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>
85 /*---------------------------- Module parameters ----------------------------*/
87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 #define BOND_LINK_MON_INTERV 0
89 #define BOND_LINK_ARP_INTERV 0
91 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
92 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
93 static int num_peer_notif
= 1;
94 static int miimon
= BOND_LINK_MON_INTERV
;
97 static int use_carrier
= 1;
100 static char *primary_reselect
;
101 static char *lacp_rate
;
102 static int min_links
;
103 static char *ad_select
;
104 static char *xmit_hash_policy
;
105 static int arp_interval
= BOND_LINK_ARP_INTERV
;
106 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
107 static char *arp_validate
;
108 static char *arp_all_targets
;
109 static char *fail_over_mac
;
110 static int all_slaves_active
;
111 static struct bond_params bonding_defaults
;
112 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
114 module_param(max_bonds
, int, 0);
115 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
116 module_param(tx_queues
, int, 0);
117 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
118 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
119 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
120 "failover event (alias of num_unsol_na)");
121 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
122 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
123 "failover event (alias of num_grat_arp)");
124 module_param(miimon
, int, 0);
125 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
126 module_param(updelay
, int, 0);
127 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
128 module_param(downdelay
, int, 0);
129 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
131 module_param(use_carrier
, int, 0);
132 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
133 "0 for off, 1 for on (default)");
134 module_param(mode
, charp
, 0);
135 MODULE_PARM_DESC(mode
, "Mode of operation; 0 for balance-rr, "
136 "1 for active-backup, 2 for balance-xor, "
137 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
138 "6 for balance-alb");
139 module_param(primary
, charp
, 0);
140 MODULE_PARM_DESC(primary
, "Primary network device to use");
141 module_param(primary_reselect
, charp
, 0);
142 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
144 "0 for always (default), "
145 "1 for only if speed of primary is "
147 "2 for only on active slave "
149 module_param(lacp_rate
, charp
, 0);
150 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner; "
151 "0 for slow, 1 for fast");
152 module_param(ad_select
, charp
, 0);
153 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
154 "0 for stable (default), 1 for bandwidth, "
156 module_param(min_links
, int, 0);
157 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
159 module_param(xmit_hash_policy
, charp
, 0);
160 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
161 "0 for layer 2 (default), 1 for layer 3+4, "
163 module_param(arp_interval
, int, 0);
164 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
165 module_param_array(arp_ip_target
, charp
, NULL
, 0);
166 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
167 module_param(arp_validate
, charp
, 0);
168 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
169 "0 for none (default), 1 for active, "
170 "2 for backup, 3 for all");
171 module_param(arp_all_targets
, charp
, 0);
172 MODULE_PARM_DESC(arp_all_targets
, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
173 module_param(fail_over_mac
, charp
, 0);
174 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
175 "the same MAC; 0 for none (default), "
176 "1 for active, 2 for follow");
177 module_param(all_slaves_active
, int, 0);
178 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
179 "by setting active flag for all slaves; "
180 "0 for never (default), 1 for always.");
181 module_param(resend_igmp
, int, 0);
182 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
185 /*----------------------------- Global variables ----------------------------*/
187 #ifdef CONFIG_NET_POLL_CONTROLLER
188 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
191 int bond_net_id __read_mostly
;
193 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
194 static int arp_ip_count
;
195 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
196 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
197 static int lacp_fast
;
199 const struct bond_parm_tbl bond_lacp_tbl
[] = {
200 { "slow", AD_LACP_SLOW
},
201 { "fast", AD_LACP_FAST
},
205 const struct bond_parm_tbl bond_mode_tbl
[] = {
206 { "balance-rr", BOND_MODE_ROUNDROBIN
},
207 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
208 { "balance-xor", BOND_MODE_XOR
},
209 { "broadcast", BOND_MODE_BROADCAST
},
210 { "802.3ad", BOND_MODE_8023AD
},
211 { "balance-tlb", BOND_MODE_TLB
},
212 { "balance-alb", BOND_MODE_ALB
},
216 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
217 { "layer2", BOND_XMIT_POLICY_LAYER2
},
218 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
219 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
223 const struct bond_parm_tbl arp_all_targets_tbl
[] = {
224 { "any", BOND_ARP_TARGETS_ANY
},
225 { "all", BOND_ARP_TARGETS_ALL
},
229 const struct bond_parm_tbl arp_validate_tbl
[] = {
230 { "none", BOND_ARP_VALIDATE_NONE
},
231 { "active", BOND_ARP_VALIDATE_ACTIVE
},
232 { "backup", BOND_ARP_VALIDATE_BACKUP
},
233 { "all", BOND_ARP_VALIDATE_ALL
},
237 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
238 { "none", BOND_FOM_NONE
},
239 { "active", BOND_FOM_ACTIVE
},
240 { "follow", BOND_FOM_FOLLOW
},
244 const struct bond_parm_tbl pri_reselect_tbl
[] = {
245 { "always", BOND_PRI_RESELECT_ALWAYS
},
246 { "better", BOND_PRI_RESELECT_BETTER
},
247 { "failure", BOND_PRI_RESELECT_FAILURE
},
251 struct bond_parm_tbl ad_select_tbl
[] = {
252 { "stable", BOND_AD_STABLE
},
253 { "bandwidth", BOND_AD_BANDWIDTH
},
254 { "count", BOND_AD_COUNT
},
258 /*-------------------------- Forward declarations ---------------------------*/
260 static int bond_init(struct net_device
*bond_dev
);
261 static void bond_uninit(struct net_device
*bond_dev
);
263 /*---------------------------- General routines -----------------------------*/
265 const char *bond_mode_name(int mode
)
267 static const char *names
[] = {
268 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
269 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
270 [BOND_MODE_XOR
] = "load balancing (xor)",
271 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
272 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
273 [BOND_MODE_TLB
] = "transmit load balancing",
274 [BOND_MODE_ALB
] = "adaptive load balancing",
277 if (mode
< BOND_MODE_ROUNDROBIN
|| mode
> BOND_MODE_ALB
)
283 /*---------------------------------- VLAN -----------------------------------*/
286 * bond_dev_queue_xmit - Prepare skb for xmit.
288 * @bond: bond device that got this skb for tx.
289 * @skb: hw accel VLAN tagged skb to transmit
290 * @slave_dev: slave that is supposed to xmit this skbuff
292 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
293 struct net_device
*slave_dev
)
295 skb
->dev
= slave_dev
;
297 BUILD_BUG_ON(sizeof(skb
->queue_mapping
) !=
298 sizeof(qdisc_skb_cb(skb
)->slave_dev_queue_mapping
));
299 skb
->queue_mapping
= qdisc_skb_cb(skb
)->slave_dev_queue_mapping
;
301 if (unlikely(netpoll_tx_running(bond
->dev
)))
302 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
310 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
311 * We don't protect the slave list iteration with a lock because:
312 * a. This operation is performed in IOCTL context,
313 * b. The operation is protected by the RTNL semaphore in the 8021q code,
314 * c. Holding a lock with BH disabled while directly calling a base driver
315 * entry point is generally a BAD idea.
317 * The design of synchronization/protection for this operation in the 8021q
318 * module is good for one or more VLAN devices over a single physical device
319 * and cannot be extended for a teaming solution like bonding, so there is a
320 * potential race condition here where a net device from the vlan group might
321 * be referenced (either by a base driver or the 8021q code) while it is being
322 * removed from the system. However, it turns out we're not making matters
323 * worse, and if it works for regular VLAN usage it will work here too.
327 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
328 * @bond_dev: bonding net device that got called
329 * @vid: vlan id being added
331 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
,
332 __be16 proto
, u16 vid
)
334 struct bonding
*bond
= netdev_priv(bond_dev
);
338 bond_for_each_slave(bond
, slave
) {
339 res
= vlan_vid_add(slave
->dev
, proto
, vid
);
347 /* unwind from the slave that failed */
348 bond_for_each_slave_continue_reverse(bond
, slave
)
349 vlan_vid_del(slave
->dev
, proto
, vid
);
355 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
356 * @bond_dev: bonding net device that got called
357 * @vid: vlan id being removed
359 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
,
360 __be16 proto
, u16 vid
)
362 struct bonding
*bond
= netdev_priv(bond_dev
);
365 bond_for_each_slave(bond
, slave
)
366 vlan_vid_del(slave
->dev
, proto
, vid
);
368 if (bond_is_lb(bond
))
369 bond_alb_clear_vlan(bond
, vid
);
374 /*------------------------------- Link status -------------------------------*/
377 * Set the carrier state for the master according to the state of its
378 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
379 * do special 802.3ad magic.
381 * Returns zero if carrier state does not change, nonzero if it does.
383 static int bond_set_carrier(struct bonding
*bond
)
387 if (list_empty(&bond
->slave_list
))
390 if (bond
->params
.mode
== BOND_MODE_8023AD
)
391 return bond_3ad_set_carrier(bond
);
393 bond_for_each_slave(bond
, slave
) {
394 if (slave
->link
== BOND_LINK_UP
) {
395 if (!netif_carrier_ok(bond
->dev
)) {
396 netif_carrier_on(bond
->dev
);
404 if (netif_carrier_ok(bond
->dev
)) {
405 netif_carrier_off(bond
->dev
);
412 * Get link speed and duplex from the slave's base driver
413 * using ethtool. If for some reason the call fails or the
414 * values are invalid, set speed and duplex to -1,
417 static void bond_update_speed_duplex(struct slave
*slave
)
419 struct net_device
*slave_dev
= slave
->dev
;
420 struct ethtool_cmd ecmd
;
424 slave
->speed
= SPEED_UNKNOWN
;
425 slave
->duplex
= DUPLEX_UNKNOWN
;
427 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
431 slave_speed
= ethtool_cmd_speed(&ecmd
);
432 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
435 switch (ecmd
.duplex
) {
443 slave
->speed
= slave_speed
;
444 slave
->duplex
= ecmd
.duplex
;
450 * if <dev> supports MII link status reporting, check its link status.
452 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
453 * depending upon the setting of the use_carrier parameter.
455 * Return either BMSR_LSTATUS, meaning that the link is up (or we
456 * can't tell and just pretend it is), or 0, meaning that the link is
459 * If reporting is non-zero, instead of faking link up, return -1 if
460 * both ETHTOOL and MII ioctls fail (meaning the device does not
461 * support them). If use_carrier is set, return whatever it says.
462 * It'd be nice if there was a good way to tell if a driver supports
463 * netif_carrier, but there really isn't.
465 static int bond_check_dev_link(struct bonding
*bond
,
466 struct net_device
*slave_dev
, int reporting
)
468 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
469 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
471 struct mii_ioctl_data
*mii
;
473 if (!reporting
&& !netif_running(slave_dev
))
476 if (bond
->params
.use_carrier
)
477 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
479 /* Try to get link status using Ethtool first. */
480 if (slave_dev
->ethtool_ops
->get_link
)
481 return slave_dev
->ethtool_ops
->get_link(slave_dev
) ?
484 /* Ethtool can't be used, fallback to MII ioctls. */
485 ioctl
= slave_ops
->ndo_do_ioctl
;
487 /* TODO: set pointer to correct ioctl on a per team member */
488 /* bases to make this more efficient. that is, once */
489 /* we determine the correct ioctl, we will always */
490 /* call it and not the others for that team */
494 * We cannot assume that SIOCGMIIPHY will also read a
495 * register; not all network drivers (e.g., e100)
499 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
500 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
502 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
503 mii
->reg_num
= MII_BMSR
;
504 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
505 return mii
->val_out
& BMSR_LSTATUS
;
510 * If reporting, report that either there's no dev->do_ioctl,
511 * or both SIOCGMIIREG and get_link failed (meaning that we
512 * cannot report link status). If not reporting, pretend
515 return reporting
? -1 : BMSR_LSTATUS
;
518 /*----------------------------- Multicast list ------------------------------*/
521 * Push the promiscuity flag down to appropriate slaves
523 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
526 if (USES_PRIMARY(bond
->params
.mode
)) {
527 /* write lock already acquired */
528 if (bond
->curr_active_slave
) {
529 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
535 bond_for_each_slave(bond
, slave
) {
536 err
= dev_set_promiscuity(slave
->dev
, inc
);
545 * Push the allmulti flag down to all slaves
547 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
550 if (USES_PRIMARY(bond
->params
.mode
)) {
551 /* write lock already acquired */
552 if (bond
->curr_active_slave
) {
553 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
559 bond_for_each_slave(bond
, slave
) {
560 err
= dev_set_allmulti(slave
->dev
, inc
);
569 * Retrieve the list of registered multicast addresses for the bonding
570 * device and retransmit an IGMP JOIN request to the current active
573 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
575 if (!rtnl_trylock()) {
576 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
579 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, bond
->dev
);
582 /* We use curr_slave_lock to protect against concurrent access to
583 * igmp_retrans from multiple running instances of this function and
584 * bond_change_active_slave
586 write_lock_bh(&bond
->curr_slave_lock
);
587 if (bond
->igmp_retrans
> 1) {
588 bond
->igmp_retrans
--;
589 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
591 write_unlock_bh(&bond
->curr_slave_lock
);
594 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
596 struct bonding
*bond
= container_of(work
, struct bonding
,
599 bond_resend_igmp_join_requests(bond
);
602 /* Flush bond's hardware addresses from slave
604 static void bond_hw_addr_flush(struct net_device
*bond_dev
,
605 struct net_device
*slave_dev
)
607 struct bonding
*bond
= netdev_priv(bond_dev
);
609 dev_uc_unsync(slave_dev
, bond_dev
);
610 dev_mc_unsync(slave_dev
, bond_dev
);
612 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
613 /* del lacpdu mc addr from mc list */
614 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
616 dev_mc_del(slave_dev
, lacpdu_multicast
);
620 /*--------------------------- Active slave change ---------------------------*/
622 /* Update the hardware address list and promisc/allmulti for the new and
623 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
624 * slaves up date at all times; only the USES_PRIMARY modes need to call
625 * this function to swap these settings during a failover.
