree_skb_any(skb
);
3532 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3533 * @bond: bonding device to use
3535 * Based on the value of the bonding device's packets_per_slave parameter
3536 * this function generates a slave id, which is usually used as the next
3537 * slave to transmit through.
3539 static u32
bond_rr_gen_slave_id(struct bonding
*bond
)
3542 struct reciprocal_value reciprocal_packets_per_slave
;
3543 int packets_per_slave
= bond
->params
.packets_per_slave
;
3545 switch (packets_per_slave
) {
3547 slave_id
= prandom_u32();
3550 slave_id
= bond
->rr_tx_counter
;
3553 reciprocal_packets_per_slave
=
3554 bond
->params
.reciprocal_packets_per_slave
;
3555 slave_id
= reciprocal_divide(bond
->rr_tx_counter
,
3556 reciprocal_packets_per_slave
);
3559 bond
->rr_tx_counter
++;
3564 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3566 struct bonding
*bond
= netdev_priv(bond_dev
);
3567 struct iphdr
*iph
= ip_hdr(skb
);
3568 struct slave
*slave
;
3571 /* Start with the curr_active_slave that joined the bond as the
3572 * default for sending IGMP traffic. For failover purposes one
3573 * needs to maintain some consistency for the interface that will
3574 * send the join/membership reports. The curr_active_slave found
3575 * will send all of this type of traffic.
3577 if (iph
->protocol
== IPPROTO_IGMP
&& skb
->protocol
== htons(ETH_P_IP
)) {
3578 slave
= rcu_dereference(bond
->curr_active_slave
);
3579 if (slave
&& bond_slave_can_tx(slave
))
3580 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3582 bond_xmit_slave_id(bond
, skb
, 0);
3584 slave_id
= bond_rr_gen_slave_id(bond
);
3585 bond_xmit_slave_id(bond
, skb
, slave_id
% bond
->slave_cnt
);
3588 return NETDEV_TX_OK
;
3592 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3593 * the bond has a usable interface.
3595 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3597 struct bonding
*bond
= netdev_priv(bond_dev
);
3598 struct slave
*slave
;
3600 slave
= rcu_dereference(bond
->curr_active_slave
);
3602 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3604 dev_kfree_skb_any(skb
);
3606 return NETDEV_TX_OK
;
3609 /* In bond_xmit_xor() , we determine the output device by using a pre-
3610 * determined xmit_hash_policy(), If the selected device is not enabled,
3611 * find the next active slave.
3613 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3615 struct bonding
*bond
= netdev_priv(bond_dev
);
3617 bond_xmit_slave_id(bond
, skb
, bond_xmit_hash(bond
, skb
) % bond
->slave_cnt
);
3619 return NETDEV_TX_OK
;
3622 /* in broadcast mode, we send everything to all usable interfaces. */
3623 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3625 struct bonding
*bond
= netdev_priv(bond_dev
);
3626 struct slave
*slave
= NULL
;
3627 struct list_head
*iter
;
3629 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3630 if (bond_is_last_slave(bond
, slave
))
3632 if (bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
3633 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
3636 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3637 bond_dev
->name
, __func__
);
3640 /* bond_dev_queue_xmit always returns 0 */
3641 bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3644 if (slave
&& bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
)
3645 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3647 dev_kfree_skb_any(skb
);
3649 return NETDEV_TX_OK
;
3652 /*------------------------- Device initialization ---------------------------*/
3655 * Lookup the slave that corresponds to a qid
3657 static inline int bond_slave_override(struct bonding
*bond
,
3658 struct sk_buff
*skb
)
3660 struct slave
*slave
= NULL
;
3661 struct list_head
*iter
;
3663 if (!skb
->queue_mapping
)
3666 /* Find out if any slaves have the same mapping as this skb. */
3667 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3668 if (slave
->queue_id
== skb
->queue_mapping
) {
3669 if (bond_slave_can_tx(slave
)) {
3670 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3673 /* If the slave isn't UP, use default transmit policy. */
3682 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
3683 void *accel_priv
, select_queue_fallback_t fallback
)
3686 * This helper function exists to help dev_pick_tx get the correct
3687 * destination queue. Using a helper function skips a call to
3688 * skb_tx_hash and will put the skbs in the queue we expect on their
3689 * way down to the bonding driver.
