2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.h>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
91 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
92 static int num_peer_notif
= 1;
96 static int use_carrier
= 1;
99 static char *primary_reselect
;
100 static char *lacp_rate
;
101 static int min_links
;
102 static char *ad_select
;
103 static char *xmit_hash_policy
;
104 static int arp_interval
;
105 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
106 static char *arp_validate
;
107 static char *arp_all_targets
;
108 static char *fail_over_mac
;
109 static int all_slaves_active
;
110 static struct bond_params bonding_defaults
;
111 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
112 static int packets_per_slave
= 1;
113 static int lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
115 module_param(max_bonds
, int, 0);
116 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
117 module_param(tx_queues
, int, 0);
118 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125 module_param(miimon
, int, 0);
126 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
127 module_param(updelay
, int, 0);
128 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
129 module_param(downdelay
, int, 0);
130 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
132 module_param(use_carrier
, int, 0);
133 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135 module_param(mode
, charp
, 0);
136 MODULE_PARM_DESC(mode
, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140 module_param(primary
, charp
, 0);
141 MODULE_PARM_DESC(primary
, "Primary network device to use");
142 module_param(primary_reselect
, charp
, 0);
143 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
148 "2 for only on active slave "
150 module_param(lacp_rate
, charp
, 0);
151 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153 module_param(ad_select
, charp
, 0);
154 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
157 module_param(min_links
, int, 0);
158 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
160 module_param(xmit_hash_policy
, charp
, 0);
161 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165 module_param(arp_interval
, int, 0);
166 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
167 module_param_array(arp_ip_target
, charp
, NULL
, 0);
168 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
169 module_param(arp_validate
, charp
, 0);
170 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173 module_param(arp_all_targets
, charp
, 0);
174 MODULE_PARM_DESC(arp_all_targets
, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac
, charp
, 0);
176 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179 module_param(all_slaves_active
, int, 0);
180 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183 module_param(resend_igmp
, int, 0);
184 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
186 module_param(packets_per_slave
, int, 0);
187 MODULE_PARM_DESC(packets_per_slave
, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190 module_param(lp_interval
, uint
, 0);
191 MODULE_PARM_DESC(lp_interval
, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
195 /*----------------------------- Global variables ----------------------------*/
197 #ifdef CONFIG_NET_POLL_CONTROLLER
198 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
201 int bond_net_id __read_mostly
;
203 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
204 static int arp_ip_count
;
205 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
206 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
207 static int lacp_fast
;
209 /*-------------------------- Forward declarations ---------------------------*/
211 static int bond_init(struct net_device
*bond_dev
);
212 static void bond_uninit(struct net_device
*bond_dev
);
214 /*---------------------------- General routines -----------------------------*/
216 const char *bond_mode_name(int mode
)
218 static const char *names
[] = {
219 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR
] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
223 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
224 [BOND_MODE_TLB
] = "transmit load balancing",
225 [BOND_MODE_ALB
] = "adaptive load balancing",
228 if (mode
< BOND_MODE_ROUNDROBIN
|| mode
> BOND_MODE_ALB
)
234 /*---------------------------------- VLAN -----------------------------------*/
237 * bond_dev_queue_xmit - Prepare skb for xmit.
239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
243 void bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
244 struct net_device
*slave_dev
)
246 skb
->dev
= slave_dev
;
248 BUILD_BUG_ON(sizeof(skb
->queue_mapping
) !=
249 sizeof(qdisc_skb_cb(skb
)->slave_dev_queue_mapping
));
250 skb
->queue_mapping
= qdisc_skb_cb(skb
)->slave_dev_queue_mapping
;
252 if (unlikely(netpoll_tx_running(bond
->dev
)))
253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
259 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
260 * We don't protect the slave list iteration with a lock because:
261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
280 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
,
281 __be16 proto
, u16 vid
)
283 struct bonding
*bond
= netdev_priv(bond_dev
);
284 struct slave
*slave
, *rollback_slave
;
285 struct list_head
*iter
;
288 bond_for_each_slave(bond
, slave
, iter
) {
289 res
= vlan_vid_add(slave
->dev
, proto
, vid
);
297 /* unwind to the slave that failed */
298 bond_for_each_slave(bond
, rollback_slave
, iter
) {
299 if (rollback_slave
== slave
)
302 vlan_vid_del(rollback_slave
->dev
, proto
, vid
);
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
313 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
,
314 __be16 proto
, u16 vid
)
316 struct bonding
*bond
= netdev_priv(bond_dev
);
317 struct list_head
*iter
;
320 bond_for_each_slave(bond
, slave
, iter
)
321 vlan_vid_del(slave
->dev
, proto
, vid
);
323 if (bond_is_lb(bond
))
324 bond_alb_clear_vlan(bond
, vid
);
329 /*------------------------------- Link status -------------------------------*/
332 * Set the carrier state for the master according to the state of its
333 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
334 * do special 802.3ad magic.
336 * Returns zero if carrier state does not change, nonzero if it does.
338 static int bond_set_carrier(struct bonding
*bond
)
340 struct list_head
*iter
;
343 if (!bond_has_slaves(bond
))
346 if (bond
->params
.mode
== BOND_MODE_8023AD
)
347 return bond_3ad_set_carrier(bond
);
349 bond_for_each_slave(bond
, slave
, iter
) {
350 if (slave
->link
== BOND_LINK_UP
) {
351 if (!netif_carrier_ok(bond
->dev
)) {
352 netif_carrier_on(bond
->dev
);
360 if (netif_carrier_ok(bond
->dev
)) {
361 netif_carrier_off(bond
->dev
);
368 * Get link speed and duplex from the slave's base driver
369 * using ethtool. If for some reason the call fails or the
370 * values are invalid, set speed and duplex to -1,
373 static void bond_update_speed_duplex(struct slave
*slave
)
375 struct net_device
*slave_dev
= slave
->dev
;
376 struct ethtool_cmd ecmd
;
380 slave
->speed
= SPEED_UNKNOWN
;
381 slave
->duplex
= DUPLEX_UNKNOWN
;
383 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
387 slave_speed
= ethtool_cmd_speed(&ecmd
);
388 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
391 switch (ecmd
.duplex
) {
399 slave
->speed
= slave_speed
;
400 slave
->duplex
= ecmd
.duplex
;
405 const char *bond_slave_link_status(s8 link
)
422 * if <dev> supports MII link status reporting, check its link status.
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
425 * depending upon the setting of the use_carrier parameter.
427 * Return either BMSR_LSTATUS, meaning that the link is up (or we
428 * can't tell and just pretend it is), or 0, meaning that the link is
431 * If reporting is non-zero, instead of faking link up, return -1 if
432 * both ETHTOOL and MII ioctls fail (meaning the device does not
433 * support them). If use_carrier is set, return whatever it says.
434 * It'd be nice if there was a good way to tell if a driver supports
435 * netif_carrier, but there really isn't.
437 static int bond_check_dev_link(struct bonding
*bond
,
438 struct net_device
*slave_dev
, int reporting
)
440 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
441 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
443 struct mii_ioctl_data
*mii
;
445 if (!reporting
&& !netif_running(slave_dev
))
448 if (bond
->params
.use_carrier
)
449 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
451 /* Try to get link status using Ethtool first. */
452 if (slave_dev
->ethtool_ops
->get_link
)
453 return slave_dev
->ethtool_ops
->get_link(slave_dev
) ?
456 /* Ethtool can't be used, fallback to MII ioctls. */
457 ioctl
= slave_ops
->ndo_do_ioctl
;
459 /* TODO: set pointer to correct ioctl on a per team member */
460 /* bases to make this more efficient. that is, once */
461 /* we determine the correct ioctl, we will always */
462 /* call it and not the others for that team */
466 * We cannot assume that SIOCGMIIPHY will also read a
467 * register; not all network drivers (e.g., e100)
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
474 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
475 mii
->reg_num
= MII_BMSR
;
476 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
477 return mii
->val_out
& BMSR_LSTATUS
;
482 * If reporting, report that either there's no dev->do_ioctl,
483 * or both SIOCGMIIREG and get_link failed (meaning that we
484 * cannot report link status). If not reporting, pretend
487 return reporting
? -1 : BMSR_LSTATUS
;
490 /*----------------------------- Multicast list ------------------------------*/
493 * Push the promiscuity flag down to appropriate slaves
495 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
497 struct list_head
*iter
;
500 if (USES_PRIMARY(bond
->params
.mode
)) {
501 /* write lock already acquired */
502 if (bond
->curr_active_slave
) {
503 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
509 bond_for_each_slave(bond
, slave
, iter
) {
510 err
= dev_set_promiscuity(slave
->dev
, inc
);
519 * Push the allmulti flag down to all slaves
521 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
523 struct list_head
*iter
;
526 if (USES_PRIMARY(bond
->params
.mode
)) {
527 /* write lock already acquired */
528 if (bond
->curr_active_slave
) {
529 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
535 bond_for_each_slave(bond
, slave
, iter
) {
536 err
= dev_set_allmulti(slave
->dev
, inc
);
545 * Retrieve the list of registered multicast addresses for the bonding
546 * device and retransmit an IGMP JOIN request to the current active
549 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
551 struct bonding
*bond
= container_of(work
, struct bonding
,
554 if (!rtnl_trylock()) {
555 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
558 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, bond
->dev
);
560 if (bond
->igmp_retrans
> 1) {
561 bond
->igmp_retrans
--;
562 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
567 /* Flush bond's hardware addresses from slave
569 static void bond_hw_addr_flush(struct net_device
*bond_dev
,
570 struct net_device
*slave_dev
)
572 struct bonding
*bond
= netdev_priv(bond_dev
);
574 dev_uc_unsync(slave_dev
, bond_dev
);
575 dev_mc_unsync(slave_dev
, bond_dev
);
577 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
578 /* del lacpdu mc addr from mc list */
579 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
581 dev_mc_del(slave_dev
, lacpdu_multicast
);
585 /*--------------------------- Active slave change ---------------------------*/
587 /* Update the hardware address list and promisc/allmulti for the new and
588 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
589 * slaves up date at all times; only the USES_PRIMARY modes need to call
590 * this function to swap these settings during a failover.
592 static void bond_hw_addr_swap(struct bonding
*bond
, struct slave
*new_active
,
593 struct slave
*old_active
)
598 if (bond
->dev
->flags
& IFF_PROMISC
)
599 dev_set_promiscuity(old_active
->dev
, -1);
601 if (bond
->dev
->flags
& IFF_ALLMULTI
)
602 dev_set_allmulti(old_active
->dev
, -1);
604 bond_hw_addr_flush(bond
->dev
, old_active
->dev
);
608 /* FIXME: Signal errors upstream. */
609 if (bond
->dev
->flags
& IFF_PROMISC
)
610 dev_set_promiscuity(new_active
->dev
, 1);
612 if (bond
->dev
->flags
& IFF_ALLMULTI
)
613 dev_set_allmulti(new_active
->dev
, 1);
615 netif_addr_lock_bh(bond
->dev
);
616 dev_uc_sync(new_active
->dev
, bond
->dev
);
617 dev_mc_sync(new_active
->dev
, bond
->dev
);
618 netif_addr_unlock_bh(bond
->dev
);
623 * bond_set_dev_addr - clone slave's address to bond
624 * @bond_dev: bond net device
625 * @slave_dev: slave net device
627 * Should be called with RTNL held.
629 static void bond_set_dev_addr(struct net_device
*bond_dev
,
630 struct net_device
*slave_dev
)
632 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
633 bond_dev
, slave_dev
, slave_dev
->addr_len
);
634 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
635 bond_dev
->addr_assign_type
= NET_ADDR_STOLEN
;
636 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
640 * bond_do_fail_over_mac
642 * Perform special MAC address swapping for fail_over_mac settings
644 * Called with RTNL, curr_slave_lock for write_bh.
646 static void bond_do_fail_over_mac(struct bonding
*bond
,
647 struct slave
*new_active
,
648 struct slave
*old_active
)
649 __releases(&bond
->curr_slave_lock
)
650 __acquires(&bond
->curr_slave_lock
)
652 u8 tmp_mac
[ETH_ALEN
];
653 struct sockaddr saddr
;
656 switch (bond
->params
.fail_over_mac
) {
657 case BOND_FOM_ACTIVE
:
659 write_unlock_bh(&bond
->curr_slave_lock
);
660 bond_set_dev_addr(bond
->dev
, new_active
->dev
);
661 write_lock_bh(&bond
->curr_slave_lock
);
664 case BOND_FOM_FOLLOW
:
666 * if new_active && old_active, swap them
667 * if just old_active, do nothing (going to no active slave)
668 * if just new_active, set new_active to bond's MAC
673 write_unlock_bh(&bond
->curr_slave_lock
);
676 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
677 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
679 saddr
.sa_family
= new_active
->dev
->type
;
681 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
682 saddr
.sa_family
= bond
->dev
->type
;
685 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
687 pr_err("%s: Error %d setting MAC of slave %s\n",
688 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
695 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
696 saddr
.sa_family
= old_active
->dev
->type
;
698 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
700 pr_err("%s: Error %d setting MAC of slave %s\n",
701 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
703 write_lock_bh(&bond
->curr_slave_lock
);
706 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
707 bond
->dev
->name
, bond
->params
.fail_over_mac
);
713 static bool bond_should_change_active(struct bonding
*bond
)
715 struct slave
*prim
= bond
->primary_slave
;
716 struct slave
*curr
= bond
->curr_active_slave
;
718 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
720 if (bond
->force_primary
) {
721 bond
->force_primary
= false;
724 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
725 (prim
->speed
< curr
->speed
||
726 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
728 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
734 * find_best_interface - select the best available slave to be the active one
735 * @bond: our bonding struct
737 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
739 struct slave
*slave
, *bestslave
= NULL
;
740 struct list_head
*iter
;
741 int mintime
= bond
->params
.updelay
;
743 if (bond
->primary_slave
&& bond
->primary_slave
->link
== BOND_LINK_UP
&&
744 bond_should_change_active(bond
))
745 return bond
->primary_slave
;
747 bond_for_each_slave(bond
, slave
, iter
) {
748 if (slave
->link
== BOND_LINK_UP
)
750 if (slave
->link
== BOND_LINK_BACK
&& IS_UP(slave
->dev
) &&
751 slave
->delay
< mintime
) {
752 mintime
= slave
->delay
;
760 static bool bond_should_notify_peers(struct bonding
*bond
)
765 slave
= rcu_dereference(bond
->curr_active_slave
);
768 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
769 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
771 if (!slave
|| !bond
->send_peer_notif
||
772 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
779 * change_active_interface - change the active slave into the specified one
780 * @bond: our bonding struct
781 * @new: the new slave to make the active one
783 * Set the new slave to the bond's settings and unset them on the old
785 * Setting include flags, mc-list, promiscuity, allmulti, etc.
