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>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV 0
88 #define BOND_LINK_ARP_INTERV 0
90 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
91 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
92 static int num_peer_notif
= 1;
93 static int miimon
= BOND_LINK_MON_INTERV
;
96 static int use_carrier
= 1;
99 static char *primary_reselect
;
100 static char *lacp_rate
;
101 static int min_links
;
102 static char *ad_select
;
103 static char *xmit_hash_policy
;
104 static int arp_interval
= BOND_LINK_ARP_INTERV
;
105 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
106 static char *arp_validate
;
107 static char *arp_all_targets
;
108 static char *fail_over_mac
;
109 static int all_slaves_active
;
110 static struct bond_params bonding_defaults
;
111 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
113 module_param(max_bonds
, int, 0);
114 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
115 module_param(tx_queues
, int, 0);
116 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
117 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
118 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
119 "failover event (alias of num_unsol_na)");
120 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
121 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
122 "failover event (alias of num_grat_arp)");
123 module_param(miimon
, int, 0);
124 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
125 module_param(updelay
, int, 0);
126 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
127 module_param(downdelay
, int, 0);
128 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
130 module_param(use_carrier
, int, 0);
131 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132 "0 for off, 1 for on (default)");
133 module_param(mode
, charp
, 0);
134 MODULE_PARM_DESC(mode
, "Mode of operation; 0 for balance-rr, "
135 "1 for active-backup, 2 for balance-xor, "
136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137 "6 for balance-alb");
138 module_param(primary
, charp
, 0);
139 MODULE_PARM_DESC(primary
, "Primary network device to use");
140 module_param(primary_reselect
, charp
, 0);
141 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
143 "0 for always (default), "
144 "1 for only if speed of primary is "
146 "2 for only on active slave "
148 module_param(lacp_rate
, charp
, 0);
149 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner; "
150 "0 for slow, 1 for fast");
151 module_param(ad_select
, charp
, 0);
152 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
153 "0 for stable (default), 1 for bandwidth, "
155 module_param(min_links
, int, 0);
156 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
158 module_param(xmit_hash_policy
, charp
, 0);
159 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
160 "0 for layer 2 (default), 1 for layer 3+4, "
162 module_param(arp_interval
, int, 0);
163 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
164 module_param_array(arp_ip_target
, charp
, NULL
, 0);
165 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
166 module_param(arp_validate
, charp
, 0);
167 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
168 "0 for none (default), 1 for active, "
169 "2 for backup, 3 for all");
170 module_param(arp_all_targets
, charp
, 0);
171 MODULE_PARM_DESC(arp_all_targets
, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
172 module_param(fail_over_mac
, charp
, 0);
173 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
174 "the same MAC; 0 for none (default), "
175 "1 for active, 2 for follow");
176 module_param(all_slaves_active
, int, 0);
177 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
178 "by setting active flag for all slaves; "
179 "0 for never (default), 1 for always.");
180 module_param(resend_igmp
, int, 0);
181 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
184 /*----------------------------- Global variables ----------------------------*/
186 #ifdef CONFIG_NET_POLL_CONTROLLER
187 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
190 int bond_net_id __read_mostly
;
192 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
193 static int arp_ip_count
;
194 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
195 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
196 static int lacp_fast
;
198 const struct bond_parm_tbl bond_lacp_tbl
[] = {
199 { "slow", AD_LACP_SLOW
},
200 { "fast", AD_LACP_FAST
},
204 const struct bond_parm_tbl bond_mode_tbl
[] = {
205 { "balance-rr", BOND_MODE_ROUNDROBIN
},
206 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
207 { "balance-xor", BOND_MODE_XOR
},
208 { "broadcast", BOND_MODE_BROADCAST
},
209 { "802.3ad", BOND_MODE_8023AD
},
210 { "balance-tlb", BOND_MODE_TLB
},
211 { "balance-alb", BOND_MODE_ALB
},
215 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
216 { "layer2", BOND_XMIT_POLICY_LAYER2
},
217 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
218 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
222 const struct bond_parm_tbl arp_all_targets_tbl
[] = {
223 { "any", BOND_ARP_TARGETS_ANY
},
224 { "all", BOND_ARP_TARGETS_ALL
},
228 const struct bond_parm_tbl arp_validate_tbl
[] = {
229 { "none", BOND_ARP_VALIDATE_NONE
},
230 { "active", BOND_ARP_VALIDATE_ACTIVE
},
231 { "backup", BOND_ARP_VALIDATE_BACKUP
},
232 { "all", BOND_ARP_VALIDATE_ALL
},
236 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
237 { "none", BOND_FOM_NONE
},
238 { "active", BOND_FOM_ACTIVE
},
239 { "follow", BOND_FOM_FOLLOW
},
243 const struct bond_parm_tbl pri_reselect_tbl
[] = {
244 { "always", BOND_PRI_RESELECT_ALWAYS
},
245 { "better", BOND_PRI_RESELECT_BETTER
},
246 { "failure", BOND_PRI_RESELECT_FAILURE
},
250 struct bond_parm_tbl ad_select_tbl
[] = {
251 { "stable", BOND_AD_STABLE
},
252 { "bandwidth", BOND_AD_BANDWIDTH
},
253 { "count", BOND_AD_COUNT
},
257 /*-------------------------- Forward declarations ---------------------------*/
259 static int bond_init(struct net_device
*bond_dev
);
260 static void bond_uninit(struct net_device
*bond_dev
);
262 /*---------------------------- General routines -----------------------------*/
264 const char *bond_mode_name(int mode
)
266 static const char *names
[] = {
267 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
268 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
269 [BOND_MODE_XOR
] = "load balancing (xor)",
270 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
271 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
272 [BOND_MODE_TLB
] = "transmit load balancing",
273 [BOND_MODE_ALB
] = "adaptive load balancing",
276 if (mode
< BOND_MODE_ROUNDROBIN
|| mode
> BOND_MODE_ALB
)
282 /*---------------------------------- VLAN -----------------------------------*/
285 * bond_add_vlan - add a new vlan id on bond
286 * @bond: bond that got the notification
287 * @vlan_id: the vlan id to add
289 * Returns -ENOMEM if allocation failed.
291 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
293 struct vlan_entry
*vlan
;
295 pr_debug("bond: %s, vlan id %d\n",
296 (bond
? bond
->dev
->name
: "None"), vlan_id
);
298 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
302 INIT_LIST_HEAD(&vlan
->vlan_list
);
303 vlan
->vlan_id
= vlan_id
;
305 write_lock_bh(&bond
->lock
);
307 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
309 write_unlock_bh(&bond
->lock
);
311 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
317 * bond_del_vlan - delete a vlan id from bond
318 * @bond: bond that got the notification
319 * @vlan_id: the vlan id to delete
321 * returns -ENODEV if @vlan_id was not found in @bond.
323 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
325 struct vlan_entry
*vlan
;
328 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
331 write_lock_bh(&bond
->lock
);
333 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
334 if (vlan
->vlan_id
== vlan_id
) {
335 list_del(&vlan
->vlan_list
);
337 if (bond_is_lb(bond
))
338 bond_alb_clear_vlan(bond
, vlan_id
);
340 pr_debug("removed VLAN ID %d from bond %s\n",
341 vlan_id
, bond
->dev
->name
);
350 pr_debug("couldn't find VLAN ID %d in bond %s\n",
351 vlan_id
, bond
->dev
->name
);
354 write_unlock_bh(&bond
->lock
);
355 unblock_netpoll_tx();
360 * bond_next_vlan - safely skip to the next item in the vlans list.
361 * @bond: the bond we're working on
362 * @curr: item we're advancing from
364 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
365 * or @curr->next otherwise (even if it is @curr itself again).
367 * Caller must hold bond->lock
369 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
371 struct vlan_entry
*next
, *last
;
373 if (list_empty(&bond
->vlan_list
))
377 next
= list_entry(bond
->vlan_list
.next
,
378 struct vlan_entry
, vlan_list
);
380 last
= list_entry(bond
->vlan_list
.prev
,
381 struct vlan_entry
, vlan_list
);
383 next
= list_entry(bond
->vlan_list
.next
,
384 struct vlan_entry
, vlan_list
);
386 next
= list_entry(curr
->vlan_list
.next
,
387 struct vlan_entry
, vlan_list
);
395 * bond_dev_queue_xmit - Prepare skb for xmit.
397 * @bond: bond device that got this skb for tx.
398 * @skb: hw accel VLAN tagged skb to transmit
399 * @slave_dev: slave that is supposed to xmit this skbuff
401 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
402 struct net_device
*slave_dev
)
404 skb
->dev
= slave_dev
;
406 BUILD_BUG_ON(sizeof(skb
->queue_mapping
) !=
407 sizeof(qdisc_skb_cb(skb
)->slave_dev_queue_mapping
));
408 skb
->queue_mapping
= qdisc_skb_cb(skb
)->slave_dev_queue_mapping
;
410 if (unlikely(netpoll_tx_running(bond
->dev
)))
411 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
419 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
420 * We don't protect the slave list iteration with a lock because:
421 * a. This operation is performed in IOCTL context,
422 * b. The operation is protected by the RTNL semaphore in the 8021q code,
423 * c. Holding a lock with BH disabled while directly calling a base driver
424 * entry point is generally a BAD idea.
426 * The design of synchronization/protection for this operation in the 8021q
427 * module is good for one or more VLAN devices over a single physical device
428 * and cannot be extended for a teaming solution like bonding, so there is a
429 * potential race condition here where a net device from the vlan group might
430 * be referenced (either by a base driver or the 8021q code) while it is being
431 * removed from the system. However, it turns out we're not making matters
432 * worse, and if it works for regular VLAN usage it will work here too.
436 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
437 * @bond_dev: bonding net device that got called
438 * @vid: vlan id being added
440 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
,
441 __be16 proto
, u16 vid
)
443 struct bonding
*bond
= netdev_priv(bond_dev
);
447 bond_for_each_slave(bond
, slave
) {
448 res
= vlan_vid_add(slave
->dev
, proto
, vid
);
453 res
= bond_add_vlan(bond
, vid
);
455 pr_err("%s: Error: Failed to add vlan id %d\n",
456 bond_dev
->name
, vid
);
463 /* unwind from head to the slave that failed */
464 bond_for_each_slave_continue_reverse(bond
, slave
)
465 vlan_vid_del(slave
->dev
, proto
, vid
);
471 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
472 * @bond_dev: bonding net device that got called
473 * @vid: vlan id being removed
475 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
,
476 __be16 proto
, u16 vid
)
478 struct bonding
*bond
= netdev_priv(bond_dev
);
482 bond_for_each_slave(bond
, slave
)
483 vlan_vid_del(slave
->dev
, proto
, vid
);
485 res
= bond_del_vlan(bond
, vid
);
487 pr_err("%s: Error: Failed to remove vlan id %d\n",
488 bond_dev
->name
, vid
);
495 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
497 struct vlan_entry
*vlan
;
500 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
501 res
= vlan_vid_add(slave_dev
, htons(ETH_P_8021Q
),
504 pr_warning("%s: Failed to add vlan id %d to device %s\n",
505 bond
->dev
->name
, vlan
->vlan_id
,
510 static void bond_del_vlans_from_slave(struct bonding
*bond
,
511 struct net_device
*slave_dev
)
513 struct vlan_entry
*vlan
;
515 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
518 vlan_vid_del(slave_dev
, htons(ETH_P_8021Q
), vlan
->vlan_id
);
522 /*------------------------------- Link status -------------------------------*/
525 * Set the carrier state for the master according to the state of its
526 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
527 * do special 802.3ad magic.