3691 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
3694 * Save the original txq to restore before passing to the driver
3696 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
3698 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
3700 txq
-= dev
->real_num_tx_queues
;
3701 } while (txq
>= dev
->real_num_tx_queues
);
3706 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3708 struct bonding
*bond
= netdev_priv(dev
);
3710 if (bond_should_override_tx_queue(bond
) &&
3711 !bond_slave_override(bond
, skb
))
3712 return NETDEV_TX_OK
;
3714 switch (BOND_MODE(bond
)) {
3715 case BOND_MODE_ROUNDROBIN
:
3716 return bond_xmit_roundrobin(skb
, dev
);
3717 case BOND_MODE_ACTIVEBACKUP
:
3718 return bond_xmit_activebackup(skb
, dev
);
3720 return bond_xmit_xor(skb
, dev
);
3721 case BOND_MODE_BROADCAST
:
3722 return bond_xmit_broadcast(skb
, dev
);
3723 case BOND_MODE_8023AD
:
3724 return bond_3ad_xmit_xor(skb
, dev
);
3726 return bond_alb_xmit(skb
, dev
);
3728 return bond_tlb_xmit(skb
, dev
);
3730 /* Should never happen, mode already checked */
3731 netdev_err(dev
, "Unknown bonding mode %d\n", BOND_MODE(bond
));
3733 dev_kfree_skb_any(skb
);
3734 return NETDEV_TX_OK
;
3738 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3740 struct bonding
*bond
= netdev_priv(dev
);
3741 netdev_tx_t ret
= NETDEV_TX_OK
;
3744 * If we risk deadlock from transmitting this in the
3745 * netpoll path, tell netpoll to queue the frame for later tx
3747 if (unlikely(is_netpoll_tx_blocked(dev
)))
3748 return NETDEV_TX_BUSY
;
3751 if (bond_has_slaves(bond
))
3752 ret
= __bond_start_xmit(skb
, dev
);
3754 dev_kfree_skb_any(skb
);
3760 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
3761 struct ethtool_cmd
*ecmd
)
3763 struct bonding
*bond
= netdev_priv(bond_dev
);
3764 unsigned long speed
= 0;
3765 struct list_head
*iter
;
3766 struct slave
*slave
;
3768 ecmd
->duplex
= DUPLEX_UNKNOWN
;
3769 ecmd
->port
= PORT_OTHER
;
3771 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3772 * do not need to check mode. Though link speed might not represent
3773 * the true receive or transmit bandwidth (not all modes are symmetric)
3774 * this is an accurate maximum.