787 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
788 * because it is apparently the best available slave we have, even though its
789 * updelay hasn't timed out yet.
791 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
793 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
795 struct slave
*old_active
= bond
->curr_active_slave
;
797 if (old_active
== new_active
)
801 new_active
->jiffies
= jiffies
;
803 if (new_active
->link
== BOND_LINK_BACK
) {
804 if (USES_PRIMARY(bond
->params
.mode
)) {
805 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
806 bond
->dev
->name
, new_active
->dev
->name
,
807 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
810 new_active
->delay
= 0;
811 new_active
->link
= BOND_LINK_UP
;
813 if (bond
->params
.mode
== BOND_MODE_8023AD
)
814 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
816 if (bond_is_lb(bond
))
817 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
819 if (USES_PRIMARY(bond
->params
.mode
)) {
820 pr_info("%s: making interface %s the new active one.\n",
821 bond
->dev
->name
, new_active
->dev
->name
);
826 if (USES_PRIMARY(bond
->params
.mode
))
827 bond_hw_addr_swap(bond
, new_active
, old_active
);
829 if (bond_is_lb(bond
)) {
830 bond_alb_handle_active_change(bond
, new_active
);
832 bond_set_slave_inactive_flags(old_active
);
834 bond_set_slave_active_flags(new_active
);
836 rcu_assign_pointer(bond
->curr_active_slave
, new_active
);
839 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
841 bond_set_slave_inactive_flags(old_active
);
844 bool should_notify_peers
= false;
846 bond_set_slave_active_flags(new_active
);
848 if (bond
->params
.fail_over_mac
)
849 bond_do_fail_over_mac(bond
, new_active
,
852 if (netif_running(bond
->dev
)) {
853 bond
->send_peer_notif
=
854 bond
->params
.num_peer_notif
;
855 should_notify_peers
=
856 bond_should_notify_peers(bond
);
859 write_unlock_bh(&bond
->curr_slave_lock
);
861 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
862 if (should_notify_peers
)
863 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
866 write_lock_bh(&bond
->curr_slave_lock
);
870 /* resend IGMP joins since active slave has changed or
871 * all were sent on curr_active_slave.
872 * resend only if bond is brought up with the affected
873 * bonding modes and the retransmission is enabled */
874 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
875 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
876 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
877 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
878 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
883 * bond_select_active_slave - select a new active slave, if needed
884 * @bond: our bonding struct
886 * This functions should be called when one of the following occurs:
887 * - The old curr_active_slave has been released or lost its link.
888 * - The primary_slave has got its link back.
889 * - A slave has got its link back and there's no old curr_active_slave.
891 * Caller must hold curr_slave_lock for write_bh.
893 void bond_select_active_slave(struct bonding
*bond
)
895 struct slave
*best_slave
;
898 best_slave
= bond_find_best_slave(bond
);
899 if (best_slave
!= bond
->curr_active_slave
) {
900 bond_change_active_slave(bond
, best_slave
);
901 rv
= bond_set_carrier(bond
);
905 if (netif_carrier_ok(bond
->dev
)) {
906 pr_info("%s: first active interface up!\n",
909 pr_info("%s: now running without any active interface !\n",
915 #ifdef CONFIG_NET_POLL_CONTROLLER
916 static inline int slave_enable_netpoll(struct slave
*slave
)
921 np
= kzalloc(sizeof(*np
), GFP_ATOMIC
);
926 err
= __netpoll_setup(np
, slave
->dev
, GFP_ATOMIC
);
935 static inline void slave_disable_netpoll(struct slave
*slave
)
937 struct netpoll
*np
= slave
->np
;
943 __netpoll_free_async(np
);
945 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
947 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
949 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
954 static void bond_poll_controller(struct net_device
*bond_dev
)
958 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
960 struct bonding
*bond
= netdev_priv(bond_dev
);
961 struct list_head
*iter
;
964 bond_for_each_slave(bond
, slave
, iter
)
965 if (IS_UP(slave
->dev
))
966 slave_disable_netpoll(slave
);
969 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
, gfp_t gfp
)
971 struct bonding
*bond
= netdev_priv(dev
);
972 struct list_head
*iter
;
976 bond_for_each_slave(bond
, slave
, iter
) {
977 err
= slave_enable_netpoll(slave
);
979 bond_netpoll_cleanup(dev
);
986 static inline int slave_enable_netpoll(struct slave
*slave
)
990 static inline void slave_disable_netpoll(struct slave
*slave
)
993 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
998 /*---------------------------------- IOCTL ----------------------------------*/
1000 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1001 netdev_features_t features
)
1003 struct bonding
*bond
= netdev_priv(dev
);
1004 struct list_head
*iter
;
1005 netdev_features_t mask
;
1006 struct slave
*slave
;
1008 if (!bond_has_slaves(bond
)) {
1009 /* Disable adding VLANs to empty bond. But why? --mq */
1010 features
|= NETIF_F_VLAN_CHALLENGED
;
1015 features
&= ~NETIF_F_ONE_FOR_ALL
;
1016 features
|= NETIF_F_ALL_FOR_ALL
;
1018 bond_for_each_slave(bond
, slave
, iter
) {
1019 features
= netdev_increment_features(features
,
1020 slave
->dev
->features
,
1023 features
= netdev_add_tso_features(features
, mask
);
1028 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1029 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1030 NETIF_F_HIGHDMA | NETIF_F_LRO)
1032 static void bond_compute_features(struct bonding
*bond
)
1034 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1035 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1036 struct net_device
*bond_dev
= bond
->dev
;
1037 struct list_head
*iter
;
1038 struct slave
*slave
;
1039 unsigned short max_hard_header_len
= ETH_HLEN
;
1040 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1041 u16 gso_max_segs
= GSO_MAX_SEGS
;
1043 if (!bond_has_slaves(bond
))
1046 bond_for_each_slave(bond
, slave
, iter
) {
1047 vlan_features
= netdev_increment_features(vlan_features
,
1048 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1050 dst_release_flag
&= slave
->dev
->priv_flags
;
1051 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1052 max_hard_header_len
= slave
->dev
->hard_header_len
;
1054 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1055 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1059 bond_dev
->vlan_features
= vlan_features
;
1060 bond_dev
->hard_header_len
= max_hard_header_len
;
1061 bond_dev
->gso_max_segs
= gso_max_segs
;
1062 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1064 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1065 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1067 netdev_change_features(bond_dev
);
1070 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1071 struct net_device
*slave_dev
)
1073 bond_dev
->header_ops
= slave_dev
->header_ops
;
1075 bond_dev
->type
= slave_dev
->type
;
1076 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1077 bond_dev
->addr_len
= slave_dev
->addr_len
;
1079 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1080 slave_dev
->addr_len
);
1083 /* On bonding slaves other than the currently active slave, suppress
1084 * duplicates except for alb non-mcast/bcast.
1086 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1087 struct slave
*slave
,
1088 struct bonding
*bond
)
1090 if (bond_is_slave_inactive(slave
)) {
1091 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1092 skb
->pkt_type
!= PACKET_BROADCAST
&&
1093 skb
->pkt_type
!= PACKET_MULTICAST
)
1100 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1102 struct sk_buff
*skb
= *pskb
;
1103 struct slave
*slave
;
1104 struct bonding
*bond
;
1105 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1107 int ret
= RX_HANDLER_ANOTHER
;
1109 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1111 return RX_HANDLER_CONSUMED
;
1115 slave
= bond_slave_get_rcu(skb
->dev
);
1118 if (bond
->params
.arp_interval
)
1119 slave
->dev
->last_rx
= jiffies
;
1121 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1123 ret
= recv_probe(skb
, bond
, slave
);
1124 if (ret
== RX_HANDLER_CONSUMED
) {
1130 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1131 return RX_HANDLER_EXACT
;
1134 skb
->dev
= bond
->dev
;
1136 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1137 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1138 skb
->pkt_type
== PACKET_HOST
) {
1140 if (unlikely(skb_cow_head(skb
,
1141 skb
->data
- skb_mac_header(skb
)))) {
1143 return RX_HANDLER_CONSUMED
;
1145 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1151 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1152 struct net_device
*slave_dev
,
1153 struct slave
*slave
)
1157 err
= netdev_master_upper_dev_link_private(slave_dev
, bond_dev
, slave
);
1160 slave_dev
->flags
|= IFF_SLAVE
;
1161 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1165 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1166 struct net_device
*slave_dev
)
1168 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1169 slave_dev
->flags
&= ~IFF_SLAVE
;
1170 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1173 /* enslave device <slave> to bond device <master> */
1174 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1176 struct bonding
*bond
= netdev_priv(bond_dev
);
1177 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1178 struct slave
*new_slave
= NULL
, *prev_slave
;
1179 struct sockaddr addr
;
1183 if (!bond
->params
.use_carrier
&&
1184 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1185 slave_ops
->ndo_do_ioctl
== NULL
) {
1186 pr_warning("%s: Warning: no link monitoring support for %s\n",
1187 bond_dev
->name
, slave_dev
->name
);
1190 /* already enslaved */
1191 if (slave_dev
->flags
& IFF_SLAVE
) {
1192 pr_debug("Error, Device was already enslaved\n");
1196 /* vlan challenged mutual exclusion */
1197 /* no need to lock since we're protected by rtnl_lock */
1198 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1199 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1200 if (vlan_uses_dev(bond_dev
)) {
1201 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1202 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1205 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1206 bond_dev
->name
, slave_dev
->name
,
1207 slave_dev
->name
, bond_dev
->name
);
1210 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1214 * Old ifenslave binaries are no longer supported. These can
1215 * be identified with moderate accuracy by the state of the slave:
1216 * the current ifenslave will set the interface down prior to
1217 * enslaving it; the old ifenslave will not.
1219 if ((slave_dev
->flags
& IFF_UP
)) {
1220 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1223 goto err_undo_flags
;
1226 /* set bonding device ether type by slave - bonding netdevices are
1227 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1228 * there is a need to override some of the type dependent attribs/funcs.
1230 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1231 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1233 if (!bond_has_slaves(bond
)) {
1234 if (bond_dev
->type
!= slave_dev
->type
) {
1235 pr_debug("%s: change device type from %d to %d\n",
1237 bond_dev
->type
, slave_dev
->type
);
1239 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1241 res
= notifier_to_errno(res
);
1243 pr_err("%s: refused to change device type\n",
1246 goto err_undo_flags
;
1249 /* Flush unicast and multicast addresses */
1250 dev_uc_flush(bond_dev
);
1251 dev_mc_flush(bond_dev
);
1253 if (slave_dev
->type
!= ARPHRD_ETHER
)
1254 bond_setup_by_slave(bond_dev
, slave_dev
);
1256 ether_setup(bond_dev
);
1257 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1260 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1263 } else if (bond_dev
->type
!= slave_dev
->type
) {
1264 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1266 slave_dev
->type
, bond_dev
->type
);
1268 goto err_undo_flags
;
1271 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1272 if (!bond_has_slaves(bond
)) {
1273 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1275 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1276 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1277 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1280 goto err_undo_flags
;
1284 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1286 /* If this is the first slave, then we need to set the master's hardware
1287 * address to be the same as the slave's. */
1288 if (!bond_has_slaves(bond
) &&
1289 bond
->dev
->addr_assign_type
== NET_ADDR_RANDOM
)
1290 bond_set_dev_addr(bond
->dev
, slave_dev
);
1292 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1295 goto err_undo_flags
;
1298 * Set the new_slave's queue_id to be zero. Queue ID mapping
1299 * is set via sysfs or module option if desired.
1301 new_slave
->queue_id
= 0;
1303 /* Save slave's original mtu and then set it to match the bond */
1304 new_slave
->original_mtu
= slave_dev
->mtu
;
1305 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1307 pr_debug("Error %d calling dev_set_mtu\n", res
);
1312 * Save slave's original ("permanent") mac address for modes
1313 * that need it, and for restoring it upon release, and then
1314 * set it to the master's address
1316 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1318 if (!bond
->params
.fail_over_mac
) {
1320 * Set slave to master's mac address. The application already
1321 * set the master's mac address to that of the first slave
1323 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1324 addr
.sa_family
= slave_dev
->type
;
1325 res
= dev_set_mac_address(slave_dev
, &addr
);
1327 pr_debug("Error %d calling set_mac_address\n", res
);
1328 goto err_restore_mtu
;
1332 /* open the slave since the application closed it */
1333 res
= dev_open(slave_dev
);
1335 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1336 goto err_restore_mac
;
1339 new_slave
->bond
= bond
;
1340 new_slave
->dev
= slave_dev
;
1341 slave_dev
->priv_flags
|= IFF_BONDING
;
1343 if (bond_is_lb(bond
)) {
1344 /* bond_alb_init_slave() must be called before all other stages since
1345 * it might fail and we do not want to have to undo everything
1347 res
= bond_alb_init_slave(bond
, new_slave
);
1352 /* If the mode USES_PRIMARY, then the following is handled by
1353 * bond_change_active_slave().
