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 */
89 #define BOND_LINK_MON_INTERV 0
90 #define BOND_LINK_ARP_INTERV 0
92 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
93 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
94 static int num_peer_notif
= 1;
95 static int miimon
= BOND_LINK_MON_INTERV
;
98 static int use_carrier
= 1;
100 static char *primary
;
101 static char *primary_reselect
;
102 static char *lacp_rate
;
103 static int min_links
;
104 static char *ad_select
;
105 static char *xmit_hash_policy
;
106 static int arp_interval
= BOND_LINK_ARP_INTERV
;
107 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
108 static char *arp_validate
;
109 static char *arp_all_targets
;
110 static char *fail_over_mac
;
111 static int all_slaves_active
;
112 static struct bond_params bonding_defaults
;
113 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
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 "
187 /*----------------------------- Global variables ----------------------------*/
189 #ifdef CONFIG_NET_POLL_CONTROLLER
190 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
193 int bond_net_id __read_mostly
;
195 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
196 static int arp_ip_count
;
197 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
198 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
199 static int lacp_fast
;
201 const struct bond_parm_tbl bond_lacp_tbl
[] = {
202 { "slow", AD_LACP_SLOW
},
203 { "fast", AD_LACP_FAST
},
207 const struct bond_parm_tbl bond_mode_tbl
[] = {
208 { "balance-rr", BOND_MODE_ROUNDROBIN
},
209 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
210 { "balance-xor", BOND_MODE_XOR
},
211 { "broadcast", BOND_MODE_BROADCAST
},
212 { "802.3ad", BOND_MODE_8023AD
},
213 { "balance-tlb", BOND_MODE_TLB
},
214 { "balance-alb", BOND_MODE_ALB
},
218 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
219 { "layer2", BOND_XMIT_POLICY_LAYER2
},
220 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
221 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
222 { "encap2+3", BOND_XMIT_POLICY_ENCAP23
},
223 { "encap3+4", BOND_XMIT_POLICY_ENCAP34
},
227 const struct bond_parm_tbl arp_all_targets_tbl
[] = {
228 { "any", BOND_ARP_TARGETS_ANY
},
229 { "all", BOND_ARP_TARGETS_ALL
},
233 const struct bond_parm_tbl arp_validate_tbl
[] = {
234 { "none", BOND_ARP_VALIDATE_NONE
},
235 { "active", BOND_ARP_VALIDATE_ACTIVE
},
236 { "backup", BOND_ARP_VALIDATE_BACKUP
},
237 { "all", BOND_ARP_VALIDATE_ALL
},
241 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
242 { "none", BOND_FOM_NONE
},
243 { "active", BOND_FOM_ACTIVE
},
244 { "follow", BOND_FOM_FOLLOW
},
248 const struct bond_parm_tbl pri_reselect_tbl
[] = {
249 { "always", BOND_PRI_RESELECT_ALWAYS
},
250 { "better", BOND_PRI_RESELECT_BETTER
},
251 { "failure", BOND_PRI_RESELECT_FAILURE
},
255 struct bond_parm_tbl ad_select_tbl
[] = {
256 { "stable", BOND_AD_STABLE
},
257 { "bandwidth", BOND_AD_BANDWIDTH
},
258 { "count", BOND_AD_COUNT
},
262 /*-------------------------- Forward declarations ---------------------------*/
264 static int bond_init(struct net_device
*bond_dev
);
265 static void bond_uninit(struct net_device
*bond_dev
);
267 /*---------------------------- General routines -----------------------------*/
269 const char *bond_mode_name(int mode
)
271 static const char *names
[] = {
272 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
273 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
274 [BOND_MODE_XOR
] = "load balancing (xor)",
275 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
276 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
277 [BOND_MODE_TLB
] = "transmit load balancing",
278 [BOND_MODE_ALB
] = "adaptive load balancing",
281 if (mode
< BOND_MODE_ROUNDROBIN
|| mode
> BOND_MODE_ALB
)
287 /*---------------------------------- VLAN -----------------------------------*/
290 * bond_dev_queue_xmit - Prepare skb for xmit.
292 * @bond: bond device that got this skb for tx.
293 * @skb: hw accel VLAN tagged skb to transmit
294 * @slave_dev: slave that is supposed to xmit this skbuff
296 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
297 struct net_device
*slave_dev
)
299 skb
->dev
= slave_dev
;
301 BUILD_BUG_ON(sizeof(skb
->queue_mapping
) !=
302 sizeof(qdisc_skb_cb(skb
)->slave_dev_queue_mapping
));
303 skb
->queue_mapping
= qdisc_skb_cb(skb
)->slave_dev_queue_mapping
;
305 if (unlikely(netpoll_tx_running(bond
->dev
)))
306 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
314 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
315 * We don't protect the slave list iteration with a lock because:
316 * a. This operation is performed in IOCTL context,
317 * b. The operation is protected by the RTNL semaphore in the 8021q code,
318 * c. Holding a lock with BH disabled while directly calling a base driver
319 * entry point is generally a BAD idea.
321 * The design of synchronization/protection for this operation in the 8021q
322 * module is good for one or more VLAN devices over a single physical device
323 * and cannot be extended for a teaming solution like bonding, so there is a
324 * potential race condition here where a net device from the vlan group might
325 * be referenced (either by a base driver or the 8021q code) while it is being
326 * removed from the system. However, it turns out we're not making matters
327 * worse, and if it works for regular VLAN usage it will work here too.
331 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
332 * @bond_dev: bonding net device that got called
333 * @vid: vlan id being added
335 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
,
336 __be16 proto
, u16 vid
)
338 struct bonding
*bond
= netdev_priv(bond_dev
);
339 struct slave
*slave
, *rollback_slave
;
340 struct list_head
*iter
;
343 bond_for_each_slave(bond
, slave
, iter
) {
344 res
= vlan_vid_add(slave
->dev
, proto
, vid
);
352 /* unwind to the slave that failed */
353 bond_for_each_slave(bond
, rollback_slave
, iter
) {
354 if (rollback_slave
== slave
)
357 vlan_vid_del(rollback_slave
->dev
, proto
, vid
);
364 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
365 * @bond_dev: bonding net device that got called
366 * @vid: vlan id being removed
368 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
,
369 __be16 proto
, u16 vid
)
371 struct bonding
*bond
= netdev_priv(bond_dev
);
372 struct list_head
*iter
;
375 bond_for_each_slave(bond
, slave
, iter
)
376 vlan_vid_del(slave
->dev
, proto
, vid
);
378 if (bond_is_lb(bond
))
379 bond_alb_clear_vlan(bond
, vid
);
384 /*------------------------------- Link status -------------------------------*/
387 * Set the carrier state for the master according to the state of its
388 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
389 * do special 802.3ad magic.
391 * Returns zero if carrier state does not change, nonzero if it does.
393 static int bond_set_carrier(struct bonding
*bond
)
395 struct list_head
*iter
;
398 if (!bond_has_slaves(bond
))
401 if (bond
->params
.mode
== BOND_MODE_8023AD
)
402 return bond_3ad_set_carrier(bond
);
404 bond_for_each_slave(bond
, slave
, iter
) {
405 if (slave
->link
== BOND_LINK_UP
) {
406 if (!netif_carrier_ok(bond
->dev
)) {
407 netif_carrier_on(bond
->dev
);
415 if (netif_carrier_ok(bond
->dev
)) {
416 netif_carrier_off(bond
->dev
);
423 * Get link speed and duplex from the slave's base driver
424 * using ethtool. If for some reason the call fails or the
425 * values are invalid, set speed and duplex to -1,
428 static void bond_update_speed_duplex(struct slave
*slave
)
430 struct net_device
*slave_dev
= slave
->dev
;
431 struct ethtool_cmd ecmd
;
435 slave
->speed
= SPEED_UNKNOWN
;
436 slave
->duplex
= DUPLEX_UNKNOWN
;
438 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
442 slave_speed
= ethtool_cmd_speed(&ecmd
);
443 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
446 switch (ecmd
.duplex
) {
454 slave
->speed
= slave_speed
;
455 slave
->duplex
= ecmd
.duplex
;
461 * if <dev> supports MII link status reporting, check its link status.
463 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
464 * depending upon the setting of the use_carrier parameter.
466 * Return either BMSR_LSTATUS, meaning that the link is up (or we
467 * can't tell and just pretend it is), or 0, meaning that the link is
470 * If reporting is non-zero, instead of faking link up, return -1 if
471 * both ETHTOOL and MII ioctls fail (meaning the device does not
472 * support them). If use_carrier is set, return whatever it says.
473 * It'd be nice if there was a good way to tell if a driver supports
474 * netif_carrier, but there really isn't.
476 static int bond_check_dev_link(struct bonding
*bond
,
477 struct net_device
*slave_dev
, int reporting
)
479 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
480 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
482 struct mii_ioctl_data
*mii
;
484 if (!reporting
&& !netif_running(slave_dev
))
487 if (bond
->params
.use_carrier
)
488 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
490 /* Try to get link status using Ethtool first. */
491 if (slave_dev
->ethtool_ops
->get_link
)
492 return slave_dev
->ethtool_ops
->get_link(slave_dev
) ?
495 /* Ethtool can't be used, fallback to MII ioctls. */
496 ioctl
= slave_ops
->ndo_do_ioctl
;
498 /* TODO: set pointer to correct ioctl on a per team member */
499 /* bases to make this more efficient. that is, once */
500 /* we determine the correct ioctl, we will always */
501 /* call it and not the others for that team */
505 * We cannot assume that SIOCGMIIPHY will also read a
506 * register; not all network drivers (e.g., e100)
510 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
511 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
513 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
514 mii
->reg_num
= MII_BMSR
;
515 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
516 return mii
->val_out
& BMSR_LSTATUS
;
521 * If reporting, report that either there's no dev->do_ioctl,
522 * or both SIOCGMIIREG and get_link failed (meaning that we
523 * cannot report link status). If not reporting, pretend
526 return reporting
? -1 : BMSR_LSTATUS
;
529 /*----------------------------- Multicast list ------------------------------*/
532 * Push the promiscuity flag down to appropriate slaves
534 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
536 struct list_head
*iter
;
539 if (USES_PRIMARY(bond
->params
.mode
)) {
540 /* write lock already acquired */
541 if (bond
->curr_active_slave
) {
542 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
548 bond_for_each_slave(bond
, slave
, iter
) {
549 err
= dev_set_promiscuity(slave
->dev
, inc
);
558 * Push the allmulti flag down to all slaves
560 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
562 struct list_head
*iter
;
565 if (USES_PRIMARY(bond
->params
.mode
)) {
566 /* write lock already acquired */
567 if (bond
->curr_active_slave
) {
568 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
574 bond_for_each_slave(bond
, slave
, iter
) {
575 err
= dev_set_allmulti(slave
->dev
, inc
);
584 * Retrieve the list of registered multicast addresses for the bonding
585 * device and retransmit an IGMP JOIN request to the current active
588 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
590 if (!rtnl_trylock()) {
591 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
594 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, bond
->dev
);
597 /* We use curr_slave_lock to protect against concurrent access to
598 * igmp_retrans from multiple running instances of this function and
599 * bond_change_active_slave
601 write_lock_bh(&bond
->curr_slave_lock
);
602 if (bond
->igmp_retrans
> 1) {
603 bond
->igmp_retrans
--;
604 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
606 write_unlock_bh(&bond
->curr_slave_lock
);
609 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
611 struct bonding
*bond
= container_of(work
, struct bonding
,
614 bond_resend_igmp_join_requests(bond
);
617 /* Flush bond's hardware addresses from slave
619 static void bond_hw_addr_flush(struct net_device
*bond_dev
,
620 struct net_device
*slave_dev
)
622 struct bonding
*bond
= netdev_priv(bond_dev
);
624 dev_uc_unsync(slave_dev
, bond_dev
);
625 dev_mc_unsync(slave_dev
, bond_dev
);
627 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
628 /* del lacpdu mc addr from mc list */
629 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
631 dev_mc_del(slave_dev
, lacpdu_multicast
);
635 /*--------------------------- Active slave change ---------------------------*/
637 /* Update the hardware address list and promisc/allmulti for the new and
638 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
639 * slaves up date at all times; only the USES_PRIMARY modes need to call
640 * this function to swap these settings during a failover.
642 static void bond_hw_addr_swap(struct bonding
*bond
, struct slave
*new_active
,
643 struct slave
*old_active
)
648 if (bond
->dev
->flags
& IFF_PROMISC
)
649 dev_set_promiscuity(old_active
->dev
, -1);
651 if (bond
->dev
->flags
& IFF_ALLMULTI
)
652 dev_set_allmulti(old_active
->dev
, -1);
654 bond_hw_addr_flush(bond
->dev
, old_active
->dev
);
658 /* FIXME: Signal errors upstream. */
659 if (bond
->dev
->flags
& IFF_PROMISC
)
660 dev_set_promiscuity(new_active
->dev
, 1);
662 if (bond
->dev
->flags
& IFF_ALLMULTI
)
663 dev_set_allmulti(new_active
->dev
, 1);
665 netif_addr_lock_bh(bond
->dev
);
666 dev_uc_sync(new_active
->dev
, bond
->dev
);
667 dev_mc_sync(new_active
->dev
, bond
->dev
);
668 netif_addr_unlock_bh(bond
->dev
);
673 * bond_set_dev_addr - clone slave's address to bond
674 * @bond_dev: bond net device
675 * @slave_dev: slave net device
677 * Should be called with RTNL held.