529 * Returns zero if carrier state does not change, nonzero if it does.
531 static int bond_set_carrier(struct bonding
*bond
)
535 if (list_empty(&bond
->slave_list
))
538 if (bond
->params
.mode
== BOND_MODE_8023AD
)
539 return bond_3ad_set_carrier(bond
);
541 bond_for_each_slave(bond
, slave
) {
542 if (slave
->link
== BOND_LINK_UP
) {
543 if (!netif_carrier_ok(bond
->dev
)) {
544 netif_carrier_on(bond
->dev
);
552 if (netif_carrier_ok(bond
->dev
)) {
553 netif_carrier_off(bond
->dev
);
560 * Get link speed and duplex from the slave's base driver
561 * using ethtool. If for some reason the call fails or the
562 * values are invalid, set speed and duplex to -1,
565 static void bond_update_speed_duplex(struct slave
*slave
)
567 struct net_device
*slave_dev
= slave
->dev
;
568 struct ethtool_cmd ecmd
;
572 slave
->speed
= SPEED_UNKNOWN
;
573 slave
->duplex
= DUPLEX_UNKNOWN
;
575 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
579 slave_speed
= ethtool_cmd_speed(&ecmd
);
580 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
583 switch (ecmd
.duplex
) {
591 slave
->speed
= slave_speed
;
592 slave
->duplex
= ecmd
.duplex
;
598 * if <dev> supports MII link status reporting, check its link status.
600 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
601 * depending upon the setting of the use_carrier parameter.
603 * Return either BMSR_LSTATUS, meaning that the link is up (or we
604 * can't tell and just pretend it is), or 0, meaning that the link is
607 * If reporting is non-zero, instead of faking link up, return -1 if
608 * both ETHTOOL and MII ioctls fail (meaning the device does not
609 * support them). If use_carrier is set, return whatever it says.
610 * It'd be nice if there was a good way to tell if a driver supports
611 * netif_carrier, but there really isn't.
613 static int bond_check_dev_link(struct bonding
*bond
,
614 struct net_device
*slave_dev
, int reporting
)
616 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
617 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
619 struct mii_ioctl_data
*mii
;
621 if (!reporting
&& !netif_running(slave_dev
))
624 if (bond
->params
.use_carrier
)
625 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
627 /* Try to get link status using Ethtool first. */
628 if (slave_dev
->ethtool_ops
->get_link
)
629 return slave_dev
->ethtool_ops
->get_link(slave_dev
) ?
632 /* Ethtool can't be used, fallback to MII ioctls. */
633 ioctl
= slave_ops
->ndo_do_ioctl
;
635 /* TODO: set pointer to correct ioctl on a per team member */
636 /* bases to make this more efficient. that is, once */
637 /* we determine the correct ioctl, we will always */
638 /* call it and not the others for that team */
642 * We cannot assume that SIOCGMIIPHY will also read a
643 * register; not all network drivers (e.g., e100)
647 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
648 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
650 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
651 mii
->reg_num
= MII_BMSR
;
652 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
653 return mii
->val_out
& BMSR_LSTATUS
;
658 * If reporting, report that either there's no dev->do_ioctl,
659 * or both SIOCGMIIREG and get_link failed (meaning that we
660 * cannot report link status). If not reporting, pretend
663 return reporting
? -1 : BMSR_LSTATUS
;
666 /*----------------------------- Multicast list ------------------------------*/
669 * Push the promiscuity flag down to appropriate slaves
671 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
674 if (USES_PRIMARY(bond
->params
.mode
)) {
675 /* write lock already acquired */
676 if (bond
->curr_active_slave
) {
677 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
683 bond_for_each_slave(bond
, slave
) {
684 err
= dev_set_promiscuity(slave
->dev
, inc
);
693 * Push the allmulti flag down to all slaves
695 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
698 if (USES_PRIMARY(bond
->params
.mode
)) {
699 /* write lock already acquired */
700 if (bond
->curr_active_slave
) {
701 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
707 bond_for_each_slave(bond
, slave
) {
708 err
= dev_set_allmulti(slave
->dev
, inc
);
717 * Retrieve the list of registered multicast addresses for the bonding
718 * device and retransmit an IGMP JOIN request to the current active
721 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
723 if (!rtnl_trylock()) {
724 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
727 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, bond
->dev
);
730 /* We use curr_slave_lock to protect against concurrent access to
731 * igmp_retrans from multiple running instances of this function and
732 * bond_change_active_slave
734 write_lock_bh(&bond
->curr_slave_lock
);
735 if (bond
->igmp_retrans
> 1) {
736 bond
->igmp_retrans
--;
737 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
739 write_unlock_bh(&bond
->curr_slave_lock
);
742 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
744 struct bonding
*bond
= container_of(work
, struct bonding
,
747 bond_resend_igmp_join_requests(bond
);
750 /* Flush bond's hardware addresses from slave
752 static void bond_hw_addr_flush(struct net_device
*bond_dev
,
753 struct net_device
*slave_dev
)
755 struct bonding
*bond
= netdev_priv(bond_dev
);
757 dev_uc_unsync(slave_dev
, bond_dev
);
758 dev_mc_unsync(slave_dev
, bond_dev
);
760 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
761 /* del lacpdu mc addr from mc list */
762 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
764 dev_mc_del(slave_dev
, lacpdu_multicast
);
768 /*--------------------------- Active slave change ---------------------------*/
770 /* Update the hardware address list and promisc/allmulti for the new and
771 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
772 * slaves up date at all times; only the USES_PRIMARY modes need to call
773 * this function to swap these settings during a failover.
775 static void bond_hw_addr_swap(struct bonding
*bond
, struct slave
*new_active
,
776 struct slave
*old_active
)
779 if (bond
->dev
->flags
& IFF_PROMISC
)
780 dev_set_promiscuity(old_active
->dev
, -1);
782 if (bond
->dev
->flags
& IFF_ALLMULTI
)
783 dev_set_allmulti(old_active
->dev
, -1);
785 bond_hw_addr_flush(bond
->dev
, old_active
->dev
);
789 /* FIXME: Signal errors upstream. */
790 if (bond
->dev
->flags
& IFF_PROMISC
)
791 dev_set_promiscuity(new_active
->dev
, 1);
793 if (bond
->dev
->flags
& IFF_ALLMULTI
)
794 dev_set_allmulti(new_active
->dev
, 1);
796 netif_addr_lock_bh(bond
->dev
);
797 dev_uc_sync(new_active
->dev
, bond
->dev
);
798 dev_mc_sync(new_active
->dev
, bond
->dev
);
799 netif_addr_unlock_bh(bond
->dev
);
804 * bond_set_dev_addr - clone slave's address to bond
805 * @bond_dev: bond net device
806 * @slave_dev: slave net device
808 * Should be called with RTNL held.
810 static void bond_set_dev_addr(struct net_device
*bond_dev
,
811 struct net_device
*slave_dev
)
813 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
814 bond_dev
, slave_dev
, slave_dev
->addr_len
);
815 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
816 bond_dev
->addr_assign_type
= NET_ADDR_STOLEN
;
817 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
821 * bond_do_fail_over_mac
823 * Perform special MAC address swapping for fail_over_mac settings
825 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
827 static void bond_do_fail_over_mac(struct bonding
*bond
,
828 struct slave
*new_active
,
829 struct slave
*old_active
)
830 __releases(&bond
->curr_slave_lock
)
831 __releases(&bond
->lock
)
832 __acquires(&bond
->lock
)
833 __acquires(&bond
->curr_slave_lock
)
835 u8 tmp_mac
[ETH_ALEN
];
836 struct sockaddr saddr
;
839 switch (bond
->params
.fail_over_mac
) {
840 case BOND_FOM_ACTIVE
:
842 write_unlock_bh(&bond
->curr_slave_lock
);
843 read_unlock(&bond
->lock
);
844 bond_set_dev_addr(bond
->dev
, new_active
->dev
);
845 read_lock(&bond
->lock
);
846 write_lock_bh(&bond
->curr_slave_lock
);
849 case BOND_FOM_FOLLOW
:
851 * if new_active && old_active, swap them
852 * if just old_active, do nothing (going to no active slave)
853 * if just new_active, set new_active to bond's MAC
858 write_unlock_bh(&bond
->curr_slave_lock
);
859 read_unlock(&bond
->lock
);
862 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
863 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
865 saddr
.sa_family
= new_active
->dev
->type
;
867 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
868 saddr
.sa_family
= bond
->dev
->type
;
871 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
873 pr_err("%s: Error %d setting MAC of slave %s\n",
874 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
881 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
882 saddr
.sa_family
= old_active
->dev
->type
;
884 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
886 pr_err("%s: Error %d setting MAC of slave %s\n",
887 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
889 read_lock(&bond
->lock
);
890 write_lock_bh(&bond
->curr_slave_lock
);
893 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
894 bond
->dev
->name
, bond
->params
.fail_over_mac
);
900 static bool bond_should_change_active(struct bonding
*bond
)
902 struct slave
*prim
= bond
->primary_slave
;
903 struct slave
*curr
= bond
->curr_active_slave
;
905 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
907 if (bond
->force_primary
) {
908 bond
->force_primary
= false;
911 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
912 (prim
->speed
< curr
->speed
||
913 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
915 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
921 * find_best_interface - select the best available slave to be the active one
922 * @bond: our bonding struct
924 * Warning: Caller must hold curr_slave_lock for writing.
926 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
928 struct slave
*new_active
, *old_active
;
929 struct slave
*bestslave
= NULL
;
930 int mintime
= bond
->params
.updelay
;
933 new_active
= bond
->curr_active_slave
;
935 if (!new_active
) { /* there were no active slaves left */
936 new_active
= bond_first_slave(bond
);
938 return NULL
; /* still no slave, return NULL */
941 if ((bond
->primary_slave
) &&
942 bond
->primary_slave
->link
== BOND_LINK_UP
&&
943 bond_should_change_active(bond
)) {
944 new_active
= bond
->primary_slave
;
947 /* remember where to stop iterating over the slaves */
948 old_active
= new_active
;
950 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
951 if (new_active
->link
== BOND_LINK_UP
) {
953 } else if (new_active
->link
== BOND_LINK_BACK
&&
954 IS_UP(new_active
->dev
)) {
955 /* link up, but waiting for stabilization */
956 if (new_active
->delay
< mintime
) {
957 mintime
= new_active
->delay
;
958 bestslave
= new_active
;
966 static bool bond_should_notify_peers(struct bonding
*bond
)
968 struct slave
*slave
= bond
->curr_active_slave
;
970 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
971 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
973 if (!slave
|| !bond
->send_peer_notif
||
974 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
977 bond
->send_peer_notif
--;
982 * change_active_interface - change the active slave into the specified one
983 * @bond: our bonding struct
984 * @new: the new slave to make the active one
986 * Set the new slave to the bond's settings and unset them on the old
988 * Setting include flags, mc-list, promiscuity, allmulti, etc.
990 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
991 * because it is apparently the best available slave we have, even though its
992 * updelay hasn't timed out yet.
994 * If new_active is not NULL, caller must hold bond->lock for read and
995 * curr_slave_lock for write_bh.