1355 if (!USES_PRIMARY(bond
->params
.mode
)) {
1356 /* set promiscuity level to new slave */
1357 if (bond_dev
->flags
& IFF_PROMISC
) {
1358 res
= dev_set_promiscuity(slave_dev
, 1);
1363 /* set allmulti level to new slave */
1364 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1365 res
= dev_set_allmulti(slave_dev
, 1);
1370 netif_addr_lock_bh(bond_dev
);
1372 dev_mc_sync_multiple(slave_dev
, bond_dev
);
1373 dev_uc_sync_multiple(slave_dev
, bond_dev
);
1375 netif_addr_unlock_bh(bond_dev
);
1378 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1379 /* add lacpdu mc addr to mc list */
1380 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1382 dev_mc_add(slave_dev
, lacpdu_multicast
);
1385 res
= vlan_vids_add_by_dev(slave_dev
, bond_dev
);
1387 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1388 bond_dev
->name
, slave_dev
->name
);
1392 prev_slave
= bond_last_slave(bond
);
1394 new_slave
->delay
= 0;
1395 new_slave
->link_failure_count
= 0;
1397 bond_update_speed_duplex(new_slave
);
1399 new_slave
->last_arp_rx
= jiffies
-
1400 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1401 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
1402 new_slave
->target_last_arp_rx
[i
] = new_slave
->last_arp_rx
;
1404 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1405 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1407 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1409 * miimon is set but a bonded network driver
1410 * does not support ETHTOOL/MII and
1411 * arp_interval is not set. Note: if
1412 * use_carrier is enabled, we will never go
1413 * here (because netif_carrier is always
1414 * supported); thus, we don't need to change
1415 * the messages for netif_carrier.
1417 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1418 bond_dev
->name
, slave_dev
->name
);
1419 } else if (link_reporting
== -1) {
1420 /* unable get link status using mii/ethtool */
1421 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1422 bond_dev
->name
, slave_dev
->name
);
1426 /* check for initial state */
1427 if (bond
->params
.miimon
) {
1428 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1429 if (bond
->params
.updelay
) {
1430 new_slave
->link
= BOND_LINK_BACK
;
1431 new_slave
->delay
= bond
->params
.updelay
;
1433 new_slave
->link
= BOND_LINK_UP
;
1436 new_slave
->link
= BOND_LINK_DOWN
;
1438 } else if (bond
->params
.arp_interval
) {
1439 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1440 BOND_LINK_UP
: BOND_LINK_DOWN
);
1442 new_slave
->link
= BOND_LINK_UP
;
1445 if (new_slave
->link
!= BOND_LINK_DOWN
)
1446 new_slave
->jiffies
= jiffies
;
1447 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1448 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1449 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1451 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1452 /* if there is a primary slave, remember it */
1453 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1454 bond
->primary_slave
= new_slave
;
1455 bond
->force_primary
= true;
1459 switch (bond
->params
.mode
) {
1460 case BOND_MODE_ACTIVEBACKUP
:
1461 bond_set_slave_inactive_flags(new_slave
);
1463 case BOND_MODE_8023AD
:
1464 /* in 802.3ad mode, the internal mechanism
1465 * will activate the slaves in the selected
1468 bond_set_slave_inactive_flags(new_slave
);
1469 /* if this is the first slave */
1471 SLAVE_AD_INFO(new_slave
).id
= 1;
1472 /* Initialize AD with the number of times that the AD timer is called in 1 second
1473 * can be called only after the mac address of the bond is set
1475 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1477 SLAVE_AD_INFO(new_slave
).id
=
1478 SLAVE_AD_INFO(prev_slave
).id
+ 1;
1481 bond_3ad_bind_slave(new_slave
);
1485 bond_set_active_slave(new_slave
);
1486 bond_set_slave_inactive_flags(new_slave
);
1489 pr_debug("This slave is always active in trunk mode\n");
1491 /* always active in trunk mode */
1492 bond_set_active_slave(new_slave
);
1494 /* In trunking mode there is little meaning to curr_active_slave
1495 * anyway (it holds no special properties of the bond device),
1496 * so we can change it without calling change_active_interface()
1498 if (!bond
->curr_active_slave
&& new_slave
->link
== BOND_LINK_UP
)
1499 rcu_assign_pointer(bond
->curr_active_slave
, new_slave
);
1502 } /* switch(bond_mode) */
1504 #ifdef CONFIG_NET_POLL_CONTROLLER
1505 slave_dev
->npinfo
= bond
->dev
->npinfo
;
1506 if (slave_dev
->npinfo
) {
1507 if (slave_enable_netpoll(new_slave
)) {
1508 read_unlock(&bond
->lock
);
1509 pr_info("Error, %s: master_dev is using netpoll, "
1510 "but new slave device does not support netpoll.\n",
1518 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1521 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1525 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
, new_slave
);
1527 pr_debug("Error %d calling bond_master_upper_dev_link\n", res
);
1528 goto err_unregister
;
1531 res
= bond_sysfs_slave_add(new_slave
);
1533 pr_debug("Error %d calling bond_sysfs_slave_add\n", res
);
1534 goto err_upper_unlink
;
1538 bond_compute_features(bond
);
1539 bond_set_carrier(bond
);
1541 if (USES_PRIMARY(bond
->params
.mode
)) {
1542 write_lock_bh(&bond
->curr_slave_lock
);
1543 bond_select_active_slave(bond
);
1544 write_unlock_bh(&bond
->curr_slave_lock
);
1547 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1548 bond_dev
->name
, slave_dev
->name
,
1549 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1550 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1552 /* enslave is successful */
1555 /* Undo stages on error */
1557 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1560 netdev_rx_handler_unregister(slave_dev
);
1563 if (!USES_PRIMARY(bond
->params
.mode
))
1564 bond_hw_addr_flush(bond_dev
, slave_dev
);
1566 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1567 if (bond
->primary_slave
== new_slave
)
1568 bond
->primary_slave
= NULL
;
1569 if (bond
->curr_active_slave
== new_slave
) {
1570 write_lock_bh(&bond
->curr_slave_lock
);
1571 bond_change_active_slave(bond
, NULL
);
1572 bond_select_active_slave(bond
);
1573 write_unlock_bh(&bond
->curr_slave_lock
);
1575 slave_disable_netpoll(new_slave
);
1578 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1579 dev_close(slave_dev
);
1582 if (!bond
->params
.fail_over_mac
) {
1583 /* XXX TODO - fom follow mode needs to change master's
1584 * MAC if this slave's MAC is in use by the bond, or at
1585 * least print a warning.
1587 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1588 addr
.sa_family
= slave_dev
->type
;
1589 dev_set_mac_address(slave_dev
, &addr
);
1593 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1599 /* Enslave of first slave has failed and we need to fix master's mac */
1600 if (!bond_has_slaves(bond
) &&
1601 ether_addr_equal_64bits(bond_dev
->dev_addr
, slave_dev
->dev_addr
))
1602 eth_hw_addr_random(bond_dev
);
1608 * Try to release the slave device <slave> from the bond device <master>
1609 * It is legal to access curr_active_slave without a lock because all the function
1610 * is write-locked. If "all" is true it means that the function is being called
1611 * while destroying a bond interface and all slaves are being released.
1613 * The rules for slave state should be:
1614 * for Active/Backup:
1615 * Active stays on all backups go down
1616 * for Bonded connections:
1617 * The first up interface should be left on and all others downed.
1619 static int __bond_release_one(struct net_device
*bond_dev
,
1620 struct net_device
*slave_dev
,
1623 struct bonding
*bond
= netdev_priv(bond_dev
);
1624 struct slave
*slave
, *oldcurrent
;
1625 struct sockaddr addr
;
1626 int old_flags
= bond_dev
->flags
;
1627 netdev_features_t old_features
= bond_dev
->features
;
1629 /* slave is not a slave or master is not master of this slave */
1630 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1631 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1632 pr_err("%s: Error: cannot release %s.\n",
1633 bond_dev
->name
, slave_dev
->name
);
1639 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1641 /* not a slave of this bond */
1642 pr_info("%s: %s not enslaved\n",
1643 bond_dev
->name
, slave_dev
->name
);
1644 unblock_netpoll_tx();
1648 /* release the slave from its bond */
1651 bond_sysfs_slave_del(slave
);
1653 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1654 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1655 * for this slave anymore.
1657 netdev_rx_handler_unregister(slave_dev
);
1658 write_lock_bh(&bond
->lock
);
1660 /* Inform AD package of unbinding of slave. */
1661 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1662 bond_3ad_unbind_slave(slave
);
1664 write_unlock_bh(&bond
->lock
);
1666 pr_info("%s: releasing %s interface %s\n",
1668 bond_is_active_slave(slave
) ? "active" : "backup",
1671 oldcurrent
= bond
->curr_active_slave
;
1673 bond
->current_arp_slave
= NULL
;
1675 if (!all
&& !bond
->params
.fail_over_mac
) {
1676 if (ether_addr_equal_64bits(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1677 bond_has_slaves(bond
))
1678 pr_warn("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1679 bond_dev
->name
, slave_dev
->name
,
1681 bond_dev
->name
, slave_dev
->name
);
1684 if (bond
->primary_slave
== slave
)
1685 bond
->primary_slave
= NULL
;
1687 if (oldcurrent
== slave
) {
1688 write_lock_bh(&bond
->curr_slave_lock
);
1689 bond_change_active_slave(bond
, NULL
);
1690 write_unlock_bh(&bond
->curr_slave_lock
);
1693 if (bond_is_lb(bond
)) {
1694 /* Must be called only after the slave has been
1695 * detached from the list and the curr_active_slave
1696 * has been cleared (if our_slave == old_current),
1697 * but before a new active slave is selected.
1699 bond_alb_deinit_slave(bond
, slave
);
1703 RCU_INIT_POINTER(bond
->curr_active_slave
, NULL
);
1704 } else if (oldcurrent
== slave
) {
1706 * Note that we hold RTNL over this sequence, so there
1707 * is no concern that another slave add/remove event
1710 write_lock_bh(&bond
->curr_slave_lock
);
1712 bond_select_active_slave(bond
);
1714 write_unlock_bh(&bond
->curr_slave_lock
);
1717 if (!bond_has_slaves(bond
)) {
1718 bond_set_carrier(bond
);
1719 eth_hw_addr_random(bond_dev
);
1721 if (vlan_uses_dev(bond_dev
)) {
1722 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1723 bond_dev
->name
, bond_dev
->name
);
1724 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1729 unblock_netpoll_tx();
1732 if (!bond_has_slaves(bond
)) {
1733 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
1734 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
1737 bond_compute_features(bond
);
1738 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1739 (old_features
& NETIF_F_VLAN_CHALLENGED
))
1740 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1741 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1743 /* must do this from outside any spinlocks */
1744 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1746 /* If the mode USES_PRIMARY, then this cases was handled above by
1747 * bond_change_active_slave(..., NULL)
1749 if (!USES_PRIMARY(bond
->params
.mode
)) {
1750 /* unset promiscuity level from slave
1751 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1752 * of the IFF_PROMISC flag in the bond_dev, but we need the
1753 * value of that flag before that change, as that was the value
1754 * when this slave was attached, so we cache at the start of the
1755 * function and use it here. Same goes for ALLMULTI below
1757 if (old_flags
& IFF_PROMISC
)
1758 dev_set_promiscuity(slave_dev
, -1);
1760 /* unset allmulti level from slave */
1761 if (old_flags
& IFF_ALLMULTI
)
1762 dev_set_allmulti(slave_dev
, -1);
1764 bond_hw_addr_flush(bond_dev
, slave_dev
);
1767 slave_disable_netpoll(slave
);
1769 /* close slave before restoring its mac address */
1770 dev_close(slave_dev
);
1772 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1773 /* restore original ("permanent") mac address */
1774 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1775 addr
.sa_family
= slave_dev
->type
;
1776 dev_set_mac_address(slave_dev
, &addr
);
1779 dev_set_mtu(slave_dev
, slave
->original_mtu
);
1781 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1785 return 0; /* deletion OK */
1788 /* A wrapper used because of ndo_del_link */
1789 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1791 return __bond_release_one(bond_dev
, slave_dev
, false);
1795 * First release a slave and then destroy the bond if no more slaves are left.
1796 * Must be under rtnl_lock when this function is called.