679 static void bond_set_dev_addr(struct net_device
*bond_dev
,
680 struct net_device
*slave_dev
)
682 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
683 bond_dev
, slave_dev
, slave_dev
->addr_len
);
684 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
685 bond_dev
->addr_assign_type
= NET_ADDR_STOLEN
;
686 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
690 * bond_do_fail_over_mac
692 * Perform special MAC address swapping for fail_over_mac settings
694 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
696 static void bond_do_fail_over_mac(struct bonding
*bond
,
697 struct slave
*new_active
,
698 struct slave
*old_active
)
699 __releases(&bond
->curr_slave_lock
)
700 __releases(&bond
->lock
)
701 __acquires(&bond
->lock
)
702 __acquires(&bond
->curr_slave_lock
)
704 u8 tmp_mac
[ETH_ALEN
];
705 struct sockaddr saddr
;
708 switch (bond
->params
.fail_over_mac
) {
709 case BOND_FOM_ACTIVE
:
711 write_unlock_bh(&bond
->curr_slave_lock
);
712 read_unlock(&bond
->lock
);
713 bond_set_dev_addr(bond
->dev
, new_active
->dev
);
714 read_lock(&bond
->lock
);
715 write_lock_bh(&bond
->curr_slave_lock
);
718 case BOND_FOM_FOLLOW
:
720 * if new_active && old_active, swap them
721 * if just old_active, do nothing (going to no active slave)
722 * if just new_active, set new_active to bond's MAC
727 write_unlock_bh(&bond
->curr_slave_lock
);
728 read_unlock(&bond
->lock
);
731 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
732 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
734 saddr
.sa_family
= new_active
->dev
->type
;
736 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
737 saddr
.sa_family
= bond
->dev
->type
;
740 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
742 pr_err("%s: Error %d setting MAC of slave %s\n",
743 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
750 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
751 saddr
.sa_family
= old_active
->dev
->type
;
753 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
755 pr_err("%s: Error %d setting MAC of slave %s\n",
756 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
758 read_lock(&bond
->lock
);
759 write_lock_bh(&bond
->curr_slave_lock
);
762 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
763 bond
->dev
->name
, bond
->params
.fail_over_mac
);
769 static bool bond_should_change_active(struct bonding
*bond
)
771 struct slave
*prim
= bond
->primary_slave
;
772 struct slave
*curr
= bond
->curr_active_slave
;
774 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
776 if (bond
->force_primary
) {
777 bond
->force_primary
= false;
780 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
781 (prim
->speed
< curr
->speed
||
782 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
784 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
790 * find_best_interface - select the best available slave to be the active one
791 * @bond: our bonding struct
793 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
795 struct slave
*slave
, *bestslave
= NULL
;
796 struct list_head
*iter
;
797 int mintime
= bond
->params
.updelay
;
799 if (bond
->primary_slave
&& bond
->primary_slave
->link
== BOND_LINK_UP
&&
800 bond_should_change_active(bond
))
801 return bond
->primary_slave
;
803 bond_for_each_slave(bond
, slave
, iter
) {
804 if (slave
->link
== BOND_LINK_UP
)
806 if (slave
->link
== BOND_LINK_BACK
&& IS_UP(slave
->dev
) &&
807 slave
->delay
< mintime
) {
808 mintime
= slave
->delay
;
816 static bool bond_should_notify_peers(struct bonding
*bond
)
818 struct slave
*slave
= bond
->curr_active_slave
;
820 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
821 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
823 if (!slave
|| !bond
->send_peer_notif
||
824 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
831 * change_active_interface - change the active slave into the specified one
832 * @bond: our bonding struct
833 * @new: the new slave to make the active one
835 * Set the new slave to the bond's settings and unset them on the old
837 * Setting include flags, mc-list, promiscuity, allmulti, etc.
839 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
840 * because it is apparently the best available slave we have, even though its
841 * updelay hasn't timed out yet.
843 * If new_active is not NULL, caller must hold bond->lock for read and
844 * curr_slave_lock for write_bh.
846 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
848 struct slave
*old_active
= bond
->curr_active_slave
;
850 if (old_active
== new_active
)
854 new_active
->jiffies
= jiffies
;
856 if (new_active
->link
== BOND_LINK_BACK
) {
857 if (USES_PRIMARY(bond
->params
.mode
)) {
858 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
859 bond
->dev
->name
, new_active
->dev
->name
,
860 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
863 new_active
->delay
= 0;
864 new_active
->link
= BOND_LINK_UP
;
866 if (bond
->params
.mode
== BOND_MODE_8023AD
)
867 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
869 if (bond_is_lb(bond
))
870 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
872 if (USES_PRIMARY(bond
->params
.mode
)) {
873 pr_info("%s: making interface %s the new active one.\n",
874 bond
->dev
->name
, new_active
->dev
->name
);
879 if (USES_PRIMARY(bond
->params
.mode
))
880 bond_hw_addr_swap(bond
, new_active
, old_active
);
882 if (bond_is_lb(bond
)) {
883 bond_alb_handle_active_change(bond
, new_active
);
885 bond_set_slave_inactive_flags(old_active
);
887 bond_set_slave_active_flags(new_active
);
889 rcu_assign_pointer(bond
->curr_active_slave
, new_active
);
892 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
894 bond_set_slave_inactive_flags(old_active
);
897 bool should_notify_peers
= false;
899 bond_set_slave_active_flags(new_active
);
901 if (bond
->params
.fail_over_mac
)
902 bond_do_fail_over_mac(bond
, new_active
,
905 if (netif_running(bond
->dev
)) {
906 bond
->send_peer_notif
=
907 bond
->params
.num_peer_notif
;
908 should_notify_peers
=
909 bond_should_notify_peers(bond
);
912 write_unlock_bh(&bond
->curr_slave_lock
);
913 read_unlock(&bond
->lock
);
915 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
916 if (should_notify_peers
)
917 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
920 read_lock(&bond
->lock
);
921 write_lock_bh(&bond
->curr_slave_lock
);
925 /* resend IGMP joins since active slave has changed or
926 * all were sent on curr_active_slave.
927 * resend only if bond is brought up with the affected
928 * bonding modes and the retransmission is enabled */
929 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
930 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
931 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
932 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
933 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
938 * bond_select_active_slave - select a new active slave, if needed
939 * @bond: our bonding struct
941 * This functions should be called when one of the following occurs:
942 * - The old curr_active_slave has been released or lost its link.
943 * - The primary_slave has got its link back.
944 * - A slave has got its link back and there's no old curr_active_slave.
946 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
948 void bond_select_active_slave(struct bonding
*bond
)
950 struct slave
*best_slave
;
953 best_slave
= bond_find_best_slave(bond
);
954 if (best_slave
!= bond
->curr_active_slave
) {
955 bond_change_active_slave(bond
, best_slave
);
956 rv
= bond_set_carrier(bond
);
960 if (netif_carrier_ok(bond
->dev
)) {
961 pr_info("%s: first active interface up!\n",
964 pr_info("%s: now running without any active interface !\n",
970 #ifdef CONFIG_NET_POLL_CONTROLLER
971 static inline int slave_enable_netpoll(struct slave
*slave
)
976 np
= kzalloc(sizeof(*np
), GFP_ATOMIC
);
981 err
= __netpoll_setup(np
, slave
->dev
, GFP_ATOMIC
);
990 static inline void slave_disable_netpoll(struct slave
*slave
)
992 struct netpoll
*np
= slave
->np
;
998 __netpoll_free_async(np
);
1000 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1002 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1004 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1009 static void bond_poll_controller(struct net_device
*bond_dev
)
1013 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1015 struct bonding
*bond
= netdev_priv(bond_dev
);
1016 struct list_head
*iter
;
1017 struct slave
*slave
;
1019 bond_for_each_slave(bond
, slave
, iter
)
1020 if (IS_UP(slave
->dev
))
1021 slave_disable_netpoll(slave
);
1024 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
, gfp_t gfp
)
1026 struct bonding
*bond
= netdev_priv(dev
);
1027 struct list_head
*iter
;
1028 struct slave
*slave
;
1031 bond_for_each_slave(bond
, slave
, iter
) {
1032 err
= slave_enable_netpoll(slave
);
1034 bond_netpoll_cleanup(dev
);
1041 static inline int slave_enable_netpoll(struct slave
*slave
)
1045 static inline void slave_disable_netpoll(struct slave
*slave
)
1048 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1053 /*---------------------------------- IOCTL ----------------------------------*/
1055 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1056 netdev_features_t features
)
1058 struct bonding
*bond
= netdev_priv(dev
);
1059 struct list_head
*iter
;
1060 netdev_features_t mask
;
1061 struct slave
*slave
;
1063 if (!bond_has_slaves(bond
)) {
1064 /* Disable adding VLANs to empty bond. But why? --mq */
1065 features
|= NETIF_F_VLAN_CHALLENGED
;
1070 features
&= ~NETIF_F_ONE_FOR_ALL
;
1071 features
|= NETIF_F_ALL_FOR_ALL
;
1073 bond_for_each_slave(bond
, slave
, iter
) {
1074 features
= netdev_increment_features(features
,
1075 slave
->dev
->features
,
1078 features
= netdev_add_tso_features(features
, mask
);
1083 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1084 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1085 NETIF_F_HIGHDMA | NETIF_F_LRO)
1087 static void bond_compute_features(struct bonding
*bond
)
1089 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1090 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1091 struct net_device
*bond_dev
= bond
->dev
;
1092 struct list_head
*iter
;
1093 struct slave
*slave
;
1094 unsigned short max_hard_header_len
= ETH_HLEN
;
1095 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1096 u16 gso_max_segs
= GSO_MAX_SEGS
;
1098 if (!bond_has_slaves(bond
))
1101 bond_for_each_slave(bond
, slave
, iter
) {
1102 vlan_features
= netdev_increment_features(vlan_features
,
1103 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1105 dst_release_flag
&= slave
->dev
->priv_flags
;
1106 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1107 max_hard_header_len
= slave
->dev
->hard_header_len
;
1109 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1110 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1114 bond_dev
->vlan_features
= vlan_features
;
1115 bond_dev
->hard_header_len
= max_hard_header_len
;
1116 bond_dev
->gso_max_segs
= gso_max_segs
;
1117 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1119 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1120 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1122 netdev_change_features(bond_dev
);
1125 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1126 struct net_device
*slave_dev
)
1128 bond_dev
->header_ops
= slave_dev
->header_ops
;
1130 bond_dev
->type
= slave_dev
->type
;
1131 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1132 bond_dev
->addr_len
= slave_dev
->addr_len
;
1134 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1135 slave_dev
->addr_len
);
1138 /* On bonding slaves other than the currently active slave, suppress
1139 * duplicates except for alb non-mcast/bcast.
1141 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1142 struct slave
*slave
,
1143 struct bonding
*bond
)
1145 if (bond_is_slave_inactive(slave
)) {
1146 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1147 skb
->pkt_type
!= PACKET_BROADCAST
&&
1148 skb
->pkt_type
!= PACKET_MULTICAST
)
1155 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1157 struct sk_buff
*skb
= *pskb
;
1158 struct slave
*slave
;
1159 struct bonding
*bond
;
1160 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1162 int ret
= RX_HANDLER_ANOTHER
;
1164 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1166 return RX_HANDLER_CONSUMED
;
1170 slave
= bond_slave_get_rcu(skb
->dev
);
1173 if (bond
->params
.arp_interval
)
1174 slave
->dev
->last_rx
= jiffies
;
1176 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1178 ret
= recv_probe(skb
, bond
, slave
);
1179 if (ret
== RX_HANDLER_CONSUMED
) {
1185 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1186 return RX_HANDLER_EXACT
;
1189 skb
->dev
= bond
->dev
;
1191 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1192 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1193 skb
->pkt_type
== PACKET_HOST
) {
1195 if (unlikely(skb_cow_head(skb
,
1196 skb
->data
- skb_mac_header(skb
)))) {
1198 return RX_HANDLER_CONSUMED
;
1200 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1206 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1207 struct net_device
*slave_dev
,
1208 struct slave
*slave
)
1212 err
= netdev_master_upper_dev_link_private(slave_dev
, bond_dev
, slave
);
1215 slave_dev
->flags
|= IFF_SLAVE
;
1216 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1220 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1221 struct net_device
*slave_dev
)
1223 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1224 slave_dev
->flags
&= ~IFF_SLAVE
;
1225 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1228 /* enslave device <slave> to bond device <master> */
1229 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1231 struct bonding
*bond
= netdev_priv(bond_dev
);
1232 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1233 struct slave
*new_slave
= NULL
, *prev_slave
;
1234 struct sockaddr addr
;
1238 if (!bond
->params
.use_carrier
&&
1239 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1240 slave_ops
->ndo_do_ioctl
== NULL
) {
1241 pr_warning("%s: Warning: no link monitoring support for %s\n",
1242 bond_dev
->name
, slave_dev
->name
);
1245 /* already enslaved */
1246 if (slave_dev
->flags
& IFF_SLAVE
) {
1247 pr_debug("Error, Device was already enslaved\n");
1251 /* vlan challenged mutual exclusion */
1252 /* no need to lock since we're protected by rtnl_lock */
1253 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1254 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1255 if (vlan_uses_dev(bond_dev
)) {
1256 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1257 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1260 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1261 bond_dev
->name
, slave_dev
->name
,
1262 slave_dev
->name
, bond_dev
->name
);
1265 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1269 * Old ifenslave binaries are no longer supported. These can
1270 * be identified with moderate accuracy by the state of the slave:
1271 * the current ifenslave will set the interface down prior to
1272 * enslaving it; the old ifenslave will not.