997 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
999 struct slave
*old_active
= bond
->curr_active_slave
;
1001 if (old_active
== new_active
)
1005 new_active
->jiffies
= jiffies
;
1007 if (new_active
->link
== BOND_LINK_BACK
) {
1008 if (USES_PRIMARY(bond
->params
.mode
)) {
1009 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1010 bond
->dev
->name
, new_active
->dev
->name
,
1011 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1014 new_active
->delay
= 0;
1015 new_active
->link
= BOND_LINK_UP
;
1017 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1018 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1020 if (bond_is_lb(bond
))
1021 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1023 if (USES_PRIMARY(bond
->params
.mode
)) {
1024 pr_info("%s: making interface %s the new active one.\n",
1025 bond
->dev
->name
, new_active
->dev
->name
);
1030 if (USES_PRIMARY(bond
->params
.mode
))
1031 bond_hw_addr_swap(bond
, new_active
, old_active
);
1033 if (bond_is_lb(bond
)) {
1034 bond_alb_handle_active_change(bond
, new_active
);
1036 bond_set_slave_inactive_flags(old_active
);
1038 bond_set_slave_active_flags(new_active
);
1040 bond
->curr_active_slave
= new_active
;
1043 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1045 bond_set_slave_inactive_flags(old_active
);
1048 bool should_notify_peers
= false;
1050 bond_set_slave_active_flags(new_active
);
1052 if (bond
->params
.fail_over_mac
)
1053 bond_do_fail_over_mac(bond
, new_active
,
1056 if (netif_running(bond
->dev
)) {
1057 bond
->send_peer_notif
=
1058 bond
->params
.num_peer_notif
;
1059 should_notify_peers
=
1060 bond_should_notify_peers(bond
);
1063 write_unlock_bh(&bond
->curr_slave_lock
);
1064 read_unlock(&bond
->lock
);
1066 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
1067 if (should_notify_peers
)
1068 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
1071 read_lock(&bond
->lock
);
1072 write_lock_bh(&bond
->curr_slave_lock
);
1076 /* resend IGMP joins since active slave has changed or
1077 * all were sent on curr_active_slave.
1078 * resend only if bond is brought up with the affected
1079 * bonding modes and the retransmission is enabled */
1080 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
1081 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
1082 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
1083 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
1084 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
1089 * bond_select_active_slave - select a new active slave, if needed
1090 * @bond: our bonding struct
1092 * This functions should be called when one of the following occurs:
1093 * - The old curr_active_slave has been released or lost its link.
1094 * - The primary_slave has got its link back.
1095 * - A slave has got its link back and there's no old curr_active_slave.
1097 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1099 void bond_select_active_slave(struct bonding
*bond
)
1101 struct slave
*best_slave
;
1104 best_slave
= bond_find_best_slave(bond
);
1105 if (best_slave
!= bond
->curr_active_slave
) {
1106 bond_change_active_slave(bond
, best_slave
);
1107 rv
= bond_set_carrier(bond
);
1111 if (netif_carrier_ok(bond
->dev
)) {
1112 pr_info("%s: first active interface up!\n",
1115 pr_info("%s: now running without any active interface !\n",
1121 /*--------------------------- slave list handling ---------------------------*/
1124 * This function attaches the slave to the end of list.
1126 * bond->lock held for writing by caller.
1128 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1130 list_add_tail(&new_slave
->list
, &bond
->slave_list
);
1135 * This function detaches the slave from the list.
1136 * WARNING: no check is made to verify if the slave effectively
1137 * belongs to <bond>.
1138 * Nothing is freed on return, structures are just unchained.
1139 * If any slave pointer in bond was pointing to <slave>,
1140 * it should be changed by the calling function.
1142 * bond->lock held for writing by caller.
1144 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1146 list_del(&slave
->list
);
1150 #ifdef CONFIG_NET_POLL_CONTROLLER
1151 static inline int slave_enable_netpoll(struct slave
*slave
)
1156 np
= kzalloc(sizeof(*np
), GFP_ATOMIC
);
1161 err
= __netpoll_setup(np
, slave
->dev
, GFP_ATOMIC
);
1170 static inline void slave_disable_netpoll(struct slave
*slave
)
1172 struct netpoll
*np
= slave
->np
;
1178 __netpoll_free_async(np
);
1180 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1182 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1184 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1189 static void bond_poll_controller(struct net_device
*bond_dev
)
1193 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1195 struct bonding
*bond
= netdev_priv(bond_dev
);
1196 struct slave
*slave
;
1198 bond_for_each_slave(bond
, slave
)
1199 if (IS_UP(slave
->dev
))
1200 slave_disable_netpoll(slave
);
1203 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
, gfp_t gfp
)
1205 struct bonding
*bond
= netdev_priv(dev
);
1206 struct slave
*slave
;
1209 bond_for_each_slave(bond
, slave
) {
1210 err
= slave_enable_netpoll(slave
);
1212 bond_netpoll_cleanup(dev
);
1219 static inline int slave_enable_netpoll(struct slave
*slave
)
1223 static inline void slave_disable_netpoll(struct slave
*slave
)
1226 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1231 /*---------------------------------- IOCTL ----------------------------------*/
1233 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1234 netdev_features_t features
)
1236 struct slave
*slave
;
1237 struct bonding
*bond
= netdev_priv(dev
);
1238 netdev_features_t mask
;
1240 read_lock(&bond
->lock
);
1242 if (list_empty(&bond
->slave_list
)) {
1243 /* Disable adding VLANs to empty bond. But why? --mq */
1244 features
|= NETIF_F_VLAN_CHALLENGED
;
1249 features
&= ~NETIF_F_ONE_FOR_ALL
;
1250 features
|= NETIF_F_ALL_FOR_ALL
;
1252 bond_for_each_slave(bond
, slave
) {
1253 features
= netdev_increment_features(features
,
1254 slave
->dev
->features
,
1257 features
= netdev_add_tso_features(features
, mask
);
1260 read_unlock(&bond
->lock
);
1264 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1265 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1266 NETIF_F_HIGHDMA | NETIF_F_LRO)
1268 static void bond_compute_features(struct bonding
*bond
)
1270 struct slave
*slave
;
1271 struct net_device
*bond_dev
= bond
->dev
;
1272 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1273 unsigned short max_hard_header_len
= ETH_HLEN
;
1274 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1275 u16 gso_max_segs
= GSO_MAX_SEGS
;
1276 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1278 read_lock(&bond
->lock
);
1280 if (list_empty(&bond
->slave_list
))
1283 bond_for_each_slave(bond
, slave
) {
1284 vlan_features
= netdev_increment_features(vlan_features
,
1285 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1287 dst_release_flag
&= slave
->dev
->priv_flags
;
1288 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1289 max_hard_header_len
= slave
->dev
->hard_header_len
;
1291 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1292 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1296 bond_dev
->vlan_features
= vlan_features
;
1297 bond_dev
->hard_header_len
= max_hard_header_len
;
1298 bond_dev
->gso_max_segs
= gso_max_segs
;
1299 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1301 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1302 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1304 read_unlock(&bond
->lock
);
1306 netdev_change_features(bond_dev
);
1309 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1310 struct net_device
*slave_dev
)
1312 bond_dev
->header_ops
= slave_dev
->header_ops
;
1314 bond_dev
->type
= slave_dev
->type
;
1315 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1316 bond_dev
->addr_len
= slave_dev
->addr_len
;
1318 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1319 slave_dev
->addr_len
);
1322 /* On bonding slaves other than the currently active slave, suppress
1323 * duplicates except for alb non-mcast/bcast.
1325 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1326 struct slave
*slave
,
1327 struct bonding
*bond
)
1329 if (bond_is_slave_inactive(slave
)) {
1330 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1331 skb
->pkt_type
!= PACKET_BROADCAST
&&
1332 skb
->pkt_type
!= PACKET_MULTICAST
)
1339 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1341 struct sk_buff
*skb
= *pskb
;
1342 struct slave
*slave
;
1343 struct bonding
*bond
;
1344 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1346 int ret
= RX_HANDLER_ANOTHER
;
1348 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1350 return RX_HANDLER_CONSUMED
;
1354 slave
= bond_slave_get_rcu(skb
->dev
);
1357 if (bond
->params
.arp_interval
)
1358 slave
->dev
->last_rx
= jiffies
;
1360 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1362 ret
= recv_probe(skb
, bond
, slave
);
1363 if (ret
== RX_HANDLER_CONSUMED
) {
1369 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1370 return RX_HANDLER_EXACT
;
1373 skb
->dev
= bond
->dev
;
1375 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1376 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1377 skb
->pkt_type
== PACKET_HOST
) {
1379 if (unlikely(skb_cow_head(skb
,
1380 skb
->data
- skb_mac_header(skb
)))) {
1382 return RX_HANDLER_CONSUMED
;
1384 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1390 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1391 struct net_device
*slave_dev
)
1395 err
= netdev_master_upper_dev_link(slave_dev
, bond_dev
);
1398 slave_dev
->flags
|= IFF_SLAVE
;
1399 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
);
1403 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1404 struct net_device
*slave_dev
)
1406 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1407 slave_dev
->flags
&= ~IFF_SLAVE
;
1408 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
);
1411 /* enslave device <slave> to bond device <master> */
1412 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1414 struct bonding
*bond
= netdev_priv(bond_dev
);
1415 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1416 struct slave
*new_slave
= NULL
;
1417 struct sockaddr addr
;
1421 if (!bond
->params
.use_carrier
&&
1422 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1423 slave_ops
->ndo_do_ioctl
== NULL
) {
1424 pr_warning("%s: Warning: no link monitoring support for %s\n",
1425 bond_dev
->name
, slave_dev
->name
);
1428 /* already enslaved */
1429 if (slave_dev
->flags
& IFF_SLAVE
) {
1430 pr_debug("Error, Device was already enslaved\n");
1434 /* vlan challenged mutual exclusion */
1435 /* no need to lock since we're protected by rtnl_lock */
1436 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1437 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1438 if (vlan_uses_dev(bond_dev
)) {
1439 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1440 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1443 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1444 bond_dev
->name
, slave_dev
->name
,
1445 slave_dev
->name
, bond_dev
->name
);
1448 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1452 * Old ifenslave binaries are no longer supported. These can
1453 * be identified with moderate accuracy by the state of the slave:
1454 * the current ifenslave will set the interface down prior to
1455 * enslaving it; the old ifenslave will not.
1457 if ((slave_dev
->flags
& IFF_UP
)) {
1458 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1461 goto err_undo_flags
;
1464 /* set bonding device ether type by slave - bonding netdevices are
1465 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1466 * there is a need to override some of the type dependent attribs/funcs.
1468 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1469 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1471 if (list_empty(&bond
->slave_list
)) {
1472 if (bond_dev
->type
!= slave_dev
->type
) {
1473 pr_debug("%s: change device type from %d to %d\n",
1475 bond_dev
->type
, slave_dev
->type
);
1477 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1479 res
= notifier_to_errno(res
);
1481 pr_err("%s: refused to change device type\n",
1484 goto err_undo_flags
;
1487 /* Flush unicast and multicast addresses */
1488 dev_uc_flush(bond_dev
);
1489 dev_mc_flush(bond_dev
);
1491 if (slave_dev
->type
!= ARPHRD_ETHER
)
1492 bond_setup_by_slave(bond_dev
, slave_dev
);
1494 ether_setup(bond_dev
);
1495 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1498 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1501 } else if (bond_dev
->type
!= slave_dev
->type
) {
1502 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1504 slave_dev
->type
, bond_dev
->type
);
1506 goto err_undo_flags
;
1509 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1510 if (list_empty(&bond
->slave_list
)) {
1511 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1513 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1514 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1515 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",
1518 goto err_undo_flags
;
1522 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1524 /* If this is the first slave, then we need to set the master's hardware
1525 * address to be the same as the slave's. */
1526 if (list_empty(&bond
->slave_list
) &&
1527 bond
->dev
->addr_assign_type
== NET_ADDR_RANDOM
)
1528 bond_set_dev_addr(bond
->dev
, slave_dev
);
1530 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1533 goto err_undo_flags
;
1535 INIT_LIST_HEAD(&new_slave
->list
);
1537 * Set the new_slave's queue_id to be zero. Queue ID mapping
1538 * is set via sysfs or module option if desired.