1798 static int bond_release_and_destroy(struct net_device
*bond_dev
,
1799 struct net_device
*slave_dev
)
1801 struct bonding
*bond
= netdev_priv(bond_dev
);
1804 ret
= bond_release(bond_dev
, slave_dev
);
1805 if (ret
== 0 && !bond_has_slaves(bond
)) {
1806 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1807 pr_info("%s: destroying bond %s.\n",
1808 bond_dev
->name
, bond_dev
->name
);
1809 unregister_netdevice(bond_dev
);
1814 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1816 struct bonding
*bond
= netdev_priv(bond_dev
);
1818 info
->bond_mode
= bond
->params
.mode
;
1819 info
->miimon
= bond
->params
.miimon
;
1821 read_lock(&bond
->lock
);
1822 info
->num_slaves
= bond
->slave_cnt
;
1823 read_unlock(&bond
->lock
);
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 read_lock(&bond
->lock
);
1836 bond_for_each_slave(bond
, slave
, iter
) {
1837 if (i
++ == (int)info
->slave_id
) {
1839 strcpy(info
->slave_name
, slave
->dev
->name
);
1840 info
->link
= slave
->link
;
1841 info
->state
= bond_slave_state(slave
);
1842 info
->link_failure_count
= slave
->link_failure_count
;
1846 read_unlock(&bond
->lock
);
1851 /*-------------------------------- Monitoring -------------------------------*/
1854 static int bond_miimon_inspect(struct bonding
*bond
)
1856 int link_state
, commit
= 0;
1857 struct list_head
*iter
;
1858 struct slave
*slave
;
1859 bool ignore_updelay
;
1861 ignore_updelay
= !bond
->curr_active_slave
? true : false;
1863 bond_for_each_slave_rcu(bond
, slave
, iter
) {
1864 slave
->new_link
= BOND_LINK_NOCHANGE
;
1866 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
1868 switch (slave
->link
) {
1873 slave
->link
= BOND_LINK_FAIL
;
1874 slave
->delay
= bond
->params
.downdelay
;
1876 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1878 (bond
->params
.mode
==
1879 BOND_MODE_ACTIVEBACKUP
) ?
1880 (bond_is_active_slave(slave
) ?
1881 "active " : "backup ") : "",
1883 bond
->params
.downdelay
* bond
->params
.miimon
);
1886 case BOND_LINK_FAIL
:
1889 * recovered before downdelay expired
1891 slave
->link
= BOND_LINK_UP
;
1892 slave
->jiffies
= jiffies
;
1893 pr_info("%s: link status up again after %d ms for interface %s.\n",
1895 (bond
->params
.downdelay
- slave
->delay
) *
1896 bond
->params
.miimon
,
1901 if (slave
->delay
<= 0) {
1902 slave
->new_link
= BOND_LINK_DOWN
;
1910 case BOND_LINK_DOWN
:
1914 slave
->link
= BOND_LINK_BACK
;
1915 slave
->delay
= bond
->params
.updelay
;
1918 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1919 bond
->dev
->name
, slave
->dev
->name
,
1920 ignore_updelay
? 0 :
1921 bond
->params
.updelay
*
1922 bond
->params
.miimon
);
1925 case BOND_LINK_BACK
:
1927 slave
->link
= BOND_LINK_DOWN
;
1928 pr_info("%s: link status down again after %d ms for interface %s.\n",
1930 (bond
->params
.updelay
- slave
->delay
) *
1931 bond
->params
.miimon
,
1940 if (slave
->delay
<= 0) {
1941 slave
->new_link
= BOND_LINK_UP
;
1943 ignore_updelay
= false;
1955 static void bond_miimon_commit(struct bonding
*bond
)
1957 struct list_head
*iter
;
1958 struct slave
*slave
;
1960 bond_for_each_slave(bond
, slave
, iter
) {
1961 switch (slave
->new_link
) {
1962 case BOND_LINK_NOCHANGE
:
1966 slave
->link
= BOND_LINK_UP
;
1967 slave
->jiffies
= jiffies
;
1969 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1970 /* prevent it from being the active one */
1971 bond_set_backup_slave(slave
);
1972 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
1973 /* make it immediately active */
1974 bond_set_active_slave(slave
);
1975 } else if (slave
!= bond
->primary_slave
) {
1976 /* prevent it from being the active one */
1977 bond_set_backup_slave(slave
);
1980 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
1981 bond
->dev
->name
, slave
->dev
->name
,
1982 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
1983 slave
->duplex
? "full" : "half");
1985 /* notify ad that the link status has changed */
1986 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1987 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
1989 if (bond_is_lb(bond
))
1990 bond_alb_handle_link_change(bond
, slave
,
1993 if (!bond
->curr_active_slave
||
1994 (slave
== bond
->primary_slave
))
1999 case BOND_LINK_DOWN
:
2000 if (slave
->link_failure_count
< UINT_MAX
)
2001 slave
->link_failure_count
++;
2003 slave
->link
= BOND_LINK_DOWN
;
2005 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2006 bond
->params
.mode
== BOND_MODE_8023AD
)
2007 bond_set_slave_inactive_flags(slave
);
2009 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2010 bond
->dev
->name
, slave
->dev
->name
);
2012 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2013 bond_3ad_handle_link_change(slave
,
2016 if (bond_is_lb(bond
))
2017 bond_alb_handle_link_change(bond
, slave
,
2020 if (slave
== bond
->curr_active_slave
)
2026 pr_err("%s: invalid new link %d on slave %s\n",
2027 bond
->dev
->name
, slave
->new_link
,
2029 slave
->new_link
= BOND_LINK_NOCHANGE
;
2037 write_lock_bh(&bond
->curr_slave_lock
);
2038 bond_select_active_slave(bond
);
2039 write_unlock_bh(&bond
->curr_slave_lock
);
2040 unblock_netpoll_tx();
2043 bond_set_carrier(bond
);
2049 * Really a wrapper that splits the mii monitor into two phases: an
2050 * inspection, then (if inspection indicates something needs to be done)
2051 * an acquisition of appropriate locks followed by a commit phase to
2052 * implement whatever link state changes are indicated.
2054 static void bond_mii_monitor(struct work_struct
*work
)
2056 struct bonding
*bond
= container_of(work
, struct bonding
,
2058 bool should_notify_peers
= false;
2059 unsigned long delay
;
2061 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2063 if (!bond_has_slaves(bond
))
2068 should_notify_peers
= bond_should_notify_peers(bond
);
2070 if (bond_miimon_inspect(bond
)) {
2073 /* Race avoidance with bond_close cancel of workqueue */
2074 if (!rtnl_trylock()) {
2076 should_notify_peers
= false;
2080 bond_miimon_commit(bond
);
2082 rtnl_unlock(); /* might sleep, hold no other locks */
2087 if (bond
->params
.miimon
)
2088 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2090 if (should_notify_peers
) {
2091 if (!rtnl_trylock())
2093 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2098 static bool bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2100 struct net_device
*upper
;
2101 struct list_head
*iter
;
2104 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2108 netdev_for_each_all_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2109 if (ip
== bond_confirm_addr(upper
, 0, ip
)) {
2120 * We go to the (large) trouble of VLAN tagging ARP frames because
2121 * switches in VLAN mode (especially if ports are configured as
2122 * "native" to a VLAN) might not pass non-tagged frames.
2124 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2126 struct sk_buff
*skb
;
2128 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op
,
2129 slave_dev
->name
, &dest_ip
, &src_ip
, vlan_id
);
2131 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2132 NULL
, slave_dev
->dev_addr
, NULL
);
2135 pr_err("ARP packet allocation failed\n");
2139 skb
= vlan_put_tag(skb
, htons(ETH_P_8021Q
), vlan_id
);
2141 pr_err("failed to insert VLAN tag\n");
2149 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2151 struct net_device
*upper
, *vlan_upper
;
2152 struct list_head
*iter
, *vlan_iter
;
2154 __be32
*targets
= bond
->params
.arp_targets
, addr
;
2157 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
&& targets
[i
]; i
++) {
2158 pr_debug("basa: target %pI4\n", &targets
[i
]);
2160 /* Find out through which dev should the packet go */
2161 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2164 pr_debug("%s: no route to arp_ip_target %pI4\n",
2165 bond
->dev
->name
, &targets
[i
]);
2171 /* bond device itself */
2172 if (rt
->dst
.dev
== bond
->dev
)
2176 /* first we search only for vlan devices. for every vlan
2177 * found we verify its upper dev list, searching for the
2178 * rt->dst.dev. If found we save the tag of the vlan and
2179 * proceed to send the packet.
2183 netdev_for_each_all_upper_dev_rcu(bond
->dev
, vlan_upper
,
2185 if (!is_vlan_dev(vlan_upper
))
2187 netdev_for_each_all_upper_dev_rcu(vlan_upper
, upper
,
2189 if (upper
== rt
->dst
.dev
) {
2190 vlan_id
= vlan_dev_vlan_id(vlan_upper
);
2197 /* if the device we're looking for is not on top of any of
2198 * our upper vlans, then just search for any dev that
2199 * matches, and in case it's a vlan - save the id
2201 netdev_for_each_all_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2202 if (upper
== rt
->dst
.dev
) {
2203 /* if it's a vlan - get its VID */
2204 if (is_vlan_dev(upper
))
2205 vlan_id
= vlan_dev_vlan_id(upper
);
2213 /* Not our device - skip */
2214 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2215 bond
->dev
->name
, &targets
[i
],
2216 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2222 addr
= bond_confirm_addr(rt
->dst
.dev
, targets
[i
], 0);
2224 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2229 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2233 if (!sip
|| !bond_has_this_ip(bond
, tip
)) {
2234 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip
, &tip
);
2238 i
= bond_get_targets_ip(bond
->params
.arp_targets
, sip
);
2240 pr_debug("bva: sip %pI4 not found in targets\n", &sip
);
2243 slave
->last_arp_rx
= jiffies
;
2244 slave
->target_last_arp_rx
[i
] = jiffies
;
2247 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2248 struct slave
*slave
)
2250 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2251 unsigned char *arp_ptr
;
2255 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2256 return RX_HANDLER_ANOTHER
;
2258 read_lock(&bond
->lock
);
2260 if (!slave_do_arp_validate(bond
, slave
))
2263 alen
= arp_hdr_len(bond
->dev
);
2265 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2266 bond
->dev
->name
, skb
->dev
->name
);
2268 if (alen
> skb_headlen(skb
)) {
2269 arp
= kmalloc(alen
, GFP_ATOMIC
);
2272 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2276 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2277 skb
->pkt_type
== PACKET_OTHERHOST
||
2278 skb
->pkt_type
== PACKET_LOOPBACK
||
2279 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2280 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2284 arp_ptr
= (unsigned char *)(arp
+ 1);
2285 arp_ptr
+= bond
->dev
->addr_len
;
2286 memcpy(&sip
, arp_ptr
, 4);
2287 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2288 memcpy(&tip
, arp_ptr
, 4);
2290 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2291 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2292 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2296 * Backup slaves won't see the ARP reply, but do come through
2297 * here for each ARP probe (so we swap the sip/tip to validate
2298 * the probe). In a "redundant switch, common router" type of
2299 * configuration, the ARP probe will (hopefully) travel from
2300 * the active, through one switch, the router, then the other
2301 * switch before reaching the backup.
2303 * We 'trust' the arp requests if there is an active slave and
2304 * it received valid arp reply(s) after it became active. This
2305 * is done to avoid endless looping when we can't reach the
2306 * arp_ip_target and fool ourselves with our own arp requests.
2308 if (bond_is_active_slave(slave
))
2309 bond_validate_arp(bond
, slave
, sip
, tip
);
2310 else if (bond
->curr_active_slave
&&
2311 time_after(slave_last_rx(bond
, bond
->curr_active_slave
),
2312 bond
->curr_active_slave
->jiffies
))
2313 bond_validate_arp(bond
, slave
, tip
, sip
);
2316 read_unlock(&bond
->lock
);
2317 if (arp
!= (struct arphdr
*)skb
->data
)
2319 return RX_HANDLER_ANOTHER
;
2322 /* function to verify if we're in the arp_interval timeslice, returns true if
2323 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2324 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2326 static bool bond_time_in_interval(struct bonding
*bond
, unsigned long last_act
,
2329 int delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2331 return time_in_range(jiffies
,
2332 last_act
- delta_in_ticks
,
2333 last_act
+ mod
* delta_in_ticks
+ delta_in_ticks
/2);
2337 * this function is called regularly to monitor each slave's link
2338 * ensuring that traffic is being sent and received when arp monitoring
2339 * is used in load-balancing mode. if the adapter has been dormant, then an
2340 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2341 * arp monitoring in active backup mode.
2343 static void bond_loadbalance_arp_mon(struct work_struct
*work
)
2345 struct bonding
*bond
= container_of(work
, struct bonding
,
2347 struct slave
*slave
, *oldcurrent
;
2348 struct list_head
*iter
;
2349 int do_failover
= 0, slave_state_changed
= 0;
2351 if (!bond_has_slaves(bond
))
2356 oldcurrent
= ACCESS_ONCE(bond
->curr_active_slave
);
2357 /* see if any of the previous devices are up now (i.e. they have
2358 * xmt and rcv traffic). the curr_active_slave does not come into
2359 * the picture unless it is null. also, slave->jiffies is not needed
2360 * here because we send an arp on each slave and give a slave as
2361 * long as it needs to get the tx/rx within the delta.