1274 if ((slave_dev
->flags
& IFF_UP
)) {
1275 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1278 goto err_undo_flags
;
1281 /* set bonding device ether type by slave - bonding netdevices are
1282 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1283 * there is a need to override some of the type dependent attribs/funcs.
1285 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1286 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1288 if (!bond_has_slaves(bond
)) {
1289 if (bond_dev
->type
!= slave_dev
->type
) {
1290 pr_debug("%s: change device type from %d to %d\n",
1292 bond_dev
->type
, slave_dev
->type
);
1294 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1296 res
= notifier_to_errno(res
);
1298 pr_err("%s: refused to change device type\n",
1301 goto err_undo_flags
;
1304 /* Flush unicast and multicast addresses */
1305 dev_uc_flush(bond_dev
);
1306 dev_mc_flush(bond_dev
);
1308 if (slave_dev
->type
!= ARPHRD_ETHER
)
1309 bond_setup_by_slave(bond_dev
, slave_dev
);
1311 ether_setup(bond_dev
);
1312 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1315 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1318 } else if (bond_dev
->type
!= slave_dev
->type
) {
1319 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1321 slave_dev
->type
, bond_dev
->type
);
1323 goto err_undo_flags
;
1326 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1327 if (!bond_has_slaves(bond
)) {
1328 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1330 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1331 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1332 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",
1335 goto err_undo_flags
;
1339 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1341 /* If this is the first slave, then we need to set the master's hardware
1342 * address to be the same as the slave's. */
1343 if (!bond_has_slaves(bond
) &&
1344 bond
->dev
->addr_assign_type
== NET_ADDR_RANDOM
)
1345 bond_set_dev_addr(bond
->dev
, slave_dev
);
1347 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1350 goto err_undo_flags
;
1353 * Set the new_slave's queue_id to be zero. Queue ID mapping
1354 * is set via sysfs or module option if desired.
1356 new_slave
->queue_id
= 0;
1358 /* Save slave's original mtu and then set it to match the bond */
1359 new_slave
->original_mtu
= slave_dev
->mtu
;
1360 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1362 pr_debug("Error %d calling dev_set_mtu\n", res
);
1367 * Save slave's original ("permanent") mac address for modes
1368 * that need it, and for restoring it upon release, and then
1369 * set it to the master's address
1371 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1373 if (!bond
->params
.fail_over_mac
) {
1375 * Set slave to master's mac address. The application already
1376 * set the master's mac address to that of the first slave
1378 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1379 addr
.sa_family
= slave_dev
->type
;
1380 res
= dev_set_mac_address(slave_dev
, &addr
);
1382 pr_debug("Error %d calling set_mac_address\n", res
);
1383 goto err_restore_mtu
;
1387 /* open the slave since the application closed it */
1388 res
= dev_open(slave_dev
);
1390 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1391 goto err_restore_mac
;
1394 new_slave
->bond
= bond
;
1395 new_slave
->dev
= slave_dev
;
1396 slave_dev
->priv_flags
|= IFF_BONDING
;
1398 if (bond_is_lb(bond
)) {
1399 /* bond_alb_init_slave() must be called before all other stages since
1400 * it might fail and we do not want to have to undo everything
1402 res
= bond_alb_init_slave(bond
, new_slave
);
1407 /* If the mode USES_PRIMARY, then the following is handled by
1408 * bond_change_active_slave().
1410 if (!USES_PRIMARY(bond
->params
.mode
)) {
1411 /* set promiscuity level to new slave */
1412 if (bond_dev
->flags
& IFF_PROMISC
) {
1413 res
= dev_set_promiscuity(slave_dev
, 1);
1418 /* set allmulti level to new slave */
1419 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1420 res
= dev_set_allmulti(slave_dev
, 1);
1425 netif_addr_lock_bh(bond_dev
);
1427 dev_mc_sync_multiple(slave_dev
, bond_dev
);
1428 dev_uc_sync_multiple(slave_dev
, bond_dev
);
1430 netif_addr_unlock_bh(bond_dev
);
1433 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1434 /* add lacpdu mc addr to mc list */
1435 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1437 dev_mc_add(slave_dev
, lacpdu_multicast
);
1440 res
= vlan_vids_add_by_dev(slave_dev
, bond_dev
);
1442 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1443 bond_dev
->name
, slave_dev
->name
);
1447 prev_slave
= bond_last_slave(bond
);
1449 new_slave
->delay
= 0;
1450 new_slave
->link_failure_count
= 0;
1452 bond_update_speed_duplex(new_slave
);
1454 new_slave
->last_arp_rx
= jiffies
-
1455 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1456 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
1457 new_slave
->target_last_arp_rx
[i
] = new_slave
->last_arp_rx
;
1459 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1460 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1462 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1464 * miimon is set but a bonded network driver
1465 * does not support ETHTOOL/MII and
1466 * arp_interval is not set. Note: if
1467 * use_carrier is enabled, we will never go
1468 * here (because netif_carrier is always
1469 * supported); thus, we don't need to change
1470 * the messages for netif_carrier.
1472 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",
1473 bond_dev
->name
, slave_dev
->name
);
1474 } else if (link_reporting
== -1) {
1475 /* unable get link status using mii/ethtool */
1476 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",
1477 bond_dev
->name
, slave_dev
->name
);
1481 /* check for initial state */
1482 if (bond
->params
.miimon
) {
1483 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1484 if (bond
->params
.updelay
) {
1485 new_slave
->link
= BOND_LINK_BACK
;
1486 new_slave
->delay
= bond
->params
.updelay
;
1488 new_slave
->link
= BOND_LINK_UP
;
1491 new_slave
->link
= BOND_LINK_DOWN
;
1493 } else if (bond
->params
.arp_interval
) {
1494 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1495 BOND_LINK_UP
: BOND_LINK_DOWN
);
1497 new_slave
->link
= BOND_LINK_UP
;
1500 if (new_slave
->link
!= BOND_LINK_DOWN
)
1501 new_slave
->jiffies
= jiffies
;
1502 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1503 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1504 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1506 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1507 /* if there is a primary slave, remember it */
1508 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1509 bond
->primary_slave
= new_slave
;
1510 bond
->force_primary
= true;
1514 switch (bond
->params
.mode
) {
1515 case BOND_MODE_ACTIVEBACKUP
:
1516 bond_set_slave_inactive_flags(new_slave
);
1518 case BOND_MODE_8023AD
:
1519 /* in 802.3ad mode, the internal mechanism
1520 * will activate the slaves in the selected
1523 bond_set_slave_inactive_flags(new_slave
);
1524 /* if this is the first slave */
1526 SLAVE_AD_INFO(new_slave
).id
= 1;
1527 /* Initialize AD with the number of times that the AD timer is called in 1 second
1528 * can be called only after the mac address of the bond is set
1530 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1532 SLAVE_AD_INFO(new_slave
).id
=
1533 SLAVE_AD_INFO(prev_slave
).id
+ 1;
1536 bond_3ad_bind_slave(new_slave
);
1540 bond_set_active_slave(new_slave
);
1541 bond_set_slave_inactive_flags(new_slave
);
1544 pr_debug("This slave is always active in trunk mode\n");
1546 /* always active in trunk mode */
1547 bond_set_active_slave(new_slave
);
1549 /* In trunking mode there is little meaning to curr_active_slave
1550 * anyway (it holds no special properties of the bond device),
1551 * so we can change it without calling change_active_interface()
1553 if (!bond
->curr_active_slave
&& new_slave
->link
== BOND_LINK_UP
)
1554 rcu_assign_pointer(bond
->curr_active_slave
, new_slave
);
1557 } /* switch(bond_mode) */
1559 #ifdef CONFIG_NET_POLL_CONTROLLER
1560 slave_dev
->npinfo
= bond
->dev
->npinfo
;
1561 if (slave_dev
->npinfo
) {
1562 if (slave_enable_netpoll(new_slave
)) {
1563 read_unlock(&bond
->lock
);
1564 pr_info("Error, %s: master_dev is using netpoll, "
1565 "but new slave device does not support netpoll.\n",
1573 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1576 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1580 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
, new_slave
);
1582 pr_debug("Error %d calling bond_master_upper_dev_link\n", res
);
1583 goto err_unregister
;
1587 bond_compute_features(bond
);
1588 bond_set_carrier(bond
);
1590 if (USES_PRIMARY(bond
->params
.mode
)) {
1591 read_lock(&bond
->lock
);
1592 write_lock_bh(&bond
->curr_slave_lock
);
1593 bond_select_active_slave(bond
);
1594 write_unlock_bh(&bond
->curr_slave_lock
);
1595 read_unlock(&bond
->lock
);
1598 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1599 bond_dev
->name
, slave_dev
->name
,
1600 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1601 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1603 /* enslave is successful */
1606 /* Undo stages on error */
1608 netdev_rx_handler_unregister(slave_dev
);
1611 if (!USES_PRIMARY(bond
->params
.mode
))
1612 bond_hw_addr_flush(bond_dev
, slave_dev
);
1614 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1615 write_lock_bh(&bond
->lock
);
1616 if (bond
->primary_slave
== new_slave
)
1617 bond
->primary_slave
= NULL
;
1618 if (bond
->curr_active_slave
== new_slave
) {
1619 bond_change_active_slave(bond
, NULL
);
1620 write_unlock_bh(&bond
->lock
);
1621 read_lock(&bond
->lock
);
1622 write_lock_bh(&bond
->curr_slave_lock
);
1623 bond_select_active_slave(bond
);
1624 write_unlock_bh(&bond
->curr_slave_lock
);
1625 read_unlock(&bond
->lock
);
1627 write_unlock_bh(&bond
->lock
);
1629 slave_disable_netpoll(new_slave
);
1632 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1633 dev_close(slave_dev
);
1636 if (!bond
->params
.fail_over_mac
) {
1637 /* XXX TODO - fom follow mode needs to change master's
1638 * MAC if this slave's MAC is in use by the bond, or at
1639 * least print a warning.
1641 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1642 addr
.sa_family
= slave_dev
->type
;
1643 dev_set_mac_address(slave_dev
, &addr
);
1647 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1653 /* Enslave of first slave has failed and we need to fix master's mac */
1654 if (!bond_has_slaves(bond
) &&
1655 ether_addr_equal(bond_dev
->dev_addr
, slave_dev
->dev_addr
))
1656 eth_hw_addr_random(bond_dev
);
1662 * Try to release the slave device <slave> from the bond device <master>
1663 * It is legal to access curr_active_slave without a lock because all the function
1664 * is write-locked. If "all" is true it means that the function is being called
1665 * while destroying a bond interface and all slaves are being released.
1667 * The rules for slave state should be:
1668 * for Active/Backup:
1669 * Active stays on all backups go down
1670 * for Bonded connections:
1671 * The first up interface should be left on and all others downed.
1673 static int __bond_release_one(struct net_device
*bond_dev
,
1674 struct net_device
*slave_dev
,
1677 struct bonding
*bond
= netdev_priv(bond_dev
);
1678 struct slave
*slave
, *oldcurrent
;
1679 struct sockaddr addr
;
1680 int old_flags
= bond_dev
->flags
;
1681 netdev_features_t old_features
= bond_dev
->features
;
1683 /* slave is not a slave or master is not master of this slave */
1684 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1685 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1686 pr_err("%s: Error: cannot release %s.\n",
1687 bond_dev
->name
, slave_dev
->name
);
1692 write_lock_bh(&bond
->lock
);
1694 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1696 /* not a slave of this bond */
1697 pr_info("%s: %s not enslaved\n",
1698 bond_dev
->name
, slave_dev
->name
);
1699 write_unlock_bh(&bond
->lock
);
1700 unblock_netpoll_tx();
1704 write_unlock_bh(&bond
->lock
);
1706 /* release the slave from its bond */
1709 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1710 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1711 * for this slave anymore.
1713 netdev_rx_handler_unregister(slave_dev
);
1714 write_lock_bh(&bond
->lock
);
1716 /* Inform AD package of unbinding of slave. */
1717 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1718 /* must be called before the slave is
1719 * detached from the list
1721 bond_3ad_unbind_slave(slave
);
1724 pr_info("%s: releasing %s interface %s\n",
1726 bond_is_active_slave(slave
) ? "active" : "backup",
1729 oldcurrent
= bond
->curr_active_slave
;
1731 bond
->current_arp_slave
= NULL
;
1733 if (!all
&& !bond
->params
.fail_over_mac
) {
1734 if (ether_addr_equal(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1735 bond_has_slaves(bond
))
1736 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",
1737 bond_dev
->name
, slave_dev
->name
,
1739 bond_dev
->name
, slave_dev
->name
);
1742 if (bond
->primary_slave
== slave
)
1743 bond
->primary_slave
= NULL
;
1745 if (oldcurrent
== slave
)
1746 bond_change_active_slave(bond
, NULL
);
1748 if (bond_is_lb(bond
)) {
1749 /* Must be called only after the slave has been
1750 * detached from the list and the curr_active_slave
1751 * has been cleared (if our_slave == old_current),
1752 * but before a new active slave is selected.