627 static void bond_hw_addr_swap(struct bonding
*bond
, struct slave
*new_active
,
628 struct slave
*old_active
)
633 if (bond
->dev
->flags
& IFF_PROMISC
)
634 dev_set_promiscuity(old_active
->dev
, -1);
636 if (bond
->dev
->flags
& IFF_ALLMULTI
)
637 dev_set_allmulti(old_active
->dev
, -1);
639 bond_hw_addr_flush(bond
->dev
, old_active
->dev
);
643 /* FIXME: Signal errors upstream. */
644 if (bond
->dev
->flags
& IFF_PROMISC
)
645 dev_set_promiscuity(new_active
->dev
, 1);
647 if (bond
->dev
->flags
& IFF_ALLMULTI
)
648 dev_set_allmulti(new_active
->dev
, 1);
650 netif_addr_lock_bh(bond
->dev
);
651 dev_uc_sync(new_active
->dev
, bond
->dev
);
652 dev_mc_sync(new_active
->dev
, bond
->dev
);
653 netif_addr_unlock_bh(bond
->dev
);
658 * bond_set_dev_addr - clone slave's address to bond
659 * @bond_dev: bond net device
660 * @slave_dev: slave net device
662 * Should be called with RTNL held.
664 static void bond_set_dev_addr(struct net_device
*bond_dev
,
665 struct net_device
*slave_dev
)
667 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
668 bond_dev
, slave_dev
, slave_dev
->addr_len
);
669 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
670 bond_dev
->addr_assign_type
= NET_ADDR_STOLEN
;
671 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
675 * bond_do_fail_over_mac
677 * Perform special MAC address swapping for fail_over_mac settings
679 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
681 static void bond_do_fail_over_mac(struct bonding
*bond
,
682 struct slave
*new_active
,
683 struct slave
*old_active
)
684 __releases(&bond
->curr_slave_lock
)
685 __releases(&bond
->lock
)
686 __acquires(&bond
->lock
)
687 __acquires(&bond
->curr_slave_lock
)
689 u8 tmp_mac
[ETH_ALEN
];
690 struct sockaddr saddr
;
693 switch (bond
->params
.fail_over_mac
) {
694 case BOND_FOM_ACTIVE
:
696 write_unlock_bh(&bond
->curr_slave_lock
);
697 read_unlock(&bond
->lock
);
698 bond_set_dev_addr(bond
->dev
, new_active
->dev
);
699 read_lock(&bond
->lock
);
700 write_lock_bh(&bond
->curr_slave_lock
);
703 case BOND_FOM_FOLLOW
:
705 * if new_active && old_active, swap them
706 * if just old_active, do nothing (going to no active slave)
707 * if just new_active, set new_active to bond's MAC
712 write_unlock_bh(&bond
->curr_slave_lock
);
713 read_unlock(&bond
->lock
);
716 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
717 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
719 saddr
.sa_family
= new_active
->dev
->type
;
721 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
722 saddr
.sa_family
= bond
->dev
->type
;
725 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
727 pr_err("%s: Error %d setting MAC of slave %s\n",
728 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
735 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
736 saddr
.sa_family
= old_active
->dev
->type
;
738 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
740 pr_err("%s: Error %d setting MAC of slave %s\n",
741 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
743 read_lock(&bond
->lock
);
744 write_lock_bh(&bond
->curr_slave_lock
);
747 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
748 bond
->dev
->name
, bond
->params
.fail_over_mac
);
754 static bool bond_should_change_active(struct bonding
*bond
)
756 struct slave
*prim
= bond
->primary_slave
;
757 struct slave
*curr
= bond
->curr_active_slave
;
759 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
761 if (bond
->force_primary
) {
762 bond
->force_primary
= false;
765 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
766 (prim
->speed
< curr
->speed
||
767 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
769 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
775 * find_best_interface - select the best available slave to be the active one
776 * @bond: our bonding struct
778 * Warning: Caller must hold curr_slave_lock for writing.
780 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
782 struct slave
*new_active
, *old_active
;
783 struct slave
*bestslave
= NULL
;
784 int mintime
= bond
->params
.updelay
;
787 new_active
= bond
->curr_active_slave
;
789 if (!new_active
) { /* there were no active slaves left */
790 new_active
= bond_first_slave(bond
);
792 return NULL
; /* still no slave, return NULL */
795 if ((bond
->primary_slave
) &&
796 bond
->primary_slave
->link
== BOND_LINK_UP
&&
797 bond_should_change_active(bond
)) {
798 new_active
= bond
->primary_slave
;
801 /* remember where to stop iterating over the slaves */
802 old_active
= new_active
;
804 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
805 if (new_active
->link
== BOND_LINK_UP
) {
807 } else if (new_active
->link
== BOND_LINK_BACK
&&
808 IS_UP(new_active
->dev
)) {
809 /* link up, but waiting for stabilization */
810 if (new_active
->delay
< mintime
) {
811 mintime
= new_active
->delay
;
812 bestslave
= new_active
;
820 static bool bond_should_notify_peers(struct bonding
*bond
)
822 struct slave
*slave
= bond
->curr_active_slave
;
824 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
825 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
827 if (!slave
|| !bond
->send_peer_notif
||
828 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
835 * change_active_interface - change the active slave into the specified one
836 * @bond: our bonding struct
837 * @new: the new slave to make the active one
839 * Set the new slave to the bond's settings and unset them on the old
841 * Setting include flags, mc-list, promiscuity, allmulti, etc.
843 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
844 * because it is apparently the best available slave we have, even though its
845 * updelay hasn't timed out yet.
847 * If new_active is not NULL, caller must hold bond->lock for read and
848 * curr_slave_lock for write_bh.
850 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
852 struct slave
*old_active
= bond
->curr_active_slave
;
854 if (old_active
== new_active
)
858 new_active
->jiffies
= jiffies
;
860 if (new_active
->link
== BOND_LINK_BACK
) {
861 if (USES_PRIMARY(bond
->params
.mode
)) {
862 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
863 bond
->dev
->name
, new_active
->dev
->name
,
864 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
867 new_active
->delay
= 0;
868 new_active
->link
= BOND_LINK_UP
;
870 if (bond
->params
.mode
== BOND_MODE_8023AD
)
871 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
873 if (bond_is_lb(bond
))
874 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
876 if (USES_PRIMARY(bond
->params
.mode
)) {
877 pr_info("%s: making interface %s the new active one.\n",
878 bond
->dev
->name
, new_active
->dev
->name
);
883 if (USES_PRIMARY(bond
->params
.mode
))
884 bond_hw_addr_swap(bond
, new_active
, old_active
);
886 if (bond_is_lb(bond
)) {
887 bond_alb_handle_active_change(bond
, new_active
);
889 bond_set_slave_inactive_flags(old_active
);
891 bond_set_slave_active_flags(new_active
);
893 rcu_assign_pointer(bond
->curr_active_slave
, new_active
);
896 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
898 bond_set_slave_inactive_flags(old_active
);
901 bool should_notify_peers
= false;
903 bond_set_slave_active_flags(new_active
);
905 if (bond
->params
.fail_over_mac
)
906 bond_do_fail_over_mac(bond
, new_active
,
909 if (netif_running(bond
->dev
)) {
910 bond
->send_peer_notif
=
911 bond
->params
.num_peer_notif
;
912 should_notify_peers
=
913 bond_should_notify_peers(bond
);
916 write_unlock_bh(&bond
->curr_slave_lock
);
917 read_unlock(&bond
->lock
);
919 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
920 if (should_notify_peers
)
921 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
924 read_lock(&bond
->lock
);
925 write_lock_bh(&bond
->curr_slave_lock
);
929 /* resend IGMP joins since active slave has changed or
930 * all were sent on curr_active_slave.
931 * resend only if bond is brought up with the affected
932 * bonding modes and the retransmission is enabled */
933 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
934 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
935 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
936 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
937 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
942 * bond_select_active_slave - select a new active slave, if needed
943 * @bond: our bonding struct
945 * This functions should be called when one of the following occurs:
946 * - The old curr_active_slave has been released or lost its link.
947 * - The primary_slave has got its link back.
948 * - A slave has got its link back and there's no old curr_active_slave.
950 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
952 void bond_select_active_slave(struct bonding
*bond
)
954 struct slave
*best_slave
;
957 best_slave
= bond_find_best_slave(bond
);
958 if (best_slave
!= bond
->curr_active_slave
) {
959 bond_change_active_slave(bond
, best_slave
);
960 rv
= bond_set_carrier(bond
);
964 if (netif_carrier_ok(bond
->dev
)) {
965 pr_info("%s: first active interface up!\n",
968 pr_info("%s: now running without any active interface !\n",
974 /*--------------------------- slave list handling ---------------------------*/
977 * This function attaches the slave to the end of list.
979 * bond->lock held for writing by caller.
981 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
983 list_add_tail_rcu(&new_slave
->list
, &bond
->slave_list
);
988 * This function detaches the slave from the list.
989 * WARNING: no check is made to verify if the slave effectively
991 * Nothing is freed on return, structures are just unchained.
992 * If any slave pointer in bond was pointing to <slave>,
993 * it should be changed by the calling function.
995 * bond->lock held for writing by caller.
997 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
999 list_del_rcu(&slave
->list
);
1003 #ifdef CONFIG_NET_POLL_CONTROLLER
1004 static inline int slave_enable_netpoll(struct slave
*slave
)
1009 np
= kzalloc(sizeof(*np
), GFP_ATOMIC
);
1014 err
= __netpoll_setup(np
, slave
->dev
, GFP_ATOMIC
);
1023 static inline void slave_disable_netpoll(struct slave
*slave
)
1025 struct netpoll
*np
= slave
->np
;
1031 __netpoll_free_async(np
);
1033 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1035 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1037 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1042 static void bond_poll_controller(struct net_device
*bond_dev
)
1046 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1048 struct bonding
*bond
= netdev_priv(bond_dev
);
1049 struct slave
*slave
;
1051 bond_for_each_slave(bond
, slave
)
1052 if (IS_UP(slave
->dev
))
1053 slave_disable_netpoll(slave
);
1056 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
, gfp_t gfp
)
1058 struct bonding
*bond
= netdev_priv(dev
);
1059 struct slave
*slave
;
1062 bond_for_each_slave(bond
, slave
) {
1063 err
= slave_enable_netpoll(slave
);
1065 bond_netpoll_cleanup(dev
);
1072 static inline int slave_enable_netpoll(struct slave
*slave
)
1076 static inline void slave_disable_netpoll(struct slave
*slave
)
1079 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1084 /*---------------------------------- IOCTL ----------------------------------*/
1086 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1087 netdev_features_t features
)
1089 struct bonding
*bond
= netdev_priv(dev
);
1090 netdev_features_t mask
;
1091 struct slave
*slave
;
1093 if (list_empty(&bond
->slave_list
)) {
1094 /* Disable adding VLANs to empty bond. But why? --mq */
1095 features
|= NETIF_F_VLAN_CHALLENGED
;
1100 features
&= ~NETIF_F_ONE_FOR_ALL
;
1101 features
|= NETIF_F_ALL_FOR_ALL
;
1103 bond_for_each_slave(bond
, slave
) {
1104 features
= netdev_increment_features(features
,
1105 slave
->dev
->features
,
1108 features
= netdev_add_tso_features(features
, mask
);
1113 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1114 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1115 NETIF_F_HIGHDMA | NETIF_F_LRO)
1117 static void bond_compute_features(struct bonding
*bond
)
1119 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1120 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1121 unsigned short max_hard_header_len
= ETH_HLEN
;
1122 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1123 struct net_device
*bond_dev
= bond
->dev
;
1124 u16 gso_max_segs
= GSO_MAX_SEGS
;
1125 struct slave
*slave
;
1127 if (list_empty(&bond
->slave_list
))
1130 bond_for_each_slave(bond
, slave
) {
1131 vlan_features
= netdev_increment_features(vlan_features
,
1132 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1134 dst_release_flag
&= slave
->dev
->priv_flags
;
1135 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1136 max_hard_header_len
= slave
->dev
->hard_header_len
;
1138 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1139 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1143 bond_dev
->vlan_features
= vlan_features
;
1144 bond_dev
->hard_header_len
= max_hard_header_len
;
1145 bond_dev
->gso_max_segs
= gso_max_segs
;
1146 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1148 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1149 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1151 netdev_change_features(bond_dev
);
1154 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1155 struct net_device
*slave_dev
)
1157 bond_dev
->header_ops
= slave_dev
->header_ops
;
1159 bond_dev
->type
= slave_dev
->type
;
1160 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1161 bond_dev
->addr_len
= slave_dev
->addr_len
;
1163 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1164 slave_dev
->addr_len
);
1167 /* On bonding slaves other than the currently active slave, suppress
1168 * duplicates except for alb non-mcast/bcast.
1170 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1171 struct slave
*slave
,
1172 struct bonding
*bond
)
1174 if (bond_is_slave_inactive(slave
)) {
1175 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1176 skb
->pkt_type
!= PACKET_BROADCAST
&&
1177 skb
->pkt_type
!= PACKET_MULTICAST
)
1184 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1186 struct sk_buff
*skb
= *pskb
;
1187 struct slave
*slave
;
1188 struct bonding
*bond
;
1189 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1191 int ret
= RX_HANDLER_ANOTHER
;
1193 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1195 return RX_HANDLER_CONSUMED
;
1199 slave
= bond_slave_get_rcu(skb
->dev
);
1202 if (bond
->params
.arp_interval
)
1203 slave
->dev
->last_rx
= jiffies
;
1205 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1207 ret
= recv_probe(skb
, bond
, slave
);
1208 if (ret
== RX_HANDLER_CONSUMED
) {
1214 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1215 return RX_HANDLER_EXACT
;
1218 skb
->dev
= bond
->dev
;
1220 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1221 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1222 skb
->pkt_type
== PACKET_HOST
) {
1224 if (unlikely(skb_cow_head(skb
,
1225 skb
->data
- skb_mac_header(skb
)))) {
1227 return RX_HANDLER_CONSUMED
;
1229 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1235 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1236 struct net_device
*slave_dev
)
1240 err
= netdev_master_upper_dev_link(slave_dev
, bond_dev
);
1243 slave_dev
->flags
|= IFF_SLAVE
;
1244 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
);
1248 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1249 struct net_device
*slave_dev
)
1251 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1252 slave_dev
->flags
&= ~IFF_SLAVE
;
1253 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
);
1256 /* enslave device <slave> to bond device <master> */
1257 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1259 struct bonding
*bond
= netdev_priv(bond_dev
);
1260 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1261 struct slave
*new_slave
= NULL
;
1262 struct sockaddr addr
;
1266 if (!bond
->params
.use_carrier
&&
1267 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1268 slave_ops
->ndo_do_ioctl
== NULL
) {
1269 pr_warning("%s: Warning: no link monitoring support for %s\n",
1270 bond_dev
->name
, slave_dev
->name
);
1273 /* already enslaved */
1274 if (slave_dev
->flags
& IFF_SLAVE
) {
1275 pr_debug("Error, Device was already enslaved\n");
1279 /* vlan challenged mutual exclusion */
1280 /* no need to lock since we're protected by rtnl_lock */
1281 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1282 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1283 if (vlan_uses_dev(bond_dev
)) {
1284 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1285 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1288 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1289 bond_dev
->name
, slave_dev
->name
,
1290 slave_dev
->name
, bond_dev
->name
);
1293 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1297 * Old ifenslave binaries are no longer supported. These can
1298 * be identified with moderate accuracy by the state of the slave:
1299 * the current ifenslave will set the interface down prior to
1300 * enslaving it; the old ifenslave will not.