3776 bond_for_each_slave(bond
, slave
, iter
) {
3777 if (bond_slave_can_tx(slave
)) {
3778 if (slave
->speed
!= SPEED_UNKNOWN
)
3779 speed
+= slave
->speed
;
3780 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
3781 slave
->duplex
!= DUPLEX_UNKNOWN
)
3782 ecmd
->duplex
= slave
->duplex
;
3785 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
3790 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
3791 struct ethtool_drvinfo
*drvinfo
)
3793 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
3794 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
3795 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
3799 static const struct ethtool_ops bond_ethtool_ops
= {
3800 .get_drvinfo
= bond_ethtool_get_drvinfo
,
3801 .get_settings
= bond_ethtool_get_settings
,
3802 .get_link
= ethtool_op_get_link
,
3805 static const struct net_device_ops bond_netdev_ops
= {
3806 .ndo_init
= bond_init
,
3807 .ndo_uninit
= bond_uninit
,
3808 .ndo_open
= bond_open
,
3809 .ndo_stop
= bond_close
,
3810 .ndo_start_xmit
= bond_start_xmit
,
3811 .ndo_select_queue
= bond_select_queue
,
3812 .ndo_get_stats64
= bond_get_stats
,
3813 .ndo_do_ioctl
= bond_do_ioctl
,
3814 .ndo_change_rx_flags
= bond_change_rx_flags
,
3815 .ndo_set_rx_mode
= bond_set_rx_mode
,
3816 .ndo_change_mtu
= bond_change_mtu
,
3817 .ndo_set_mac_address
= bond_set_mac_address
,
3818 .ndo_neigh_setup
= bond_neigh_setup
,
3819 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
3820 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
3821 #ifdef CONFIG_NET_POLL_CONTROLLER
3822 .ndo_netpoll_setup
= bond_netpoll_setup
,
3823 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
3824 .ndo_poll_controller
= bond_poll_controller
,
3826 .ndo_add_slave
= bond_enslave
,
3827 .ndo_del_slave
= bond_release
,
3828 .ndo_fix_features
= bond_fix_features
,
3831 static const struct device_type bond_type
= {
3835 static void bond_destructor(struct net_device
*bond_dev
)
3837 struct bonding
*bond
= netdev_priv(bond_dev
);
3839 destroy_workqueue(bond
->wq
);
3840 free_netdev(bond_dev
);
3843 void bond_setup(struct net_device
*bond_dev
)
3845 struct bonding
*bond
= netdev_priv(bond_dev
);
3847 spin_lock_init(&bond
->mode_lock
);
3848 bond
->params
= bonding_defaults
;
3850 /* Initialize pointers */
3851 bond
->dev
= bond_dev
;
3853 /* Initialize the device entry points */
3854 ether_setup(bond_dev
);
3855 bond_dev
->netdev_ops
= &bond_netdev_ops
;
3856 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
3858 bond_dev
->destructor
= bond_destructor
;
3860 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
3862 /* Initialize the device options */
3863 bond_dev
->tx_queue_len
= 0;
3864 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
3865 bond_dev
->priv_flags
|= IFF_BONDING
| IFF_UNICAST_FLT
;
3866 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
3868 /* don't acquire bond device's netif_tx_lock when
3870 bond_dev
->features
|= NETIF_F_LLTX
;
3872 /* By default, we declare the bond to be fully
3873 * VLAN hardware accelerated capable. Special
3874 * care is taken in the various xmit functions
3875 * when there are slaves that are not hw accel
3879 /* Don't allow bond devices to change network namespaces. */
3880 bond_dev
->features
|= NETIF_F_NETNS_LOCAL
;
3882 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
3883 NETIF_F_HW_VLAN_CTAG_TX
|
3884 NETIF_F_HW_VLAN_CTAG_RX
|
3885 NETIF_F_HW_VLAN_CTAG_FILTER
;
3887 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
3888 bond_dev
->hw_features
|= NETIF_F_GSO_UDP_TUNNEL
;
3889 bond_dev
->features
|= bond_dev
->hw_features
;
3893 * Destroy a bonding device.
3894 * Must be under rtnl_lock when this function is called.
3896 static void bond_uninit(struct net_device
*bond_dev
)
3898 struct bonding
*bond
= netdev_priv(bond_dev
);
3899 struct list_head
*iter
;
3900 struct slave
*slave
;
3902 bond_netpoll_cleanup(bond_dev
);
3904 /* Release the bonded slaves */
3905 bond_for_each_slave(bond
, slave
, iter
)
3906 __bond_release_one(bond_dev
, slave
->dev
, true);
3907 netdev_info(bond_dev
, "Released all slaves\n");
3909 list_del(&bond
->bond_list
);
3911 bond_debug_unregister(bond
);
3914 /*------------------------- Module initialization ---------------------------*/
3916 static int bond_check_params(struct bond_params
*params
)
3918 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
3919 struct bond_opt_value newval
;
3920 const struct bond_opt_value
*valptr
;
3921 int arp_all_targets_value
;
3924 * Convert string parameters.