1540 new_slave
->queue_id
= 0;
1542 /* Save slave's original mtu and then set it to match the bond */
1543 new_slave
->original_mtu
= slave_dev
->mtu
;
1544 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1546 pr_debug("Error %d calling dev_set_mtu\n", res
);
1551 * Save slave's original ("permanent") mac address for modes
1552 * that need it, and for restoring it upon release, and then
1553 * set it to the master's address
1555 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1557 if (!bond
->params
.fail_over_mac
) {
1559 * Set slave to master's mac address. The application already
1560 * set the master's mac address to that of the first slave
1562 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1563 addr
.sa_family
= slave_dev
->type
;
1564 res
= dev_set_mac_address(slave_dev
, &addr
);
1566 pr_debug("Error %d calling set_mac_address\n", res
);
1567 goto err_restore_mtu
;
1571 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
);
1573 pr_debug("Error %d calling bond_master_upper_dev_link\n", res
);
1574 goto err_restore_mac
;
1577 /* open the slave since the application closed it */
1578 res
= dev_open(slave_dev
);
1580 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1581 goto err_unset_master
;
1584 new_slave
->bond
= bond
;
1585 new_slave
->dev
= slave_dev
;
1586 slave_dev
->priv_flags
|= IFF_BONDING
;
1588 if (bond_is_lb(bond
)) {
1589 /* bond_alb_init_slave() must be called before all other stages since
1590 * it might fail and we do not want to have to undo everything
1592 res
= bond_alb_init_slave(bond
, new_slave
);
1597 /* If the mode USES_PRIMARY, then the following is handled by
1598 * bond_change_active_slave().
1600 if (!USES_PRIMARY(bond
->params
.mode
)) {
1601 /* set promiscuity level to new slave */
1602 if (bond_dev
->flags
& IFF_PROMISC
) {
1603 res
= dev_set_promiscuity(slave_dev
, 1);
1608 /* set allmulti level to new slave */
1609 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1610 res
= dev_set_allmulti(slave_dev
, 1);
1615 netif_addr_lock_bh(bond_dev
);
1617 dev_mc_sync_multiple(slave_dev
, bond_dev
);
1618 dev_uc_sync_multiple(slave_dev
, bond_dev
);
1620 netif_addr_unlock_bh(bond_dev
);
1623 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1624 /* add lacpdu mc addr to mc list */
1625 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1627 dev_mc_add(slave_dev
, lacpdu_multicast
);
1630 bond_add_vlans_on_slave(bond
, slave_dev
);
1632 write_lock_bh(&bond
->lock
);
1634 bond_attach_slave(bond
, new_slave
);
1636 new_slave
->delay
= 0;
1637 new_slave
->link_failure_count
= 0;
1639 write_unlock_bh(&bond
->lock
);
1641 bond_compute_features(bond
);
1643 bond_update_speed_duplex(new_slave
);
1645 read_lock(&bond
->lock
);
1647 new_slave
->last_arp_rx
= jiffies
-
1648 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1649 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
1650 new_slave
->target_last_arp_rx
[i
] = new_slave
->last_arp_rx
;
1652 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1653 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1655 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1657 * miimon is set but a bonded network driver
1658 * does not support ETHTOOL/MII and
1659 * arp_interval is not set. Note: if
1660 * use_carrier is enabled, we will never go
1661 * here (because netif_carrier is always
1662 * supported); thus, we don't need to change
1663 * the messages for netif_carrier.
1665 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",
1666 bond_dev
->name
, slave_dev
->name
);
1667 } else if (link_reporting
== -1) {
1668 /* unable get link status using mii/ethtool */
1669 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",
1670 bond_dev
->name
, slave_dev
->name
);
1674 /* check for initial state */
1675 if (bond
->params
.miimon
) {
1676 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1677 if (bond
->params
.updelay
) {
1678 new_slave
->link
= BOND_LINK_BACK
;
1679 new_slave
->delay
= bond
->params
.updelay
;
1681 new_slave
->link
= BOND_LINK_UP
;
1684 new_slave
->link
= BOND_LINK_DOWN
;
1686 } else if (bond
->params
.arp_interval
) {
1687 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1688 BOND_LINK_UP
: BOND_LINK_DOWN
);
1690 new_slave
->link
= BOND_LINK_UP
;
1693 if (new_slave
->link
!= BOND_LINK_DOWN
)
1694 new_slave
->jiffies
= jiffies
;
1695 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1696 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1697 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1699 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1700 /* if there is a primary slave, remember it */
1701 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1702 bond
->primary_slave
= new_slave
;
1703 bond
->force_primary
= true;
1707 write_lock_bh(&bond
->curr_slave_lock
);
1709 switch (bond
->params
.mode
) {
1710 case BOND_MODE_ACTIVEBACKUP
:
1711 bond_set_slave_inactive_flags(new_slave
);
1712 bond_select_active_slave(bond
);
1714 case BOND_MODE_8023AD
:
1715 /* in 802.3ad mode, the internal mechanism
1716 * will activate the slaves in the selected
1719 bond_set_slave_inactive_flags(new_slave
);
1720 /* if this is the first slave */
1721 if (bond_first_slave(bond
) == new_slave
) {
1722 SLAVE_AD_INFO(new_slave
).id
= 1;
1723 /* Initialize AD with the number of times that the AD timer is called in 1 second
1724 * can be called only after the mac address of the bond is set
1726 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1728 struct slave
*prev_slave
;
1730 prev_slave
= bond_prev_slave(bond
, new_slave
);
1731 SLAVE_AD_INFO(new_slave
).id
=
1732 SLAVE_AD_INFO(prev_slave
).id
+ 1;
1735 bond_3ad_bind_slave(new_slave
);
1739 bond_set_active_slave(new_slave
);
1740 bond_set_slave_inactive_flags(new_slave
);
1741 bond_select_active_slave(bond
);
1744 pr_debug("This slave is always active in trunk mode\n");
1746 /* always active in trunk mode */
1747 bond_set_active_slave(new_slave
);
1749 /* In trunking mode there is little meaning to curr_active_slave
1750 * anyway (it holds no special properties of the bond device),
1751 * so we can change it without calling change_active_interface()
1753 if (!bond
->curr_active_slave
&& new_slave
->link
== BOND_LINK_UP
)
1754 bond
->curr_active_slave
= new_slave
;
1757 } /* switch(bond_mode) */
1759 write_unlock_bh(&bond
->curr_slave_lock
);
1761 bond_set_carrier(bond
);
1763 #ifdef CONFIG_NET_POLL_CONTROLLER
1764 slave_dev
->npinfo
= bond
->dev
->npinfo
;
1765 if (slave_dev
->npinfo
) {
1766 if (slave_enable_netpoll(new_slave
)) {
1767 read_unlock(&bond
->lock
);
1768 pr_info("Error, %s: master_dev is using netpoll, "
1769 "but new slave device does not support netpoll.\n",
1777 read_unlock(&bond
->lock
);
1779 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1783 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1786 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1787 goto err_dest_symlinks
;
1790 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1791 bond_dev
->name
, slave_dev
->name
,
1792 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1793 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1795 /* enslave is successful */
1798 /* Undo stages on error */
1800 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1803 if (!USES_PRIMARY(bond
->params
.mode
))
1804 bond_hw_addr_flush(bond_dev
, slave_dev
);
1806 bond_del_vlans_from_slave(bond
, slave_dev
);
1807 write_lock_bh(&bond
->lock
);
1808 bond_detach_slave(bond
, new_slave
);
1809 if (bond
->primary_slave
== new_slave
)
1810 bond
->primary_slave
= NULL
;
1811 if (bond
->curr_active_slave
== new_slave
) {
1812 bond_change_active_slave(bond
, NULL
);
1813 write_unlock_bh(&bond
->lock
);
1814 read_lock(&bond
->lock
);
1815 write_lock_bh(&bond
->curr_slave_lock
);
1816 bond_select_active_slave(bond
);
1817 write_unlock_bh(&bond
->curr_slave_lock
);
1818 read_unlock(&bond
->lock
);
1820 write_unlock_bh(&bond
->lock
);
1822 slave_disable_netpoll(new_slave
);
1825 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1826 dev_close(slave_dev
);
1829 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1832 if (!bond
->params
.fail_over_mac
) {
1833 /* XXX TODO - fom follow mode needs to change master's
1834 * MAC if this slave's MAC is in use by the bond, or at
1835 * least print a warning.
1837 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1838 addr
.sa_family
= slave_dev
->type
;
1839 dev_set_mac_address(slave_dev
, &addr
);
1843 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1849 bond_compute_features(bond
);
1850 /* Enslave of first slave has failed and we need to fix master's mac */
1851 if (list_empty(&bond
->slave_list
) &&
1852 ether_addr_equal(bond_dev
->dev_addr
, slave_dev
->dev_addr
))
1853 eth_hw_addr_random(bond_dev
);
1859 * Try to release the slave device <slave> from the bond device <master>
1860 * It is legal to access curr_active_slave without a lock because all the function
1861 * is write-locked. If "all" is true it means that the function is being called
1862 * while destroying a bond interface and all slaves are being released.
1864 * The rules for slave state should be:
1865 * for Active/Backup:
1866 * Active stays on all backups go down
1867 * for Bonded connections:
1868 * The first up interface should be left on and all others downed.
1870 static int __bond_release_one(struct net_device
*bond_dev
,
1871 struct net_device
*slave_dev
,
1874 struct bonding
*bond
= netdev_priv(bond_dev
);
1875 struct slave
*slave
, *oldcurrent
;
1876 struct sockaddr addr
;
1877 netdev_features_t old_features
= bond_dev
->features
;
1879 /* slave is not a slave or master is not master of this slave */
1880 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1881 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1882 pr_err("%s: Error: cannot release %s.\n",
1883 bond_dev
->name
, slave_dev
->name
);
1888 write_lock_bh(&bond
->lock
);
1890 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1892 /* not a slave of this bond */
1893 pr_info("%s: %s not enslaved\n",
1894 bond_dev
->name
, slave_dev
->name
);
1895 write_unlock_bh(&bond
->lock
);
1896 unblock_netpoll_tx();
1900 write_unlock_bh(&bond
->lock
);
1901 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1902 * for this slave anymore.