1754 write_unlock_bh(&bond
->lock
);
1755 bond_alb_deinit_slave(bond
, slave
);
1756 write_lock_bh(&bond
->lock
);
1760 rcu_assign_pointer(bond
->curr_active_slave
, NULL
);
1761 } else if (oldcurrent
== slave
) {
1763 * Note that we hold RTNL over this sequence, so there
1764 * is no concern that another slave add/remove event
1767 write_unlock_bh(&bond
->lock
);
1768 read_lock(&bond
->lock
);
1769 write_lock_bh(&bond
->curr_slave_lock
);
1771 bond_select_active_slave(bond
);
1773 write_unlock_bh(&bond
->curr_slave_lock
);
1774 read_unlock(&bond
->lock
);
1775 write_lock_bh(&bond
->lock
);
1778 if (!bond_has_slaves(bond
)) {
1779 bond_set_carrier(bond
);
1780 eth_hw_addr_random(bond_dev
);
1782 if (vlan_uses_dev(bond_dev
)) {
1783 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1784 bond_dev
->name
, bond_dev
->name
);
1785 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1790 write_unlock_bh(&bond
->lock
);
1791 unblock_netpoll_tx();
1794 if (!bond_has_slaves(bond
)) {
1795 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
1796 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
1799 bond_compute_features(bond
);
1800 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1801 (old_features
& NETIF_F_VLAN_CHALLENGED
))
1802 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1803 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1805 /* must do this from outside any spinlocks */
1806 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1808 /* If the mode USES_PRIMARY, then this cases was handled above by
1809 * bond_change_active_slave(..., NULL)
1811 if (!USES_PRIMARY(bond
->params
.mode
)) {
1812 /* unset promiscuity level from slave
1813 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1814 * of the IFF_PROMISC flag in the bond_dev, but we need the
1815 * value of that flag before that change, as that was the value
1816 * when this slave was attached, so we cache at the start of the
1817 * function and use it here. Same goes for ALLMULTI below
1819 if (old_flags
& IFF_PROMISC
)
1820 dev_set_promiscuity(slave_dev
, -1);
1822 /* unset allmulti level from slave */
1823 if (old_flags
& IFF_ALLMULTI
)
1824 dev_set_allmulti(slave_dev
, -1);
1826 bond_hw_addr_flush(bond_dev
, slave_dev
);
1829 slave_disable_netpoll(slave
);
1831 /* close slave before restoring its mac address */
1832 dev_close(slave_dev
);
1834 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1835 /* restore original ("permanent") mac address */
1836 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1837 addr
.sa_family
= slave_dev
->type
;
1838 dev_set_mac_address(slave_dev
, &addr
);
1841 dev_set_mtu(slave_dev
, slave
->original_mtu
);
1843 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1847 return 0; /* deletion OK */
1850 /* A wrapper used because of ndo_del_link */
1851 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1853 return __bond_release_one(bond_dev
, slave_dev
, false);
1857 * First release a slave and then destroy the bond if no more slaves are left.
1858 * Must be under rtnl_lock when this function is called.
1860 static int bond_release_and_destroy(struct net_device
*bond_dev
,
1861 struct net_device
*slave_dev
)
1863 struct bonding
*bond
= netdev_priv(bond_dev
);
1866 ret
= bond_release(bond_dev
, slave_dev
);
1867 if (ret
== 0 && !bond_has_slaves(bond
)) {
1868 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1869 pr_info("%s: destroying bond %s.\n",
1870 bond_dev
->name
, bond_dev
->name
);
1871 unregister_netdevice(bond_dev
);
1876 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1878 struct bonding
*bond
= netdev_priv(bond_dev
);
1880 info
->bond_mode
= bond
->params
.mode
;
1881 info
->miimon
= bond
->params
.miimon
;
1883 read_lock(&bond
->lock
);
1884 info
->num_slaves
= bond
->slave_cnt
;
1885 read_unlock(&bond
->lock
);
1890 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
1892 struct bonding
*bond
= netdev_priv(bond_dev
);
1893 struct list_head
*iter
;
1894 int i
= 0, res
= -ENODEV
;
1895 struct slave
*slave
;
1897 read_lock(&bond
->lock
);
1898 bond_for_each_slave(bond
, slave
, iter
) {
1899 if (i
++ == (int)info
->slave_id
) {
1901 strcpy(info
->slave_name
, slave
->dev
->name
);
1902 info
->link
= slave
->link
;
1903 info
->state
= bond_slave_state(slave
);
1904 info
->link_failure_count
= slave
->link_failure_count
;
1908 read_unlock(&bond
->lock
);
1913 /*-------------------------------- Monitoring -------------------------------*/
1916 static int bond_miimon_inspect(struct bonding
*bond
)
1918 int link_state
, commit
= 0;
1919 struct list_head
*iter
;
1920 struct slave
*slave
;
1921 bool ignore_updelay
;
1923 ignore_updelay
= !bond
->curr_active_slave
? true : false;
1925 bond_for_each_slave(bond
, slave
, iter
) {
1926 slave
->new_link
= BOND_LINK_NOCHANGE
;
1928 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
1930 switch (slave
->link
) {
1935 slave
->link
= BOND_LINK_FAIL
;
1936 slave
->delay
= bond
->params
.downdelay
;
1938 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1940 (bond
->params
.mode
==
1941 BOND_MODE_ACTIVEBACKUP
) ?
1942 (bond_is_active_slave(slave
) ?
1943 "active " : "backup ") : "",
1945 bond
->params
.downdelay
* bond
->params
.miimon
);
1948 case BOND_LINK_FAIL
:
1951 * recovered before downdelay expired
1953 slave
->link
= BOND_LINK_UP
;
1954 slave
->jiffies
= jiffies
;
1955 pr_info("%s: link status up again after %d ms for interface %s.\n",
1957 (bond
->params
.downdelay
- slave
->delay
) *
1958 bond
->params
.miimon
,
1963 if (slave
->delay
<= 0) {
1964 slave
->new_link
= BOND_LINK_DOWN
;
1972 case BOND_LINK_DOWN
:
1976 slave
->link
= BOND_LINK_BACK
;
1977 slave
->delay
= bond
->params
.updelay
;
1980 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1981 bond
->dev
->name
, slave
->dev
->name
,
1982 ignore_updelay
? 0 :
1983 bond
->params
.updelay
*
1984 bond
->params
.miimon
);
1987 case BOND_LINK_BACK
:
1989 slave
->link
= BOND_LINK_DOWN
;
1990 pr_info("%s: link status down again after %d ms for interface %s.\n",
1992 (bond
->params
.updelay
- slave
->delay
) *
1993 bond
->params
.miimon
,
2002 if (slave
->delay
<= 0) {
2003 slave
->new_link
= BOND_LINK_UP
;
2005 ignore_updelay
= false;
2017 static void bond_miimon_commit(struct bonding
*bond
)
2019 struct list_head
*iter
;
2020 struct slave
*slave
;
2022 bond_for_each_slave(bond
, slave
, iter
) {
2023 switch (slave
->new_link
) {
2024 case BOND_LINK_NOCHANGE
:
2028 slave
->link
= BOND_LINK_UP
;
2029 slave
->jiffies
= jiffies
;
2031 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2032 /* prevent it from being the active one */
2033 bond_set_backup_slave(slave
);
2034 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2035 /* make it immediately active */
2036 bond_set_active_slave(slave
);
2037 } else if (slave
!= bond
->primary_slave
) {
2038 /* prevent it from being the active one */
2039 bond_set_backup_slave(slave
);
2042 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2043 bond
->dev
->name
, slave
->dev
->name
,
2044 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
2045 slave
->duplex
? "full" : "half");
2047 /* notify ad that the link status has changed */
2048 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2049 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2051 if (bond_is_lb(bond
))
2052 bond_alb_handle_link_change(bond
, slave
,
2055 if (!bond
->curr_active_slave
||
2056 (slave
== bond
->primary_slave
))
2061 case BOND_LINK_DOWN
:
2062 if (slave
->link_failure_count
< UINT_MAX
)
2063 slave
->link_failure_count
++;
2065 slave
->link
= BOND_LINK_DOWN
;
2067 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2068 bond
->params
.mode
== BOND_MODE_8023AD
)
2069 bond_set_slave_inactive_flags(slave
);
2071 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2072 bond
->dev
->name
, slave
->dev
->name
);
2074 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2075 bond_3ad_handle_link_change(slave
,
2078 if (bond_is_lb(bond
))
2079 bond_alb_handle_link_change(bond
, slave
,
2082 if (slave
== bond
->curr_active_slave
)
2088 pr_err("%s: invalid new link %d on slave %s\n",
2089 bond
->dev
->name
, slave
->new_link
,
2091 slave
->new_link
= BOND_LINK_NOCHANGE
;
2099 write_lock_bh(&bond
->curr_slave_lock
);
2100 bond_select_active_slave(bond
);
2101 write_unlock_bh(&bond
->curr_slave_lock
);
2102 unblock_netpoll_tx();
2105 bond_set_carrier(bond
);
2111 * Really a wrapper that splits the mii monitor into two phases: an
2112 * inspection, then (if inspection indicates something needs to be done)
2113 * an acquisition of appropriate locks followed by a commit phase to
2114 * implement whatever link state changes are indicated.
2116 void bond_mii_monitor(struct work_struct
*work
)
2118 struct bonding
*bond
= container_of(work
, struct bonding
,
2120 bool should_notify_peers
= false;
2122 if (!rtnl_trylock())
2125 if (!bond_has_slaves(bond
)) {
2130 should_notify_peers
= bond_should_notify_peers(bond
);
2132 if (bond_miimon_inspect(bond
))
2133 bond_miimon_commit(bond
);
2135 if (should_notify_peers
)
2136 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2141 if (bond
->params
.miimon
)
2142 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2143 msecs_to_jiffies(bond
->params
.miimon
));
2146 static bool bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2148 struct net_device
*upper
;
2149 struct list_head
*iter
;
2152 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2156 netdev_for_each_all_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2157 if (ip
== bond_confirm_addr(upper
, 0, ip
)) {
2168 * We go to the (large) trouble of VLAN tagging ARP frames because
2169 * switches in VLAN mode (especially if ports are configured as
2170 * "native" to a VLAN) might not pass non-tagged frames.
2172 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2174 struct sk_buff
*skb
;
2176 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op
,
2177 slave_dev
->name
, &dest_ip
, &src_ip
, vlan_id
);
2179 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2180 NULL
, slave_dev
->dev_addr
, NULL
);
2183 pr_err("ARP packet allocation failed\n");
2187 skb
= vlan_put_tag(skb
, htons(ETH_P_8021Q
), vlan_id
);
2189 pr_err("failed to insert VLAN tag\n");
2197 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2199 struct net_device
*upper
, *vlan_upper
;
2200 struct list_head
*iter
, *vlan_iter
;
2202 __be32
*targets
= bond
->params
.arp_targets
, addr
;
2205 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
&& targets
[i
]; i
++) {
2206 pr_debug("basa: target %pI4\n", &targets
[i
]);
2208 /* Find out through which dev should the packet go */
2209 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2212 pr_debug("%s: no route to arp_ip_target %pI4\n",
2213 bond
->dev
->name
, &targets
[i
]);
2219 /* bond device itself */
2220 if (rt
->dst
.dev
== bond
->dev
)
2224 /* first we search only for vlan devices. for every vlan
2225 * found we verify its upper dev list, searching for the
2226 * rt->dst.dev. If found we save the tag of the vlan and
2227 * proceed to send the packet.
2231 netdev_for_each_all_upper_dev_rcu(bond
->dev
, vlan_upper
,
2233 if (!is_vlan_dev(vlan_upper
))
2235 netdev_for_each_all_upper_dev_rcu(vlan_upper
, upper
,
2237 if (upper
== rt
->dst
.dev
) {
2238 vlan_id
= vlan_dev_vlan_id(vlan_upper
);
2245 /* if the device we're looking for is not on top of any of
2246 * our upper vlans, then just search for any dev that
2247 * matches, and in case it's a vlan - save the id
2249 netdev_for_each_all_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2250 if (upper
== rt
->dst
.dev
) {
2251 /* if it's a vlan - get its VID */
2252 if (is_vlan_dev(upper
))
2253 vlan_id
= vlan_dev_vlan_id(upper
);
2261 /* Not our device - skip */
2262 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2263 bond
->dev
->name
, &targets
[i
],
2264 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2270 addr
= bond_confirm_addr(rt
->dst
.dev
, targets
[i
], 0);
2272 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2277 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2281 if (!sip
|| !bond_has_this_ip(bond
, tip
)) {
2282 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip
, &tip
);
2286 i
= bond_get_targets_ip(bond
->params
.arp_targets
, sip
);
2288 pr_debug("bva: sip %pI4 not found in targets\n", &sip
);
2291 slave
->last_arp_rx
= jiffies
;
2292 slave
->target_last_arp_rx
[i
] = jiffies
;
2295 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2296 struct slave
*slave
)
2298 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2299 unsigned char *arp_ptr
;
2303 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2304 return RX_HANDLER_ANOTHER
;
2306 read_lock(&bond
->lock
);
2308 if (!slave_do_arp_validate(bond
, slave
))
2311 alen
= arp_hdr_len(bond
->dev
);
2313 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2314 bond
->dev
->name
, skb
->dev
->name
);
2316 if (alen
> skb_headlen(skb
)) {
2317 arp
= kmalloc(alen
, GFP_ATOMIC
);
2320 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2324 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2325 skb
->pkt_type
== PACKET_OTHERHOST
||
2326 skb
->pkt_type
== PACKET_LOOPBACK
||
2327 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2328 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2332 arp_ptr
= (unsigned char *)(arp
+ 1);
2333 arp_ptr
+= bond
->dev
->addr_len
;
2334 memcpy(&sip
, arp_ptr
, 4);
2335 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2336 memcpy(&tip
, arp_ptr
, 4);
2338 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2339 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2340 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2344 * Backup slaves won't see the ARP reply, but do come through
2345 * here for each ARP probe (so we swap the sip/tip to validate
2346 * the probe). In a "redundant switch, common router" type of
2347 * configuration, the ARP probe will (hopefully) travel from
2348 * the active, through one switch, the router, then the other
2349 * switch before reaching the backup.