3927 bond_opt_initstr(&newval
, mode
);
3928 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_MODE
), &newval
);
3930 pr_err("Error: Invalid bonding mode \"%s\"\n", mode
);
3933 bond_mode
= valptr
->value
;
3936 if (xmit_hash_policy
) {
3937 if ((bond_mode
!= BOND_MODE_XOR
) &&
3938 (bond_mode
!= BOND_MODE_8023AD
) &&
3939 (bond_mode
!= BOND_MODE_TLB
)) {
3940 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3941 bond_mode_name(bond_mode
));
3943 bond_opt_initstr(&newval
, xmit_hash_policy
);
3944 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH
),
3947 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
3951 xmit_hashtype
= valptr
->value
;
3956 if (bond_mode
!= BOND_MODE_8023AD
) {
3957 pr_info("lacp_rate param is irrelevant in mode %s\n",
3958 bond_mode_name(bond_mode
));
3960 bond_opt_initstr(&newval
, lacp_rate
);
3961 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE
),
3964 pr_err("Error: Invalid lacp rate \"%s\"\n",
3968 lacp_fast
= valptr
->value
;
3973 bond_opt_initstr(&newval
, ad_select
);
3974 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT
),
3977 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select
);
3980 params
->ad_select
= valptr
->value
;
3981 if (bond_mode
!= BOND_MODE_8023AD
)
3982 pr_warn("ad_select param only affects 802.3ad mode\n");
3984 params
->ad_select
= BOND_AD_STABLE
;
3987 if (max_bonds
< 0) {
3988 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
3989 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
3990 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
3994 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4000 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4005 if (downdelay
< 0) {
4006 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4007 downdelay
, INT_MAX
);
4011 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4012 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4017 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4018 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4023 /* reset values for 802.3ad/TLB/ALB */
4024 if (!bond_mode_uses_arp(bond_mode
)) {
4026 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");
4027 pr_warn("Forcing miimon to 100msec\n");
4028 miimon
= BOND_DEFAULT_MIIMON
;
4032 if (tx_queues
< 1 || tx_queues
> 255) {
4033 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4034 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4035 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4038 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4039 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4041 all_slaves_active
= 0;
4044 if (resend_igmp
< 0 || resend_igmp
> 255) {
4045 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4046 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4047 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4050 bond_opt_initval(&newval
, packets_per_slave
);
4051 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE
), &newval
)) {
4052 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4053 packets_per_slave
, USHRT_MAX
);
4054 packets_per_slave
= 1;
4057 if (bond_mode
== BOND_MODE_ALB
) {
4058 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",
4063 if (updelay
|| downdelay
) {
4064 /* just warn the user the up/down delay will have
4065 * no effect since miimon is zero...
4067 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",
4068 updelay
, downdelay
);
4071 /* don't allow arp monitoring */
4073 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4074 miimon
, arp_interval
);
4078 if ((updelay
% miimon
) != 0) {
4079 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4080 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4085 if ((downdelay
% miimon
) != 0) {
4086 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4088 (downdelay
/ miimon
) * miimon
);
4091 downdelay
/= miimon
;
4094 if (arp_interval
< 0) {
4095 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4096 arp_interval
, INT_MAX
);
4100 for (arp_ip_count
= 0, i
= 0;
4101 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4102 /* not complete check, but should be good enough to
4105 if (!in4_pton(arp_ip_target
[i
], -1, (u8
*)&ip
, -1, NULL
) ||
4106 !bond_is_ip_target_ok(ip
)) {
4107 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4111 if (bond_get_targets_ip(arp_target
, ip
) == -1)
4112 arp_target
[arp_ip_count
++] = ip
;
4114 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4119 if (arp_interval
&& !arp_ip_count
) {
4120 /* don't allow arping if no arp_ip_target given... */
4121 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4127 if (!arp_interval
) {
4128 pr_err("arp_validate requires arp_interval\n");
4132 bond_opt_initstr(&newval
, arp_validate
);
4133 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE
),
4136 pr_err("Error: invalid arp_validate \"%s\"\n",
4140 arp_validate_value
= valptr
->value
;
4142 arp_validate_value
= 0;