1302 if ((slave_dev
->flags
& IFF_UP
)) {
1303 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1306 goto err_undo_flags
;
1309 /* set bonding device ether type by slave - bonding netdevices are
1310 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1311 * there is a need to override some of the type dependent attribs/funcs.
1313 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1314 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1316 if (list_empty(&bond
->slave_list
)) {
1317 if (bond_dev
->type
!= slave_dev
->type
) {
1318 pr_debug("%s: change device type from %d to %d\n",
1320 bond_dev
->type
, slave_dev
->type
);
1322 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1324 res
= notifier_to_errno(res
);
1326 pr_err("%s: refused to change device type\n",
1329 goto err_undo_flags
;
1332 /* Flush unicast and multicast addresses */
1333 dev_uc_flush(bond_dev
);
1334 dev_mc_flush(bond_dev
);
1336 if (slave_dev
->type
!= ARPHRD_ETHER
)
1337 bond_setup_by_slave(bond_dev
, slave_dev
);
1339 ether_setup(bond_dev
);
1340 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1343 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1346 } else if (bond_dev
->type
!= slave_dev
->type
) {
1347 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1349 slave_dev
->type
, bond_dev
->type
);
1351 goto err_undo_flags
;
1354 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1355 if (list_empty(&bond
->slave_list
)) {
1356 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1358 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1359 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1360 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",
1363 goto err_undo_flags
;
1367 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1369 /* If this is the first slave, then we need to set the master's hardware
1370 * address to be the same as the slave's. */
1371 if (list_empty(&bond
->slave_list
) &&
1372 bond
->dev
->addr_assign_type
== NET_ADDR_RANDOM
)
1373 bond_set_dev_addr(bond
->dev
, slave_dev
);
1375 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1378 goto err_undo_flags
;
1380 INIT_LIST_HEAD(&new_slave
->list
);
1382 * Set the new_slave's queue_id to be zero. Queue ID mapping
1383 * is set via sysfs or module option if desired.
1385 new_slave
->queue_id
= 0;
1387 /* Save slave's original mtu and then set it to match the bond */
1388 new_slave
->original_mtu
= slave_dev
->mtu
;
1389 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1391 pr_debug("Error %d calling dev_set_mtu\n", res
);
1396 * Save slave's original ("permanent") mac address for modes
1397 * that need it, and for restoring it upon release, and then
1398 * set it to the master's address
1400 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1402 if (!bond
->params
.fail_over_mac
) {
1404 * Set slave to master's mac address. The application already
1405 * set the master's mac address to that of the first slave
1407 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1408 addr
.sa_family
= slave_dev
->type
;
1409 res
= dev_set_mac_address(slave_dev
, &addr
);
1411 pr_debug("Error %d calling set_mac_address\n", res
);
1412 goto err_restore_mtu
;
1416 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
);
1418 pr_debug("Error %d calling bond_master_upper_dev_link\n", res
);
1419 goto err_restore_mac
;
1422 /* open the slave since the application closed it */
1423 res
= dev_open(slave_dev
);
1425 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1426 goto err_unset_master
;
1429 new_slave
->bond
= bond
;
1430 new_slave
->dev
= slave_dev
;
1431 slave_dev
->priv_flags
|= IFF_BONDING
;
1433 if (bond_is_lb(bond
)) {
1434 /* bond_alb_init_slave() must be called before all other stages since
1435 * it might fail and we do not want to have to undo everything
1437 res
= bond_alb_init_slave(bond
, new_slave
);
1442 /* If the mode USES_PRIMARY, then the following is handled by
1443 * bond_change_active_slave().
1445 if (!USES_PRIMARY(bond
->params
.mode
)) {
1446 /* set promiscuity level to new slave */
1447 if (bond_dev
->flags
& IFF_PROMISC
) {
1448 res
= dev_set_promiscuity(slave_dev
, 1);
1453 /* set allmulti level to new slave */
1454 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1455 res
= dev_set_allmulti(slave_dev
, 1);
1460 netif_addr_lock_bh(bond_dev
);
1462 dev_mc_sync_multiple(slave_dev
, bond_dev
);
1463 dev_uc_sync_multiple(slave_dev
, bond_dev
);
1465 netif_addr_unlock_bh(bond_dev
);
1468 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1469 /* add lacpdu mc addr to mc list */
1470 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1472 dev_mc_add(slave_dev
, lacpdu_multicast
);
1475 res
= vlan_vids_add_by_dev(slave_dev
, bond_dev
);
1477 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1478 bond_dev
->name
, slave_dev
->name
);
1482 write_lock_bh(&bond
->lock
);
1484 bond_attach_slave(bond
, new_slave
);
1486 new_slave
->delay
= 0;
1487 new_slave
->link_failure_count
= 0;
1489 write_unlock_bh(&bond
->lock
);
1491 bond_compute_features(bond
);
1493 bond_update_speed_duplex(new_slave
);
1495 read_lock(&bond
->lock
);
1497 new_slave
->last_arp_rx
= jiffies
-
1498 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1499 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
1500 new_slave
->target_last_arp_rx
[i
] = new_slave
->last_arp_rx
;
1502 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1503 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1505 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1507 * miimon is set but a bonded network driver
1508 * does not support ETHTOOL/MII and
1509 * arp_interval is not set. Note: if
1510 * use_carrier is enabled, we will never go
1511 * here (because netif_carrier is always
1512 * supported); thus, we don't need to change
1513 * the messages for netif_carrier.
1515 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",
1516 bond_dev
->name
, slave_dev
->name
);
1517 } else if (link_reporting
== -1) {
1518 /* unable get link status using mii/ethtool */
1519 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",
1520 bond_dev
->name
, slave_dev
->name
);
1524 /* check for initial state */
1525 if (bond
->params
.miimon
) {
1526 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1527 if (bond
->params
.updelay
) {
1528 new_slave
->link
= BOND_LINK_BACK
;
1529 new_slave
->delay
= bond
->params
.updelay
;
1531 new_slave
->link
= BOND_LINK_UP
;
1534 new_slave
->link
= BOND_LINK_DOWN
;
1536 } else if (bond
->params
.arp_interval
) {
1537 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1538 BOND_LINK_UP
: BOND_LINK_DOWN
);
1540 new_slave
->link
= BOND_LINK_UP
;
1543 if (new_slave
->link
!= BOND_LINK_DOWN
)
1544 new_slave
->jiffies
= jiffies
;
1545 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1546 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1547 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1549 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1550 /* if there is a primary slave, remember it */
1551 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1552 bond
->primary_slave
= new_slave
;
1553 bond
->force_primary
= true;
1557 write_lock_bh(&bond
->curr_slave_lock
);
1559 switch (bond
->params
.mode
) {
1560 case BOND_MODE_ACTIVEBACKUP
:
1561 bond_set_slave_inactive_flags(new_slave
);
1562 bond_select_active_slave(bond
);
1564 case BOND_MODE_8023AD
:
1565 /* in 802.3ad mode, the internal mechanism
1566 * will activate the slaves in the selected
1569 bond_set_slave_inactive_flags(new_slave
);
1570 /* if this is the first slave */
1571 if (bond_first_slave(bond
) == new_slave
) {
1572 SLAVE_AD_INFO(new_slave
).id
= 1;
1573 /* Initialize AD with the number of times that the AD timer is called in 1 second
1574 * can be called only after the mac address of the bond is set
1576 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1578 struct slave
*prev_slave
;
1580 prev_slave
= bond_prev_slave(bond
, new_slave
);
1581 SLAVE_AD_INFO(new_slave
).id
=
1582 SLAVE_AD_INFO(prev_slave
).id
+ 1;
1585 bond_3ad_bind_slave(new_slave
);
1589 bond_set_active_slave(new_slave
);
1590 bond_set_slave_inactive_flags(new_slave
);
1591 bond_select_active_slave(bond
);
1594 pr_debug("This slave is always active in trunk mode\n");
1596 /* always active in trunk mode */
1597 bond_set_active_slave(new_slave
);
1599 /* In trunking mode there is little meaning to curr_active_slave
1600 * anyway (it holds no special properties of the bond device),
1601 * so we can change it without calling change_active_interface()
1603 if (!bond
->curr_active_slave
&& new_slave
->link
== BOND_LINK_UP
)
1604 rcu_assign_pointer(bond
->curr_active_slave
, new_slave
);
1607 } /* switch(bond_mode) */
1609 write_unlock_bh(&bond
->curr_slave_lock
);
1611 bond_set_carrier(bond
);
1613 #ifdef CONFIG_NET_POLL_CONTROLLER
1614 slave_dev
->npinfo
= bond
->dev
->npinfo
;
1615 if (slave_dev
->npinfo
) {
1616 if (slave_enable_netpoll(new_slave
)) {
1617 read_unlock(&bond
->lock
);
1618 pr_info("Error, %s: master_dev is using netpoll, "
1619 "but new slave device does not support netpoll.\n",
1627 read_unlock(&bond
->lock
);
1629 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1633 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1636 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1637 goto err_dest_symlinks
;
1640 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1641 bond_dev
->name
, slave_dev
->name
,
1642 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1643 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1645 /* enslave is successful */
1648 /* Undo stages on error */
1650 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1653 if (!USES_PRIMARY(bond
->params
.mode
))
1654 bond_hw_addr_flush(bond_dev
, slave_dev
);
1656 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1657 write_lock_bh(&bond
->lock
);
1658 bond_detach_slave(bond
, new_slave
);
1659 if (bond
->primary_slave
== new_slave
)
1660 bond
->primary_slave
= NULL
;
1661 if (bond
->curr_active_slave
== new_slave
) {
1662 bond_change_active_slave(bond
, NULL
);
1663 write_unlock_bh(&bond
->lock
);
1664 read_lock(&bond
->lock
);
1665 write_lock_bh(&bond
->curr_slave_lock
);
1666 bond_select_active_slave(bond
);
1667 write_unlock_bh(&bond
->curr_slave_lock
);
1668 read_unlock(&bond
->lock
);
1670 write_unlock_bh(&bond
->lock
);
1672 slave_disable_netpoll(new_slave
);
1675 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1676 dev_close(slave_dev
);
1679 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1682 if (!bond
->params
.fail_over_mac
) {
1683 /* XXX TODO - fom follow mode needs to change master's
1684 * MAC if this slave's MAC is in use by the bond, or at
1685 * least print a warning.
1687 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1688 addr
.sa_family
= slave_dev
->type
;
1689 dev_set_mac_address(slave_dev
, &addr
);
1693 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1699 bond_compute_features(bond
);
1700 /* Enslave of first slave has failed and we need to fix master's mac */
1701 if (list_empty(&bond
->slave_list
) &&
1702 ether_addr_equal(bond_dev
->dev_addr
, slave_dev
->dev_addr
))
1703 eth_hw_addr_random(bond_dev
);
1709 * Try to release the slave device <slave> from the bond device <master>
1710 * It is legal to access curr_active_slave without a lock because all the function
1711 * is write-locked. If "all" is true it means that the function is being called
1712 * while destroying a bond interface and all slaves are being released.
1714 * The rules for slave state should be:
1715 * for Active/Backup:
1716 * Active stays on all backups go down
1717 * for Bonded connections:
1718 * The first up interface should be left on and all others downed.
1720 static int __bond_release_one(struct net_device
*bond_dev
,
1721 struct net_device
*slave_dev
,
1724 struct bonding
*bond
= netdev_priv(bond_dev
);
1725 struct slave
*slave
, *oldcurrent
;
1726 struct sockaddr addr
;
1727 netdev_features_t old_features
= bond_dev
->features
;
1729 /* slave is not a slave or master is not master of this slave */
1730 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1731 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1732 pr_err("%s: Error: cannot release %s.\n",
1733 bond_dev
->name
, slave_dev
->name
);
1738 write_lock_bh(&bond
->lock
);
1740 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1742 /* not a slave of this bond */
1743 pr_info("%s: %s not enslaved\n",
1744 bond_dev
->name
, slave_dev
->name
);
1745 write_unlock_bh(&bond
->lock
);
1746 unblock_netpoll_tx();
1750 write_unlock_bh(&bond
->lock
);
1751 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1752 * for this slave anymore.
1754 netdev_rx_handler_unregister(slave_dev
);
1755 write_lock_bh(&bond
->lock
);
1757 /* Inform AD package of unbinding of slave. */
1758 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1759 /* must be called before the slave is
1760 * detached from the list
1762 bond_3ad_unbind_slave(slave
);
1765 pr_info("%s: releasing %s interface %s\n",
1767 bond_is_active_slave(slave
) ? "active" : "backup",
1770 oldcurrent
= bond
->curr_active_slave
;
1772 bond
->current_arp_slave
= NULL
;
1774 /* release the slave from its bond */
1775 bond_detach_slave(bond
, slave
);
1777 if (!all
&& !bond
->params
.fail_over_mac
) {
1778 if (ether_addr_equal(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1779 !list_empty(&bond
->slave_list
))
1780 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",
1781 bond_dev
->name
, slave_dev
->name
,
1783 bond_dev
->name
, slave_dev
->name
);
1786 if (bond
->primary_slave
== slave
)
1787 bond
->primary_slave
= NULL
;
1789 if (oldcurrent
== slave
)
1790 bond_change_active_slave(bond
, NULL
);
1792 if (bond_is_lb(bond
)) {
1793 /* Must be called only after the slave has been
1794 * detached from the list and the curr_active_slave
1795 * has been cleared (if our_slave == old_current),
1796 * but before a new active slave is selected.