2351 * We 'trust' the arp requests if there is an active slave and
2352 * it received valid arp reply(s) after it became active. This
2353 * is done to avoid endless looping when we can't reach the
2354 * arp_ip_target and fool ourselves with our own arp requests.
2356 if (bond_is_active_slave(slave
))
2357 bond_validate_arp(bond
, slave
, sip
, tip
);
2358 else if (bond
->curr_active_slave
&&
2359 time_after(slave_last_rx(bond
, bond
->curr_active_slave
),
2360 bond
->curr_active_slave
->jiffies
))
2361 bond_validate_arp(bond
, slave
, tip
, sip
);
2364 read_unlock(&bond
->lock
);
2365 if (arp
!= (struct arphdr
*)skb
->data
)
2367 return RX_HANDLER_ANOTHER
;
2370 /* function to verify if we're in the arp_interval timeslice, returns true if
2371 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2372 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2374 static bool bond_time_in_interval(struct bonding
*bond
, unsigned long last_act
,
2377 int delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2379 return time_in_range(jiffies
,
2380 last_act
- delta_in_ticks
,
2381 last_act
+ mod
* delta_in_ticks
+ delta_in_ticks
/2);
2385 * this function is called regularly to monitor each slave's link
2386 * ensuring that traffic is being sent and received when arp monitoring
2387 * is used in load-balancing mode. if the adapter has been dormant, then an
2388 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2389 * arp monitoring in active backup mode.
2391 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2393 struct bonding
*bond
= container_of(work
, struct bonding
,
2395 struct slave
*slave
, *oldcurrent
;
2396 struct list_head
*iter
;
2397 int do_failover
= 0;
2399 if (!rtnl_trylock())
2402 if (!bond_has_slaves(bond
)) {
2407 oldcurrent
= bond
->curr_active_slave
;
2408 /* see if any of the previous devices are up now (i.e. they have
2409 * xmt and rcv traffic). the curr_active_slave does not come into
2410 * the picture unless it is null. also, slave->jiffies is not needed
2411 * here because we send an arp on each slave and give a slave as
2412 * long as it needs to get the tx/rx within the delta.
2413 * TODO: what about up/down delay in arp mode? it wasn't here before
2416 bond_for_each_slave(bond
, slave
, iter
) {
2417 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2419 if (slave
->link
!= BOND_LINK_UP
) {
2420 if (bond_time_in_interval(bond
, trans_start
, 1) &&
2421 bond_time_in_interval(bond
, slave
->dev
->last_rx
, 1)) {
2423 slave
->link
= BOND_LINK_UP
;
2424 bond_set_active_slave(slave
);
2426 /* primary_slave has no meaning in round-robin
2427 * mode. the window of a slave being up and
2428 * curr_active_slave being null after enslaving
2432 pr_info("%s: link status definitely up for interface %s, ",
2437 pr_info("%s: interface %s is now up\n",
2443 /* slave->link == BOND_LINK_UP */
2445 /* not all switches will respond to an arp request
2446 * when the source ip is 0, so don't take the link down
2447 * if we don't know our ip yet
2449 if (!bond_time_in_interval(bond
, trans_start
, 2) ||
2450 !bond_time_in_interval(bond
, slave
->dev
->last_rx
, 2)) {
2452 slave
->link
= BOND_LINK_DOWN
;
2453 bond_set_backup_slave(slave
);
2455 if (slave
->link_failure_count
< UINT_MAX
)
2456 slave
->link_failure_count
++;
2458 pr_info("%s: interface %s is now down.\n",
2462 if (slave
== oldcurrent
)
2467 /* note: if switch is in round-robin mode, all links
2468 * must tx arp to ensure all links rx an arp - otherwise
2469 * links may oscillate or not come up at all; if switch is
2470 * in something like xor mode, there is nothing we can
2471 * do - all replies will be rx'ed on same link causing slaves
2472 * to be unstable during low/no traffic periods
2474 if (IS_UP(slave
->dev
))
2475 bond_arp_send_all(bond
, slave
);
2480 write_lock_bh(&bond
->curr_slave_lock
);
2482 bond_select_active_slave(bond
);
2484 write_unlock_bh(&bond
->curr_slave_lock
);
2485 unblock_netpoll_tx();
2490 if (bond
->params
.arp_interval
)
2491 queue_delayed_work(bond
->wq
, &bond
->arp_work
,
2492 msecs_to_jiffies(bond
->params
.arp_interval
));
2496 * Called to inspect slaves for active-backup mode ARP monitor link state
2497 * changes. Sets new_link in slaves to specify what action should take
2498 * place for the slave. Returns 0 if no changes are found, >0 if changes
2499 * to link states must be committed.
2501 * Called with bond->lock held for read.
2503 static int bond_ab_arp_inspect(struct bonding
*bond
)
2505 unsigned long trans_start
, last_rx
;
2506 struct list_head
*iter
;
2507 struct slave
*slave
;
2510 bond_for_each_slave(bond
, slave
, iter
) {
2511 slave
->new_link
= BOND_LINK_NOCHANGE
;
2512 last_rx
= slave_last_rx(bond
, slave
);
2514 if (slave
->link
!= BOND_LINK_UP
) {
2515 if (bond_time_in_interval(bond
, last_rx
, 1)) {
2516 slave
->new_link
= BOND_LINK_UP
;
2523 * Give slaves 2*delta after being enslaved or made
2524 * active. This avoids bouncing, as the last receive
2525 * times need a full ARP monitor cycle to be updated.
2527 if (bond_time_in_interval(bond
, slave
->jiffies
, 2))
2531 * Backup slave is down if:
2532 * - No current_arp_slave AND
2533 * - more than 3*delta since last receive AND
2534 * - the bond has an IP address
2536 * Note: a non-null current_arp_slave indicates
2537 * the curr_active_slave went down and we are
2538 * searching for a new one; under this condition
2539 * we only take the curr_active_slave down - this
2540 * gives each slave a chance to tx/rx traffic
2541 * before being taken out
2543 if (!bond_is_active_slave(slave
) &&
2544 !bond
->current_arp_slave
&&
2545 !bond_time_in_interval(bond
, last_rx
, 3)) {
2546 slave
->new_link
= BOND_LINK_DOWN
;
2551 * Active slave is down if:
2552 * - more than 2*delta since transmitting OR
2553 * - (more than 2*delta since receive AND
2554 * the bond has an IP address)
2556 trans_start
= dev_trans_start(slave
->dev
);
2557 if (bond_is_active_slave(slave
) &&
2558 (!bond_time_in_interval(bond
, trans_start
, 2) ||
2559 !bond_time_in_interval(bond
, last_rx
, 2))) {
2560 slave
->new_link
= BOND_LINK_DOWN
;
2569 * Called to commit link state changes noted by inspection step of
2570 * active-backup mode ARP monitor.
2572 * Called with RTNL and bond->lock for read.
2574 static void bond_ab_arp_commit(struct bonding
*bond
)
2576 unsigned long trans_start
;
2577 struct list_head
*iter
;
2578 struct slave
*slave
;
2580 bond_for_each_slave(bond
, slave
, iter
) {
2581 switch (slave
->new_link
) {
2582 case BOND_LINK_NOCHANGE
:
2586 trans_start
= dev_trans_start(slave
->dev
);
2587 if (bond
->curr_active_slave
!= slave
||
2588 (!bond
->curr_active_slave
&&
2589 bond_time_in_interval(bond
, trans_start
, 1))) {
2590 slave
->link
= BOND_LINK_UP
;
2591 if (bond
->current_arp_slave
) {
2592 bond_set_slave_inactive_flags(
2593 bond
->current_arp_slave
);
2594 bond
->current_arp_slave
= NULL
;
2597 pr_info("%s: link status definitely up for interface %s.\n",
2598 bond
->dev
->name
, slave
->dev
->name
);
2600 if (!bond
->curr_active_slave
||
2601 (slave
== bond
->primary_slave
))
2608 case BOND_LINK_DOWN
:
2609 if (slave
->link_failure_count
< UINT_MAX
)
2610 slave
->link_failure_count
++;
2612 slave
->link
= BOND_LINK_DOWN
;
2613 bond_set_slave_inactive_flags(slave
);
2615 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2616 bond
->dev
->name
, slave
->dev
->name
);
2618 if (slave
== bond
->curr_active_slave
) {
2619 bond
->current_arp_slave
= NULL
;
2626 pr_err("%s: impossible: new_link %d on slave %s\n",
2627 bond
->dev
->name
, slave
->new_link
,
2635 write_lock_bh(&bond
->curr_slave_lock
);
2636 bond_select_active_slave(bond
);
2637 write_unlock_bh(&bond
->curr_slave_lock
);
2638 unblock_netpoll_tx();
2641 bond_set_carrier(bond
);
2645 * Send ARP probes for active-backup mode ARP monitor.
2647 * Called with bond->lock held for read.
2649 static void bond_ab_arp_probe(struct bonding
*bond
)
2651 struct slave
*slave
, *before
= NULL
, *new_slave
= NULL
;
2652 struct list_head
*iter
;
2655 read_lock(&bond
->curr_slave_lock
);
2657 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
2658 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2659 bond
->current_arp_slave
->dev
->name
,
2660 bond
->curr_active_slave
->dev
->name
);
2662 if (bond
->curr_active_slave
) {
2663 bond_arp_send_all(bond
, bond
->curr_active_slave
);
2664 read_unlock(&bond
->curr_slave_lock
);
2668 read_unlock(&bond
->curr_slave_lock
);
2670 /* if we don't have a curr_active_slave, search for the next available
2671 * backup slave from the current_arp_slave and make it the candidate
2672 * for becoming the curr_active_slave
2675 if (!bond
->current_arp_slave
) {
2676 bond
->current_arp_slave
= bond_first_slave(bond
);
2677 if (!bond
->current_arp_slave
)
2681 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
2683 bond_for_each_slave(bond
, slave
, iter
) {
2684 if (!found
&& !before
&& IS_UP(slave
->dev
))
2687 if (found
&& !new_slave
&& IS_UP(slave
->dev
))
2689 /* if the link state is up at this point, we
2690 * mark it down - this can happen if we have
2691 * simultaneous link failures and
2692 * reselect_active_interface doesn't make this
2693 * one the current slave so it is still marked
2694 * up when it is actually down
2696 if (!IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
) {
2697 slave
->link
= BOND_LINK_DOWN
;
2698 if (slave
->link_failure_count
< UINT_MAX
)
2699 slave
->link_failure_count
++;
2701 bond_set_slave_inactive_flags(slave
);
2703 pr_info("%s: backup interface %s is now down.\n",
2704 bond
->dev
->name
, slave
->dev
->name
);
2706 if (slave
== bond
->current_arp_slave
)
2710 if (!new_slave
&& before
)
2716 new_slave
->link
= BOND_LINK_BACK
;
2717 bond_set_slave_active_flags(new_slave
);
2718 bond_arp_send_all(bond
, new_slave
);
2719 new_slave
->jiffies
= jiffies
;
2720 bond
->current_arp_slave
= new_slave
;
2724 void bond_activebackup_arp_mon(struct work_struct
*work
)
2726 struct bonding
*bond
= container_of(work
, struct bonding
,
2728 bool should_notify_peers
= false;
2731 read_lock(&bond
->lock
);
2733 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2735 if (!bond_has_slaves(bond
))
2738 should_notify_peers
= bond_should_notify_peers(bond
);
2740 if (bond_ab_arp_inspect(bond
)) {
2741 read_unlock(&bond
->lock
);
2743 /* Race avoidance with bond_close flush of workqueue */
2744 if (!rtnl_trylock()) {
2745 read_lock(&bond
->lock
);
2747 should_notify_peers
= false;
2751 read_lock(&bond
->lock
);
2753 bond_ab_arp_commit(bond
);
2755 read_unlock(&bond
->lock
);
2757 read_lock(&bond
->lock
);
2760 bond_ab_arp_probe(bond
);
2763 if (bond
->params
.arp_interval
)
2764 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2766 read_unlock(&bond
->lock
);
2768 if (should_notify_peers
) {
2769 if (!rtnl_trylock())
2771 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2776 /*-------------------------- netdev event handling --------------------------*/
2779 * Change device name
2781 static int bond_event_changename(struct bonding
*bond
)
2783 bond_remove_proc_entry(bond
);
2784 bond_create_proc_entry(bond
);
2786 bond_debug_reregister(bond
);
2791 static int bond_master_netdev_event(unsigned long event
,
2792 struct net_device
*bond_dev
)
2794 struct bonding
*event_bond
= netdev_priv(bond_dev
);
2797 case NETDEV_CHANGENAME
:
2798 return bond_event_changename(event_bond
);
2799 case NETDEV_UNREGISTER
:
2800 bond_remove_proc_entry(event_bond
);
2802 case NETDEV_REGISTER
:
2803 bond_create_proc_entry(event_bond
);
2805 case NETDEV_NOTIFY_PEERS
:
2806 if (event_bond
->send_peer_notif
)
2807 event_bond
->send_peer_notif
--;
2816 static int bond_slave_netdev_event(unsigned long event
,
2817 struct net_device
*slave_dev
)
2819 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
);
2820 struct bonding
*bond
;
2821 struct net_device
*bond_dev
;
2825 /* A netdev event can be generated while enslaving a device
2826 * before netdev_rx_handler_register is called in which case
2827 * slave will be NULL
2831 bond_dev
= slave
->bond
->dev
;
2835 case NETDEV_UNREGISTER
:
2836 if (bond_dev
->type
!= ARPHRD_ETHER
)
2837 bond_release_and_destroy(bond_dev
, slave_dev
);
2839 bond_release(bond_dev
, slave_dev
);
2843 old_speed
= slave
->speed
;
2844 old_duplex
= slave
->duplex
;
2846 bond_update_speed_duplex(slave
);
2848 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2849 if (old_speed
!= slave
->speed
)
2850 bond_3ad_adapter_speed_changed(slave
);
2851 if (old_duplex
!= slave
->duplex
)
2852 bond_3ad_adapter_duplex_changed(slave
);
2857 * ... Or is it this?