2362 * TODO: what about up/down delay in arp mode? it wasn't here before
2365 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2366 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2368 if (slave
->link
!= BOND_LINK_UP
) {
2369 if (bond_time_in_interval(bond
, trans_start
, 1) &&
2370 bond_time_in_interval(bond
, slave
->dev
->last_rx
, 1)) {
2372 slave
->link
= BOND_LINK_UP
;
2373 slave_state_changed
= 1;
2375 /* primary_slave has no meaning in round-robin
2376 * mode. the window of a slave being up and
2377 * curr_active_slave being null after enslaving
2381 pr_info("%s: link status definitely up for interface %s, ",
2386 pr_info("%s: interface %s is now up\n",
2392 /* slave->link == BOND_LINK_UP */
2394 /* not all switches will respond to an arp request
2395 * when the source ip is 0, so don't take the link down
2396 * if we don't know our ip yet
2398 if (!bond_time_in_interval(bond
, trans_start
, 2) ||
2399 !bond_time_in_interval(bond
, slave
->dev
->last_rx
, 2)) {
2401 slave
->link
= BOND_LINK_DOWN
;
2402 slave_state_changed
= 1;
2404 if (slave
->link_failure_count
< UINT_MAX
)
2405 slave
->link_failure_count
++;
2407 pr_info("%s: interface %s is now down.\n",
2411 if (slave
== oldcurrent
)
2416 /* note: if switch is in round-robin mode, all links
2417 * must tx arp to ensure all links rx an arp - otherwise
2418 * links may oscillate or not come up at all; if switch is
2419 * in something like xor mode, there is nothing we can
2420 * do - all replies will be rx'ed on same link causing slaves
2421 * to be unstable during low/no traffic periods
2423 if (IS_UP(slave
->dev
))
2424 bond_arp_send_all(bond
, slave
);
2429 if (do_failover
|| slave_state_changed
) {
2430 if (!rtnl_trylock())
2433 if (slave_state_changed
) {
2434 bond_slave_state_change(bond
);
2435 } else if (do_failover
) {
2436 /* the bond_select_active_slave must hold RTNL
2437 * and curr_slave_lock for write.
2440 write_lock_bh(&bond
->curr_slave_lock
);
2442 bond_select_active_slave(bond
);
2444 write_unlock_bh(&bond
->curr_slave_lock
);
2445 unblock_netpoll_tx();
2451 if (bond
->params
.arp_interval
)
2452 queue_delayed_work(bond
->wq
, &bond
->arp_work
,
2453 msecs_to_jiffies(bond
->params
.arp_interval
));
2457 * Called to inspect slaves for active-backup mode ARP monitor link state
2458 * changes. Sets new_link in slaves to specify what action should take
2459 * place for the slave. Returns 0 if no changes are found, >0 if changes
2460 * to link states must be committed.
2462 * Called with rcu_read_lock hold.
2464 static int bond_ab_arp_inspect(struct bonding
*bond
)
2466 unsigned long trans_start
, last_rx
;
2467 struct list_head
*iter
;
2468 struct slave
*slave
;
2471 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2472 slave
->new_link
= BOND_LINK_NOCHANGE
;
2473 last_rx
= slave_last_rx(bond
, slave
);
2475 if (slave
->link
!= BOND_LINK_UP
) {
2476 if (bond_time_in_interval(bond
, last_rx
, 1)) {
2477 slave
->new_link
= BOND_LINK_UP
;
2484 * Give slaves 2*delta after being enslaved or made
2485 * active. This avoids bouncing, as the last receive
2486 * times need a full ARP monitor cycle to be updated.
2488 if (bond_time_in_interval(bond
, slave
->jiffies
, 2))
2492 * Backup slave is down if:
2493 * - No current_arp_slave AND
2494 * - more than 3*delta since last receive AND
2495 * - the bond has an IP address
2497 * Note: a non-null current_arp_slave indicates
2498 * the curr_active_slave went down and we are
2499 * searching for a new one; under this condition
2500 * we only take the curr_active_slave down - this
2501 * gives each slave a chance to tx/rx traffic
2502 * before being taken out
2504 if (!bond_is_active_slave(slave
) &&
2505 !bond
->current_arp_slave
&&
2506 !bond_time_in_interval(bond
, last_rx
, 3)) {
2507 slave
->new_link
= BOND_LINK_DOWN
;
2512 * Active slave is down if:
2513 * - more than 2*delta since transmitting OR
2514 * - (more than 2*delta since receive AND
2515 * the bond has an IP address)
2517 trans_start
= dev_trans_start(slave
->dev
);
2518 if (bond_is_active_slave(slave
) &&
2519 (!bond_time_in_interval(bond
, trans_start
, 2) ||
2520 !bond_time_in_interval(bond
, last_rx
, 2))) {
2521 slave
->new_link
= BOND_LINK_DOWN
;
2530 * Called to commit link state changes noted by inspection step of
2531 * active-backup mode ARP monitor.
2533 * Called with RTNL hold.
2535 static void bond_ab_arp_commit(struct bonding
*bond
)
2537 unsigned long trans_start
;
2538 struct list_head
*iter
;
2539 struct slave
*slave
;
2541 bond_for_each_slave(bond
, slave
, iter
) {
2542 switch (slave
->new_link
) {
2543 case BOND_LINK_NOCHANGE
:
2547 trans_start
= dev_trans_start(slave
->dev
);
2548 if (bond
->curr_active_slave
!= slave
||
2549 (!bond
->curr_active_slave
&&
2550 bond_time_in_interval(bond
, trans_start
, 1))) {
2551 slave
->link
= BOND_LINK_UP
;
2552 if (bond
->current_arp_slave
) {
2553 bond_set_slave_inactive_flags(
2554 bond
->current_arp_slave
);
2555 bond
->current_arp_slave
= NULL
;
2558 pr_info("%s: link status definitely up for interface %s.\n",
2559 bond
->dev
->name
, slave
->dev
->name
);
2561 if (!bond
->curr_active_slave
||
2562 (slave
== bond
->primary_slave
))
2569 case BOND_LINK_DOWN
:
2570 if (slave
->link_failure_count
< UINT_MAX
)
2571 slave
->link_failure_count
++;
2573 slave
->link
= BOND_LINK_DOWN
;
2574 bond_set_slave_inactive_flags(slave
);
2576 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2577 bond
->dev
->name
, slave
->dev
->name
);
2579 if (slave
== bond
->curr_active_slave
) {
2580 bond
->current_arp_slave
= NULL
;
2587 pr_err("%s: impossible: new_link %d on slave %s\n",
2588 bond
->dev
->name
, slave
->new_link
,
2596 write_lock_bh(&bond
->curr_slave_lock
);
2597 bond_select_active_slave(bond
);
2598 write_unlock_bh(&bond
->curr_slave_lock
);
2599 unblock_netpoll_tx();
2602 bond_set_carrier(bond
);
2606 * Send ARP probes for active-backup mode ARP monitor.
2608 static bool bond_ab_arp_probe(struct bonding
*bond
)
2610 struct slave
*slave
, *before
= NULL
, *new_slave
= NULL
,
2611 *curr_arp_slave
, *curr_active_slave
;
2612 struct list_head
*iter
;
2616 curr_arp_slave
= rcu_dereference(bond
->current_arp_slave
);
2617 curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2619 if (curr_arp_slave
&& curr_active_slave
)
2620 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2621 curr_arp_slave
->dev
->name
,
2622 curr_active_slave
->dev
->name
);
2624 if (curr_active_slave
) {
2625 bond_arp_send_all(bond
, curr_active_slave
);
2631 /* if we don't have a curr_active_slave, search for the next available
2632 * backup slave from the current_arp_slave and make it the candidate
2633 * for becoming the curr_active_slave
2636 if (!rtnl_trylock())
2638 /* curr_arp_slave might have gone away */
2639 curr_arp_slave
= ACCESS_ONCE(bond
->current_arp_slave
);
2641 if (!curr_arp_slave
) {
2642 curr_arp_slave
= bond_first_slave(bond
);
2643 if (!curr_arp_slave
) {
2649 bond_set_slave_inactive_flags(curr_arp_slave
);
2651 bond_for_each_slave(bond
, slave
, iter
) {
2652 if (!found
&& !before
&& IS_UP(slave
->dev
))
2655 if (found
&& !new_slave
&& IS_UP(slave
->dev
))
2657 /* if the link state is up at this point, we
2658 * mark it down - this can happen if we have
2659 * simultaneous link failures and
2660 * reselect_active_interface doesn't make this
2661 * one the current slave so it is still marked
2662 * up when it is actually down
2664 if (!IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
) {
2665 slave
->link
= BOND_LINK_DOWN
;
2666 if (slave
->link_failure_count
< UINT_MAX
)
2667 slave
->link_failure_count
++;
2669 bond_set_slave_inactive_flags(slave
);
2671 pr_info("%s: backup interface %s is now down.\n",
2672 bond
->dev
->name
, slave
->dev
->name
);
2674 if (slave
== curr_arp_slave
)
2678 if (!new_slave
&& before
)
2686 new_slave
->link
= BOND_LINK_BACK
;
2687 bond_set_slave_active_flags(new_slave
);
2688 bond_arp_send_all(bond
, new_slave
);
2689 new_slave
->jiffies
= jiffies
;
2690 rcu_assign_pointer(bond
->current_arp_slave
, new_slave
);
2696 static void bond_activebackup_arp_mon(struct work_struct
*work
)
2698 struct bonding
*bond
= container_of(work
, struct bonding
,
2700 bool should_notify_peers
= false, should_commit
= false;
2703 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2705 if (!bond_has_slaves(bond
))
2709 should_notify_peers
= bond_should_notify_peers(bond
);
2710 should_commit
= bond_ab_arp_inspect(bond
);
2713 if (should_commit
) {
2714 /* Race avoidance with bond_close flush of workqueue */
2715 if (!rtnl_trylock()) {
2717 should_notify_peers
= false;
2721 bond_ab_arp_commit(bond
);
2725 if (!bond_ab_arp_probe(bond
)) {
2726 /* rtnl locking failed, re-arm */
2728 should_notify_peers
= false;
2732 if (bond
->params
.arp_interval
)
2733 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2735 if (should_notify_peers
) {
2736 if (!rtnl_trylock())
2738 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2743 /*-------------------------- netdev event handling --------------------------*/
2746 * Change device name
2748 static int bond_event_changename(struct bonding
*bond
)
2750 bond_remove_proc_entry(bond
);
2751 bond_create_proc_entry(bond
);
2753 bond_debug_reregister(bond
);
2758 static int bond_master_netdev_event(unsigned long event
,
2759 struct net_device
*bond_dev
)
2761 struct bonding
*event_bond
= netdev_priv(bond_dev
);
2764 case NETDEV_CHANGENAME
:
2765 return bond_event_changename(event_bond
);
2766 case NETDEV_UNREGISTER
:
2767 bond_remove_proc_entry(event_bond
);
2769 case NETDEV_REGISTER
:
2770 bond_create_proc_entry(event_bond
);
2772 case NETDEV_NOTIFY_PEERS
:
2773 if (event_bond
->send_peer_notif
)
2774 event_bond
->send_peer_notif
--;
2783 static int bond_slave_netdev_event(unsigned long event
,
2784 struct net_device
*slave_dev
)
2786 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
);
2787 struct bonding
*bond
;
2788 struct net_device
*bond_dev
;
2792 /* A netdev event can be generated while enslaving a device
2793 * before netdev_rx_handler_register is called in which case
2794 * slave will be NULL
2798 bond_dev
= slave
->bond
->dev
;
2802 case NETDEV_UNREGISTER
:
2803 if (bond_dev
->type
!= ARPHRD_ETHER
)
2804 bond_release_and_destroy(bond_dev
, slave_dev
);
2806 bond_release(bond_dev
, slave_dev
);
2810 old_speed
= slave
->speed
;
2811 old_duplex
= slave
->duplex
;
2813 bond_update_speed_duplex(slave
);
2815 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2816 if (old_speed
!= slave
->speed
)
2817 bond_3ad_adapter_speed_changed(slave
);
2818 if (old_duplex
!= slave
->duplex
)
2819 bond_3ad_adapter_duplex_changed(slave
);
2824 * ... Or is it this?
2827 case NETDEV_CHANGEMTU
:
2829 * TODO: Should slaves be allowed to
2830 * independently alter their MTU? For
2831 * an active-backup bond, slaves need
2832 * not be the same type of device, so
2833 * MTUs may vary. For other modes,
2834 * slaves arguably should have the
2835 * same MTUs. To do this, we'd need to
2836 * take over the slave's change_mtu
2837 * function for the duration of their
2841 case NETDEV_CHANGENAME
:
2842 /* we don't care if we don't have primary set */
2843 if (!USES_PRIMARY(bond
->params
.mode
) ||
2844 !bond
->params
.primary
[0])
2847 if (slave
== bond
->primary_slave
) {
2848 /* slave's name changed - he's no longer primary */
2849 bond
->primary_slave
= NULL
;
2850 } else if (!strcmp(slave_dev
->name
, bond
->params
.primary
)) {
2851 /* we have a new primary slave */
2852 bond
->primary_slave
= slave
;
2853 } else { /* we didn't change primary - exit */
2857 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2858 bond
->dev
->name
, bond
->primary_slave
? slave_dev
->name
:
2860 write_lock_bh(&bond
->curr_slave_lock
);
2861 bond_select_active_slave(bond
);
2862 write_unlock_bh(&bond
->curr_slave_lock
);
2864 case NETDEV_FEAT_CHANGE
:
2865 bond_compute_features(bond
);
2867 case NETDEV_RESEND_IGMP
:
2868 /* Propagate to master device */
2869 call_netdevice_notifiers(event
, slave
->bond
->dev
);
2879 * bond_netdev_event: handle netdev notifier chain events.