1798 write_unlock_bh(&bond
->lock
);
1799 bond_alb_deinit_slave(bond
, slave
);
1800 write_lock_bh(&bond
->lock
);
1804 rcu_assign_pointer(bond
->curr_active_slave
, NULL
);
1805 } else if (oldcurrent
== slave
) {
1807 * Note that we hold RTNL over this sequence, so there
1808 * is no concern that another slave add/remove event
1811 write_unlock_bh(&bond
->lock
);
1812 read_lock(&bond
->lock
);
1813 write_lock_bh(&bond
->curr_slave_lock
);
1815 bond_select_active_slave(bond
);
1817 write_unlock_bh(&bond
->curr_slave_lock
);
1818 read_unlock(&bond
->lock
);
1819 write_lock_bh(&bond
->lock
);
1822 if (list_empty(&bond
->slave_list
)) {
1823 bond_set_carrier(bond
);
1824 eth_hw_addr_random(bond_dev
);
1826 if (vlan_uses_dev(bond_dev
)) {
1827 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1828 bond_dev
->name
, bond_dev
->name
);
1829 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1834 write_unlock_bh(&bond
->lock
);
1835 unblock_netpoll_tx();
1838 if (list_empty(&bond
->slave_list
)) {
1839 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
1840 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
1843 bond_compute_features(bond
);
1844 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1845 (old_features
& NETIF_F_VLAN_CHALLENGED
))
1846 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1847 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1849 /* must do this from outside any spinlocks */
1850 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1852 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1854 /* If the mode USES_PRIMARY, then this cases was handled above by
1855 * bond_change_active_slave(..., NULL)
1857 if (!USES_PRIMARY(bond
->params
.mode
)) {
1858 /* unset promiscuity level from slave */
1859 if (bond_dev
->flags
& IFF_PROMISC
)
1860 dev_set_promiscuity(slave_dev
, -1);
1862 /* unset allmulti level from slave */
1863 if (bond_dev
->flags
& IFF_ALLMULTI
)
1864 dev_set_allmulti(slave_dev
, -1);
1866 bond_hw_addr_flush(bond_dev
, slave_dev
);
1869 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1871 slave_disable_netpoll(slave
);
1873 /* close slave before restoring its mac address */
1874 dev_close(slave_dev
);
1876 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1877 /* restore original ("permanent") mac address */
1878 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1879 addr
.sa_family
= slave_dev
->type
;
1880 dev_set_mac_address(slave_dev
, &addr
);
1883 dev_set_mtu(slave_dev
, slave
->original_mtu
);
1885 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1889 return 0; /* deletion OK */
1892 /* A wrapper used because of ndo_del_link */
1893 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1895 return __bond_release_one(bond_dev
, slave_dev
, false);
1899 * First release a slave and then destroy the bond if no more slaves are left.
1900 * Must be under rtnl_lock when this function is called.
1902 static int bond_release_and_destroy(struct net_device
*bond_dev
,
1903 struct net_device
*slave_dev
)
1905 struct bonding
*bond
= netdev_priv(bond_dev
);
1908 ret
= bond_release(bond_dev
, slave_dev
);
1909 if (ret
== 0 && list_empty(&bond
->slave_list
)) {
1910 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1911 pr_info("%s: destroying bond %s.\n",
1912 bond_dev
->name
, bond_dev
->name
);
1913 unregister_netdevice(bond_dev
);
1919 * This function changes the active slave to slave <slave_dev>.
1920 * It returns -EINVAL in the following cases.
1921 * - <slave_dev> is not found in the list.
1922 * - There is not active slave now.
1923 * - <slave_dev> is already active.
1924 * - The link state of <slave_dev> is not BOND_LINK_UP.
1925 * - <slave_dev> is not running.
1926 * In these cases, this function does nothing.
1927 * In the other cases, current_slave pointer is changed and 0 is returned.
1929 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1931 struct bonding
*bond
= netdev_priv(bond_dev
);
1932 struct slave
*old_active
= NULL
;
1933 struct slave
*new_active
= NULL
;
1936 if (!USES_PRIMARY(bond
->params
.mode
))
1939 /* Verify that bond_dev is indeed the master of slave_dev */
1940 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1941 !netdev_has_upper_dev(slave_dev
, bond_dev
))
1944 read_lock(&bond
->lock
);
1946 old_active
= bond
->curr_active_slave
;
1947 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
1949 * Changing to the current active: do nothing; return success.
1951 if (new_active
&& new_active
== old_active
) {
1952 read_unlock(&bond
->lock
);
1958 new_active
->link
== BOND_LINK_UP
&&
1959 IS_UP(new_active
->dev
)) {
1961 write_lock_bh(&bond
->curr_slave_lock
);
1962 bond_change_active_slave(bond
, new_active
);
1963 write_unlock_bh(&bond
->curr_slave_lock
);
1964 unblock_netpoll_tx();
1968 read_unlock(&bond
->lock
);
1973 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1975 struct bonding
*bond
= netdev_priv(bond_dev
);
1977 info
->bond_mode
= bond
->params
.mode
;
1978 info
->miimon
= bond
->params
.miimon
;
1980 read_lock(&bond
->lock
);
1981 info
->num_slaves
= bond
->slave_cnt
;
1982 read_unlock(&bond
->lock
);
1987 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
1989 struct bonding
*bond
= netdev_priv(bond_dev
);
1990 int i
= 0, res
= -ENODEV
;
1991 struct slave
*slave
;
1993 read_lock(&bond
->lock
);
1994 bond_for_each_slave(bond
, slave
) {
1995 if (i
++ == (int)info
->slave_id
) {
1997 strcpy(info
->slave_name
, slave
->dev
->name
);
1998 info
->link
= slave
->link
;
1999 info
->state
= bond_slave_state(slave
);
2000 info
->link_failure_count
= slave
->link_failure_count
;
2004 read_unlock(&bond
->lock
);
2009 /*-------------------------------- Monitoring -------------------------------*/
2012 static int bond_miimon_inspect(struct bonding
*bond
)
2014 int link_state
, commit
= 0;
2015 struct slave
*slave
;
2016 bool ignore_updelay
;
2018 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2020 bond_for_each_slave(bond
, slave
) {
2021 slave
->new_link
= BOND_LINK_NOCHANGE
;
2023 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2025 switch (slave
->link
) {
2030 slave
->link
= BOND_LINK_FAIL
;
2031 slave
->delay
= bond
->params
.downdelay
;
2033 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2035 (bond
->params
.mode
==
2036 BOND_MODE_ACTIVEBACKUP
) ?
2037 (bond_is_active_slave(slave
) ?
2038 "active " : "backup ") : "",
2040 bond
->params
.downdelay
* bond
->params
.miimon
);
2043 case BOND_LINK_FAIL
:
2046 * recovered before downdelay expired
2048 slave
->link
= BOND_LINK_UP
;
2049 slave
->jiffies
= jiffies
;
2050 pr_info("%s: link status up again after %d ms for interface %s.\n",
2052 (bond
->params
.downdelay
- slave
->delay
) *
2053 bond
->params
.miimon
,
2058 if (slave
->delay
<= 0) {
2059 slave
->new_link
= BOND_LINK_DOWN
;
2067 case BOND_LINK_DOWN
:
2071 slave
->link
= BOND_LINK_BACK
;
2072 slave
->delay
= bond
->params
.updelay
;
2075 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2076 bond
->dev
->name
, slave
->dev
->name
,
2077 ignore_updelay
? 0 :
2078 bond
->params
.updelay
*
2079 bond
->params
.miimon
);
2082 case BOND_LINK_BACK
:
2084 slave
->link
= BOND_LINK_DOWN
;
2085 pr_info("%s: link status down again after %d ms for interface %s.\n",
2087 (bond
->params
.updelay
- slave
->delay
) *
2088 bond
->params
.miimon
,
2097 if (slave
->delay
<= 0) {
2098 slave
->new_link
= BOND_LINK_UP
;
2100 ignore_updelay
= false;
2112 static void bond_miimon_commit(struct bonding
*bond
)
2114 struct slave
*slave
;
2116 bond_for_each_slave(bond
, slave
) {
2117 switch (slave
->new_link
) {
2118 case BOND_LINK_NOCHANGE
:
2122 slave
->link
= BOND_LINK_UP
;
2123 slave
->jiffies
= jiffies
;
2125 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2126 /* prevent it from being the active one */
2127 bond_set_backup_slave(slave
);
2128 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2129 /* make it immediately active */
2130 bond_set_active_slave(slave
);
2131 } else if (slave
!= bond
->primary_slave
) {
2132 /* prevent it from being the active one */
2133 bond_set_backup_slave(slave
);
2136 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2137 bond
->dev
->name
, slave
->dev
->name
,
2138 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
2139 slave
->duplex
? "full" : "half");
2141 /* notify ad that the link status has changed */
2142 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2143 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2145 if (bond_is_lb(bond
))
2146 bond_alb_handle_link_change(bond
, slave
,
2149 if (!bond
->curr_active_slave
||
2150 (slave
== bond
->primary_slave
))
2155 case BOND_LINK_DOWN
:
2156 if (slave
->link_failure_count
< UINT_MAX
)
2157 slave
->link_failure_count
++;
2159 slave
->link
= BOND_LINK_DOWN
;
2161 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2162 bond
->params
.mode
== BOND_MODE_8023AD
)
2163 bond_set_slave_inactive_flags(slave
);
2165 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2166 bond
->dev
->name
, slave
->dev
->name
);
2168 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2169 bond_3ad_handle_link_change(slave
,
2172 if (bond_is_lb(bond
))
2173 bond_alb_handle_link_change(bond
, slave
,
2176 if (slave
== bond
->curr_active_slave
)
2182 pr_err("%s: invalid new link %d on slave %s\n",
2183 bond
->dev
->name
, slave
->new_link
,
2185 slave
->new_link
= BOND_LINK_NOCHANGE
;
2193 write_lock_bh(&bond
->curr_slave_lock
);
2194 bond_select_active_slave(bond
);
2195 write_unlock_bh(&bond
->curr_slave_lock
);
2196 unblock_netpoll_tx();
2199 bond_set_carrier(bond
);
2205 * Really a wrapper that splits the mii monitor into two phases: an
2206 * inspection, then (if inspection indicates something needs to be done)
2207 * an acquisition of appropriate locks followed by a commit phase to
2208 * implement whatever link state changes are indicated.
2210 void bond_mii_monitor(struct work_struct
*work
)
2212 struct bonding
*bond
= container_of(work
, struct bonding
,
2214 bool should_notify_peers
= false;
2215 unsigned long delay
;
2217 read_lock(&bond
->lock
);
2219 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2221 if (list_empty(&bond
->slave_list
))
2224 should_notify_peers
= bond_should_notify_peers(bond
);
2226 if (bond_miimon_inspect(bond
)) {
2227 read_unlock(&bond
->lock
);
2229 /* Race avoidance with bond_close cancel of workqueue */
2230 if (!rtnl_trylock()) {
2231 read_lock(&bond
->lock
);
2233 should_notify_peers
= false;
2237 read_lock(&bond
->lock
);
2239 bond_miimon_commit(bond
);
2241 read_unlock(&bond
->lock
);
2242 rtnl_unlock(); /* might sleep, hold no other locks */
2243 read_lock(&bond
->lock
);
2247 if (bond
->params
.miimon
)
2248 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2250 read_unlock(&bond
->lock
);
2252 if (should_notify_peers
) {
2253 if (!rtnl_trylock())
2255 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2260 static bool bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2262 struct net_device
*upper
;
2263 struct list_head
*iter
;
2266 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2270 netdev_for_each_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2271 if (ip
== bond_confirm_addr(upper
, 0, ip
)) {
2282 * We go to the (large) trouble of VLAN tagging ARP frames because
2283 * switches in VLAN mode (especially if ports are configured as
2284 * "native" to a VLAN) might not pass non-tagged frames.
2286 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2288 struct sk_buff
*skb
;
2290 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op
,
2291 slave_dev
->name
, &dest_ip
, &src_ip
, vlan_id
);
2293 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2294 NULL
, slave_dev
->dev_addr
, NULL
);
2297 pr_err("ARP packet allocation failed\n");
2301 skb
= vlan_put_tag(skb
, htons(ETH_P_8021Q
), vlan_id
);
2303 pr_err("failed to insert VLAN tag\n");
2311 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2313 struct net_device
*upper
, *vlan_upper
;
2314 struct list_head
*iter
, *vlan_iter
;
2316 __be32
*targets
= bond
->params
.arp_targets
, addr
;
2319 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
&& targets
[i
]; i
++) {
2320 pr_debug("basa: target %pI4\n", &targets
[i
]);
2322 /* Find out through which dev should the packet go */
2323 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2326 pr_debug("%s: no route to arp_ip_target %pI4\n",
2327 bond
->dev
->name
, &targets
[i
]);
2333 /* bond device itself */
2334 if (rt
->dst
.dev
== bond
->dev
)
2338 /* first we search only for vlan devices. for every vlan
2339 * found we verify its upper dev list, searching for the
2340 * rt->dst.dev. If found we save the tag of the vlan and
2341 * proceed to send the packet.