2860 case NETDEV_CHANGEMTU
:
2862 * TODO: Should slaves be allowed to
2863 * independently alter their MTU? For
2864 * an active-backup bond, slaves need
2865 * not be the same type of device, so
2866 * MTUs may vary. For other modes,
2867 * slaves arguably should have the
2868 * same MTUs. To do this, we'd need to
2869 * take over the slave's change_mtu
2870 * function for the duration of their
2874 case NETDEV_CHANGENAME
:
2876 * TODO: handle changing the primary's name
2879 case NETDEV_FEAT_CHANGE
:
2880 bond_compute_features(bond
);
2882 case NETDEV_RESEND_IGMP
:
2883 /* Propagate to master device */
2884 call_netdevice_notifiers(event
, slave
->bond
->dev
);
2894 * bond_netdev_event: handle netdev notifier chain events.
2896 * This function receives events for the netdev chain. The caller (an
2897 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2898 * locks for us to safely manipulate the slave devices (RTNL lock,
2901 static int bond_netdev_event(struct notifier_block
*this,
2902 unsigned long event
, void *ptr
)
2904 struct net_device
*event_dev
= netdev_notifier_info_to_dev(ptr
);
2906 pr_debug("event_dev: %s, event: %lx\n",
2907 event_dev
? event_dev
->name
: "None",
2910 if (!(event_dev
->priv_flags
& IFF_BONDING
))
2913 if (event_dev
->flags
& IFF_MASTER
) {
2914 pr_debug("IFF_MASTER\n");
2915 return bond_master_netdev_event(event
, event_dev
);
2918 if (event_dev
->flags
& IFF_SLAVE
) {
2919 pr_debug("IFF_SLAVE\n");
2920 return bond_slave_netdev_event(event
, event_dev
);
2926 static struct notifier_block bond_netdev_notifier
= {
2927 .notifier_call
= bond_netdev_event
,
2930 /*---------------------------- Hashing Policies -----------------------------*/
2932 /* L2 hash helper */
2933 static inline u32
bond_eth_hash(struct sk_buff
*skb
)
2935 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
2937 if (skb_headlen(skb
) >= offsetof(struct ethhdr
, h_proto
))
2938 return data
->h_dest
[5] ^ data
->h_source
[5];
2943 /* Extract the appropriate headers based on bond's xmit policy */
2944 static bool bond_flow_dissect(struct bonding
*bond
, struct sk_buff
*skb
,
2945 struct flow_keys
*fk
)
2947 const struct ipv6hdr
*iph6
;
2948 const struct iphdr
*iph
;
2949 int noff
, proto
= -1;
2951 if (bond
->params
.xmit_policy
> BOND_XMIT_POLICY_LAYER23
)
2952 return skb_flow_dissect(skb
, fk
);
2955 noff
= skb_network_offset(skb
);
2956 if (skb
->protocol
== htons(ETH_P_IP
)) {
2957 if (!pskb_may_pull(skb
, noff
+ sizeof(*iph
)))
2960 fk
->src
= iph
->saddr
;
2961 fk
->dst
= iph
->daddr
;
2962 noff
+= iph
->ihl
<< 2;
2963 if (!ip_is_fragment(iph
))
2964 proto
= iph
->protocol
;
2965 } else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
2966 if (!pskb_may_pull(skb
, noff
+ sizeof(*iph6
)))
2968 iph6
= ipv6_hdr(skb
);
2969 fk
->src
= (__force __be32
)ipv6_addr_hash(&iph6
->saddr
);
2970 fk
->dst
= (__force __be32
)ipv6_addr_hash(&iph6
->daddr
);
2971 noff
+= sizeof(*iph6
);
2972 proto
= iph6
->nexthdr
;
2976 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
&& proto
>= 0)
2977 fk
->ports
= skb_flow_get_ports(skb
, noff
, proto
);
2983 * bond_xmit_hash - generate a hash value based on the xmit policy
2984 * @bond: bonding device
2985 * @skb: buffer to use for headers
2986 * @count: modulo value
2988 * This function will extract the necessary headers from the skb buffer and use
2989 * them to generate a hash based on the xmit_policy set in the bonding device
2990 * which will be reduced modulo count before returning.
2992 int bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
, int count
)
2994 struct flow_keys flow
;
2997 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER2
||
2998 !bond_flow_dissect(bond
, skb
, &flow
))
2999 return bond_eth_hash(skb
) % count
;
3001 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER23
||
3002 bond
->params
.xmit_policy
== BOND_XMIT_POLICY_ENCAP23
)
3003 hash
= bond_eth_hash(skb
);
3005 hash
= (__force u32
)flow
.ports
;
3006 hash
^= (__force u32
)flow
.dst
^ (__force u32
)flow
.src
;
3007 hash
^= (hash
>> 16);
3008 hash
^= (hash
>> 8);
3010 return hash
% count
;
3013 /*-------------------------- Device entry points ----------------------------*/
3015 static void bond_work_init_all(struct bonding
*bond
)
3017 INIT_DELAYED_WORK(&bond
->mcast_work
,
3018 bond_resend_igmp_join_requests_delayed
);
3019 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3020 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3021 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3022 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3024 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3025 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3028 static void bond_work_cancel_all(struct bonding
*bond
)
3030 cancel_delayed_work_sync(&bond
->mii_work
);
3031 cancel_delayed_work_sync(&bond
->arp_work
);
3032 cancel_delayed_work_sync(&bond
->alb_work
);
3033 cancel_delayed_work_sync(&bond
->ad_work
);
3034 cancel_delayed_work_sync(&bond
->mcast_work
);
3037 static int bond_open(struct net_device
*bond_dev
)
3039 struct bonding
*bond
= netdev_priv(bond_dev
);
3040 struct list_head
*iter
;
3041 struct slave
*slave
;
3043 /* reset slave->backup and slave->inactive */
3044 read_lock(&bond
->lock
);
3045 if (bond_has_slaves(bond
)) {
3046 read_lock(&bond
->curr_slave_lock
);
3047 bond_for_each_slave(bond
, slave
, iter
) {
3048 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3049 && (slave
!= bond
->curr_active_slave
)) {
3050 bond_set_slave_inactive_flags(slave
);
3052 bond_set_slave_active_flags(slave
);
3055 read_unlock(&bond
->curr_slave_lock
);
3057 read_unlock(&bond
->lock
);
3059 bond_work_init_all(bond
);
3061 if (bond_is_lb(bond
)) {
3062 /* bond_alb_initialize must be called before the timer
3065 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
)))
3067 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3070 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3071 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3073 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3074 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3075 if (bond
->params
.arp_validate
)
3076 bond
->recv_probe
= bond_arp_rcv
;
3079 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3080 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3081 /* register to receive LACPDUs */
3082 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3083 bond_3ad_initiate_agg_selection(bond
, 1);
3089 static int bond_close(struct net_device
*bond_dev
)
3091 struct bonding
*bond
= netdev_priv(bond_dev
);
3093 bond_work_cancel_all(bond
);
3094 bond
->send_peer_notif
= 0;
3095 if (bond_is_lb(bond
))
3096 bond_alb_deinitialize(bond
);
3097 bond
->recv_probe
= NULL
;
3102 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3103 struct rtnl_link_stats64
*stats
)
3105 struct bonding
*bond
= netdev_priv(bond_dev
);
3106 struct rtnl_link_stats64 temp
;
3107 struct list_head
*iter
;
3108 struct slave
*slave
;
3110 memset(stats
, 0, sizeof(*stats
));
3112 read_lock_bh(&bond
->lock
);
3113 bond_for_each_slave(bond
, slave
, iter
) {
3114 const struct rtnl_link_stats64
*sstats
=
3115 dev_get_stats(slave
->dev
, &temp
);
3117 stats
->rx_packets
+= sstats
->rx_packets
;
3118 stats
->rx_bytes
+= sstats
->rx_bytes
;
3119 stats
->rx_errors
+= sstats
->rx_errors
;
3120 stats
->rx_dropped
+= sstats
->rx_dropped
;
3122 stats
->tx_packets
+= sstats
->tx_packets
;
3123 stats
->tx_bytes
+= sstats
->tx_bytes
;
3124 stats
->tx_errors
+= sstats
->tx_errors
;
3125 stats
->tx_dropped
+= sstats
->tx_dropped
;
3127 stats
->multicast
+= sstats
->multicast
;
3128 stats
->collisions
+= sstats
->collisions
;
3130 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3131 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3132 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3133 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3134 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3135 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3137 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3138 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3139 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3140 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3141 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3143 read_unlock_bh(&bond
->lock
);
3148 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3150 struct bonding
*bond
= netdev_priv(bond_dev
);
3151 struct net_device
*slave_dev
= NULL
;
3152 struct ifbond k_binfo
;
3153 struct ifbond __user
*u_binfo
= NULL
;
3154 struct ifslave k_sinfo
;
3155 struct ifslave __user
*u_sinfo
= NULL
;
3156 struct mii_ioctl_data
*mii
= NULL
;
3160 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3172 * We do this again just in case we were called by SIOCGMIIREG
3173 * instead of SIOCGMIIPHY.
3180 if (mii
->reg_num
== 1) {
3182 read_lock(&bond
->lock
);
3183 read_lock(&bond
->curr_slave_lock
);
3184 if (netif_carrier_ok(bond
->dev
))
3185 mii
->val_out
= BMSR_LSTATUS
;
3187 read_unlock(&bond
->curr_slave_lock
);
3188 read_unlock(&bond
->lock
);
3192 case BOND_INFO_QUERY_OLD
:
3193 case SIOCBONDINFOQUERY
:
3194 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3196 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3199 res
= bond_info_query(bond_dev
, &k_binfo
);
3201 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3205 case BOND_SLAVE_INFO_QUERY_OLD
:
3206 case SIOCBONDSLAVEINFOQUERY
:
3207 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3209 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3212 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3214 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3223 net
= dev_net(bond_dev
);
3225 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3228 slave_dev
= dev_get_by_name(net
, ifr
->ifr_slave
);
3230 pr_debug("slave_dev=%p:\n", slave_dev
);
3235 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3237 case BOND_ENSLAVE_OLD
:
3238 case SIOCBONDENSLAVE
:
3239 res
= bond_enslave(bond_dev
, slave_dev
);
3241 case BOND_RELEASE_OLD
:
3242 case SIOCBONDRELEASE
:
3243 res
= bond_release(bond_dev
, slave_dev
);
3245 case BOND_SETHWADDR_OLD
:
3246 case SIOCBONDSETHWADDR
:
3247 bond_set_dev_addr(bond_dev
, slave_dev
);
3250 case BOND_CHANGE_ACTIVE_OLD
:
3251 case SIOCBONDCHANGEACTIVE
:
3252 res
= bond_option_active_slave_set(bond
, slave_dev
);
3264 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3266 struct bonding
*bond
= netdev_priv(bond_dev
);
3268 if (change
& IFF_PROMISC
)
3269 bond_set_promiscuity(bond
,
3270 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3272 if (change
& IFF_ALLMULTI
)
3273 bond_set_allmulti(bond
,
3274 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3277 static void bond_set_rx_mode(struct net_device
*bond_dev
)
3279 struct bonding
*bond
= netdev_priv(bond_dev
);
3280 struct list_head
*iter
;
3281 struct slave
*slave
;
3285 if (USES_PRIMARY(bond
->params
.mode
)) {
3286 slave
= rcu_dereference(bond
->curr_active_slave
);
3288 dev_uc_sync(slave
->dev
, bond_dev
);
3289 dev_mc_sync(slave
->dev
, bond_dev
);
3292 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3293 dev_uc_sync_multiple(slave
->dev
, bond_dev
);
3294 dev_mc_sync_multiple(slave
->dev
, bond_dev
);
3300 static int bond_neigh_init(struct neighbour
*n
)
3302 struct bonding
*bond
= netdev_priv(n
->dev
);
3303 const struct net_device_ops
*slave_ops
;
3304 struct neigh_parms parms
;
3305 struct slave
*slave
;
3308 slave
= bond_first_slave(bond
);
3311 slave_ops
= slave
->dev
->netdev_ops
;
3312 if (!slave_ops
->ndo_neigh_setup
)
3315 parms
.neigh_setup
= NULL
;
3316 parms
.neigh_cleanup
= NULL
;
3317 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3322 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3323 * after the last slave has been detached. Assumes that all slaves
3324 * utilize the same neigh_cleanup (true at this writing as only user
3327 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3329 if (!parms
.neigh_setup
)
3332 return parms
.neigh_setup(n
);
3336 * The bonding ndo_neigh_setup is called at init time beofre any
3337 * slave exists. So we must declare proxy setup function which will
3338 * be used at run time to resolve the actual slave neigh param setup.