2881 * This function receives events for the netdev chain. The caller (an
2882 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2883 * locks for us to safely manipulate the slave devices (RTNL lock,
2886 static int bond_netdev_event(struct notifier_block
*this,
2887 unsigned long event
, void *ptr
)
2889 struct net_device
*event_dev
= netdev_notifier_info_to_dev(ptr
);
2891 pr_debug("event_dev: %s, event: %lx\n",
2892 event_dev
? event_dev
->name
: "None",
2895 if (!(event_dev
->priv_flags
& IFF_BONDING
))
2898 if (event_dev
->flags
& IFF_MASTER
) {
2899 pr_debug("IFF_MASTER\n");
2900 return bond_master_netdev_event(event
, event_dev
);
2903 if (event_dev
->flags
& IFF_SLAVE
) {
2904 pr_debug("IFF_SLAVE\n");
2905 return bond_slave_netdev_event(event
, event_dev
);
2911 static struct notifier_block bond_netdev_notifier
= {
2912 .notifier_call
= bond_netdev_event
,
2915 /*---------------------------- Hashing Policies -----------------------------*/
2917 /* L2 hash helper */
2918 static inline u32
bond_eth_hash(struct sk_buff
*skb
)
2920 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
2922 if (skb_headlen(skb
) >= offsetof(struct ethhdr
, h_proto
))
2923 return data
->h_dest
[5] ^ data
->h_source
[5];
2928 /* Extract the appropriate headers based on bond's xmit policy */
2929 static bool bond_flow_dissect(struct bonding
*bond
, struct sk_buff
*skb
,
2930 struct flow_keys
*fk
)
2932 const struct ipv6hdr
*iph6
;
2933 const struct iphdr
*iph
;
2934 int noff
, proto
= -1;
2936 if (bond
->params
.xmit_policy
> BOND_XMIT_POLICY_LAYER23
)
2937 return skb_flow_dissect(skb
, fk
);
2940 noff
= skb_network_offset(skb
);
2941 if (skb
->protocol
== htons(ETH_P_IP
)) {
2942 if (!pskb_may_pull(skb
, noff
+ sizeof(*iph
)))
2945 fk
->src
= iph
->saddr
;
2946 fk
->dst
= iph
->daddr
;
2947 noff
+= iph
->ihl
<< 2;
2948 if (!ip_is_fragment(iph
))
2949 proto
= iph
->protocol
;
2950 } else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
2951 if (!pskb_may_pull(skb
, noff
+ sizeof(*iph6
)))
2953 iph6
= ipv6_hdr(skb
);
2954 fk
->src
= (__force __be32
)ipv6_addr_hash(&iph6
->saddr
);
2955 fk
->dst
= (__force __be32
)ipv6_addr_hash(&iph6
->daddr
);
2956 noff
+= sizeof(*iph6
);
2957 proto
= iph6
->nexthdr
;
2961 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
&& proto
>= 0)
2962 fk
->ports
= skb_flow_get_ports(skb
, noff
, proto
);
2968 * bond_xmit_hash - generate a hash value based on the xmit policy
2969 * @bond: bonding device
2970 * @skb: buffer to use for headers
2971 * @count: modulo value
2973 * This function will extract the necessary headers from the skb buffer and use
2974 * them to generate a hash based on the xmit_policy set in the bonding device
2975 * which will be reduced modulo count before returning.
2977 int bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
, int count
)
2979 struct flow_keys flow
;
2982 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER2
||
2983 !bond_flow_dissect(bond
, skb
, &flow
))
2984 return bond_eth_hash(skb
) % count
;
2986 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER23
||
2987 bond
->params
.xmit_policy
== BOND_XMIT_POLICY_ENCAP23
)
2988 hash
= bond_eth_hash(skb
);
2990 hash
= (__force u32
)flow
.ports
;
2991 hash
^= (__force u32
)flow
.dst
^ (__force u32
)flow
.src
;
2992 hash
^= (hash
>> 16);
2993 hash
^= (hash
>> 8);
2995 return hash
% count
;
2998 /*-------------------------- Device entry points ----------------------------*/
3000 static void bond_work_init_all(struct bonding
*bond
)
3002 INIT_DELAYED_WORK(&bond
->mcast_work
,
3003 bond_resend_igmp_join_requests_delayed
);
3004 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3005 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3006 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3007 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3009 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3010 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3013 static void bond_work_cancel_all(struct bonding
*bond
)
3015 cancel_delayed_work_sync(&bond
->mii_work
);
3016 cancel_delayed_work_sync(&bond
->arp_work
);
3017 cancel_delayed_work_sync(&bond
->alb_work
);
3018 cancel_delayed_work_sync(&bond
->ad_work
);
3019 cancel_delayed_work_sync(&bond
->mcast_work
);
3022 static int bond_open(struct net_device
*bond_dev
)
3024 struct bonding
*bond
= netdev_priv(bond_dev
);
3025 struct list_head
*iter
;
3026 struct slave
*slave
;
3028 /* reset slave->backup and slave->inactive */
3029 read_lock(&bond
->lock
);
3030 if (bond_has_slaves(bond
)) {
3031 read_lock(&bond
->curr_slave_lock
);
3032 bond_for_each_slave(bond
, slave
, iter
) {
3033 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3034 && (slave
!= bond
->curr_active_slave
)) {
3035 bond_set_slave_inactive_flags(slave
);
3037 bond_set_slave_active_flags(slave
);
3040 read_unlock(&bond
->curr_slave_lock
);
3042 read_unlock(&bond
->lock
);
3044 bond_work_init_all(bond
);
3046 if (bond_is_lb(bond
)) {
3047 /* bond_alb_initialize must be called before the timer
3050 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
)))
3052 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3055 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3056 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3058 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3059 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3060 if (bond
->params
.arp_validate
)
3061 bond
->recv_probe
= bond_arp_rcv
;
3064 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3065 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3066 /* register to receive LACPDUs */
3067 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3068 bond_3ad_initiate_agg_selection(bond
, 1);
3074 static int bond_close(struct net_device
*bond_dev
)
3076 struct bonding
*bond
= netdev_priv(bond_dev
);
3078 bond_work_cancel_all(bond
);
3079 bond
->send_peer_notif
= 0;
3080 if (bond_is_lb(bond
))
3081 bond_alb_deinitialize(bond
);
3082 bond
->recv_probe
= NULL
;
3087 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3088 struct rtnl_link_stats64
*stats
)
3090 struct bonding
*bond
= netdev_priv(bond_dev
);
3091 struct rtnl_link_stats64 temp
;
3092 struct list_head
*iter
;
3093 struct slave
*slave
;
3095 memset(stats
, 0, sizeof(*stats
));
3097 read_lock_bh(&bond
->lock
);
3098 bond_for_each_slave(bond
, slave
, iter
) {
3099 const struct rtnl_link_stats64
*sstats
=
3100 dev_get_stats(slave
->dev
, &temp
);
3102 stats
->rx_packets
+= sstats
->rx_packets
;
3103 stats
->rx_bytes
+= sstats
->rx_bytes
;
3104 stats
->rx_errors
+= sstats
->rx_errors
;
3105 stats
->rx_dropped
+= sstats
->rx_dropped
;
3107 stats
->tx_packets
+= sstats
->tx_packets
;
3108 stats
->tx_bytes
+= sstats
->tx_bytes
;
3109 stats
->tx_errors
+= sstats
->tx_errors
;
3110 stats
->tx_dropped
+= sstats
->tx_dropped
;
3112 stats
->multicast
+= sstats
->multicast
;
3113 stats
->collisions
+= sstats
->collisions
;
3115 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3116 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3117 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3118 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3119 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3120 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3122 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3123 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3124 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3125 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3126 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3128 read_unlock_bh(&bond
->lock
);
3133 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3135 struct bonding
*bond
= netdev_priv(bond_dev
);
3136 struct net_device
*slave_dev
= NULL
;
3137 struct ifbond k_binfo
;
3138 struct ifbond __user
*u_binfo
= NULL
;
3139 struct ifslave k_sinfo
;
3140 struct ifslave __user
*u_sinfo
= NULL
;
3141 struct mii_ioctl_data
*mii
= NULL
;
3142 struct bond_opt_value newval
;
3146 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3158 * We do this again just in case we were called by SIOCGMIIREG
3159 * instead of SIOCGMIIPHY.
3166 if (mii
->reg_num
== 1) {
3168 read_lock(&bond
->lock
);
3169 read_lock(&bond
->curr_slave_lock
);
3170 if (netif_carrier_ok(bond
->dev
))
3171 mii
->val_out
= BMSR_LSTATUS
;
3173 read_unlock(&bond
->curr_slave_lock
);
3174 read_unlock(&bond
->lock
);
3178 case BOND_INFO_QUERY_OLD
:
3179 case SIOCBONDINFOQUERY
:
3180 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3182 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3185 res
= bond_info_query(bond_dev
, &k_binfo
);
3187 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3191 case BOND_SLAVE_INFO_QUERY_OLD
:
3192 case SIOCBONDSLAVEINFOQUERY
:
3193 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3195 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3198 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3200 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3209 net
= dev_net(bond_dev
);
3211 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3214 slave_dev
= __dev_get_by_name(net
, ifr
->ifr_slave
);
3216 pr_debug("slave_dev=%p:\n", slave_dev
);
3221 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3223 case BOND_ENSLAVE_OLD
:
3224 case SIOCBONDENSLAVE
:
3225 res
= bond_enslave(bond_dev
, slave_dev
);
3227 case BOND_RELEASE_OLD
:
3228 case SIOCBONDRELEASE
:
3229 res
= bond_release(bond_dev
, slave_dev
);
3231 case BOND_SETHWADDR_OLD
:
3232 case SIOCBONDSETHWADDR
:
3233 bond_set_dev_addr(bond_dev
, slave_dev
);
3236 case BOND_CHANGE_ACTIVE_OLD
:
3237 case SIOCBONDCHANGEACTIVE
:
3238 bond_opt_initstr(&newval
, slave_dev
->name
);
3239 res
= __bond_opt_set(bond
, BOND_OPT_ACTIVE_SLAVE
, &newval
);
3248 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3250 struct bonding
*bond
= netdev_priv(bond_dev
);
3252 if (change
& IFF_PROMISC
)
3253 bond_set_promiscuity(bond
,
3254 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3256 if (change
& IFF_ALLMULTI
)
3257 bond_set_allmulti(bond
,
3258 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3261 static void bond_set_rx_mode(struct net_device
*bond_dev
)
3263 struct bonding
*bond
= netdev_priv(bond_dev
);
3264 struct list_head
*iter
;
3265 struct slave
*slave
;
3269 if (USES_PRIMARY(bond
->params
.mode
)) {
3270 slave
= rcu_dereference(bond
->curr_active_slave
);
3272 dev_uc_sync(slave
->dev
, bond_dev
);
3273 dev_mc_sync(slave
->dev
, bond_dev
);
3276 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3277 dev_uc_sync_multiple(slave
->dev
, bond_dev
);
3278 dev_mc_sync_multiple(slave
->dev
, bond_dev
);
3284 static int bond_neigh_init(struct neighbour
*n
)
3286 struct bonding
*bond
= netdev_priv(n
->dev
);
3287 const struct net_device_ops
*slave_ops
;
3288 struct neigh_parms parms
;
3289 struct slave
*slave
;
3292 slave
= bond_first_slave(bond
);
3295 slave_ops
= slave
->dev
->netdev_ops
;
3296 if (!slave_ops
->ndo_neigh_setup
)
3299 parms
.neigh_setup
= NULL
;
3300 parms
.neigh_cleanup
= NULL
;
3301 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3306 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3307 * after the last slave has been detached. Assumes that all slaves
3308 * utilize the same neigh_cleanup (true at this writing as only user
3311 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3313 if (!parms
.neigh_setup
)
3316 return parms
.neigh_setup(n
);
3320 * The bonding ndo_neigh_setup is called at init time beofre any
3321 * slave exists. So we must declare proxy setup function which will
3322 * be used at run time to resolve the actual slave neigh param setup.
3324 * It's also called by master devices (such as vlans) to setup their
3325 * underlying devices. In that case - do nothing, we're already set up from
3328 static int bond_neigh_setup(struct net_device
*dev
,
3329 struct neigh_parms
*parms
)
3331 /* modify only our neigh_parms */
3332 if (parms
->dev
== dev
)
3333 parms
->neigh_setup
= bond_neigh_init
;
3339 * Change the MTU of all of a master's slaves to match the master
3341 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3343 struct bonding
*bond
= netdev_priv(bond_dev
);
3344 struct slave
*slave
, *rollback_slave
;
3345 struct list_head
*iter
;
3348 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3349 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3351 /* Can't hold bond->lock with bh disabled here since
3352 * some base drivers panic. On the other hand we can't
3353 * hold bond->lock without bh disabled because we'll
3354 * deadlock. The only solution is to rely on the fact
3355 * that we're under rtnl_lock here, and the slaves
3356 * list won't change. This doesn't solve the problem
3357 * of setting the slave's MTU while it is
3358 * transmitting, but the assumption is that the base
3359 * driver can handle that.
3361 * TODO: figure out a way to safely iterate the slaves
3362 * list, but without holding a lock around the actual
3363 * call to the base driver.
3366 bond_for_each_slave(bond
, slave
, iter
) {
3367 pr_debug("s %p c_m %p\n",
3369 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3371 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3374 /* If we failed to set the slave's mtu to the new value
3375 * we must abort the operation even in ACTIVE_BACKUP
3376 * mode, because if we allow the backup slaves to have
3377 * different mtu values than the active slave we'll
3378 * need to change their mtu when doing a failover. That
3379 * means changing their mtu from timer context, which
3380 * is probably not a good idea.
3382 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3387 bond_dev
->mtu
= new_mtu
;
3392 /* unwind from head to the slave that failed */
3393 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3396 if (rollback_slave
== slave
)
3399 tmp_res
= dev_set_mtu(rollback_slave
->dev
, bond_dev
->mtu
);
3401 pr_debug("unwind err %d dev %s\n",
3402 tmp_res
, rollback_slave
->dev
->name
);
3412 * Note that many devices must be down to change the HW address, and
3413 * downing the master releases all slaves. We can make bonds full of
3414 * bonding devices to test this, however.