4145 arp_all_targets_value
= 0;
4146 if (arp_all_targets
) {
4147 bond_opt_initstr(&newval
, arp_all_targets
);
4148 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS
),
4151 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4153 arp_all_targets_value
= 0;
4155 arp_all_targets_value
= valptr
->value
;
4160 pr_info("MII link monitoring set to %d ms\n", miimon
);
4161 } else if (arp_interval
) {
4162 valptr
= bond_opt_get_val(BOND_OPT_ARP_VALIDATE
,
4163 arp_validate_value
);
4164 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4165 arp_interval
, valptr
->string
, arp_ip_count
);
4167 for (i
= 0; i
< arp_ip_count
; i
++)
4168 pr_cont(" %s", arp_ip_target
[i
]);
4172 } else if (max_bonds
) {
4173 /* miimon and arp_interval not set, we need one so things
4174 * work as expected, see bonding.txt for details
4176 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");
4179 if (primary
&& !bond_mode_uses_primary(bond_mode
)) {
4180 /* currently, using a primary only makes sense
4181 * in active backup, TLB or ALB modes
4183 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4184 primary
, bond_mode_name(bond_mode
));
4188 if (primary
&& primary_reselect
) {
4189 bond_opt_initstr(&newval
, primary_reselect
);
4190 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT
),
4193 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4197 primary_reselect_value
= valptr
->value
;
4199 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4202 if (fail_over_mac
) {
4203 bond_opt_initstr(&newval
, fail_over_mac
);
4204 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC
),
4207 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4211 fail_over_mac_value
= valptr
->value
;
4212 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4213 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4215 fail_over_mac_value
= BOND_FOM_NONE
;
4218 if (lp_interval
== 0) {
4219 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4220 INT_MAX
, BOND_ALB_DEFAULT_LP_INTERVAL
);
4221 lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
4224 /* fill params struct with the proper values */
4225 params
->mode
= bond_mode
;
4226 params
->xmit_policy
= xmit_hashtype
;
4227 params
->miimon
= miimon
;
4228 params
->num_peer_notif
= num_peer_notif
;
4229 params
->arp_interval
= arp_interval
;
4230 params
->arp_validate
= arp_validate_value
;
4231 params
->arp_all_targets
= arp_all_targets_value
;
4232 params
->updelay
= updelay
;
4233 params
->downdelay
= downdelay
;
4234 params
->use_carrier
= use_carrier
;
4235 params
->lacp_fast
= lacp_fast
;
4236 params
->primary
[0] = 0;
4237 params
->primary_reselect
= primary_reselect_value
;
4238 params
->fail_over_mac
= fail_over_mac_value
;
4239 params
->tx_queues
= tx_queues
;
4240 params
->all_slaves_active
= all_slaves_active
;
4241 params
->resend_igmp
= resend_igmp
;
4242 params
->min_links
= min_links
;
4243 params
->lp_interval
= lp_interval
;
4244 params
->packets_per_slave
= packets_per_slave
;
4245 params
->tlb_dynamic_lb
= 1; /* Default value */
4246 if (packets_per_slave
> 0) {
4247 params
->reciprocal_packets_per_slave
=
4248 reciprocal_value(packets_per_slave
);
4250 /* reciprocal_packets_per_slave is unused if
4251 * packets_per_slave is 0 or 1, just initialize it
4253 params
->reciprocal_packets_per_slave
=
4254 (struct reciprocal_value
) { 0 };
4258 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4259 params
->primary
[IFNAMSIZ
- 1] = 0;
4262 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4267 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4268 static struct lock_class_key bonding_netdev_addr_lock_key
;
4269 static struct lock_class_key bonding_tx_busylock_key
;
4271 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4272 struct netdev_queue
*txq
,
4275 lockdep_set_class(&txq
->_xmit_lock
,
4276 &bonding_netdev_xmit_lock_key
);
4279 static void bond_set_lockdep_class(struct net_device
*dev
)
4281 lockdep_set_class(&dev
->addr_list_lock
,
4282 &bonding_netdev_addr_lock_key
);
4283 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4284 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4288 * Called from registration process
4290 static int bond_init(struct net_device
*bond_dev
)
4292 struct bonding
*bond
= netdev_priv(bond_dev
);
4293 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4295 netdev_dbg(bond_dev
, "Begin bond_init\n");
4297 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4301 bond_set_lockdep_class(bond_dev
);
4303 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4305 bond_prepare_sysfs_group(bond
);
4307 bond_debug_register(bond
);
4309 /* Ensure valid dev_addr */
4310 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4311 bond_dev
->addr_assign_type
== NET_ADDR_PERM
)
4312 eth_hw_addr_random(bond_dev
);
4317 unsigned int bond_get_num_tx_queues(void)
4322 /* Create a new bond based on the specified name and bonding parameters.