3340 * It's also called by master devices (such as vlans) to setup their
3341 * underlying devices. In that case - do nothing, we're already set up from
3344 static int bond_neigh_setup(struct net_device
*dev
,
3345 struct neigh_parms
*parms
)
3347 /* modify only our neigh_parms */
3348 if (parms
->dev
== dev
)
3349 parms
->neigh_setup
= bond_neigh_init
;
3355 * Change the MTU of all of a master's slaves to match the master
3357 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3359 struct bonding
*bond
= netdev_priv(bond_dev
);
3360 struct slave
*slave
, *rollback_slave
;
3361 struct list_head
*iter
;
3364 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3365 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3367 /* Can't hold bond->lock with bh disabled here since
3368 * some base drivers panic. On the other hand we can't
3369 * hold bond->lock without bh disabled because we'll
3370 * deadlock. The only solution is to rely on the fact
3371 * that we're under rtnl_lock here, and the slaves
3372 * list won't change. This doesn't solve the problem
3373 * of setting the slave's MTU while it is
3374 * transmitting, but the assumption is that the base
3375 * driver can handle that.
3377 * TODO: figure out a way to safely iterate the slaves
3378 * list, but without holding a lock around the actual
3379 * call to the base driver.
3382 bond_for_each_slave(bond
, slave
, iter
) {
3383 pr_debug("s %p c_m %p\n",
3385 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3387 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3390 /* If we failed to set the slave's mtu to the new value
3391 * we must abort the operation even in ACTIVE_BACKUP
3392 * mode, because if we allow the backup slaves to have
3393 * different mtu values than the active slave we'll
3394 * need to change their mtu when doing a failover. That
3395 * means changing their mtu from timer context, which
3396 * is probably not a good idea.
3398 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3403 bond_dev
->mtu
= new_mtu
;
3408 /* unwind from head to the slave that failed */
3409 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3412 if (rollback_slave
== slave
)
3415 tmp_res
= dev_set_mtu(rollback_slave
->dev
, bond_dev
->mtu
);
3417 pr_debug("unwind err %d dev %s\n",
3418 tmp_res
, rollback_slave
->dev
->name
);
3428 * Note that many devices must be down to change the HW address, and
3429 * downing the master releases all slaves. We can make bonds full of
3430 * bonding devices to test this, however.
3432 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3434 struct bonding
*bond
= netdev_priv(bond_dev
);
3435 struct slave
*slave
, *rollback_slave
;
3436 struct sockaddr
*sa
= addr
, tmp_sa
;
3437 struct list_head
*iter
;
3440 if (bond
->params
.mode
== BOND_MODE_ALB
)
3441 return bond_alb_set_mac_address(bond_dev
, addr
);
3444 pr_debug("bond=%p, name=%s\n",
3445 bond
, bond_dev
? bond_dev
->name
: "None");
3447 /* If fail_over_mac is enabled, do nothing and return success.
3448 * Returning an error causes ifenslave to fail.
3450 if (bond
->params
.fail_over_mac
)
3453 if (!is_valid_ether_addr(sa
->sa_data
))
3454 return -EADDRNOTAVAIL
;
3456 /* Can't hold bond->lock with bh disabled here since
3457 * some base drivers panic. On the other hand we can't
3458 * hold bond->lock without bh disabled because we'll
3459 * deadlock. The only solution is to rely on the fact
3460 * that we're under rtnl_lock here, and the slaves
3461 * list won't change. This doesn't solve the problem
3462 * of setting the slave's hw address while it is
3463 * transmitting, but the assumption is that the base
3464 * driver can handle that.
3466 * TODO: figure out a way to safely iterate the slaves
3467 * list, but without holding a lock around the actual
3468 * call to the base driver.
3471 bond_for_each_slave(bond
, slave
, iter
) {
3472 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3473 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3475 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3477 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3481 res
= dev_set_mac_address(slave
->dev
, addr
);
3483 /* TODO: consider downing the slave
3485 * User should expect communications
3486 * breakage anyway until ARP finish
3489 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3495 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3499 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3500 tmp_sa
.sa_family
= bond_dev
->type
;
3502 /* unwind from head to the slave that failed */
3503 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3506 if (rollback_slave
== slave
)
3509 tmp_res
= dev_set_mac_address(rollback_slave
->dev
, &tmp_sa
);
3511 pr_debug("unwind err %d dev %s\n",
3512 tmp_res
, rollback_slave
->dev
->name
);
3520 * bond_xmit_slave_id - transmit skb through slave with slave_id
3521 * @bond: bonding device that is transmitting
3522 * @skb: buffer to transmit
3523 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3525 * This function tries to transmit through slave with slave_id but in case
3526 * it fails, it tries to find the first available slave for transmission.
3527 * The skb is consumed in all cases, thus the function is void.
3529 void bond_xmit_slave_id(struct bonding
*bond
, struct sk_buff
*skb
, int slave_id
)
3531 struct list_head
*iter
;
3532 struct slave
*slave
;
3535 /* Here we start from the slave with slave_id */
3536 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3538 if (slave_can_tx(slave
)) {
3539 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3545 /* Here we start from the first slave up to slave_id */
3547 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3550 if (slave_can_tx(slave
)) {
3551 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3555 /* no slave that can tx has been found */
3559 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3561 struct bonding
*bond
= netdev_priv(bond_dev
);
3562 struct iphdr
*iph
= ip_hdr(skb
);
3563 struct slave
*slave
;
3566 * Start with the curr_active_slave that joined the bond as the
3567 * default for sending IGMP traffic. For failover purposes one
3568 * needs to maintain some consistency for the interface that will
3569 * send the join/membership reports. The curr_active_slave found
3570 * will send all of this type of traffic.
3572 if (iph
->protocol
== IPPROTO_IGMP
&& skb
->protocol
== htons(ETH_P_IP
)) {
3573 slave
= rcu_dereference(bond
->curr_active_slave
);
3574 if (slave
&& slave_can_tx(slave
))
3575 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3577 bond_xmit_slave_id(bond
, skb
, 0);
3579 bond_xmit_slave_id(bond
, skb
,
3580 bond
->rr_tx_counter
++ % bond
->slave_cnt
);
3583 return NETDEV_TX_OK
;
3587 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3588 * the bond has a usable interface.
3590 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3592 struct bonding
*bond
= netdev_priv(bond_dev
);
3593 struct slave
*slave
;
3595 slave
= rcu_dereference(bond
->curr_active_slave
);
3597 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3601 return NETDEV_TX_OK
;
3604 /* In bond_xmit_xor() , we determine the output device by using a pre-
3605 * determined xmit_hash_policy(), If the selected device is not enabled,
3606 * find the next active slave.
3608 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3610 struct bonding
*bond
= netdev_priv(bond_dev
);
3612 bond_xmit_slave_id(bond
, skb
, bond_xmit_hash(bond
, skb
, bond
->slave_cnt
));
3614 return NETDEV_TX_OK
;
3617 /* in broadcast mode, we send everything to all usable interfaces. */
3618 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3620 struct bonding
*bond
= netdev_priv(bond_dev
);
3621 struct slave
*slave
= NULL
;
3622 struct list_head
*iter
;
3624 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3625 if (bond_is_last_slave(bond
, slave
))
3627 if (IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
) {
3628 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
3631 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3635 /* bond_dev_queue_xmit always returns 0 */
3636 bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3639 if (slave
&& IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
)
3640 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3644 return NETDEV_TX_OK
;
3647 /*------------------------- Device initialization ---------------------------*/
3650 * Lookup the slave that corresponds to a qid
3652 static inline int bond_slave_override(struct bonding
*bond
,
3653 struct sk_buff
*skb
)
3655 struct slave
*slave
= NULL
;
3656 struct slave
*check_slave
;
3657 struct list_head
*iter
;
3660 if (!skb
->queue_mapping
)
3663 /* Find out if any slaves have the same mapping as this skb. */
3664 bond_for_each_slave_rcu(bond
, check_slave
, iter
) {
3665 if (check_slave
->queue_id
== skb
->queue_mapping
) {
3666 slave
= check_slave
;
3671 /* If the slave isn't UP, use default transmit policy. */
3672 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
3673 (slave
->link
== BOND_LINK_UP
)) {
3674 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3681 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
3684 * This helper function exists to help dev_pick_tx get the correct
3685 * destination queue. Using a helper function skips a call to
3686 * skb_tx_hash and will put the skbs in the queue we expect on their
3687 * way down to the bonding driver.
3689 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
3692 * Save the original txq to restore before passing to the driver
3694 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
3696 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
3698 txq
-= dev
->real_num_tx_queues
;
3699 } while (txq
>= dev
->real_num_tx_queues
);
3704 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3706 struct bonding
*bond
= netdev_priv(dev
);
3708 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
3709 if (!bond_slave_override(bond
, skb
))
3710 return NETDEV_TX_OK
;
3713 switch (bond
->params
.mode
) {
3714 case BOND_MODE_ROUNDROBIN
:
3715 return bond_xmit_roundrobin(skb
, dev
);
3716 case BOND_MODE_ACTIVEBACKUP
:
3717 return bond_xmit_activebackup(skb
, dev
);
3719 return bond_xmit_xor(skb
, dev
);
3720 case BOND_MODE_BROADCAST
:
3721 return bond_xmit_broadcast(skb
, dev
);
3722 case BOND_MODE_8023AD
:
3723 return bond_3ad_xmit_xor(skb
, dev
);
3726 return bond_alb_xmit(skb
, dev
);
3728 /* Should never happen, mode already checked */
3729 pr_err("%s: Error: Unknown bonding mode %d\n",
3730 dev
->name
, bond
->params
.mode
);
3733 return NETDEV_TX_OK
;
3737 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3739 struct bonding
*bond
= netdev_priv(dev
);
3740 netdev_tx_t ret
= NETDEV_TX_OK
;
3743 * If we risk deadlock from transmitting this in the
3744 * netpoll path, tell netpoll to queue the frame for later tx
3746 if (is_netpoll_tx_blocked(dev
))
3747 return NETDEV_TX_BUSY
;
3750 if (bond_has_slaves(bond
))
3751 ret
= __bond_start_xmit(skb
, dev
);
3759 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
3760 struct ethtool_cmd
*ecmd
)
3762 struct bonding
*bond
= netdev_priv(bond_dev
);
3763 unsigned long speed
= 0;
3764 struct list_head
*iter
;
3765 struct slave
*slave
;
3767 ecmd
->duplex
= DUPLEX_UNKNOWN
;
3768 ecmd
->port
= PORT_OTHER
;
3770 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3771 * do not need to check mode. Though link speed might not represent
3772 * the true receive or transmit bandwidth (not all modes are symmetric)
3773 * this is an accurate maximum.
3775 read_lock(&bond
->lock
);
3776 bond_for_each_slave(bond
, slave
, iter
) {
3777 if (SLAVE_IS_OK(slave
)) {
3778 if (slave
->speed
!= SPEED_UNKNOWN
)
3779 speed
+= slave
->speed
;
3780 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
3781 slave
->duplex
!= DUPLEX_UNKNOWN
)
3782 ecmd
->duplex
= slave
->duplex
;
3785 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
3786 read_unlock(&bond
->lock
);
3791 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
3792 struct ethtool_drvinfo
*drvinfo
)
3794 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
3795 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
3796 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
3800 static const struct ethtool_ops bond_ethtool_ops
= {
3801 .get_drvinfo
= bond_ethtool_get_drvinfo
,
3802 .get_settings
= bond_ethtool_get_settings
,
3803 .get_link
= ethtool_op_get_link
,
3806 static const struct net_device_ops bond_netdev_ops
= {
3807 .ndo_init
= bond_init
,
3808 .ndo_uninit
= bond_uninit
,
3809 .ndo_open
= bond_open
,
3810 .ndo_stop
= bond_close
,
3811 .ndo_start_xmit
= bond_start_xmit
,
3812 .ndo_select_queue
= bond_select_queue
,
3813 .ndo_get_stats64
= bond_get_stats
,
3814 .ndo_do_ioctl
= bond_do_ioctl
,
3815 .ndo_change_rx_flags
= bond_change_rx_flags
,
3816 .ndo_set_rx_mode
= bond_set_rx_mode
,
3817 .ndo_change_mtu
= bond_change_mtu
,
3818 .ndo_set_mac_address
= bond_set_mac_address
,
3819 .ndo_neigh_setup
= bond_neigh_setup
,
3820 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
3821 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
3822 #ifdef CONFIG_NET_POLL_CONTROLLER
3823 .ndo_netpoll_setup
= bond_netpoll_setup
,
3824 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
3825 .ndo_poll_controller
= bond_poll_controller
,
3827 .ndo_add_slave
= bond_enslave
,
3828 .ndo_del_slave
= bond_release
,
3829 .ndo_fix_features
= bond_fix_features
,
3832 static const struct device_type bond_type
= {
3836 static void bond_destructor(struct net_device
*bond_dev
)
3838 struct bonding
*bond
= netdev_priv(bond_dev
);
3840 destroy_workqueue(bond
->wq
);
3841 free_netdev(bond_dev
);
3844 void bond_setup(struct net_device
*bond_dev
)
3846 struct bonding
*bond
= netdev_priv(bond_dev
);
3848 /* initialize rwlocks */
3849 rwlock_init(&bond
->lock
);
3850 rwlock_init(&bond
->curr_slave_lock
);
3851 bond
->params
= bonding_defaults
;
3853 /* Initialize pointers */
3854 bond
->dev
= bond_dev
;
3856 /* Initialize the device entry points */
3857 ether_setup(bond_dev
);
3858 bond_dev
->netdev_ops
= &bond_netdev_ops
;
3859 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
3861 bond_dev
->destructor
= bond_destructor
;
3863 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
3865 /* Initialize the device options */
3866 bond_dev
->tx_queue_len
= 0;
3867 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
3868 bond_dev
->priv_flags
|= IFF_BONDING
;
3869 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
3871 /* At first, we block adding VLANs. That's the only way to
3872 * prevent problems that occur when adding VLANs over an
3873 * empty bond. The block will be removed once non-challenged
3874 * slaves are enslaved.