2345 netdev_for_each_upper_dev_rcu(bond
->dev
, vlan_upper
, vlan_iter
) {
2346 if (!is_vlan_dev(vlan_upper
))
2348 netdev_for_each_upper_dev_rcu(vlan_upper
, upper
, iter
) {
2349 if (upper
== rt
->dst
.dev
) {
2350 vlan_id
= vlan_dev_vlan_id(vlan_upper
);
2357 /* if the device we're looking for is not on top of any of
2358 * our upper vlans, then just search for any dev that
2359 * matches, and in case it's a vlan - save the id
2361 netdev_for_each_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2362 if (upper
== rt
->dst
.dev
) {
2363 /* if it's a vlan - get its VID */
2364 if (is_vlan_dev(upper
))
2365 vlan_id
= vlan_dev_vlan_id(upper
);
2373 /* Not our device - skip */
2374 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2375 bond
->dev
->name
, &targets
[i
],
2376 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2382 addr
= bond_confirm_addr(rt
->dst
.dev
, targets
[i
], 0);
2384 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2389 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2393 if (!sip
|| !bond_has_this_ip(bond
, tip
)) {
2394 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip
, &tip
);
2398 i
= bond_get_targets_ip(bond
->params
.arp_targets
, sip
);
2400 pr_debug("bva: sip %pI4 not found in targets\n", &sip
);
2403 slave
->last_arp_rx
= jiffies
;
2404 slave
->target_last_arp_rx
[i
] = jiffies
;
2407 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2408 struct slave
*slave
)
2410 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2411 unsigned char *arp_ptr
;
2415 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2416 return RX_HANDLER_ANOTHER
;
2418 read_lock(&bond
->lock
);
2420 if (!slave_do_arp_validate(bond
, slave
))
2423 alen
= arp_hdr_len(bond
->dev
);
2425 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2426 bond
->dev
->name
, skb
->dev
->name
);
2428 if (alen
> skb_headlen(skb
)) {
2429 arp
= kmalloc(alen
, GFP_ATOMIC
);
2432 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2436 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2437 skb
->pkt_type
== PACKET_OTHERHOST
||
2438 skb
->pkt_type
== PACKET_LOOPBACK
||
2439 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2440 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2444 arp_ptr
= (unsigned char *)(arp
+ 1);
2445 arp_ptr
+= bond
->dev
->addr_len
;
2446 memcpy(&sip
, arp_ptr
, 4);
2447 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2448 memcpy(&tip
, arp_ptr
, 4);
2450 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2451 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2452 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2456 * Backup slaves won't see the ARP reply, but do come through
2457 * here for each ARP probe (so we swap the sip/tip to validate
2458 * the probe). In a "redundant switch, common router" type of
2459 * configuration, the ARP probe will (hopefully) travel from
2460 * the active, through one switch, the router, then the other
2461 * switch before reaching the backup.
2463 * We 'trust' the arp requests if there is an active slave and
2464 * it received valid arp reply(s) after it became active. This
2465 * is done to avoid endless looping when we can't reach the
2466 * arp_ip_target and fool ourselves with our own arp requests.
2468 if (bond_is_active_slave(slave
))
2469 bond_validate_arp(bond
, slave
, sip
, tip
);
2470 else if (bond
->curr_active_slave
&&
2471 time_after(slave_last_rx(bond
, bond
->curr_active_slave
),
2472 bond
->curr_active_slave
->jiffies
))
2473 bond_validate_arp(bond
, slave
, tip
, sip
);
2476 read_unlock(&bond
->lock
);
2477 if (arp
!= (struct arphdr
*)skb
->data
)
2479 return RX_HANDLER_ANOTHER
;
2482 /* function to verify if we're in the arp_interval timeslice, returns true if
2483 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2484 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2486 static bool bond_time_in_interval(struct bonding
*bond
, unsigned long last_act
,
2489 int delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2491 return time_in_range(jiffies
,
2492 last_act
- delta_in_ticks
,
2493 last_act
+ mod
* delta_in_ticks
+ delta_in_ticks
/2);
2497 * this function is called regularly to monitor each slave's link
2498 * ensuring that traffic is being sent and received when arp monitoring
2499 * is used in load-balancing mode. if the adapter has been dormant, then an
2500 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2501 * arp monitoring in active backup mode.
2503 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2505 struct bonding
*bond
= container_of(work
, struct bonding
,
2507 struct slave
*slave
, *oldcurrent
;
2508 int do_failover
= 0;
2510 read_lock(&bond
->lock
);
2512 if (list_empty(&bond
->slave_list
))
2515 oldcurrent
= bond
->curr_active_slave
;
2516 /* see if any of the previous devices are up now (i.e. they have
2517 * xmt and rcv traffic). the curr_active_slave does not come into
2518 * the picture unless it is null. also, slave->jiffies is not needed
2519 * here because we send an arp on each slave and give a slave as
2520 * long as it needs to get the tx/rx within the delta.
2521 * TODO: what about up/down delay in arp mode? it wasn't here before
2524 bond_for_each_slave(bond
, slave
) {
2525 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2527 if (slave
->link
!= BOND_LINK_UP
) {
2528 if (bond_time_in_interval(bond
, trans_start
, 1) &&
2529 bond_time_in_interval(bond
, slave
->dev
->last_rx
, 1)) {
2531 slave
->link
= BOND_LINK_UP
;
2532 bond_set_active_slave(slave
);
2534 /* primary_slave has no meaning in round-robin
2535 * mode. the window of a slave being up and
2536 * curr_active_slave being null after enslaving
2540 pr_info("%s: link status definitely up for interface %s, ",
2545 pr_info("%s: interface %s is now up\n",
2551 /* slave->link == BOND_LINK_UP */
2553 /* not all switches will respond to an arp request
2554 * when the source ip is 0, so don't take the link down
2555 * if we don't know our ip yet
2557 if (!bond_time_in_interval(bond
, trans_start
, 2) ||
2558 !bond_time_in_interval(bond
, slave
->dev
->last_rx
, 2)) {
2560 slave
->link
= BOND_LINK_DOWN
;
2561 bond_set_backup_slave(slave
);
2563 if (slave
->link_failure_count
< UINT_MAX
)
2564 slave
->link_failure_count
++;
2566 pr_info("%s: interface %s is now down.\n",
2570 if (slave
== oldcurrent
)
2575 /* note: if switch is in round-robin mode, all links
2576 * must tx arp to ensure all links rx an arp - otherwise
2577 * links may oscillate or not come up at all; if switch is
2578 * in something like xor mode, there is nothing we can
2579 * do - all replies will be rx'ed on same link causing slaves
2580 * to be unstable during low/no traffic periods
2582 if (IS_UP(slave
->dev
))
2583 bond_arp_send_all(bond
, slave
);
2588 write_lock_bh(&bond
->curr_slave_lock
);
2590 bond_select_active_slave(bond
);
2592 write_unlock_bh(&bond
->curr_slave_lock
);
2593 unblock_netpoll_tx();
2597 if (bond
->params
.arp_interval
)
2598 queue_delayed_work(bond
->wq
, &bond
->arp_work
,
2599 msecs_to_jiffies(bond
->params
.arp_interval
));
2601 read_unlock(&bond
->lock
);
2605 * Called to inspect slaves for active-backup mode ARP monitor link state
2606 * changes. Sets new_link in slaves to specify what action should take
2607 * place for the slave. Returns 0 if no changes are found, >0 if changes
2608 * to link states must be committed.
2610 * Called with bond->lock held for read.
2612 static int bond_ab_arp_inspect(struct bonding
*bond
)
2614 unsigned long trans_start
, last_rx
;
2615 struct slave
*slave
;
2618 bond_for_each_slave(bond
, slave
) {
2619 slave
->new_link
= BOND_LINK_NOCHANGE
;
2620 last_rx
= slave_last_rx(bond
, slave
);
2622 if (slave
->link
!= BOND_LINK_UP
) {
2623 if (bond_time_in_interval(bond
, last_rx
, 1)) {
2624 slave
->new_link
= BOND_LINK_UP
;
2631 * Give slaves 2*delta after being enslaved or made
2632 * active. This avoids bouncing, as the last receive
2633 * times need a full ARP monitor cycle to be updated.
2635 if (bond_time_in_interval(bond
, slave
->jiffies
, 2))
2639 * Backup slave is down if:
2640 * - No current_arp_slave AND
2641 * - more than 3*delta since last receive AND
2642 * - the bond has an IP address
2644 * Note: a non-null current_arp_slave indicates
2645 * the curr_active_slave went down and we are
2646 * searching for a new one; under this condition
2647 * we only take the curr_active_slave down - this
2648 * gives each slave a chance to tx/rx traffic
2649 * before being taken out
2651 if (!bond_is_active_slave(slave
) &&
2652 !bond
->current_arp_slave
&&
2653 !bond_time_in_interval(bond
, last_rx
, 3)) {
2654 slave
->new_link
= BOND_LINK_DOWN
;
2659 * Active slave is down if:
2660 * - more than 2*delta since transmitting OR
2661 * - (more than 2*delta since receive AND
2662 * the bond has an IP address)
2664 trans_start
= dev_trans_start(slave
->dev
);
2665 if (bond_is_active_slave(slave
) &&
2666 (!bond_time_in_interval(bond
, trans_start
, 2) ||
2667 !bond_time_in_interval(bond
, last_rx
, 2))) {
2668 slave
->new_link
= BOND_LINK_DOWN
;
2677 * Called to commit link state changes noted by inspection step of
2678 * active-backup mode ARP monitor.
2680 * Called with RTNL and bond->lock for read.
2682 static void bond_ab_arp_commit(struct bonding
*bond
)
2684 unsigned long trans_start
;
2685 struct slave
*slave
;
2687 bond_for_each_slave(bond
, slave
) {
2688 switch (slave
->new_link
) {
2689 case BOND_LINK_NOCHANGE
:
2693 trans_start
= dev_trans_start(slave
->dev
);
2694 if (bond
->curr_active_slave
!= slave
||
2695 (!bond
->curr_active_slave
&&
2696 bond_time_in_interval(bond
, trans_start
, 1))) {
2697 slave
->link
= BOND_LINK_UP
;
2698 if (bond
->current_arp_slave
) {
2699 bond_set_slave_inactive_flags(
2700 bond
->current_arp_slave
);
2701 bond
->current_arp_slave
= NULL
;
2704 pr_info("%s: link status definitely up for interface %s.\n",
2705 bond
->dev
->name
, slave
->dev
->name
);
2707 if (!bond
->curr_active_slave
||
2708 (slave
== bond
->primary_slave
))
2715 case BOND_LINK_DOWN
:
2716 if (slave
->link_failure_count
< UINT_MAX
)
2717 slave
->link_failure_count
++;
2719 slave
->link
= BOND_LINK_DOWN
;
2720 bond_set_slave_inactive_flags(slave
);
2722 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2723 bond
->dev
->name
, slave
->dev
->name
);
2725 if (slave
== bond
->curr_active_slave
) {
2726 bond
->current_arp_slave
= NULL
;
2733 pr_err("%s: impossible: new_link %d on slave %s\n",
2734 bond
->dev
->name
, slave
->new_link
,
2742 write_lock_bh(&bond
->curr_slave_lock
);
2743 bond_select_active_slave(bond
);
2744 write_unlock_bh(&bond
->curr_slave_lock
);
2745 unblock_netpoll_tx();
2748 bond_set_carrier(bond
);
2752 * Send ARP probes for active-backup mode ARP monitor.
2754 * Called with bond->lock held for read.
2756 static void bond_ab_arp_probe(struct bonding
*bond
)
2758 struct slave
*slave
, *next_slave
;
2761 read_lock(&bond
->curr_slave_lock
);
2763 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
2764 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2765 bond
->current_arp_slave
->dev
->name
,
2766 bond
->curr_active_slave
->dev
->name
);
2768 if (bond
->curr_active_slave
) {
2769 bond_arp_send_all(bond
, bond
->curr_active_slave
);
2770 read_unlock(&bond
->curr_slave_lock
);
2774 read_unlock(&bond
->curr_slave_lock
);
2776 /* if we don't have a curr_active_slave, search for the next available
2777 * backup slave from the current_arp_slave and make it the candidate
2778 * for becoming the curr_active_slave
2781 if (!bond
->current_arp_slave
) {
2782 bond
->current_arp_slave
= bond_first_slave(bond
);
2783 if (!bond
->current_arp_slave
)
2787 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
2789 /* search for next candidate */
2790 next_slave
= bond_next_slave(bond
, bond
->current_arp_slave
);
2791 bond_for_each_slave_from(bond
, slave
, i
, next_slave
) {
2792 if (IS_UP(slave
->dev
)) {
2793 slave
->link
= BOND_LINK_BACK
;
2794 bond_set_slave_active_flags(slave
);
2795 bond_arp_send_all(bond
, slave
);
2796 slave
->jiffies
= jiffies
;
2797 bond
->current_arp_slave
= slave
;
2801 /* if the link state is up at this point, we
2802 * mark it down - this can happen if we have
2803 * simultaneous link failures and
2804 * reselect_active_interface doesn't make this
2805 * one the current slave so it is still marked
2806 * up when it is actually down
2808 if (slave
->link
== BOND_LINK_UP
) {
2809 slave
->link
= BOND_LINK_DOWN
;
2810 if (slave
->link_failure_count
< UINT_MAX
)
2811 slave
->link_failure_count
++;
2813 bond_set_slave_inactive_flags(slave
);
2815 pr_info("%s: backup interface %s is now down.\n",
2816 bond
->dev
->name
, slave
->dev
->name
);
2821 void bond_activebackup_arp_mon(struct work_struct
*work
)
2823 struct bonding
*bond
= container_of(work
, struct bonding
,
2825 bool should_notify_peers
= false;
2828 read_lock(&bond
->lock
);
2830 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2832 if (list_empty(&bond
->slave_list
))
2835 should_notify_peers
= bond_should_notify_peers(bond
);
2837 if (bond_ab_arp_inspect(bond
)) {
2838 read_unlock(&bond
->lock
);
2840 /* Race avoidance with bond_close flush of workqueue */
2841 if (!rtnl_trylock()) {
2842 read_lock(&bond
->lock
);
2844 should_notify_peers
= false;
2848 read_lock(&bond
->lock
);
2850 bond_ab_arp_commit(bond
);
2852 read_unlock(&bond
->lock
);
2854 read_lock(&bond
->lock
);
2857 bond_ab_arp_probe(bond
);
2860 if (bond
->params
.arp_interval
)
2861 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2863 read_unlock(&bond
->lock
);
2865 if (should_notify_peers
) {
2866 if (!rtnl_trylock())
2868 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2873 /*-------------------------- netdev event handling --------------------------*/
2876 * Change device name
2878 static int bond_event_changename(struct bonding
*bond
)
2880 bond_remove_proc_entry(bond
);
2881 bond_create_proc_entry(bond
);
2883 bond_debug_reregister(bond
);
2888 static int bond_master_netdev_event(unsigned long event
,
2889 struct net_device
*bond_dev
)
2891 struct bonding
*event_bond
= netdev_priv(bond_dev
);
2894 case NETDEV_CHANGENAME
:
2895 return bond_event_changename(event_bond
);
2896 case NETDEV_UNREGISTER
:
2897 bond_remove_proc_entry(event_bond
);
2899 case NETDEV_REGISTER
:
2900 bond_create_proc_entry(event_bond
);
2902 case NETDEV_NOTIFY_PEERS
:
2903 if (event_bond
->send_peer_notif
)
2904 event_bond
->send_peer_notif
--;
2913 static int bond_slave_netdev_event(unsigned long event
,
2914 struct net_device
*slave_dev
)
2916 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
);
2917 struct bonding
*bond
;
2918 struct net_device
*bond_dev
;
2922 /* A netdev event can be generated while enslaving a device
2923 * before netdev_rx_handler_register is called in which case
2924 * slave will be NULL
2928 bond_dev
= slave
->bond
->dev
;
2932 case NETDEV_UNREGISTER
:
2933 if (bond_dev
->type
!= ARPHRD_ETHER
)
2934 bond_release_and_destroy(bond_dev
, slave_dev
);
2936 bond_release(bond_dev
, slave_dev
);
2940 old_speed
= slave
->speed
;
2941 old_duplex
= slave
->duplex
;
2943 bond_update_speed_duplex(slave
);
2945 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2946 if (old_speed
!= slave
->speed
)
2947 bond_3ad_adapter_speed_changed(slave
);
2948 if (old_duplex
!= slave
->duplex
)
2949 bond_3ad_adapter_duplex_changed(slave
);
2954 * ... Or is it this?