3416 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3418 struct bonding
*bond
= netdev_priv(bond_dev
);
3419 struct slave
*slave
, *rollback_slave
;
3420 struct sockaddr
*sa
= addr
, tmp_sa
;
3421 struct list_head
*iter
;
3424 if (bond
->params
.mode
== BOND_MODE_ALB
)
3425 return bond_alb_set_mac_address(bond_dev
, addr
);
3428 pr_debug("bond=%p, name=%s\n",
3429 bond
, bond_dev
? bond_dev
->name
: "None");
3431 /* If fail_over_mac is enabled, do nothing and return success.
3432 * Returning an error causes ifenslave to fail.
3434 if (bond
->params
.fail_over_mac
)
3437 if (!is_valid_ether_addr(sa
->sa_data
))
3438 return -EADDRNOTAVAIL
;
3440 /* Can't hold bond->lock with bh disabled here since
3441 * some base drivers panic. On the other hand we can't
3442 * hold bond->lock without bh disabled because we'll
3443 * deadlock. The only solution is to rely on the fact
3444 * that we're under rtnl_lock here, and the slaves
3445 * list won't change. This doesn't solve the problem
3446 * of setting the slave's hw address while it is
3447 * transmitting, but the assumption is that the base
3448 * driver can handle that.
3450 * TODO: figure out a way to safely iterate the slaves
3451 * list, but without holding a lock around the actual
3452 * call to the base driver.
3455 bond_for_each_slave(bond
, slave
, iter
) {
3456 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3457 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3459 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3461 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3465 res
= dev_set_mac_address(slave
->dev
, addr
);
3467 /* TODO: consider downing the slave
3469 * User should expect communications
3470 * breakage anyway until ARP finish
3473 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3479 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3483 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3484 tmp_sa
.sa_family
= bond_dev
->type
;
3486 /* unwind from head to the slave that failed */
3487 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3490 if (rollback_slave
== slave
)
3493 tmp_res
= dev_set_mac_address(rollback_slave
->dev
, &tmp_sa
);
3495 pr_debug("unwind err %d dev %s\n",
3496 tmp_res
, rollback_slave
->dev
->name
);
3504 * bond_xmit_slave_id - transmit skb through slave with slave_id
3505 * @bond: bonding device that is transmitting
3506 * @skb: buffer to transmit
3507 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3509 * This function tries to transmit through slave with slave_id but in case
3510 * it fails, it tries to find the first available slave for transmission.
3511 * The skb is consumed in all cases, thus the function is void.
3513 static void bond_xmit_slave_id(struct bonding
*bond
, struct sk_buff
*skb
, int slave_id
)
3515 struct list_head
*iter
;
3516 struct slave
*slave
;
3519 /* Here we start from the slave with slave_id */
3520 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3522 if (slave_can_tx(slave
)) {
3523 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3529 /* Here we start from the first slave up to slave_id */
3531 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3534 if (slave_can_tx(slave
)) {
3535 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3539 /* no slave that can tx has been found */
3544 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3545 * @bond: bonding device to use
3547 * Based on the value of the bonding device's packets_per_slave parameter
3548 * this function generates a slave id, which is usually used as the next
3549 * slave to transmit through.
3551 static u32
bond_rr_gen_slave_id(struct bonding
*bond
)
3554 struct reciprocal_value reciprocal_packets_per_slave
;
3555 int packets_per_slave
= bond
->params
.packets_per_slave
;
3557 switch (packets_per_slave
) {
3559 slave_id
= prandom_u32();
3562 slave_id
= bond
->rr_tx_counter
;
3565 reciprocal_packets_per_slave
=
3566 bond
->params
.reciprocal_packets_per_slave
;
3567 slave_id
= reciprocal_divide(bond
->rr_tx_counter
,
3568 reciprocal_packets_per_slave
);
3571 bond
->rr_tx_counter
++;
3576 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3578 struct bonding
*bond
= netdev_priv(bond_dev
);
3579 struct iphdr
*iph
= ip_hdr(skb
);
3580 struct slave
*slave
;
3583 /* Start with the curr_active_slave that joined the bond as the
3584 * default for sending IGMP traffic. For failover purposes one
3585 * needs to maintain some consistency for the interface that will
3586 * send the join/membership reports. The curr_active_slave found
3587 * will send all of this type of traffic.
3589 if (iph
->protocol
== IPPROTO_IGMP
&& skb
->protocol
== htons(ETH_P_IP
)) {
3590 slave
= rcu_dereference(bond
->curr_active_slave
);
3591 if (slave
&& slave_can_tx(slave
))
3592 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3594 bond_xmit_slave_id(bond
, skb
, 0);
3596 slave_id
= bond_rr_gen_slave_id(bond
);
3597 bond_xmit_slave_id(bond
, skb
, slave_id
% bond
->slave_cnt
);
3600 return NETDEV_TX_OK
;
3604 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3605 * the bond has a usable interface.
3607 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3609 struct bonding
*bond
= netdev_priv(bond_dev
);
3610 struct slave
*slave
;
3612 slave
= rcu_dereference(bond
->curr_active_slave
);
3614 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3618 return NETDEV_TX_OK
;
3621 /* In bond_xmit_xor() , we determine the output device by using a pre-
3622 * determined xmit_hash_policy(), If the selected device is not enabled,
3623 * find the next active slave.
3625 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3627 struct bonding
*bond
= netdev_priv(bond_dev
);
3629 bond_xmit_slave_id(bond
, skb
, bond_xmit_hash(bond
, skb
, bond
->slave_cnt
));
3631 return NETDEV_TX_OK
;
3634 /* in broadcast mode, we send everything to all usable interfaces. */
3635 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3637 struct bonding
*bond
= netdev_priv(bond_dev
);
3638 struct slave
*slave
= NULL
;
3639 struct list_head
*iter
;
3641 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3642 if (bond_is_last_slave(bond
, slave
))
3644 if (IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
) {
3645 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
3648 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3652 /* bond_dev_queue_xmit always returns 0 */
3653 bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3656 if (slave
&& IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
)
3657 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3661 return NETDEV_TX_OK
;
3664 /*------------------------- Device initialization ---------------------------*/
3667 * Lookup the slave that corresponds to a qid
3669 static inline int bond_slave_override(struct bonding
*bond
,
3670 struct sk_buff
*skb
)
3672 struct slave
*slave
= NULL
;
3673 struct list_head
*iter
;
3675 if (!skb
->queue_mapping
)
3678 /* Find out if any slaves have the same mapping as this skb. */
3679 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3680 if (slave
->queue_id
== skb
->queue_mapping
) {
3681 if (slave_can_tx(slave
)) {
3682 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3685 /* If the slave isn't UP, use default transmit policy. */
3694 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
3698 * This helper function exists to help dev_pick_tx get the correct
3699 * destination queue. Using a helper function skips a call to
3700 * skb_tx_hash and will put the skbs in the queue we expect on their
3701 * way down to the bonding driver.
3703 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
3706 * Save the original txq to restore before passing to the driver
3708 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
3710 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
3712 txq
-= dev
->real_num_tx_queues
;
3713 } while (txq
>= dev
->real_num_tx_queues
);
3718 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3720 struct bonding
*bond
= netdev_priv(dev
);
3722 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
3723 if (!bond_slave_override(bond
, skb
))
3724 return NETDEV_TX_OK
;
3727 switch (bond
->params
.mode
) {
3728 case BOND_MODE_ROUNDROBIN
:
3729 return bond_xmit_roundrobin(skb
, dev
);
3730 case BOND_MODE_ACTIVEBACKUP
:
3731 return bond_xmit_activebackup(skb
, dev
);
3733 return bond_xmit_xor(skb
, dev
);
3734 case BOND_MODE_BROADCAST
:
3735 return bond_xmit_broadcast(skb
, dev
);
3736 case BOND_MODE_8023AD
:
3737 return bond_3ad_xmit_xor(skb
, dev
);
3740 return bond_alb_xmit(skb
, dev
);
3742 /* Should never happen, mode already checked */
3743 pr_err("%s: Error: Unknown bonding mode %d\n",
3744 dev
->name
, bond
->params
.mode
);
3747 return NETDEV_TX_OK
;
3751 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3753 struct bonding
*bond
= netdev_priv(dev
);
3754 netdev_tx_t ret
= NETDEV_TX_OK
;
3757 * If we risk deadlock from transmitting this in the
3758 * netpoll path, tell netpoll to queue the frame for later tx
3760 if (is_netpoll_tx_blocked(dev
))
3761 return NETDEV_TX_BUSY
;
3764 if (bond_has_slaves(bond
))
3765 ret
= __bond_start_xmit(skb
, dev
);
3773 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
3774 struct ethtool_cmd
*ecmd
)
3776 struct bonding
*bond
= netdev_priv(bond_dev
);
3777 unsigned long speed
= 0;
3778 struct list_head
*iter
;
3779 struct slave
*slave
;
3781 ecmd
->duplex
= DUPLEX_UNKNOWN
;
3782 ecmd
->port
= PORT_OTHER
;
3784 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3785 * do not need to check mode. Though link speed might not represent
3786 * the true receive or transmit bandwidth (not all modes are symmetric)
3787 * this is an accurate maximum.
3789 read_lock(&bond
->lock
);
3790 bond_for_each_slave(bond
, slave
, iter
) {
3791 if (SLAVE_IS_OK(slave
)) {
3792 if (slave
->speed
!= SPEED_UNKNOWN
)
3793 speed
+= slave
->speed
;
3794 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
3795 slave
->duplex
!= DUPLEX_UNKNOWN
)
3796 ecmd
->duplex
= slave
->duplex
;
3799 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
3800 read_unlock(&bond
->lock
);
3805 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
3806 struct ethtool_drvinfo
*drvinfo
)
3808 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
3809 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
3810 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
3814 static const struct ethtool_ops bond_ethtool_ops
= {
3815 .get_drvinfo
= bond_ethtool_get_drvinfo
,
3816 .get_settings
= bond_ethtool_get_settings
,
3817 .get_link
= ethtool_op_get_link
,
3820 static const struct net_device_ops bond_netdev_ops
= {
3821 .ndo_init
= bond_init
,
3822 .ndo_uninit
= bond_uninit
,
3823 .ndo_open
= bond_open
,
3824 .ndo_stop
= bond_close
,
3825 .ndo_start_xmit
= bond_start_xmit
,
3826 .ndo_select_queue
= bond_select_queue
,
3827 .ndo_get_stats64
= bond_get_stats
,
3828 .ndo_do_ioctl
= bond_do_ioctl
,
3829 .ndo_change_rx_flags
= bond_change_rx_flags
,
3830 .ndo_set_rx_mode
= bond_set_rx_mode
,
3831 .ndo_change_mtu
= bond_change_mtu
,
3832 .ndo_set_mac_address
= bond_set_mac_address
,
3833 .ndo_neigh_setup
= bond_neigh_setup
,
3834 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
3835 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
3836 #ifdef CONFIG_NET_POLL_CONTROLLER
3837 .ndo_netpoll_setup
= bond_netpoll_setup
,
3838 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
3839 .ndo_poll_controller
= bond_poll_controller
,
3841 .ndo_add_slave
= bond_enslave
,
3842 .ndo_del_slave
= bond_release
,
3843 .ndo_fix_features
= bond_fix_features
,
3846 static const struct device_type bond_type
= {
3850 static void bond_destructor(struct net_device
*bond_dev
)
3852 struct bonding
*bond
= netdev_priv(bond_dev
);
3854 destroy_workqueue(bond
->wq
);
3855 free_netdev(bond_dev
);
3858 void bond_setup(struct net_device
*bond_dev
)
3860 struct bonding
*bond
= netdev_priv(bond_dev
);
3862 /* initialize rwlocks */
3863 rwlock_init(&bond
->lock
);
3864 rwlock_init(&bond
->curr_slave_lock
);
3865 bond
->params
= bonding_defaults
;
3867 /* Initialize pointers */
3868 bond
->dev
= bond_dev
;
3870 /* Initialize the device entry points */
3871 ether_setup(bond_dev
);
3872 bond_dev
->netdev_ops
= &bond_netdev_ops
;
3873 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
3875 bond_dev
->destructor
= bond_destructor
;
3877 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
3879 /* Initialize the device options */
3880 bond_dev
->tx_queue_len
= 0;
3881 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
3882 bond_dev
->priv_flags
|= IFF_BONDING
;
3883 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
3885 /* At first, we block adding VLANs. That's the only way to
3886 * prevent problems that occur when adding VLANs over an
3887 * empty bond. The block will be removed once non-challenged
3888 * slaves are enslaved.
3890 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
3892 /* don't acquire bond device's netif_tx_lock when
3894 bond_dev
->features
|= NETIF_F_LLTX
;
3896 /* By default, we declare the bond to be fully
3897 * VLAN hardware accelerated capable. Special
3898 * care is taken in the various xmit functions
3899 * when there are slaves that are not hw accel
3903 /* Don't allow bond devices to change network namespaces. */
3904 bond_dev
->features
|= NETIF_F_NETNS_LOCAL
;
3906 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
3907 NETIF_F_HW_VLAN_CTAG_TX
|
3908 NETIF_F_HW_VLAN_CTAG_RX
|
3909 NETIF_F_HW_VLAN_CTAG_FILTER
;
3911 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
3912 bond_dev
->features
|= bond_dev
->hw_features
;
3916 * Destroy a bonding device.
3917 * Must be under rtnl_lock when this function is called.