3876 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
3878 /* don't acquire bond device's netif_tx_lock when
3880 bond_dev
->features
|= NETIF_F_LLTX
;
3882 /* By default, we declare the bond to be fully
3883 * VLAN hardware accelerated capable. Special
3884 * care is taken in the various xmit functions
3885 * when there are slaves that are not hw accel
3889 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
3890 NETIF_F_HW_VLAN_CTAG_TX
|
3891 NETIF_F_HW_VLAN_CTAG_RX
|
3892 NETIF_F_HW_VLAN_CTAG_FILTER
;
3894 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
3895 bond_dev
->features
|= bond_dev
->hw_features
;
3899 * Destroy a bonding device.
3900 * Must be under rtnl_lock when this function is called.
3902 static void bond_uninit(struct net_device
*bond_dev
)
3904 struct bonding
*bond
= netdev_priv(bond_dev
);
3905 struct list_head
*iter
;
3906 struct slave
*slave
;
3908 bond_netpoll_cleanup(bond_dev
);
3910 /* Release the bonded slaves */
3911 bond_for_each_slave(bond
, slave
, iter
)
3912 __bond_release_one(bond_dev
, slave
->dev
, true);
3913 pr_info("%s: released all slaves\n", bond_dev
->name
);
3915 list_del(&bond
->bond_list
);
3917 bond_debug_unregister(bond
);
3920 /*------------------------- Module initialization ---------------------------*/
3923 * Convert string input module parms. Accept either the
3924 * number of the mode or its string name. A bit complicated because
3925 * some mode names are substrings of other names, and calls from sysfs
3926 * may have whitespace in the name (trailing newlines, for example).
3928 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
3930 int modeint
= -1, i
, rv
;
3931 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
3933 for (p
= (char *)buf
; *p
; p
++)
3934 if (!(isdigit(*p
) || isspace(*p
)))
3938 rv
= sscanf(buf
, "%20s", modestr
);
3940 rv
= sscanf(buf
, "%d", &modeint
);
3945 for (i
= 0; tbl
[i
].modename
; i
++) {
3946 if (modeint
== tbl
[i
].mode
)
3948 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
3955 static int bond_check_params(struct bond_params
*params
)
3957 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
3958 int arp_all_targets_value
;
3961 * Convert string parameters.
3964 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
3965 if (bond_mode
== -1) {
3966 pr_err("Error: Invalid bonding mode \"%s\"\n",
3967 mode
== NULL
? "NULL" : mode
);
3972 if (xmit_hash_policy
) {
3973 if ((bond_mode
!= BOND_MODE_XOR
) &&
3974 (bond_mode
!= BOND_MODE_8023AD
)) {
3975 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3976 bond_mode_name(bond_mode
));
3978 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
3980 if (xmit_hashtype
== -1) {
3981 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
3982 xmit_hash_policy
== NULL
? "NULL" :
3990 if (bond_mode
!= BOND_MODE_8023AD
) {
3991 pr_info("lacp_rate param is irrelevant in mode %s\n",
3992 bond_mode_name(bond_mode
));
3994 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
3995 if (lacp_fast
== -1) {
3996 pr_err("Error: Invalid lacp rate \"%s\"\n",
3997 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4004 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4005 if (params
->ad_select
== -1) {
4006 pr_err("Error: Invalid ad_select \"%s\"\n",
4007 ad_select
== NULL
? "NULL" : ad_select
);
4011 if (bond_mode
!= BOND_MODE_8023AD
) {
4012 pr_warning("ad_select param only affects 802.3ad mode\n");
4015 params
->ad_select
= BOND_AD_STABLE
;
4018 if (max_bonds
< 0) {
4019 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4020 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4021 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4025 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4026 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4027 miimon
= BOND_LINK_MON_INTERV
;
4031 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4036 if (downdelay
< 0) {
4037 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4038 downdelay
, INT_MAX
);
4042 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4043 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4048 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4049 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4054 /* reset values for 802.3ad */
4055 if (bond_mode
== BOND_MODE_8023AD
) {
4057 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");
4058 pr_warning("Forcing miimon to 100msec\n");
4063 if (tx_queues
< 1 || tx_queues
> 255) {
4064 pr_warning("Warning: tx_queues (%d) should be between "
4065 "1 and 255, resetting to %d\n",
4066 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4067 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4070 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4071 pr_warning("Warning: all_slaves_active module parameter (%d), "
4072 "not of valid value (0/1), so it was set to "
4073 "0\n", all_slaves_active
);
4074 all_slaves_active
= 0;
4077 if (resend_igmp
< 0 || resend_igmp
> 255) {
4078 pr_warning("Warning: resend_igmp (%d) should be between "
4079 "0 and 255, resetting to %d\n",
4080 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4081 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4084 /* reset values for TLB/ALB */
4085 if ((bond_mode
== BOND_MODE_TLB
) ||
4086 (bond_mode
== BOND_MODE_ALB
)) {
4088 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4089 pr_warning("Forcing miimon to 100msec\n");
4094 if (bond_mode
== BOND_MODE_ALB
) {
4095 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",
4100 if (updelay
|| downdelay
) {
4101 /* just warn the user the up/down delay will have
4102 * no effect since miimon is zero...
4104 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",
4105 updelay
, downdelay
);
4108 /* don't allow arp monitoring */
4110 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4111 miimon
, arp_interval
);
4115 if ((updelay
% miimon
) != 0) {
4116 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4118 (updelay
/ miimon
) * miimon
);
4123 if ((downdelay
% miimon
) != 0) {
4124 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4126 (downdelay
/ miimon
) * miimon
);
4129 downdelay
/= miimon
;
4132 if (arp_interval
< 0) {
4133 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4134 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4135 arp_interval
= BOND_LINK_ARP_INTERV
;
4138 for (arp_ip_count
= 0, i
= 0;
4139 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4140 /* not complete check, but should be good enough to
4142 __be32 ip
= in_aton(arp_ip_target
[i
]);
4143 if (!isdigit(arp_ip_target
[i
][0]) || ip
== 0 ||
4144 ip
== htonl(INADDR_BROADCAST
)) {
4145 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4149 if (bond_get_targets_ip(arp_target
, ip
) == -1)
4150 arp_target
[arp_ip_count
++] = ip
;
4152 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4157 if (arp_interval
&& !arp_ip_count
) {
4158 /* don't allow arping if no arp_ip_target given... */
4159 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4165 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4166 pr_err("arp_validate only supported in active-backup mode\n");
4169 if (!arp_interval
) {
4170 pr_err("arp_validate requires arp_interval\n");
4174 arp_validate_value
= bond_parse_parm(arp_validate
,
4176 if (arp_validate_value
== -1) {
4177 pr_err("Error: invalid arp_validate \"%s\"\n",
4178 arp_validate
== NULL
? "NULL" : arp_validate
);
4182 arp_validate_value
= 0;
4184 arp_all_targets_value
= 0;
4185 if (arp_all_targets
) {
4186 arp_all_targets_value
= bond_parse_parm(arp_all_targets
,
4187 arp_all_targets_tbl
);
4189 if (arp_all_targets_value
== -1) {
4190 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4192 arp_all_targets_value
= 0;
4197 pr_info("MII link monitoring set to %d ms\n", miimon
);
4198 } else if (arp_interval
) {
4199 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4201 arp_validate_tbl
[arp_validate_value
].modename
,
4204 for (i
= 0; i
< arp_ip_count
; i
++)
4205 pr_info(" %s", arp_ip_target
[i
]);
4209 } else if (max_bonds
) {
4210 /* miimon and arp_interval not set, we need one so things
4211 * work as expected, see bonding.txt for details
4213 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");
4216 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4217 /* currently, using a primary only makes sense
4218 * in active backup, TLB or ALB modes
4220 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4221 primary
, bond_mode_name(bond_mode
));
4225 if (primary
&& primary_reselect
) {
4226 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4228 if (primary_reselect_value
== -1) {
4229 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4231 NULL
? "NULL" : primary_reselect
);
4235 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4238 if (fail_over_mac
) {
4239 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4241 if (fail_over_mac_value
== -1) {
4242 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4243 arp_validate
== NULL
? "NULL" : arp_validate
);
4247 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4248 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4250 fail_over_mac_value
= BOND_FOM_NONE
;
4253 /* fill params struct with the proper values */
4254 params
->mode
= bond_mode
;
4255 params
->xmit_policy
= xmit_hashtype
;
4256 params
->miimon
= miimon
;
4257 params
->num_peer_notif
= num_peer_notif
;
4258 params
->arp_interval
= arp_interval
;
4259 params
->arp_validate
= arp_validate_value
;
4260 params
->arp_all_targets
= arp_all_targets_value
;
4261 params
->updelay
= updelay
;
4262 params
->downdelay
= downdelay
;
4263 params
->use_carrier
= use_carrier
;
4264 params
->lacp_fast
= lacp_fast
;
4265 params
->primary
[0] = 0;
4266 params
->primary_reselect
= primary_reselect_value
;
4267 params
->fail_over_mac
= fail_over_mac_value
;
4268 params
->tx_queues
= tx_queues
;
4269 params
->all_slaves_active
= all_slaves_active
;
4270 params
->resend_igmp
= resend_igmp
;
4271 params
->min_links
= min_links
;
4272 params
->lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
4275 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4276 params
->primary
[IFNAMSIZ
- 1] = 0;
4279 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4284 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4285 static struct lock_class_key bonding_netdev_addr_lock_key
;
4286 static struct lock_class_key bonding_tx_busylock_key
;
4288 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4289 struct netdev_queue
*txq
,
4292 lockdep_set_class(&txq
->_xmit_lock
,
4293 &bonding_netdev_xmit_lock_key
);
4296 static void bond_set_lockdep_class(struct net_device
*dev
)
4298 lockdep_set_class(&dev
->addr_list_lock
,
4299 &bonding_netdev_addr_lock_key
);
4300 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4301 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4305 * Called from registration process
4307 static int bond_init(struct net_device
*bond_dev
)
4309 struct bonding
*bond
= netdev_priv(bond_dev
);
4310 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4311 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4313 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4316 * Initialize locks that may be required during
4317 * en/deslave operations. All of the bond_open work
4318 * (of which this is part) should really be moved to
4319 * a phase prior to dev_open
4321 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4322 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4324 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4328 bond_set_lockdep_class(bond_dev
);
4330 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4332 bond_prepare_sysfs_group(bond
);
4334 bond_debug_register(bond
);
4336 /* Ensure valid dev_addr */
4337 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4338 bond_dev
->addr_assign_type
== NET_ADDR_PERM
)
4339 eth_hw_addr_random(bond_dev
);
4344 unsigned int bond_get_num_tx_queues(void)
4349 /* Create a new bond based on the specified name and bonding parameters.
4350 * If name is NULL, obtain a suitable "bond%d" name for us.
4351 * Caller must NOT hold rtnl_lock; we need to release it here before we
4352 * set up our sysfs entries.
4354 int bond_create(struct net
*net
, const char *name
)
4356 struct net_device
*bond_dev
;
4361 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4362 name
? name
: "bond%d",
4363 bond_setup
, tx_queues
);
4365 pr_err("%s: eek! can't alloc netdev!\n", name
);
4370 dev_net_set(bond_dev
, net
);
4371 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4373 res
= register_netdevice(bond_dev
);
4375 netif_carrier_off(bond_dev
);
4379 bond_destructor(bond_dev
);
4383 static int __net_init
bond_net_init(struct net
*net
)
4385 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4388 INIT_LIST_HEAD(&bn
->dev_list
);
4390 bond_create_proc_dir(bn
);
4391 bond_create_sysfs(bn
);
4396 static void __net_exit
bond_net_exit(struct net
*net
)
4398 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4399 struct bonding
*bond
, *tmp_bond
;
4402 bond_destroy_sysfs(bn
);
4403 bond_destroy_proc_dir(bn
);
4405 /* Kill off any bonds created after unregistering bond rtnl ops */
4407 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4408 unregister_netdevice_queue(bond
->dev
, &list
);
4409 unregister_netdevice_many(&list
);
4413 static struct pernet_operations bond_net_ops
= {
4414 .init
= bond_net_init
,
4415 .exit
= bond_net_exit
,
4417 .size
= sizeof(struct bond_net
),
4420 static int __init
bonding_init(void)
4425 pr_info("%s", bond_version
);
4427 res
= bond_check_params(&bonding_defaults
);
4431 res
= register_pernet_subsys(&bond_net_ops
);
4435 res
= bond_netlink_init();
4439 bond_create_debugfs();
4441 for (i
= 0; i
< max_bonds
; i
++) {
4442 res
= bond_create(&init_net
, NULL
);
4447 register_netdevice_notifier(&bond_netdev_notifier
);
4451 bond_netlink_fini();
4453 unregister_pernet_subsys(&bond_net_ops
);
4458 static void __exit
bonding_exit(void)
4460 unregister_netdevice_notifier(&bond_netdev_notifier
);
4462 bond_destroy_debugfs();
4464 bond_netlink_fini();
4465 unregister_pernet_subsys(&bond_net_ops
);
4467 #ifdef CONFIG_NET_POLL_CONTROLLER
4469 * Make sure we don't have an imbalance on our netpoll blocking
4471 WARN_ON(atomic_read(&netpoll_block_tx
));
4475 module_init(bonding_init
);
4476 module_exit(bonding_exit
);
4477 MODULE_LICENSE("GPL");
4478 MODULE_VERSION(DRV_VERSION
);
4479 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4480 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");