1904 netdev_rx_handler_unregister(slave_dev
);
1905 write_lock_bh(&bond
->lock
);
1907 /* Inform AD package of unbinding of slave. */
1908 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1909 /* must be called before the slave is
1910 * detached from the list
1912 bond_3ad_unbind_slave(slave
);
1915 pr_info("%s: releasing %s interface %s\n",
1917 bond_is_active_slave(slave
) ? "active" : "backup",
1920 oldcurrent
= bond
->curr_active_slave
;
1922 bond
->current_arp_slave
= NULL
;
1924 /* release the slave from its bond */
1925 bond_detach_slave(bond
, slave
);
1927 if (!all
&& !bond
->params
.fail_over_mac
) {
1928 if (ether_addr_equal(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1929 !list_empty(&bond
->slave_list
))
1930 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",
1931 bond_dev
->name
, slave_dev
->name
,
1933 bond_dev
->name
, slave_dev
->name
);
1936 if (bond
->primary_slave
== slave
)
1937 bond
->primary_slave
= NULL
;
1939 if (oldcurrent
== slave
)
1940 bond_change_active_slave(bond
, NULL
);
1942 if (bond_is_lb(bond
)) {
1943 /* Must be called only after the slave has been
1944 * detached from the list and the curr_active_slave
1945 * has been cleared (if our_slave == old_current),
1946 * but before a new active slave is selected.
1948 write_unlock_bh(&bond
->lock
);
1949 bond_alb_deinit_slave(bond
, slave
);
1950 write_lock_bh(&bond
->lock
);
1954 bond
->curr_active_slave
= NULL
;
1955 } else if (oldcurrent
== slave
) {
1957 * Note that we hold RTNL over this sequence, so there
1958 * is no concern that another slave add/remove event
1961 write_unlock_bh(&bond
->lock
);
1962 read_lock(&bond
->lock
);
1963 write_lock_bh(&bond
->curr_slave_lock
);
1965 bond_select_active_slave(bond
);
1967 write_unlock_bh(&bond
->curr_slave_lock
);
1968 read_unlock(&bond
->lock
);
1969 write_lock_bh(&bond
->lock
);
1972 if (list_empty(&bond
->slave_list
)) {
1973 bond_set_carrier(bond
);
1974 eth_hw_addr_random(bond_dev
);
1976 if (bond_vlan_used(bond
)) {
1977 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1978 bond_dev
->name
, bond_dev
->name
);
1979 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1984 write_unlock_bh(&bond
->lock
);
1985 unblock_netpoll_tx();
1987 if (list_empty(&bond
->slave_list
)) {
1988 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
1989 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
1992 bond_compute_features(bond
);
1993 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1994 (old_features
& NETIF_F_VLAN_CHALLENGED
))
1995 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1996 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1998 /* must do this from outside any spinlocks */
1999 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2001 bond_del_vlans_from_slave(bond
, slave_dev
);
2003 /* If the mode USES_PRIMARY, then this cases was handled above by
2004 * bond_change_active_slave(..., NULL)
2006 if (!USES_PRIMARY(bond
->params
.mode
)) {
2007 /* unset promiscuity level from slave */
2008 if (bond_dev
->flags
& IFF_PROMISC
)
2009 dev_set_promiscuity(slave_dev
, -1);
2011 /* unset allmulti level from slave */
2012 if (bond_dev
->flags
& IFF_ALLMULTI
)
2013 dev_set_allmulti(slave_dev
, -1);
2015 bond_hw_addr_flush(bond_dev
, slave_dev
);
2018 bond_upper_dev_unlink(bond_dev
, slave_dev
);
2020 slave_disable_netpoll(slave
);
2022 /* close slave before restoring its mac address */
2023 dev_close(slave_dev
);
2025 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2026 /* restore original ("permanent") mac address */
2027 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2028 addr
.sa_family
= slave_dev
->type
;
2029 dev_set_mac_address(slave_dev
, &addr
);
2032 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2034 slave_dev
->priv_flags
&= ~IFF_BONDING
;
2038 return 0; /* deletion OK */
2041 /* A wrapper used because of ndo_del_link */
2042 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2044 return __bond_release_one(bond_dev
, slave_dev
, false);
2048 * First release a slave and then destroy the bond if no more slaves are left.
2049 * Must be under rtnl_lock when this function is called.
2051 static int bond_release_and_destroy(struct net_device
*bond_dev
,
2052 struct net_device
*slave_dev
)
2054 struct bonding
*bond
= netdev_priv(bond_dev
);
2057 ret
= bond_release(bond_dev
, slave_dev
);
2058 if (ret
== 0 && list_empty(&bond
->slave_list
)) {
2059 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
2060 pr_info("%s: destroying bond %s.\n",
2061 bond_dev
->name
, bond_dev
->name
);
2062 unregister_netdevice(bond_dev
);
2068 * This function changes the active slave to slave <slave_dev>.
2069 * It returns -EINVAL in the following cases.
2070 * - <slave_dev> is not found in the list.
2071 * - There is not active slave now.
2072 * - <slave_dev> is already active.
2073 * - The link state of <slave_dev> is not BOND_LINK_UP.
2074 * - <slave_dev> is not running.
2075 * In these cases, this function does nothing.
2076 * In the other cases, current_slave pointer is changed and 0 is returned.
2078 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2080 struct bonding
*bond
= netdev_priv(bond_dev
);
2081 struct slave
*old_active
= NULL
;
2082 struct slave
*new_active
= NULL
;
2085 if (!USES_PRIMARY(bond
->params
.mode
))
2088 /* Verify that bond_dev is indeed the master of slave_dev */
2089 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
2090 !netdev_has_upper_dev(slave_dev
, bond_dev
))
2093 read_lock(&bond
->lock
);
2095 old_active
= bond
->curr_active_slave
;
2096 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2098 * Changing to the current active: do nothing; return success.
2100 if (new_active
&& new_active
== old_active
) {
2101 read_unlock(&bond
->lock
);
2107 new_active
->link
== BOND_LINK_UP
&&
2108 IS_UP(new_active
->dev
)) {
2110 write_lock_bh(&bond
->curr_slave_lock
);
2111 bond_change_active_slave(bond
, new_active
);
2112 write_unlock_bh(&bond
->curr_slave_lock
);
2113 unblock_netpoll_tx();
2117 read_unlock(&bond
->lock
);
2122 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2124 struct bonding
*bond
= netdev_priv(bond_dev
);
2126 info
->bond_mode
= bond
->params
.mode
;
2127 info
->miimon
= bond
->params
.miimon
;
2129 read_lock(&bond
->lock
);
2130 info
->num_slaves
= bond
->slave_cnt
;
2131 read_unlock(&bond
->lock
);
2136 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2138 struct bonding
*bond
= netdev_priv(bond_dev
);
2139 int i
= 0, res
= -ENODEV
;
2140 struct slave
*slave
;
2142 read_lock(&bond
->lock
);
2143 bond_for_each_slave(bond
, slave
) {
2144 if (i
++ == (int)info
->slave_id
) {
2146 strcpy(info
->slave_name
, slave
->dev
->name
);
2147 info
->link
= slave
->link
;
2148 info
->state
= bond_slave_state(slave
);
2149 info
->link_failure_count
= slave
->link_failure_count
;
2153 read_unlock(&bond
->lock
);
2158 /*-------------------------------- Monitoring -------------------------------*/
2161 static int bond_miimon_inspect(struct bonding
*bond
)
2163 int link_state
, commit
= 0;
2164 struct slave
*slave
;
2165 bool ignore_updelay
;
2167 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2169 bond_for_each_slave(bond
, slave
) {
2170 slave
->new_link
= BOND_LINK_NOCHANGE
;
2172 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2174 switch (slave
->link
) {
2179 slave
->link
= BOND_LINK_FAIL
;
2180 slave
->delay
= bond
->params
.downdelay
;
2182 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2184 (bond
->params
.mode
==
2185 BOND_MODE_ACTIVEBACKUP
) ?
2186 (bond_is_active_slave(slave
) ?
2187 "active " : "backup ") : "",
2189 bond
->params
.downdelay
* bond
->params
.miimon
);
2192 case BOND_LINK_FAIL
:
2195 * recovered before downdelay expired
2197 slave
->link
= BOND_LINK_UP
;
2198 slave
->jiffies
= jiffies
;
2199 pr_info("%s: link status up again after %d ms for interface %s.\n",
2201 (bond
->params
.downdelay
- slave
->delay
) *
2202 bond
->params
.miimon
,
2207 if (slave
->delay
<= 0) {
2208 slave
->new_link
= BOND_LINK_DOWN
;
2216 case BOND_LINK_DOWN
:
2220 slave
->link
= BOND_LINK_BACK
;
2221 slave
->delay
= bond
->params
.updelay
;
2224 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2225 bond
->dev
->name
, slave
->dev
->name
,
2226 ignore_updelay
? 0 :
2227 bond
->params
.updelay
*
2228 bond
->params
.miimon
);
2231 case BOND_LINK_BACK
:
2233 slave
->link
= BOND_LINK_DOWN
;
2234 pr_info("%s: link status down again after %d ms for interface %s.\n",
2236 (bond
->params
.updelay
- slave
->delay
) *
2237 bond
->params
.miimon
,
2246 if (slave
->delay
<= 0) {
2247 slave
->new_link
= BOND_LINK_UP
;
2249 ignore_updelay
= false;
2261 static void bond_miimon_commit(struct bonding
*bond
)
2263 struct slave
*slave
;
2265 bond_for_each_slave(bond
, slave
) {
2266 switch (slave
->new_link
) {
2267 case BOND_LINK_NOCHANGE
:
2271 slave
->link
= BOND_LINK_UP
;
2272 slave
->jiffies
= jiffies
;
2274 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2275 /* prevent it from being the active one */
2276 bond_set_backup_slave(slave
);
2277 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2278 /* make it immediately active */
2279 bond_set_active_slave(slave
);
2280 } else if (slave
!= bond
->primary_slave
) {
2281 /* prevent it from being the active one */
2282 bond_set_backup_slave(slave
);
2285 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2286 bond
->dev
->name
, slave
->dev
->name
,
2287 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
2288 slave
->duplex
? "full" : "half");
2290 /* notify ad that the link status has changed */
2291 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2292 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2294 if (bond_is_lb(bond
))
2295 bond_alb_handle_link_change(bond
, slave
,
2298 if (!bond
->curr_active_slave
||
2299 (slave
== bond
->primary_slave
))
2304 case BOND_LINK_DOWN
:
2305 if (slave
->link_failure_count
< UINT_MAX
)
2306 slave
->link_failure_count
++;
2308 slave
->link
= BOND_LINK_DOWN
;
2310 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2311 bond
->params
.mode
== BOND_MODE_8023AD
)
2312 bond_set_slave_inactive_flags(slave
);
2314 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2315 bond
->dev
->name
, slave
->dev
->name
);
2317 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2318 bond_3ad_handle_link_change(slave
,
2321 if (bond_is_lb(bond
))
2322 bond_alb_handle_link_change(bond
, slave
,
2325 if (slave
== bond
->curr_active_slave
)
2331 pr_err("%s: invalid new link %d on slave %s\n",
2332 bond
->dev
->name
, slave
->new_link
,
2334 slave
->new_link
= BOND_LINK_NOCHANGE
;
2342 write_lock_bh(&bond
->curr_slave_lock
);
2343 bond_select_active_slave(bond
);
2344 write_unlock_bh(&bond
->curr_slave_lock
);
2345 unblock_netpoll_tx();
2348 bond_set_carrier(bond
);
2354 * Really a wrapper that splits the mii monitor into two phases: an
2355 * inspection, then (if inspection indicates something needs to be done)
2356 * an acquisition of appropriate locks followed by a commit phase to
2357 * implement whatever link state changes are indicated.