2957 case NETDEV_CHANGEMTU
:
2959 * TODO: Should slaves be allowed to
2960 * independently alter their MTU? For
2961 * an active-backup bond, slaves need
2962 * not be the same type of device, so
2963 * MTUs may vary. For other modes,
2964 * slaves arguably should have the
2965 * same MTUs. To do this, we'd need to
2966 * take over the slave's change_mtu
2967 * function for the duration of their
2971 case NETDEV_CHANGENAME
:
2973 * TODO: handle changing the primary's name
2976 case NETDEV_FEAT_CHANGE
:
2977 bond_compute_features(bond
);
2979 case NETDEV_RESEND_IGMP
:
2980 /* Propagate to master device */
2981 call_netdevice_notifiers(event
, slave
->bond
->dev
);
2991 * bond_netdev_event: handle netdev notifier chain events.
2993 * This function receives events for the netdev chain. The caller (an
2994 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2995 * locks for us to safely manipulate the slave devices (RTNL lock,
2998 static int bond_netdev_event(struct notifier_block
*this,
2999 unsigned long event
, void *ptr
)
3001 struct net_device
*event_dev
= netdev_notifier_info_to_dev(ptr
);
3003 pr_debug("event_dev: %s, event: %lx\n",
3004 event_dev
? event_dev
->name
: "None",
3007 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3010 if (event_dev
->flags
& IFF_MASTER
) {
3011 pr_debug("IFF_MASTER\n");
3012 return bond_master_netdev_event(event
, event_dev
);
3015 if (event_dev
->flags
& IFF_SLAVE
) {
3016 pr_debug("IFF_SLAVE\n");
3017 return bond_slave_netdev_event(event
, event_dev
);
3023 static struct notifier_block bond_netdev_notifier
= {
3024 .notifier_call
= bond_netdev_event
,
3027 /*---------------------------- Hashing Policies -----------------------------*/
3030 * Hash for the output device based upon layer 2 data
3032 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3034 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3036 if (skb_headlen(skb
) >= offsetof(struct ethhdr
, h_proto
))
3037 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3043 * Hash for the output device based upon layer 2 and layer 3 data. If
3044 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3046 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3048 const struct ethhdr
*data
;
3049 const struct iphdr
*iph
;
3050 const struct ipv6hdr
*ipv6h
;
3052 const __be32
*s
, *d
;
3054 if (skb
->protocol
== htons(ETH_P_IP
) &&
3055 pskb_network_may_pull(skb
, sizeof(*iph
))) {
3057 data
= (struct ethhdr
*)skb
->data
;
3058 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3059 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3060 } else if (skb
->protocol
== htons(ETH_P_IPV6
) &&
3061 pskb_network_may_pull(skb
, sizeof(*ipv6h
))) {
3062 ipv6h
= ipv6_hdr(skb
);
3063 data
= (struct ethhdr
*)skb
->data
;
3064 s
= &ipv6h
->saddr
.s6_addr32
[0];
3065 d
= &ipv6h
->daddr
.s6_addr32
[0];
3066 v6hash
= (s
[1] ^ d
[1]) ^ (s
[2] ^ d
[2]) ^ (s
[3] ^ d
[3]);
3067 v6hash
^= (v6hash
>> 24) ^ (v6hash
>> 16) ^ (v6hash
>> 8);
3068 return (v6hash
^ data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3071 return bond_xmit_hash_policy_l2(skb
, count
);
3075 * Hash for the output device based upon layer 3 and layer 4 data. If
3076 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3077 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3079 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3082 const struct iphdr
*iph
;
3083 const struct ipv6hdr
*ipv6h
;
3084 const __be32
*s
, *d
;
3085 const __be16
*l4
= NULL
;
3087 int noff
= skb_network_offset(skb
);
3090 if (skb
->protocol
== htons(ETH_P_IP
) &&
3091 pskb_may_pull(skb
, noff
+ sizeof(*iph
))) {
3093 poff
= proto_ports_offset(iph
->protocol
);
3095 if (!ip_is_fragment(iph
) && poff
>= 0) {
3096 l4
= skb_header_pointer(skb
, noff
+ (iph
->ihl
<< 2) + poff
,
3099 layer4_xor
= ntohs(l4
[0] ^ l4
[1]);
3101 return (layer4_xor
^
3102 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3103 } else if (skb
->protocol
== htons(ETH_P_IPV6
) &&
3104 pskb_may_pull(skb
, noff
+ sizeof(*ipv6h
))) {
3105 ipv6h
= ipv6_hdr(skb
);
3106 poff
= proto_ports_offset(ipv6h
->nexthdr
);
3108 l4
= skb_header_pointer(skb
, noff
+ sizeof(*ipv6h
) + poff
,
3111 layer4_xor
= ntohs(l4
[0] ^ l4
[1]);
3113 s
= &ipv6h
->saddr
.s6_addr32
[0];
3114 d
= &ipv6h
->daddr
.s6_addr32
[0];
3115 layer4_xor
^= (s
[1] ^ d
[1]) ^ (s
[2] ^ d
[2]) ^ (s
[3] ^ d
[3]);
3116 layer4_xor
^= (layer4_xor
>> 24) ^ (layer4_xor
>> 16) ^
3118 return layer4_xor
% count
;
3121 return bond_xmit_hash_policy_l2(skb
, count
);
3124 /*-------------------------- Device entry points ----------------------------*/
3126 static void bond_work_init_all(struct bonding
*bond
)
3128 INIT_DELAYED_WORK(&bond
->mcast_work
,
3129 bond_resend_igmp_join_requests_delayed
);
3130 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3131 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3132 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3133 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3135 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3136 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3139 static void bond_work_cancel_all(struct bonding
*bond
)
3141 cancel_delayed_work_sync(&bond
->mii_work
);
3142 cancel_delayed_work_sync(&bond
->arp_work
);
3143 cancel_delayed_work_sync(&bond
->alb_work
);
3144 cancel_delayed_work_sync(&bond
->ad_work
);
3145 cancel_delayed_work_sync(&bond
->mcast_work
);
3148 static int bond_open(struct net_device
*bond_dev
)
3150 struct bonding
*bond
= netdev_priv(bond_dev
);
3151 struct slave
*slave
;
3153 /* reset slave->backup and slave->inactive */
3154 read_lock(&bond
->lock
);
3155 if (!list_empty(&bond
->slave_list
)) {
3156 read_lock(&bond
->curr_slave_lock
);
3157 bond_for_each_slave(bond
, slave
) {
3158 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3159 && (slave
!= bond
->curr_active_slave
)) {
3160 bond_set_slave_inactive_flags(slave
);
3162 bond_set_slave_active_flags(slave
);
3165 read_unlock(&bond
->curr_slave_lock
);
3167 read_unlock(&bond
->lock
);
3169 bond_work_init_all(bond
);
3171 if (bond_is_lb(bond
)) {
3172 /* bond_alb_initialize must be called before the timer
3175 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
)))
3177 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3180 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3181 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3183 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3184 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3185 if (bond
->params
.arp_validate
)
3186 bond
->recv_probe
= bond_arp_rcv
;
3189 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3190 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3191 /* register to receive LACPDUs */
3192 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3193 bond_3ad_initiate_agg_selection(bond
, 1);
3199 static int bond_close(struct net_device
*bond_dev
)
3201 struct bonding
*bond
= netdev_priv(bond_dev
);
3203 bond_work_cancel_all(bond
);
3204 bond
->send_peer_notif
= 0;
3205 if (bond_is_lb(bond
))
3206 bond_alb_deinitialize(bond
);
3207 bond
->recv_probe
= NULL
;
3212 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3213 struct rtnl_link_stats64
*stats
)
3215 struct bonding
*bond
= netdev_priv(bond_dev
);
3216 struct rtnl_link_stats64 temp
;
3217 struct slave
*slave
;
3219 memset(stats
, 0, sizeof(*stats
));
3221 read_lock_bh(&bond
->lock
);
3222 bond_for_each_slave(bond
, slave
) {
3223 const struct rtnl_link_stats64
*sstats
=
3224 dev_get_stats(slave
->dev
, &temp
);
3226 stats
->rx_packets
+= sstats
->rx_packets
;
3227 stats
->rx_bytes
+= sstats
->rx_bytes
;
3228 stats
->rx_errors
+= sstats
->rx_errors
;
3229 stats
->rx_dropped
+= sstats
->rx_dropped
;
3231 stats
->tx_packets
+= sstats
->tx_packets
;
3232 stats
->tx_bytes
+= sstats
->tx_bytes
;
3233 stats
->tx_errors
+= sstats
->tx_errors
;
3234 stats
->tx_dropped
+= sstats
->tx_dropped
;
3236 stats
->multicast
+= sstats
->multicast
;
3237 stats
->collisions
+= sstats
->collisions
;
3239 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3240 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3241 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3242 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3243 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3244 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3246 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3247 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3248 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3249 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3250 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3252 read_unlock_bh(&bond
->lock
);
3257 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3259 struct net_device
*slave_dev
= NULL
;
3260 struct ifbond k_binfo
;
3261 struct ifbond __user
*u_binfo
= NULL
;
3262 struct ifslave k_sinfo
;
3263 struct ifslave __user
*u_sinfo
= NULL
;
3264 struct mii_ioctl_data
*mii
= NULL
;
3268 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3280 * We do this again just in case we were called by SIOCGMIIREG
3281 * instead of SIOCGMIIPHY.
3288 if (mii
->reg_num
== 1) {
3289 struct bonding
*bond
= netdev_priv(bond_dev
);
3291 read_lock(&bond
->lock
);
3292 read_lock(&bond
->curr_slave_lock
);
3293 if (netif_carrier_ok(bond
->dev
))
3294 mii
->val_out
= BMSR_LSTATUS
;
3296 read_unlock(&bond
->curr_slave_lock
);
3297 read_unlock(&bond
->lock
);
3301 case BOND_INFO_QUERY_OLD
:
3302 case SIOCBONDINFOQUERY
:
3303 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3305 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3308 res
= bond_info_query(bond_dev
, &k_binfo
);
3310 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3314 case BOND_SLAVE_INFO_QUERY_OLD
:
3315 case SIOCBONDSLAVEINFOQUERY
:
3316 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3318 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3321 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3323 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3332 net
= dev_net(bond_dev
);
3334 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3337 slave_dev
= dev_get_by_name(net
, ifr
->ifr_slave
);
3339 pr_debug("slave_dev=%p:\n", slave_dev
);
3344 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3346 case BOND_ENSLAVE_OLD
:
3347 case SIOCBONDENSLAVE
:
3348 res
= bond_enslave(bond_dev
, slave_dev
);
3350 case BOND_RELEASE_OLD
:
3351 case SIOCBONDRELEASE
:
3352 res
= bond_release(bond_dev
, slave_dev
);
3354 case BOND_SETHWADDR_OLD
:
3355 case SIOCBONDSETHWADDR
:
3356 bond_set_dev_addr(bond_dev
, slave_dev
);
3359 case BOND_CHANGE_ACTIVE_OLD
:
3360 case SIOCBONDCHANGEACTIVE
:
3361 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3373 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3375 struct bonding
*bond
= netdev_priv(bond_dev
);
3377 if (change
& IFF_PROMISC
)
3378 bond_set_promiscuity(bond
,
3379 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3381 if (change
& IFF_ALLMULTI
)
3382 bond_set_allmulti(bond
,
3383 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3386 static void bond_set_rx_mode(struct net_device
*bond_dev
)
3388 struct bonding
*bond
= netdev_priv(bond_dev
);
3389 struct slave
*slave
;
3393 if (USES_PRIMARY(bond
->params
.mode
)) {
3394 slave
= rtnl_dereference(bond
->curr_active_slave
);
3396 dev_uc_sync(slave
->dev
, bond_dev
);
3397 dev_mc_sync(slave
->dev
, bond_dev
);
3400 bond_for_each_slave(bond
, slave
) {
3401 dev_uc_sync_multiple(slave
->dev
, bond_dev
);
3402 dev_mc_sync_multiple(slave
->dev
, bond_dev
);
3407 static int bond_neigh_init(struct neighbour
*n
)
3409 struct bonding
*bond
= netdev_priv(n
->dev
);
3410 const struct net_device_ops
*slave_ops
;
3411 struct neigh_parms parms
;
3412 struct slave
*slave
;
3415 slave
= bond_first_slave(bond
);
3418 slave_ops
= slave
->dev
->netdev_ops
;
3419 if (!slave_ops
->ndo_neigh_setup
)
3422 parms
.neigh_setup
= NULL
;
3423 parms
.neigh_cleanup
= NULL
;
3424 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3429 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3430 * after the last slave has been detached. Assumes that all slaves
3431 * utilize the same neigh_cleanup (true at this writing as only user
3434 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3436 if (!parms
.neigh_setup
)
3439 return parms
.neigh_setup(n
);
3443 * The bonding ndo_neigh_setup is called at init time beofre any
3444 * slave exists. So we must declare proxy setup function which will
3445 * be used at run time to resolve the actual slave neigh param setup.