3919 static void bond_uninit(struct net_device
*bond_dev
)
3921 struct bonding
*bond
= netdev_priv(bond_dev
);
3922 struct list_head
*iter
;
3923 struct slave
*slave
;
3925 bond_netpoll_cleanup(bond_dev
);
3927 /* Release the bonded slaves */
3928 bond_for_each_slave(bond
, slave
, iter
)
3929 __bond_release_one(bond_dev
, slave
->dev
, true);
3930 pr_info("%s: released all slaves\n", bond_dev
->name
);
3932 list_del(&bond
->bond_list
);
3934 bond_debug_unregister(bond
);
3937 /*------------------------- Module initialization ---------------------------*/
3939 int bond_parm_tbl_lookup(int mode
, const struct bond_parm_tbl
*tbl
)
3943 for (i
= 0; tbl
[i
].modename
; i
++)
3944 if (mode
== tbl
[i
].mode
)
3950 static int bond_parm_tbl_lookup_name(const char *modename
,
3951 const struct bond_parm_tbl
*tbl
)
3955 for (i
= 0; tbl
[i
].modename
; i
++)
3956 if (strcmp(modename
, tbl
[i
].modename
) == 0)
3963 * Convert string input module parms. Accept either the
3964 * number of the mode or its string name. A bit complicated because
3965 * some mode names are substrings of other names, and calls from sysfs
3966 * may have whitespace in the name (trailing newlines, for example).
3968 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
3971 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1];
3973 for (p
= (char *)buf
; *p
; p
++)
3974 if (!(isdigit(*p
) || isspace(*p
)))
3977 if (*p
&& sscanf(buf
, "%20s", modestr
) != 0)
3978 return bond_parm_tbl_lookup_name(modestr
, tbl
);
3979 else if (sscanf(buf
, "%d", &modeint
) != 0)
3980 return bond_parm_tbl_lookup(modeint
, tbl
);
3985 static int bond_check_params(struct bond_params
*params
)
3987 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
3988 struct bond_opt_value newval
, *valptr
;
3989 int arp_all_targets_value
;
3992 * Convert string parameters.
3995 bond_opt_initstr(&newval
, mode
);
3996 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_MODE
), &newval
);
3998 pr_err("Error: Invalid bonding mode \"%s\"\n", mode
);
4001 bond_mode
= valptr
->value
;
4004 if (xmit_hash_policy
) {
4005 if ((bond_mode
!= BOND_MODE_XOR
) &&
4006 (bond_mode
!= BOND_MODE_8023AD
)) {
4007 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4008 bond_mode_name(bond_mode
));
4010 bond_opt_initstr(&newval
, xmit_hash_policy
);
4011 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH
),
4014 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4018 xmit_hashtype
= valptr
->value
;
4023 if (bond_mode
!= BOND_MODE_8023AD
) {
4024 pr_info("lacp_rate param is irrelevant in mode %s\n",
4025 bond_mode_name(bond_mode
));
4027 bond_opt_initstr(&newval
, lacp_rate
);
4028 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE
),
4031 pr_err("Error: Invalid lacp rate \"%s\"\n",
4035 lacp_fast
= valptr
->value
;
4040 bond_opt_initstr(&newval
, lacp_rate
);
4041 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT
),
4044 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select
);
4047 params
->ad_select
= valptr
->value
;
4048 if (bond_mode
!= BOND_MODE_8023AD
)
4049 pr_warning("ad_select param only affects 802.3ad mode\n");
4051 params
->ad_select
= BOND_AD_STABLE
;
4054 if (max_bonds
< 0) {
4055 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4056 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4057 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4061 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4067 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4072 if (downdelay
< 0) {
4073 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4074 downdelay
, INT_MAX
);
4078 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4079 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4084 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4085 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4090 /* reset values for 802.3ad/TLB/ALB */
4091 if (BOND_NO_USES_ARP(bond_mode
)) {
4093 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4094 pr_warning("Forcing miimon to 100msec\n");
4095 miimon
= BOND_DEFAULT_MIIMON
;
4099 if (tx_queues
< 1 || tx_queues
> 255) {
4100 pr_warning("Warning: tx_queues (%d) should be between "
4101 "1 and 255, resetting to %d\n",
4102 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4103 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4106 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4107 pr_warning("Warning: all_slaves_active module parameter (%d), "
4108 "not of valid value (0/1), so it was set to "
4109 "0\n", all_slaves_active
);
4110 all_slaves_active
= 0;
4113 if (resend_igmp
< 0 || resend_igmp
> 255) {
4114 pr_warning("Warning: resend_igmp (%d) should be between "
4115 "0 and 255, resetting to %d\n",
4116 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4117 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4120 bond_opt_initval(&newval
, packets_per_slave
);
4121 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE
), &newval
)) {
4122 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4123 packets_per_slave
, USHRT_MAX
);
4124 packets_per_slave
= 1;
4127 if (bond_mode
== BOND_MODE_ALB
) {
4128 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",
4133 if (updelay
|| downdelay
) {
4134 /* just warn the user the up/down delay will have
4135 * no effect since miimon is zero...
4137 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4138 updelay
, downdelay
);
4141 /* don't allow arp monitoring */
4143 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4144 miimon
, arp_interval
);
4148 if ((updelay
% miimon
) != 0) {
4149 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4151 (updelay
/ miimon
) * miimon
);
4156 if ((downdelay
% miimon
) != 0) {
4157 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4159 (downdelay
/ miimon
) * miimon
);
4162 downdelay
/= miimon
;
4165 if (arp_interval
< 0) {
4166 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to 0\n",
4167 arp_interval
, INT_MAX
);
4171 for (arp_ip_count
= 0, i
= 0;
4172 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4173 /* not complete check, but should be good enough to
4176 if (!in4_pton(arp_ip_target
[i
], -1, (u8
*)&ip
, -1, NULL
) ||
4177 IS_IP_TARGET_UNUSABLE_ADDRESS(ip
)) {
4178 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4182 if (bond_get_targets_ip(arp_target
, ip
) == -1)
4183 arp_target
[arp_ip_count
++] = ip
;
4185 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4190 if (arp_interval
&& !arp_ip_count
) {
4191 /* don't allow arping if no arp_ip_target given... */
4192 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4198 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4199 pr_err("arp_validate only supported in active-backup mode\n");
4202 if (!arp_interval
) {
4203 pr_err("arp_validate requires arp_interval\n");
4207 bond_opt_initstr(&newval
, arp_validate
);
4208 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE
),
4211 pr_err("Error: invalid arp_validate \"%s\"\n",
4215 arp_validate_value
= valptr
->value
;
4217 arp_validate_value
= 0;
4220 arp_all_targets_value
= 0;
4221 if (arp_all_targets
) {
4222 bond_opt_initstr(&newval
, arp_all_targets
);
4223 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS
),
4226 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4228 arp_all_targets_value
= 0;
4230 arp_all_targets_value
= valptr
->value
;
4235 pr_info("MII link monitoring set to %d ms\n", miimon
);
4236 } else if (arp_interval
) {
4237 valptr
= bond_opt_get_val(BOND_OPT_ARP_VALIDATE
,
4238 arp_validate_value
);
4239 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4240 arp_interval
, valptr
->string
, arp_ip_count
);
4242 for (i
= 0; i
< arp_ip_count
; i
++)
4243 pr_info(" %s", arp_ip_target
[i
]);
4247 } else if (max_bonds
) {
4248 /* miimon and arp_interval not set, we need one so things
4249 * work as expected, see bonding.txt for details
4251 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");
4254 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4255 /* currently, using a primary only makes sense
4256 * in active backup, TLB or ALB modes
4258 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4259 primary
, bond_mode_name(bond_mode
));
4263 if (primary
&& primary_reselect
) {
4264 bond_opt_initstr(&newval
, primary_reselect
);
4265 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT
),
4268 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4272 primary_reselect_value
= valptr
->value
;
4274 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4277 if (fail_over_mac
) {
4278 bond_opt_initstr(&newval
, fail_over_mac
);
4279 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC
),
4282 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4286 fail_over_mac_value
= valptr
->value
;
4287 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4288 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4290 fail_over_mac_value
= BOND_FOM_NONE
;
4293 if (lp_interval
== 0) {
4294 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4295 INT_MAX
, BOND_ALB_DEFAULT_LP_INTERVAL
);
4296 lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
4299 /* fill params struct with the proper values */
4300 params
->mode
= bond_mode
;
4301 params
->xmit_policy
= xmit_hashtype
;
4302 params
->miimon
= miimon
;
4303 params
->num_peer_notif
= num_peer_notif
;
4304 params
->arp_interval
= arp_interval
;
4305 params
->arp_validate
= arp_validate_value
;
4306 params
->arp_all_targets
= arp_all_targets_value
;
4307 params
->updelay
= updelay
;
4308 params
->downdelay
= downdelay
;
4309 params
->use_carrier
= use_carrier
;
4310 params
->lacp_fast
= lacp_fast
;
4311 params
->primary
[0] = 0;
4312 params
->primary_reselect
= primary_reselect_value
;
4313 params
->fail_over_mac
= fail_over_mac_value
;
4314 params
->tx_queues
= tx_queues
;
4315 params
->all_slaves_active
= all_slaves_active
;
4316 params
->resend_igmp
= resend_igmp
;
4317 params
->min_links
= min_links
;
4318 params
->lp_interval
= lp_interval
;
4319 params
->packets_per_slave
= packets_per_slave
;
4320 if (packets_per_slave
> 0) {
4321 params
->reciprocal_packets_per_slave
=
4322 reciprocal_value(packets_per_slave
);
4324 /* reciprocal_packets_per_slave is unused if
4325 * packets_per_slave is 0 or 1, just initialize it
4327 params
->reciprocal_packets_per_slave
=
4328 (struct reciprocal_value
) { 0 };
4332 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4333 params
->primary
[IFNAMSIZ
- 1] = 0;
4336 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4341 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4342 static struct lock_class_key bonding_netdev_addr_lock_key
;
4343 static struct lock_class_key bonding_tx_busylock_key
;
4345 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4346 struct netdev_queue
*txq
,
4349 lockdep_set_class(&txq
->_xmit_lock
,
4350 &bonding_netdev_xmit_lock_key
);
4353 static void bond_set_lockdep_class(struct net_device
*dev
)
4355 lockdep_set_class(&dev
->addr_list_lock
,
4356 &bonding_netdev_addr_lock_key
);
4357 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4358 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4362 * Called from registration process
4364 static int bond_init(struct net_device
*bond_dev
)
4366 struct bonding
*bond
= netdev_priv(bond_dev
);
4367 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4368 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4370 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4373 * Initialize locks that may be required during
4374 * en/deslave operations. All of the bond_open work
4375 * (of which this is part) should really be moved to
4376 * a phase prior to dev_open
4378 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4379 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4381 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4385 bond_set_lockdep_class(bond_dev
);
4387 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4389 bond_prepare_sysfs_group(bond
);
4391 bond_debug_register(bond
);
4393 /* Ensure valid dev_addr */
4394 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4395 bond_dev
->addr_assign_type
== NET_ADDR_PERM
)
4396 eth_hw_addr_random(bond_dev
);
4401 unsigned int bond_get_num_tx_queues(void)
4406 /* Create a new bond based on the specified name and bonding parameters.
4407 * If name is NULL, obtain a suitable "bond%d" name for us.
4408 * Caller must NOT hold rtnl_lock; we need to release it here before we
4409 * set up our sysfs entries.
4411 int bond_create(struct net
*net
, const char *name
)
4413 struct net_device
*bond_dev
;
4418 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4419 name
? name
: "bond%d",
4420 bond_setup
, tx_queues
);
4422 pr_err("%s: eek! can't alloc netdev!\n", name
);
4427 dev_net_set(bond_dev
, net
);
4428 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4430 res
= register_netdevice(bond_dev
);
4432 netif_carrier_off(bond_dev
);
4436 bond_destructor(bond_dev
);
4440 static int __net_init
bond_net_init(struct net
*net
)
4442 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4445 INIT_LIST_HEAD(&bn
->dev_list
);
4447 bond_create_proc_dir(bn
);
4448 bond_create_sysfs(bn
);
4453 static void __net_exit
bond_net_exit(struct net
*net
)
4455 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4456 struct bonding
*bond
, *tmp_bond
;
4459 bond_destroy_sysfs(bn
);
4460 bond_destroy_proc_dir(bn
);
4462 /* Kill off any bonds created after unregistering bond rtnl ops */
4464 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4465 unregister_netdevice_queue(bond
->dev
, &list
);
4466 unregister_netdevice_many(&list
);
4470 static struct pernet_operations bond_net_ops
= {
4471 .init
= bond_net_init
,
4472 .exit
= bond_net_exit
,
4474 .size
= sizeof(struct bond_net
),
4477 static int __init
bonding_init(void)
4482 pr_info("%s", bond_version
);
4484 res
= bond_check_params(&bonding_defaults
);
4488 res
= register_pernet_subsys(&bond_net_ops
);
4492 res
= bond_netlink_init();
4496 bond_create_debugfs();
4498 for (i
= 0; i
< max_bonds
; i
++) {
4499 res
= bond_create(&init_net
, NULL
);
4504 register_netdevice_notifier(&bond_netdev_notifier
);
4508 bond_netlink_fini();
4510 unregister_pernet_subsys(&bond_net_ops
);
4515 static void __exit
bonding_exit(void)
4517 unregister_netdevice_notifier(&bond_netdev_notifier
);
4519 bond_destroy_debugfs();
4521 bond_netlink_fini();
4522 unregister_pernet_subsys(&bond_net_ops
);
4524 #ifdef CONFIG_NET_POLL_CONTROLLER
4526 * Make sure we don't have an imbalance on our netpoll blocking
4528 WARN_ON(atomic_read(&netpoll_block_tx
));
4532 module_init(bonding_init
);
4533 module_exit(bonding_exit
);
4534 MODULE_LICENSE("GPL");
4535 MODULE_VERSION(DRV_VERSION
);
4536 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4537 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");