2359 void bond_mii_monitor(struct work_struct
*work
)
2361 struct bonding
*bond
= container_of(work
, struct bonding
,
2363 bool should_notify_peers
= false;
2364 unsigned long delay
;
2366 read_lock(&bond
->lock
);
2368 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2370 if (list_empty(&bond
->slave_list
))
2373 should_notify_peers
= bond_should_notify_peers(bond
);
2375 if (bond_miimon_inspect(bond
)) {
2376 read_unlock(&bond
->lock
);
2378 /* Race avoidance with bond_close cancel of workqueue */
2379 if (!rtnl_trylock()) {
2380 read_lock(&bond
->lock
);
2382 should_notify_peers
= false;
2386 read_lock(&bond
->lock
);
2388 bond_miimon_commit(bond
);
2390 read_unlock(&bond
->lock
);
2391 rtnl_unlock(); /* might sleep, hold no other locks */
2392 read_lock(&bond
->lock
);
2396 if (bond
->params
.miimon
)
2397 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2399 read_unlock(&bond
->lock
);
2401 if (should_notify_peers
) {
2402 if (!rtnl_trylock()) {
2403 read_lock(&bond
->lock
);
2404 bond
->send_peer_notif
++;
2405 read_unlock(&bond
->lock
);
2408 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2413 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2415 struct vlan_entry
*vlan
;
2416 struct net_device
*vlan_dev
;
2418 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2421 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2423 vlan_dev
= __vlan_find_dev_deep(bond
->dev
, htons(ETH_P_8021Q
),
2426 if (vlan_dev
&& ip
== bond_confirm_addr(vlan_dev
, 0, ip
))
2434 * We go to the (large) trouble of VLAN tagging ARP frames because
2435 * switches in VLAN mode (especially if ports are configured as
2436 * "native" to a VLAN) might not pass non-tagged frames.
2438 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2440 struct sk_buff
*skb
;
2442 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op
,
2443 slave_dev
->name
, &dest_ip
, &src_ip
, vlan_id
);
2445 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2446 NULL
, slave_dev
->dev_addr
, NULL
);
2449 pr_err("ARP packet allocation failed\n");
2453 skb
= vlan_put_tag(skb
, htons(ETH_P_8021Q
), vlan_id
);
2455 pr_err("failed to insert VLAN tag\n");
2463 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2466 __be32
*targets
= bond
->params
.arp_targets
;
2467 struct vlan_entry
*vlan
;
2468 struct net_device
*vlan_dev
= NULL
;
2471 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2475 pr_debug("basa: target %pI4\n", &targets
[i
]);
2476 if (!bond_vlan_used(bond
)) {
2477 pr_debug("basa: empty vlan: arp_send\n");
2478 addr
= bond_confirm_addr(bond
->dev
, targets
[i
], 0);
2479 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2485 * If VLANs are configured, we do a route lookup to
2486 * determine which VLAN interface would be used, so we
2487 * can tag the ARP with the proper VLAN tag.
2489 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2492 if (net_ratelimit()) {
2493 pr_warning("%s: no route to arp_ip_target %pI4\n",
2494 bond
->dev
->name
, &targets
[i
]);
2500 * This target is not on a VLAN
2502 if (rt
->dst
.dev
== bond
->dev
) {
2504 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2505 addr
= bond_confirm_addr(bond
->dev
, targets
[i
], 0);
2506 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2512 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2514 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
2518 if (vlan_dev
== rt
->dst
.dev
) {
2519 vlan_id
= vlan
->vlan_id
;
2520 pr_debug("basa: vlan match on %s %d\n",
2521 vlan_dev
->name
, vlan_id
);
2526 if (vlan_id
&& vlan_dev
) {
2528 addr
= bond_confirm_addr(vlan_dev
, targets
[i
], 0);
2529 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2534 if (net_ratelimit()) {
2535 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2536 bond
->dev
->name
, &targets
[i
],
2537 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2543 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2547 if (!sip
|| !bond_has_this_ip(bond
, tip
)) {
2548 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip
, &tip
);
2552 i
= bond_get_targets_ip(bond
->params
.arp_targets
, sip
);
2554 pr_debug("bva: sip %pI4 not found in targets\n", &sip
);
2557 slave
->last_arp_rx
= jiffies
;
2558 slave
->target_last_arp_rx
[i
] = jiffies
;
2561 static int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2562 struct slave
*slave
)
2564 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2565 unsigned char *arp_ptr
;
2569 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2570 return RX_HANDLER_ANOTHER
;
2572 read_lock(&bond
->lock
);
2574 if (!slave_do_arp_validate(bond
, slave
))
2577 alen
= arp_hdr_len(bond
->dev
);
2579 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2580 bond
->dev
->name
, skb
->dev
->name
);
2582 if (alen
> skb_headlen(skb
)) {
2583 arp
= kmalloc(alen
, GFP_ATOMIC
);
2586 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2590 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2591 skb
->pkt_type
== PACKET_OTHERHOST
||
2592 skb
->pkt_type
== PACKET_LOOPBACK
||
2593 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2594 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2598 arp_ptr
= (unsigned char *)(arp
+ 1);
2599 arp_ptr
+= bond
->dev
->addr_len
;
2600 memcpy(&sip
, arp_ptr
, 4);
2601 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2602 memcpy(&tip
, arp_ptr
, 4);
2604 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2605 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2606 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2610 * Backup slaves won't see the ARP reply, but do come through
2611 * here for each ARP probe (so we swap the sip/tip to validate
2612 * the probe). In a "redundant switch, common router" type of
2613 * configuration, the ARP probe will (hopefully) travel from
2614 * the active, through one switch, the router, then the other
2615 * switch before reaching the backup.
2617 * We 'trust' the arp requests if there is an active slave and
2618 * it received valid arp reply(s) after it became active. This
2619 * is done to avoid endless looping when we can't reach the
2620 * arp_ip_target and fool ourselves with our own arp requests.
2622 if (bond_is_active_slave(slave
))
2623 bond_validate_arp(bond
, slave
, sip
, tip
);
2624 else if (bond
->curr_active_slave
&&
2625 time_after(slave_last_rx(bond
, bond
->curr_active_slave
),
2626 bond
->curr_active_slave
->jiffies
))
2627 bond_validate_arp(bond
, slave
, tip
, sip
);
2630 read_unlock(&bond
->lock
);
2631 if (arp
!= (struct arphdr
*)skb
->data
)
2633 return RX_HANDLER_ANOTHER
;
2637 * this function is called regularly to monitor each slave's link
2638 * ensuring that traffic is being sent and received when arp monitoring
2639 * is used in load-balancing mode. if the adapter has been dormant, then an
2640 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2641 * arp monitoring in active backup mode.
2643 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2645 struct bonding
*bond
= container_of(work
, struct bonding
,
2647 struct slave
*slave
, *oldcurrent
;
2648 int do_failover
= 0;
2649 int delta_in_ticks
, extra_ticks
;
2651 read_lock(&bond
->lock
);
2653 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2654 extra_ticks
= delta_in_ticks
/ 2;
2656 if (list_empty(&bond
->slave_list
))
2659 oldcurrent
= bond
->curr_active_slave
;
2660 /* see if any of the previous devices are up now (i.e. they have
2661 * xmt and rcv traffic). the curr_active_slave does not come into
2662 * the picture unless it is null. also, slave->jiffies is not needed
2663 * here because we send an arp on each slave and give a slave as
2664 * long as it needs to get the tx/rx within the delta.
2665 * TODO: what about up/down delay in arp mode? it wasn't here before
2668 bond_for_each_slave(bond
, slave
) {
2669 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2671 if (slave
->link
!= BOND_LINK_UP
) {
2672 if (time_in_range(jiffies
,
2673 trans_start
- delta_in_ticks
,
2674 trans_start
+ delta_in_ticks
+ extra_ticks
) &&
2675 time_in_range(jiffies
,
2676 slave
->dev
->last_rx
- delta_in_ticks
,
2677 slave
->dev
->last_rx
+ delta_in_ticks
+ extra_ticks
)) {
2679 slave
->link
= BOND_LINK_UP
;
2680 bond_set_active_slave(slave
);
2682 /* primary_slave has no meaning in round-robin
2683 * mode. the window of a slave being up and
2684 * curr_active_slave being null after enslaving
2688 pr_info("%s: link status definitely up for interface %s, ",
2693 pr_info("%s: interface %s is now up\n",
2699 /* slave->link == BOND_LINK_UP */
2701 /* not all switches will respond to an arp request
2702 * when the source ip is 0, so don't take the link down
2703 * if we don't know our ip yet
2705 if (!time_in_range(jiffies
,
2706 trans_start
- delta_in_ticks
,
2707 trans_start
+ 2 * delta_in_ticks
+ extra_ticks
) ||
2708 !time_in_range(jiffies
,
2709 slave
->dev
->last_rx
- delta_in_ticks
,
2710 slave
->dev
->last_rx
+ 2 * delta_in_ticks
+ extra_ticks
)) {
2712 slave
->link
= BOND_LINK_DOWN
;
2713 bond_set_backup_slave(slave
);
2715 if (slave
->link_failure_count
< UINT_MAX
)
2716 slave
->link_failure_count
++;
2718 pr_info("%s: interface %s is now down.\n",
2722 if (slave
== oldcurrent
)
2727 /* note: if switch is in round-robin mode, all links
2728 * must tx arp to ensure all links rx an arp - otherwise
2729 * links may oscillate or not come up at all; if switch is
2730 * in something like xor mode, there is nothing we can
2731 * do - all replies will be rx'ed on same link causing slaves
2732 * to be unstable during low/no traffic periods
2734 if (IS_UP(slave
->dev
))
2735 bond_arp_send_all(bond
, slave
);
2740 write_lock_bh(&bond
->curr_slave_lock
);
2742 bond_select_active_slave(bond
);
2744 write_unlock_bh(&bond
->curr_slave_lock
);
2745 unblock_netpoll_tx();
2749 if (bond
->params
.arp_interval
)
2750 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2752 read_unlock(&bond
->lock
);
2756 * Called to inspect slaves for active-backup mode ARP monitor link state
2757 * changes. Sets new_link in slaves to specify what action should take
2758 * place for the slave. Returns 0 if no changes are found, >0 if changes
2759 * to link states must be committed.
2761 * Called with bond->lock held for read.
2763 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2765 unsigned long trans_start
;
2766 struct slave
*slave
;
2770 /* All the time comparisons below need some extra time. Otherwise, on
2771 * fast networks the ARP probe/reply may arrive within the same jiffy
2772 * as it was sent. Then, the next time the ARP monitor is run, one
2773 * arp_interval will already have passed in the comparisons.
2775 extra_ticks
= delta_in_ticks
/ 2;
2777 bond_for_each_slave(bond
, slave
) {
2778 slave
->new_link
= BOND_LINK_NOCHANGE
;
2780 if (slave
->link
!= BOND_LINK_UP
) {
2781 if (time_in_range(jiffies
,
2782 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2783 slave_last_rx(bond
, slave
) + delta_in_ticks
+ extra_ticks
)) {
2785 slave
->new_link
= BOND_LINK_UP
;
2793 * Give slaves 2*delta after being enslaved or made
2794 * active. This avoids bouncing, as the last receive
2795 * times need a full ARP monitor cycle to be updated.