3447 * It's also called by master devices (such as vlans) to setup their
3448 * underlying devices. In that case - do nothing, we're already set up from
3451 static int bond_neigh_setup(struct net_device
*dev
,
3452 struct neigh_parms
*parms
)
3454 /* modify only our neigh_parms */
3455 if (parms
->dev
== dev
)
3456 parms
->neigh_setup
= bond_neigh_init
;
3462 * Change the MTU of all of a master's slaves to match the master
3464 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3466 struct bonding
*bond
= netdev_priv(bond_dev
);
3467 struct slave
*slave
;
3470 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3471 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3473 /* Can't hold bond->lock with bh disabled here since
3474 * some base drivers panic. On the other hand we can't
3475 * hold bond->lock without bh disabled because we'll
3476 * deadlock. The only solution is to rely on the fact
3477 * that we're under rtnl_lock here, and the slaves
3478 * list won't change. This doesn't solve the problem
3479 * of setting the slave's MTU while it is
3480 * transmitting, but the assumption is that the base
3481 * driver can handle that.
3483 * TODO: figure out a way to safely iterate the slaves
3484 * list, but without holding a lock around the actual
3485 * call to the base driver.
3488 bond_for_each_slave(bond
, slave
) {
3489 pr_debug("s %p s->p %p c_m %p\n",
3491 bond_prev_slave(bond
, slave
),
3492 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3494 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3497 /* If we failed to set the slave's mtu to the new value
3498 * we must abort the operation even in ACTIVE_BACKUP
3499 * mode, because if we allow the backup slaves to have
3500 * different mtu values than the active slave we'll
3501 * need to change their mtu when doing a failover. That
3502 * means changing their mtu from timer context, which
3503 * is probably not a good idea.
3505 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3510 bond_dev
->mtu
= new_mtu
;
3515 /* unwind from head to the slave that failed */
3516 bond_for_each_slave_continue_reverse(bond
, slave
) {
3519 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3521 pr_debug("unwind err %d dev %s\n",
3522 tmp_res
, slave
->dev
->name
);
3532 * Note that many devices must be down to change the HW address, and
3533 * downing the master releases all slaves. We can make bonds full of
3534 * bonding devices to test this, however.
3536 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3538 struct bonding
*bond
= netdev_priv(bond_dev
);
3539 struct sockaddr
*sa
= addr
, tmp_sa
;
3540 struct slave
*slave
;
3543 if (bond
->params
.mode
== BOND_MODE_ALB
)
3544 return bond_alb_set_mac_address(bond_dev
, addr
);
3547 pr_debug("bond=%p, name=%s\n",
3548 bond
, bond_dev
? bond_dev
->name
: "None");
3550 /* If fail_over_mac is enabled, do nothing and return success.
3551 * Returning an error causes ifenslave to fail.
3553 if (bond
->params
.fail_over_mac
)
3556 if (!is_valid_ether_addr(sa
->sa_data
))
3557 return -EADDRNOTAVAIL
;
3559 /* Can't hold bond->lock with bh disabled here since
3560 * some base drivers panic. On the other hand we can't
3561 * hold bond->lock without bh disabled because we'll
3562 * deadlock. The only solution is to rely on the fact
3563 * that we're under rtnl_lock here, and the slaves
3564 * list won't change. This doesn't solve the problem
3565 * of setting the slave's hw address while it is
3566 * transmitting, but the assumption is that the base
3567 * driver can handle that.
3569 * TODO: figure out a way to safely iterate the slaves
3570 * list, but without holding a lock around the actual
3571 * call to the base driver.
3574 bond_for_each_slave(bond
, slave
) {
3575 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3576 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3578 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3580 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3584 res
= dev_set_mac_address(slave
->dev
, addr
);
3586 /* TODO: consider downing the slave
3588 * User should expect communications
3589 * breakage anyway until ARP finish
3592 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3598 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3602 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3603 tmp_sa
.sa_family
= bond_dev
->type
;
3605 /* unwind from head to the slave that failed */
3606 bond_for_each_slave_continue_reverse(bond
, slave
) {
3609 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3611 pr_debug("unwind err %d dev %s\n",
3612 tmp_res
, slave
->dev
->name
);
3620 * bond_xmit_slave_id - transmit skb through slave with slave_id
3621 * @bond: bonding device that is transmitting
3622 * @skb: buffer to transmit
3623 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3625 * This function tries to transmit through slave with slave_id but in case
3626 * it fails, it tries to find the first available slave for transmission.
3627 * The skb is consumed in all cases, thus the function is void.
3629 void bond_xmit_slave_id(struct bonding
*bond
, struct sk_buff
*skb
, int slave_id
)
3631 struct slave
*slave
;
3634 /* Here we start from the slave with slave_id */
3635 bond_for_each_slave_rcu(bond
, slave
) {
3637 if (slave_can_tx(slave
)) {
3638 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3644 /* Here we start from the first slave up to slave_id */
3646 bond_for_each_slave_rcu(bond
, slave
) {
3649 if (slave_can_tx(slave
)) {
3650 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3654 /* no slave that can tx has been found */
3658 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3660 struct bonding
*bond
= netdev_priv(bond_dev
);
3661 struct iphdr
*iph
= ip_hdr(skb
);
3662 struct slave
*slave
;
3665 * Start with the curr_active_slave that joined the bond as the
3666 * default for sending IGMP traffic. For failover purposes one
3667 * needs to maintain some consistency for the interface that will
3668 * send the join/membership reports. The curr_active_slave found
3669 * will send all of this type of traffic.
3671 if (iph
->protocol
== IPPROTO_IGMP
&& skb
->protocol
== htons(ETH_P_IP
)) {
3672 slave
= rcu_dereference(bond
->curr_active_slave
);
3673 if (slave
&& slave_can_tx(slave
))
3674 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3676 bond_xmit_slave_id(bond
, skb
, 0);
3678 bond_xmit_slave_id(bond
, skb
,
3679 bond
->rr_tx_counter
++ % bond
->slave_cnt
);
3682 return NETDEV_TX_OK
;
3686 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3687 * the bond has a usable interface.
3689 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3691 struct bonding
*bond
= netdev_priv(bond_dev
);
3692 struct slave
*slave
;
3694 slave
= rcu_dereference(bond
->curr_active_slave
);
3696 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3700 return NETDEV_TX_OK
;
3704 * In bond_xmit_xor() , we determine the output device by using a pre-
3705 * determined xmit_hash_policy(), If the selected device is not enabled,
3706 * find the next active slave.
3708 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3710 struct bonding
*bond
= netdev_priv(bond_dev
);
3712 bond_xmit_slave_id(bond
, skb
,
3713 bond
->xmit_hash_policy(skb
, bond
->slave_cnt
));
3715 return NETDEV_TX_OK
;
3718 /* in broadcast mode, we send everything to all usable interfaces. */
3719 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3721 struct bonding
*bond
= netdev_priv(bond_dev
);
3722 struct slave
*slave
= NULL
;
3724 bond_for_each_slave_rcu(bond
, slave
) {
3725 if (bond_is_last_slave(bond
, slave
))
3727 if (IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
) {
3728 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
3731 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3735 /* bond_dev_queue_xmit always returns 0 */
3736 bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3739 if (slave
&& IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
)
3740 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3744 return NETDEV_TX_OK
;
3747 /*------------------------- Device initialization ---------------------------*/
3749 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
3751 switch (bond
->params
.xmit_policy
) {
3752 case BOND_XMIT_POLICY_LAYER23
:
3753 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
3755 case BOND_XMIT_POLICY_LAYER34
:
3756 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
3758 case BOND_XMIT_POLICY_LAYER2
:
3760 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
3766 * Lookup the slave that corresponds to a qid
3768 static inline int bond_slave_override(struct bonding
*bond
,
3769 struct sk_buff
*skb
)
3771 struct slave
*slave
= NULL
;
3772 struct slave
*check_slave
;
3775 if (!skb
->queue_mapping
)
3778 /* Find out if any slaves have the same mapping as this skb. */
3779 bond_for_each_slave_rcu(bond
, check_slave
) {
3780 if (check_slave
->queue_id
== skb
->queue_mapping
) {
3781 slave
= check_slave
;
3786 /* If the slave isn't UP, use default transmit policy. */
3787 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
3788 (slave
->link
== BOND_LINK_UP
)) {
3789 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3796 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
3799 * This helper function exists to help dev_pick_tx get the correct
3800 * destination queue. Using a helper function skips a call to
3801 * skb_tx_hash and will put the skbs in the queue we expect on their
3802 * way down to the bonding driver.
3804 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
3807 * Save the original txq to restore before passing to the driver
3809 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
3811 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
3813 txq
-= dev
->real_num_tx_queues
;
3814 } while (txq
>= dev
->real_num_tx_queues
);
3819 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3821 struct bonding
*bond
= netdev_priv(dev
);
3823 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
3824 if (!bond_slave_override(bond
, skb
))
3825 return NETDEV_TX_OK
;
3828 switch (bond
->params
.mode
) {
3829 case BOND_MODE_ROUNDROBIN
:
3830 return bond_xmit_roundrobin(skb
, dev
);
3831 case BOND_MODE_ACTIVEBACKUP
:
3832 return bond_xmit_activebackup(skb
, dev
);
3834 return bond_xmit_xor(skb
, dev
);
3835 case BOND_MODE_BROADCAST
:
3836 return bond_xmit_broadcast(skb
, dev
);
3837 case BOND_MODE_8023AD
:
3838 return bond_3ad_xmit_xor(skb
, dev
);
3841 return bond_alb_xmit(skb
, dev
);
3843 /* Should never happen, mode already checked */
3844 pr_err("%s: Error: Unknown bonding mode %d\n",
3845 dev
->name
, bond
->params
.mode
);
3848 return NETDEV_TX_OK
;
3852 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3854 struct bonding
*bond
= netdev_priv(dev
);
3855 netdev_tx_t ret
= NETDEV_TX_OK
;
3858 * If we risk deadlock from transmitting this in the
3859 * netpoll path, tell netpoll to queue the frame for later tx
3861 if (is_netpoll_tx_blocked(dev
))
3862 return NETDEV_TX_BUSY
;
3865 if (!list_empty(&bond
->slave_list
))
3866 ret
= __bond_start_xmit(skb
, dev
);
3875 * set bond mode specific net device operations
3877 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
3879 struct net_device
*bond_dev
= bond
->dev
;
3882 case BOND_MODE_ROUNDROBIN
:
3884 case BOND_MODE_ACTIVEBACKUP
:
3887 bond_set_xmit_hash_policy(bond
);
3889 case BOND_MODE_BROADCAST
:
3891 case BOND_MODE_8023AD
:
3892 bond_set_xmit_hash_policy(bond
);
3899 /* Should never happen, mode already checked */
3900 pr_err("%s: Error: Unknown bonding mode %d\n",
3901 bond_dev
->name
, mode
);
3906 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
3907 struct ethtool_cmd
*ecmd
)
3909 struct bonding
*bond
= netdev_priv(bond_dev
);
3910 unsigned long speed
= 0;
3911 struct slave
*slave
;
3913 ecmd
->duplex
= DUPLEX_UNKNOWN
;
3914 ecmd
->port
= PORT_OTHER
;
3916 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3917 * do not need to check mode. Though link speed might not represent
3918 * the true receive or transmit bandwidth (not all modes are symmetric)
3919 * this is an accurate maximum.
3921 read_lock(&bond
->lock
);
3922 bond_for_each_slave(bond
, slave
) {
3923 if (SLAVE_IS_OK(slave
)) {
3924 if (slave
->speed
!= SPEED_UNKNOWN
)
3925 speed
+= slave
->speed
;
3926 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
3927 slave
->duplex
!= DUPLEX_UNKNOWN
)
3928 ecmd
->duplex
= slave
->duplex
;
3931 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
3932 read_unlock(&bond
->lock
);
3937 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
3938 struct ethtool_drvinfo
*drvinfo
)
3940 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
3941 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
3942 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
3946 static const struct ethtool_ops bond_ethtool_ops
= {
3947 .get_drvinfo
= bond_ethtool_get_drvinfo
,
3948 .get_settings
= bond_ethtool_get_settings
,
3949 .get_link
= ethtool_op_get_link
,
3952 static const struct net_device_ops bond_netdev_ops
= {
3953 .ndo_init
= bond_init
,
3954 .ndo_uninit
= bond_uninit
,
3955 .ndo_open
= bond_open
,
3956 .ndo_stop
= bond_close
,
3957 .ndo_start_xmit
= bond_start_xmit
,
3958 .ndo_select_queue
= bond_select_queue
,
3959 .ndo_get_stats64
= bond_get_stats
,
3960 .ndo_do_ioctl
= bond_do_ioctl
,
3961 .ndo_change_rx_flags
= bond_change_rx_flags
,
3962 .ndo_set_rx_mode
= bond_set_rx_mode
,
3963 .ndo_change_mtu
= bond_change_mtu
,
3964 .ndo_set_mac_address
= bond_set_mac_address
,
3965 .ndo_neigh_setup
= bond_neigh_setup
,
3966 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
3967 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
3968 #ifdef CONFIG_NET_POLL_CONTROLLER
3969 .ndo_netpoll_setup
= bond_netpoll_setup
,
3970 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
3971 .ndo_poll_controller
= bond_poll_controller
,
3973 .ndo_add_slave
= bond_enslave
,
3974 .ndo_del_slave
= bond_release
,
3975 .ndo_fix_features
= bond_fix_features
,
3978 static const struct device_type bond_type
= {
3982 static void bond_destructor(struct net_device
*bond_dev
)
3984 struct bonding
*bond
= netdev_priv(bond_dev
);
3986 destroy_workqueue(bond
->wq
);
3987 free_netdev(bond_dev
);
3990 static void bond_setup(struct net_device
*bond_dev
)
3992 struct bonding
*bond
= netdev_priv(bond_dev
);
3994 /* initialize rwlocks */
3995 rwlock_init(&bond
->lock
);
3996 rwlock_init(&bond
->curr_slave_lock
);
3997 INIT_LIST_HEAD(&bond
->slave_list
);
3998 bond
->params
= bonding_defaults
;
4000 /* Initialize pointers */
4001 bond
->dev
= bond_dev
;
4003 /* Initialize the device entry points */
4004 ether_setup(bond_dev
);
4005 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4006 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4007 bond_set_mode_ops(bond
, bond
->params
.mode
);
4009 bond_dev
->destructor
= bond_destructor
;
4011 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
4013 /* Initialize the device options */
4014 bond_dev
->tx_queue_len
= 0;
4015 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4016 bond_dev
->priv_flags
|= IFF_BONDING
;
4017 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
4019 /* At first, we block adding VLANs. That's the only way to
4020 * prevent problems that occur when adding VLANs over an
4021 * empty bond. The block will be removed once non-challenged
4022 * slaves are enslaved.