4323 * If name is NULL, obtain a suitable "bond%d" name for us.
4324 * Caller must NOT hold rtnl_lock; we need to release it here before we
4325 * set up our sysfs entries.
4327 int bond_create(struct net
*net
, const char *name
)
4329 struct net_device
*bond_dev
;
4334 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4335 name
? name
: "bond%d", NET_NAME_UNKNOWN
,
4336 bond_setup
, tx_queues
);
4338 pr_err("%s: eek! can't alloc netdev!\n", name
);
4343 dev_net_set(bond_dev
, net
);
4344 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4346 res
= register_netdevice(bond_dev
);
4348 netif_carrier_off(bond_dev
);
4352 bond_destructor(bond_dev
);
4356 static int __net_init
bond_net_init(struct net
*net
)
4358 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4361 INIT_LIST_HEAD(&bn
->dev_list
);
4363 bond_create_proc_dir(bn
);
4364 bond_create_sysfs(bn
);
4369 static void __net_exit
bond_net_exit(struct net
*net
)
4371 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4372 struct bonding
*bond
, *tmp_bond
;
4375 bond_destroy_sysfs(bn
);
4377 /* Kill off any bonds created after unregistering bond rtnl ops */
4379 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4380 unregister_netdevice_queue(bond
->dev
, &list
);
4381 unregister_netdevice_many(&list
);
4384 bond_destroy_proc_dir(bn
);
4387 static struct pernet_operations bond_net_ops
= {
4388 .init
= bond_net_init
,
4389 .exit
= bond_net_exit
,
4391 .size
= sizeof(struct bond_net
),
4394 static int __init
bonding_init(void)
4399 pr_info("%s", bond_version
);
4401 res
= bond_check_params(&bonding_defaults
);
4405 res
= register_pernet_subsys(&bond_net_ops
);
4409 res
= bond_netlink_init();
4413 bond_create_debugfs();
4415 for (i
= 0; i
< max_bonds
; i
++) {
4416 res
= bond_create(&init_net
, NULL
);
4421 register_netdevice_notifier(&bond_netdev_notifier
);
4425 bond_destroy_debugfs();
4426 bond_netlink_fini();
4428 unregister_pernet_subsys(&bond_net_ops
);
4433 static void __exit
bonding_exit(void)
4435 unregister_netdevice_notifier(&bond_netdev_notifier
);
4437 bond_destroy_debugfs();
4439 bond_netlink_fini();
4440 unregister_pernet_subsys(&bond_net_ops
);
4442 #ifdef CONFIG_NET_POLL_CONTROLLER
4444 * Make sure we don't have an imbalance on our netpoll blocking
4446 WARN_ON(atomic_read(&netpoll_block_tx
));
4450 module_init(bonding_init
);
4451 module_exit(bonding_exit
);
4452 MODULE_LICENSE("GPL");
4453 MODULE_VERSION(DRV_VERSION
);
4454 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4455 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");