2797 if (time_in_range(jiffies
,
2798 slave
->jiffies
- delta_in_ticks
,
2799 slave
->jiffies
+ 2 * delta_in_ticks
+ extra_ticks
))
2803 * Backup slave is down if:
2804 * - No current_arp_slave AND
2805 * - more than 3*delta since last receive AND
2806 * - the bond has an IP address
2808 * Note: a non-null current_arp_slave indicates
2809 * the curr_active_slave went down and we are
2810 * searching for a new one; under this condition
2811 * we only take the curr_active_slave down - this
2812 * gives each slave a chance to tx/rx traffic
2813 * before being taken out
2815 if (!bond_is_active_slave(slave
) &&
2816 !bond
->current_arp_slave
&&
2817 !time_in_range(jiffies
,
2818 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2819 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
+ extra_ticks
)) {
2821 slave
->new_link
= BOND_LINK_DOWN
;
2826 * Active slave is down if:
2827 * - more than 2*delta since transmitting OR
2828 * - (more than 2*delta since receive AND
2829 * the bond has an IP address)
2831 trans_start
= dev_trans_start(slave
->dev
);
2832 if (bond_is_active_slave(slave
) &&
2833 (!time_in_range(jiffies
,
2834 trans_start
- delta_in_ticks
,
2835 trans_start
+ 2 * delta_in_ticks
+ extra_ticks
) ||
2836 !time_in_range(jiffies
,
2837 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2838 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
+ extra_ticks
))) {
2840 slave
->new_link
= BOND_LINK_DOWN
;
2849 * Called to commit link state changes noted by inspection step of
2850 * active-backup mode ARP monitor.
2852 * Called with RTNL and bond->lock for read.
2854 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2856 unsigned long trans_start
;
2857 struct slave
*slave
;
2859 bond_for_each_slave(bond
, slave
) {
2860 switch (slave
->new_link
) {
2861 case BOND_LINK_NOCHANGE
:
2865 trans_start
= dev_trans_start(slave
->dev
);
2866 if ((!bond
->curr_active_slave
&&
2867 time_in_range(jiffies
,
2868 trans_start
- delta_in_ticks
,
2869 trans_start
+ delta_in_ticks
+ delta_in_ticks
/ 2)) ||
2870 bond
->curr_active_slave
!= slave
) {
2871 slave
->link
= BOND_LINK_UP
;
2872 if (bond
->current_arp_slave
) {
2873 bond_set_slave_inactive_flags(
2874 bond
->current_arp_slave
);
2875 bond
->current_arp_slave
= NULL
;
2878 pr_info("%s: link status definitely up for interface %s.\n",
2879 bond
->dev
->name
, slave
->dev
->name
);
2881 if (!bond
->curr_active_slave
||
2882 (slave
== bond
->primary_slave
))
2889 case BOND_LINK_DOWN
:
2890 if (slave
->link_failure_count
< UINT_MAX
)
2891 slave
->link_failure_count
++;
2893 slave
->link
= BOND_LINK_DOWN
;
2894 bond_set_slave_inactive_flags(slave
);
2896 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2897 bond
->dev
->name
, slave
->dev
->name
);
2899 if (slave
== bond
->curr_active_slave
) {
2900 bond
->current_arp_slave
= NULL
;
2907 pr_err("%s: impossible: new_link %d on slave %s\n",
2908 bond
->dev
->name
, slave
->new_link
,
2916 write_lock_bh(&bond
->curr_slave_lock
);
2917 bond_select_active_slave(bond
);
2918 write_unlock_bh(&bond
->curr_slave_lock
);
2919 unblock_netpoll_tx();
2922 bond_set_carrier(bond
);
2926 * Send ARP probes for active-backup mode ARP monitor.
2928 * Called with bond->lock held for read.
2930 static void bond_ab_arp_probe(struct bonding
*bond
)
2932 struct slave
*slave
, *next_slave
;
2935 read_lock(&bond
->curr_slave_lock
);
2937 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
2938 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2939 bond
->current_arp_slave
->dev
->name
,
2940 bond
->curr_active_slave
->dev
->name
);
2942 if (bond
->curr_active_slave
) {
2943 bond_arp_send_all(bond
, bond
->curr_active_slave
);
2944 read_unlock(&bond
->curr_slave_lock
);
2948 read_unlock(&bond
->curr_slave_lock
);
2950 /* if we don't have a curr_active_slave, search for the next available
2951 * backup slave from the current_arp_slave and make it the candidate
2952 * for becoming the curr_active_slave
2955 if (!bond
->current_arp_slave
) {
2956 bond
->current_arp_slave
= bond_first_slave(bond
);
2957 if (!bond
->current_arp_slave
)
2961 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
2963 /* search for next candidate */
2964 next_slave
= bond_next_slave(bond
, bond
->current_arp_slave
);
2965 bond_for_each_slave_from(bond
, slave
, i
, next_slave
) {
2966 if (IS_UP(slave
->dev
)) {
2967 slave
->link
= BOND_LINK_BACK
;
2968 bond_set_slave_active_flags(slave
);
2969 bond_arp_send_all(bond
, slave
);
2970 slave
->jiffies
= jiffies
;
2971 bond
->current_arp_slave
= slave
;
2975 /* if the link state is up at this point, we
2976 * mark it down - this can happen if we have
2977 * simultaneous link failures and
2978 * reselect_active_interface doesn't make this
2979 * one the current slave so it is still marked
2980 * up when it is actually down
2982 if (slave
->link
== BOND_LINK_UP
) {
2983 slave
->link
= BOND_LINK_DOWN
;
2984 if (slave
->link_failure_count
< UINT_MAX
)
2985 slave
->link_failure_count
++;
2987 bond_set_slave_inactive_flags(slave
);
2989 pr_info("%s: backup interface %s is now down.\n",
2990 bond
->dev
->name
, slave
->dev
->name
);
2995 void bond_activebackup_arp_mon(struct work_struct
*work
)
2997 struct bonding
*bond
= container_of(work
, struct bonding
,
2999 bool should_notify_peers
= false;
3002 read_lock(&bond
->lock
);
3004 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3006 if (list_empty(&bond
->slave_list
))
3009 should_notify_peers
= bond_should_notify_peers(bond
);
3011 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3012 read_unlock(&bond
->lock
);
3014 /* Race avoidance with bond_close flush of workqueue */
3015 if (!rtnl_trylock()) {
3016 read_lock(&bond
->lock
);
3018 should_notify_peers
= false;
3022 read_lock(&bond
->lock
);
3024 bond_ab_arp_commit(bond
, delta_in_ticks
);
3026 read_unlock(&bond
->lock
);
3028 read_lock(&bond
->lock
);
3031 bond_ab_arp_probe(bond
);
3034 if (bond
->params
.arp_interval
)
3035 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3037 read_unlock(&bond
->lock
);
3039 if (should_notify_peers
) {
3040 if (!rtnl_trylock()) {
3041 read_lock(&bond
->lock
);
3042 bond
->send_peer_notif
++;
3043 read_unlock(&bond
->lock
);
3046 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
3051 /*-------------------------- netdev event handling --------------------------*/
3054 * Change device name
3056 static int bond_event_changename(struct bonding
*bond
)
3058 bond_remove_proc_entry(bond
);
3059 bond_create_proc_entry(bond
);
3061 bond_debug_reregister(bond
);
3066 static int bond_master_netdev_event(unsigned long event
,
3067 struct net_device
*bond_dev
)
3069 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3072 case NETDEV_CHANGENAME
:
3073 return bond_event_changename(event_bond
);
3074 case NETDEV_UNREGISTER
:
3075 bond_remove_proc_entry(event_bond
);
3077 case NETDEV_REGISTER
:
3078 bond_create_proc_entry(event_bond
);
3087 static int bond_slave_netdev_event(unsigned long event
,
3088 struct net_device
*slave_dev
)
3090 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
);
3091 struct bonding
*bond
;
3092 struct net_device
*bond_dev
;
3096 /* A netdev event can be generated while enslaving a device
3097 * before netdev_rx_handler_register is called in which case
3098 * slave will be NULL
3102 bond_dev
= slave
->bond
->dev
;
3106 case NETDEV_UNREGISTER
:
3107 if (bond_dev
->type
!= ARPHRD_ETHER
)
3108 bond_release_and_destroy(bond_dev
, slave_dev
);
3110 bond_release(bond_dev
, slave_dev
);
3114 old_speed
= slave
->speed
;
3115 old_duplex
= slave
->duplex
;
3117 bond_update_speed_duplex(slave
);
3119 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3120 if (old_speed
!= slave
->speed
)
3121 bond_3ad_adapter_speed_changed(slave
);
3122 if (old_duplex
!= slave
->duplex
)
3123 bond_3ad_adapter_duplex_changed(slave
);
3128 * ... Or is it this?
3131 case NETDEV_CHANGEMTU
:
3133 * TODO: Should slaves be allowed to
3134 * independently alter their MTU? For
3135 * an active-backup bond, slaves need
3136 * not be the same type of device, so
3137 * MTUs may vary. For other modes,
3138 * slaves arguably should have the
3139 * same MTUs. To do this, we'd need to
3140 * take over the slave's change_mtu
3141 * function for the duration of their
3145 case NETDEV_CHANGENAME
:
3147 * TODO: handle changing the primary's name
3150 case NETDEV_FEAT_CHANGE
:
3151 bond_compute_features(bond
);
3153 case NETDEV_RESEND_IGMP
:
3154 /* Propagate to master device */
3155 call_netdevice_notifiers(event
, slave
->bond
->dev
);
3165 * bond_netdev_event: handle netdev notifier chain events.