4024 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4026 /* don't acquire bond device's netif_tx_lock when
4028 bond_dev
->features
|= NETIF_F_LLTX
;
4030 /* By default, we declare the bond to be fully
4031 * VLAN hardware accelerated capable. Special
4032 * care is taken in the various xmit functions
4033 * when there are slaves that are not hw accel
4037 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4038 NETIF_F_HW_VLAN_CTAG_TX
|
4039 NETIF_F_HW_VLAN_CTAG_RX
|
4040 NETIF_F_HW_VLAN_CTAG_FILTER
;
4042 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
4043 bond_dev
->features
|= bond_dev
->hw_features
;
4047 * Destroy a bonding device.
4048 * Must be under rtnl_lock when this function is called.
4050 static void bond_uninit(struct net_device
*bond_dev
)
4052 struct bonding
*bond
= netdev_priv(bond_dev
);
4053 struct slave
*slave
, *tmp_slave
;
4055 bond_netpoll_cleanup(bond_dev
);
4057 /* Release the bonded slaves */
4058 list_for_each_entry_safe(slave
, tmp_slave
, &bond
->slave_list
, list
)
4059 __bond_release_one(bond_dev
, slave
->dev
, true);
4060 pr_info("%s: released all slaves\n", bond_dev
->name
);
4062 list_del(&bond
->bond_list
);
4064 bond_debug_unregister(bond
);
4067 /*------------------------- Module initialization ---------------------------*/
4070 * Convert string input module parms. Accept either the
4071 * number of the mode or its string name. A bit complicated because
4072 * some mode names are substrings of other names, and calls from sysfs
4073 * may have whitespace in the name (trailing newlines, for example).
4075 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4077 int modeint
= -1, i
, rv
;
4078 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4080 for (p
= (char *)buf
; *p
; p
++)
4081 if (!(isdigit(*p
) || isspace(*p
)))
4085 rv
= sscanf(buf
, "%20s", modestr
);
4087 rv
= sscanf(buf
, "%d", &modeint
);
4092 for (i
= 0; tbl
[i
].modename
; i
++) {
4093 if (modeint
== tbl
[i
].mode
)
4095 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4102 static int bond_check_params(struct bond_params
*params
)
4104 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
4105 int arp_all_targets_value
;
4108 * Convert string parameters.
4111 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4112 if (bond_mode
== -1) {
4113 pr_err("Error: Invalid bonding mode \"%s\"\n",
4114 mode
== NULL
? "NULL" : mode
);
4119 if (xmit_hash_policy
) {
4120 if ((bond_mode
!= BOND_MODE_XOR
) &&
4121 (bond_mode
!= BOND_MODE_8023AD
)) {
4122 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4123 bond_mode_name(bond_mode
));
4125 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4127 if (xmit_hashtype
== -1) {
4128 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4129 xmit_hash_policy
== NULL
? "NULL" :
4137 if (bond_mode
!= BOND_MODE_8023AD
) {
4138 pr_info("lacp_rate param is irrelevant in mode %s\n",
4139 bond_mode_name(bond_mode
));
4141 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4142 if (lacp_fast
== -1) {
4143 pr_err("Error: Invalid lacp rate \"%s\"\n",
4144 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4151 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4152 if (params
->ad_select
== -1) {
4153 pr_err("Error: Invalid ad_select \"%s\"\n",
4154 ad_select
== NULL
? "NULL" : ad_select
);
4158 if (bond_mode
!= BOND_MODE_8023AD
) {
4159 pr_warning("ad_select param only affects 802.3ad mode\n");
4162 params
->ad_select
= BOND_AD_STABLE
;
4165 if (max_bonds
< 0) {
4166 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4167 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4168 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4172 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4173 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4174 miimon
= BOND_LINK_MON_INTERV
;
4178 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4183 if (downdelay
< 0) {
4184 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4185 downdelay
, INT_MAX
);
4189 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4190 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4195 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4196 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4201 /* reset values for 802.3ad */
4202 if (bond_mode
== BOND_MODE_8023AD
) {
4204 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");
4205 pr_warning("Forcing miimon to 100msec\n");
4210 if (tx_queues
< 1 || tx_queues
> 255) {
4211 pr_warning("Warning: tx_queues (%d) should be between "
4212 "1 and 255, resetting to %d\n",
4213 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4214 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4217 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4218 pr_warning("Warning: all_slaves_active module parameter (%d), "
4219 "not of valid value (0/1), so it was set to "
4220 "0\n", all_slaves_active
);
4221 all_slaves_active
= 0;
4224 if (resend_igmp
< 0 || resend_igmp
> 255) {
4225 pr_warning("Warning: resend_igmp (%d) should be between "
4226 "0 and 255, resetting to %d\n",
4227 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4228 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4231 /* reset values for TLB/ALB */
4232 if ((bond_mode
== BOND_MODE_TLB
) ||
4233 (bond_mode
== BOND_MODE_ALB
)) {
4235 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");
4236 pr_warning("Forcing miimon to 100msec\n");
4241 if (bond_mode
== BOND_MODE_ALB
) {
4242 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",
4247 if (updelay
|| downdelay
) {
4248 /* just warn the user the up/down delay will have
4249 * no effect since miimon is zero...
4251 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",
4252 updelay
, downdelay
);
4255 /* don't allow arp monitoring */
4257 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4258 miimon
, arp_interval
);
4262 if ((updelay
% miimon
) != 0) {
4263 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4265 (updelay
/ miimon
) * miimon
);
4270 if ((downdelay
% miimon
) != 0) {
4271 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4273 (downdelay
/ miimon
) * miimon
);
4276 downdelay
/= miimon
;
4279 if (arp_interval
< 0) {
4280 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4281 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4282 arp_interval
= BOND_LINK_ARP_INTERV
;
4285 for (arp_ip_count
= 0, i
= 0;
4286 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4287 /* not complete check, but should be good enough to
4289 __be32 ip
= in_aton(arp_ip_target
[i
]);
4290 if (!isdigit(arp_ip_target
[i
][0]) || ip
== 0 ||
4291 ip
== htonl(INADDR_BROADCAST
)) {
4292 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4296 if (bond_get_targets_ip(arp_target
, ip
) == -1)
4297 arp_target
[arp_ip_count
++] = ip
;
4299 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4304 if (arp_interval
&& !arp_ip_count
) {
4305 /* don't allow arping if no arp_ip_target given... */
4306 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4312 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4313 pr_err("arp_validate only supported in active-backup mode\n");
4316 if (!arp_interval
) {
4317 pr_err("arp_validate requires arp_interval\n");
4321 arp_validate_value
= bond_parse_parm(arp_validate
,
4323 if (arp_validate_value
== -1) {
4324 pr_err("Error: invalid arp_validate \"%s\"\n",
4325 arp_validate
== NULL
? "NULL" : arp_validate
);
4329 arp_validate_value
= 0;
4331 arp_all_targets_value
= 0;
4332 if (arp_all_targets
) {
4333 arp_all_targets_value
= bond_parse_parm(arp_all_targets
,
4334 arp_all_targets_tbl
);
4336 if (arp_all_targets_value
== -1) {
4337 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4339 arp_all_targets_value
= 0;
4344 pr_info("MII link monitoring set to %d ms\n", miimon
);
4345 } else if (arp_interval
) {
4346 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4348 arp_validate_tbl
[arp_validate_value
].modename
,
4351 for (i
= 0; i
< arp_ip_count
; i
++)
4352 pr_info(" %s", arp_ip_target
[i
]);
4356 } else if (max_bonds
) {
4357 /* miimon and arp_interval not set, we need one so things
4358 * work as expected, see bonding.txt for details
4360 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");
4363 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4364 /* currently, using a primary only makes sense
4365 * in active backup, TLB or ALB modes
4367 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4368 primary
, bond_mode_name(bond_mode
));
4372 if (primary
&& primary_reselect
) {
4373 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4375 if (primary_reselect_value
== -1) {
4376 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4378 NULL
? "NULL" : primary_reselect
);
4382 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4385 if (fail_over_mac
) {
4386 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4388 if (fail_over_mac_value
== -1) {
4389 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4390 arp_validate
== NULL
? "NULL" : arp_validate
);
4394 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4395 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4397 fail_over_mac_value
= BOND_FOM_NONE
;
4400 /* fill params struct with the proper values */
4401 params
->mode
= bond_mode
;
4402 params
->xmit_policy
= xmit_hashtype
;
4403 params
->miimon
= miimon
;
4404 params
->num_peer_notif
= num_peer_notif
;
4405 params
->arp_interval
= arp_interval
;
4406 params
->arp_validate
= arp_validate_value
;
4407 params
->arp_all_targets
= arp_all_targets_value
;
4408 params
->updelay
= updelay
;
4409 params
->downdelay
= downdelay
;
4410 params
->use_carrier
= use_carrier
;
4411 params
->lacp_fast
= lacp_fast
;
4412 params
->primary
[0] = 0;
4413 params
->primary_reselect
= primary_reselect_value
;
4414 params
->fail_over_mac
= fail_over_mac_value
;
4415 params
->tx_queues
= tx_queues
;
4416 params
->all_slaves_active
= all_slaves_active
;
4417 params
->resend_igmp
= resend_igmp
;
4418 params
->min_links
= min_links
;
4421 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4422 params
->primary
[IFNAMSIZ
- 1] = 0;
4425 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4430 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4431 static struct lock_class_key bonding_netdev_addr_lock_key
;
4432 static struct lock_class_key bonding_tx_busylock_key
;
4434 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4435 struct netdev_queue
*txq
,
4438 lockdep_set_class(&txq
->_xmit_lock
,
4439 &bonding_netdev_xmit_lock_key
);
4442 static void bond_set_lockdep_class(struct net_device
*dev
)
4444 lockdep_set_class(&dev
->addr_list_lock
,
4445 &bonding_netdev_addr_lock_key
);
4446 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4447 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4451 * Called from registration process
4453 static int bond_init(struct net_device
*bond_dev
)
4455 struct bonding
*bond
= netdev_priv(bond_dev
);
4456 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4457 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4459 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4462 * Initialize locks that may be required during
4463 * en/deslave operations. All of the bond_open work
4464 * (of which this is part) should really be moved to
4465 * a phase prior to dev_open
4467 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4468 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4470 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4474 bond_set_lockdep_class(bond_dev
);
4476 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4478 bond_prepare_sysfs_group(bond
);
4480 bond_debug_register(bond
);
4482 /* Ensure valid dev_addr */
4483 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4484 bond_dev
->addr_assign_type
== NET_ADDR_PERM
)
4485 eth_hw_addr_random(bond_dev
);
4490 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4492 if (tb
[IFLA_ADDRESS
]) {
4493 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4495 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4496 return -EADDRNOTAVAIL
;
4501 static unsigned int bond_get_num_tx_queues(void)
4506 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
4508 .priv_size
= sizeof(struct bonding
),
4509 .setup
= bond_setup
,
4510 .validate
= bond_validate
,
4511 .get_num_tx_queues
= bond_get_num_tx_queues
,
4512 .get_num_rx_queues
= bond_get_num_tx_queues
, /* Use the same number
4516 /* Create a new bond based on the specified name and bonding parameters.
4517 * If name is NULL, obtain a suitable "bond%d" name for us.
4518 * Caller must NOT hold rtnl_lock; we need to release it here before we
4519 * set up our sysfs entries.
4521 int bond_create(struct net
*net
, const char *name
)
4523 struct net_device
*bond_dev
;
4528 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4529 name
? name
: "bond%d",
4530 bond_setup
, tx_queues
);
4532 pr_err("%s: eek! can't alloc netdev!\n", name
);
4537 dev_net_set(bond_dev
, net
);
4538 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4540 res
= register_netdevice(bond_dev
);
4542 netif_carrier_off(bond_dev
);
4546 bond_destructor(bond_dev
);
4550 static int __net_init
bond_net_init(struct net
*net
)
4552 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4555 INIT_LIST_HEAD(&bn
->dev_list
);
4557 bond_create_proc_dir(bn
);
4558 bond_create_sysfs(bn
);
4563 static void __net_exit
bond_net_exit(struct net
*net
)
4565 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4566 struct bonding
*bond
, *tmp_bond
;
4569 bond_destroy_sysfs(bn
);
4570 bond_destroy_proc_dir(bn
);
4572 /* Kill off any bonds created after unregistering bond rtnl ops */
4574 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4575 unregister_netdevice_queue(bond
->dev
, &list
);
4576 unregister_netdevice_many(&list
);
4580 static struct pernet_operations bond_net_ops
= {
4581 .init
= bond_net_init
,
4582 .exit
= bond_net_exit
,
4584 .size
= sizeof(struct bond_net
),
4587 static int __init
bonding_init(void)
4592 pr_info("%s", bond_version
);
4594 res
= bond_check_params(&bonding_defaults
);
4598 res
= register_pernet_subsys(&bond_net_ops
);
4602 res
= rtnl_link_register(&bond_link_ops
);
4606 bond_create_debugfs();
4608 for (i
= 0; i
< max_bonds
; i
++) {
4609 res
= bond_create(&init_net
, NULL
);
4614 register_netdevice_notifier(&bond_netdev_notifier
);
4618 rtnl_link_unregister(&bond_link_ops
);
4620 unregister_pernet_subsys(&bond_net_ops
);
4625 static void __exit
bonding_exit(void)
4627 unregister_netdevice_notifier(&bond_netdev_notifier
);
4629 bond_destroy_debugfs();
4631 rtnl_link_unregister(&bond_link_ops
);
4632 unregister_pernet_subsys(&bond_net_ops
);
4634 #ifdef CONFIG_NET_POLL_CONTROLLER
4636 * Make sure we don't have an imbalance on our netpoll blocking
4638 WARN_ON(atomic_read(&netpoll_block_tx
));
4642 module_init(bonding_init
);
4643 module_exit(bonding_exit
);
4644 MODULE_LICENSE("GPL");
4645 MODULE_VERSION(DRV_VERSION
);
4646 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4647 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4648 MODULE_ALIAS_RTNL_LINK("bond");