3167 * This function receives events for the netdev chain. The caller (an
3168 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3169 * locks for us to safely manipulate the slave devices (RTNL lock,
3172 static int bond_netdev_event(struct notifier_block
*this,
3173 unsigned long event
, void *ptr
)
3175 struct net_device
*event_dev
= netdev_notifier_info_to_dev(ptr
);
3177 pr_debug("event_dev: %s, event: %lx\n",
3178 event_dev
? event_dev
->name
: "None",
3181 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3184 if (event_dev
->flags
& IFF_MASTER
) {
3185 pr_debug("IFF_MASTER\n");
3186 return bond_master_netdev_event(event
, event_dev
);
3189 if (event_dev
->flags
& IFF_SLAVE
) {
3190 pr_debug("IFF_SLAVE\n");
3191 return bond_slave_netdev_event(event
, event_dev
);
3197 static struct notifier_block bond_netdev_notifier
= {
3198 .notifier_call
= bond_netdev_event
,
3201 /*---------------------------- Hashing Policies -----------------------------*/
3204 * Hash for the output device based upon layer 2 data
3206 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3208 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3210 if (skb_headlen(skb
) >= offsetof(struct ethhdr
, h_proto
))
3211 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3217 * Hash for the output device based upon layer 2 and layer 3 data. If
3218 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3220 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3222 const struct ethhdr
*data
;
3223 const struct iphdr
*iph
;
3224 const struct ipv6hdr
*ipv6h
;
3226 const __be32
*s
, *d
;
3228 if (skb
->protocol
== htons(ETH_P_IP
) &&
3229 pskb_network_may_pull(skb
, sizeof(*iph
))) {
3231 data
= (struct ethhdr
*)skb
->data
;
3232 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3233 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3234 } else if (skb
->protocol
== htons(ETH_P_IPV6
) &&
3235 pskb_network_may_pull(skb
, sizeof(*ipv6h
))) {
3236 ipv6h
= ipv6_hdr(skb
);
3237 data
= (struct ethhdr
*)skb
->data
;
3238 s
= &ipv6h
->saddr
.s6_addr32
[0];
3239 d
= &ipv6h
->daddr
.s6_addr32
[0];
3240 v6hash
= (s
[1] ^ d
[1]) ^ (s
[2] ^ d
[2]) ^ (s
[3] ^ d
[3]);
3241 v6hash
^= (v6hash
>> 24) ^ (v6hash
>> 16) ^ (v6hash
>> 8);
3242 return (v6hash
^ data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3245 return bond_xmit_hash_policy_l2(skb
, count
);
3249 * Hash for the output device based upon layer 3 and layer 4 data. If
3250 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3251 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3253 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3256 const struct iphdr
*iph
;
3257 const struct ipv6hdr
*ipv6h
;
3258 const __be32
*s
, *d
;
3259 const __be16
*l4
= NULL
;
3261 int noff
= skb_network_offset(skb
);
3264 if (skb
->protocol
== htons(ETH_P_IP
) &&
3265 pskb_may_pull(skb
, noff
+ sizeof(*iph
))) {
3267 poff
= proto_ports_offset(iph
->protocol
);
3269 if (!ip_is_fragment(iph
) && poff
>= 0) {
3270 l4
= skb_header_pointer(skb
, noff
+ (iph
->ihl
<< 2) + poff
,
3273 layer4_xor
= ntohs(l4
[0] ^ l4
[1]);
3275 return (layer4_xor
^
3276 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3277 } else if (skb
->protocol
== htons(ETH_P_IPV6
) &&
3278 pskb_may_pull(skb
, noff
+ sizeof(*ipv6h
))) {
3279 ipv6h
= ipv6_hdr(skb
);
3280 poff
= proto_ports_offset(ipv6h
->nexthdr
);
3282 l4
= skb_header_pointer(skb
, noff
+ sizeof(*ipv6h
) + poff
,
3285 layer4_xor
= ntohs(l4
[0] ^ l4
[1]);
3287 s
= &ipv6h
->saddr
.s6_addr32
[0];
3288 d
= &ipv6h
->daddr
.s6_addr32
[0];
3289 layer4_xor
^= (s
[1] ^ d
[1]) ^ (s
[2] ^ d
[2]) ^ (s
[3] ^ d
[3]);
3290 layer4_xor
^= (layer4_xor
>> 24) ^ (layer4_xor
>> 16) ^
3292 return layer4_xor
% count
;
3295 return bond_xmit_hash_policy_l2(skb
, count
);
3298 /*-------------------------- Device entry points ----------------------------*/
3300 static void bond_work_init_all(struct bonding
*bond
)
3302 INIT_DELAYED_WORK(&bond
->mcast_work
,
3303 bond_resend_igmp_join_requests_delayed
);
3304 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3305 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3306 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3307 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3309 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3310 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3313 static void bond_work_cancel_all(struct bonding
*bond
)
3315 cancel_delayed_work_sync(&bond
->mii_work
);
3316 cancel_delayed_work_sync(&bond
->arp_work
);
3317 cancel_delayed_work_sync(&bond
->alb_work
);
3318 cancel_delayed_work_sync(&bond
->ad_work
);
3319 cancel_delayed_work_sync(&bond
->mcast_work
);
3322 static int bond_open(struct net_device
*bond_dev
)
3324 struct bonding
*bond
= netdev_priv(bond_dev
);
3325 struct slave
*slave
;
3327 /* reset slave->backup and slave->inactive */
3328 read_lock(&bond
->lock
);
3329 if (!list_empty(&bond
->slave_list
)) {
3330 read_lock(&bond
->curr_slave_lock
);
3331 bond_for_each_slave(bond
, slave
) {
3332 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3333 && (slave
!= bond
->curr_active_slave
)) {
3334 bond_set_slave_inactive_flags(slave
);
3336 bond_set_slave_active_flags(slave
);
3339 read_unlock(&bond
->curr_slave_lock
);
3341 read_unlock(&bond
->lock
);
3343 bond_work_init_all(bond
);
3345 if (bond_is_lb(bond
)) {
3346 /* bond_alb_initialize must be called before the timer
3349 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
)))
3351 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3354 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3355 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3357 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3358 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3359 if (bond
->params
.arp_validate
)
3360 bond
->recv_probe
= bond_arp_rcv
;
3363 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3364 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3365 /* register to receive LACPDUs */
3366 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3367 bond_3ad_initiate_agg_selection(bond
, 1);
3373 static int bond_close(struct net_device
*bond_dev
)
3375 struct bonding
*bond
= netdev_priv(bond_dev
);
3377 write_lock_bh(&bond
->lock
);
3378 bond
->send_peer_notif
= 0;
3379 write_unlock_bh(&bond
->lock
);
3381 bond_work_cancel_all(bond
);
3382 if (bond_is_lb(bond
)) {
3383 /* Must be called only after all
3384 * slaves have been released
3386 bond_alb_deinitialize(bond
);
3388 bond
->recv_probe
= NULL
;
3393 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3394 struct rtnl_link_stats64
*stats
)
3396 struct bonding
*bond
= netdev_priv(bond_dev
);
3397 struct rtnl_link_stats64 temp
;
3398 struct slave
*slave
;
3400 memset(stats
, 0, sizeof(*stats
));
3402 read_lock_bh(&bond
->lock
);
3403 bond_for_each_slave(bond
, slave
) {
3404 const struct rtnl_link_stats64
*sstats
=
3405 dev_get_stats(slave
->dev
, &temp
);
3407 stats
->rx_packets
+= sstats
->rx_packets
;
3408 stats
->rx_bytes
+= sstats
->rx_bytes
;
3409 stats
->rx_errors
+= sstats
->rx_errors
;
3410 stats
->rx_dropped
+= sstats
->rx_dropped
;
3412 stats
->tx_packets
+= sstats
->tx_packets
;
3413 stats
->tx_bytes
+= sstats
->tx_bytes
;
3414 stats
->tx_errors
+= sstats
->tx_errors
;
3415 stats
->tx_dropped
+= sstats
->tx_dropped
;
3417 stats
->multicast
+= sstats
->multicast
;
3418 stats
->collisions
+= sstats
->collisions
;
3420 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3421 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3422 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3423 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3424 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3425 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3427 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3428 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3429 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3430 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3431 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3433 read_unlock_bh(&bond
->lock
);
3438 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3440 struct net_device
*slave_dev
= NULL
;
3441 struct ifbond k_binfo
;
3442 struct ifbond __user
*u_binfo
= NULL
;
3443 struct ifslave k_sinfo
;
3444 struct ifslave __user
*u_sinfo
= NULL
;
3445 struct mii_ioctl_data
*mii
= NULL
;
3449 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3461 * We do this again just in case we were called by SIOCGMIIREG
3462 * instead of SIOCGMIIPHY.
3469 if (mii
->reg_num
== 1) {
3470 struct bonding
*bond
= netdev_priv(bond_dev
);
3472 read_lock(&bond
->lock
);
3473 read_lock(&bond
->curr_slave_lock
);
3474 if (netif_carrier_ok(bond
->dev
))
3475 mii
->val_out
= BMSR_LSTATUS
;
3477 read_unlock(&bond
->curr_slave_lock
);
3478 read_unlock(&bond
->lock
);
3482 case BOND_INFO_QUERY_OLD
:
3483 case SIOCBONDINFOQUERY
:
3484 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3486 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3489 res
= bond_info_query(bond_dev
, &k_binfo
);
3491 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3495 case BOND_SLAVE_INFO_QUERY_OLD
:
3496 case SIOCBONDSLAVEINFOQUERY
:
3497 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3499 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3502 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3504 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3513 net
= dev_net(bond_dev
);
3515 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3518 slave_dev
= dev_get_by_name(net
, ifr
->ifr_slave
);
3520 pr_debug("slave_dev=%p:\n", slave_dev
);
3525 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3527 case BOND_ENSLAVE_OLD
:
3528 case SIOCBONDENSLAVE
:
3529 res
= bond_enslave(bond_dev
, slave_dev
);
3531 case BOND_RELEASE_OLD
:
3532 case SIOCBONDRELEASE
:
3533 res
= bond_release(bond_dev
, slave_dev
);
3535 case BOND_SETHWADDR_OLD
:
3536 case SIOCBONDSETHWADDR
:
3537 bond_set_dev_addr(bond_dev
, slave_dev
);
3540 case BOND_CHANGE_ACTIVE_OLD
:
3541 case SIOCBONDCHANGEACTIVE
:
3542 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3554 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3556 struct bonding
*bond
= netdev_priv(bond_dev
);
3558 if (change
& IFF_PROMISC
)
3559 bond_set_promiscuity(bond
,
3560 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3562 if (change
& IFF_ALLMULTI
)
3563 bond_set_allmulti(bond
,
3564 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3567 static void bond_set_rx_mode(struct net_device
*bond_dev
)
3569 struct bonding
*bond
= netdev_priv(bond_dev
);
3570 struct slave
*slave
;
3572 read_lock(&bond
->lock
);
3574 if (USES_PRIMARY(bond
->params
.mode
)) {
3575 read_lock(&bond
->curr_slave_lock
);
3576 slave
= bond
->curr_active_slave
;
3578 dev_uc_sync(slave
->dev
, bond_dev
);
3579 dev_mc_sync(slave
->dev
, bond_dev
);
3581 read_unlock(&bond
->curr_slave_lock
);
3583 bond_for_each_slave(bond
, slave
) {
3584 dev_uc_sync_multiple(slave
->dev
, bond_dev
);
3585 dev_mc_sync_multiple(slave
->dev
, bond_dev
);
3589 read_unlock(&bond
->lock
);
3592 static int bond_neigh_init(struct neighbour
*n
)
3594 struct bonding
*bond
= netdev_priv(n
->dev
);
3595 const struct net_device_ops
*slave_ops
;
3596 struct neigh_parms parms
;
3597 struct slave
*slave
;
3600 slave
= bond_first_slave(bond
);
3603 slave_ops
= slave
->dev
->netdev_ops
;
3604 if (!slave_ops
->ndo_neigh_setup
)
3607 parms
.neigh_setup
= NULL
;
3608 parms
.neigh_cleanup
= NULL
;
3609 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3614 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3615 * after the last slave has been detached. Assumes that all slaves
3616 * utilize the same neigh_cleanup (true at this writing as only user
3619 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3621 if (!parms
.neigh_setup
)
3624 return parms
.neigh_setup(n
);
3628 * The bonding ndo_neigh_setup is called at init time beofre any
3629 * slave exists. So we must declare proxy setup function which will
3630 * be used at run time to resolve the actual slave neigh param setup.
3632 static int bond_neigh_setup(struct net_device
*dev
,
3633 struct neigh_parms
*parms
)
3635 parms
->neigh_setup
= bond_neigh_init
;
3641 * Change the MTU of all of a master's slaves to match the master
3643 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3645 struct bonding
*bond
= netdev_priv(bond_dev
);
3646 struct slave
*slave
;
3649 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3650 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3652 /* Can't hold bond->lock with bh disabled here since
3653 * some base drivers panic. On the other hand we can't
3654 * hold bond->lock without bh disabled because we'll
3655 * deadlock. The only solution is to rely on the fact
3656 * that we're under rtnl_lock here, and the slaves
3657 * list won't change. This doesn't solve the problem
3658 * of setting the slave's MTU while it is
3659 * transmitting, but the assumption is that the base
3660 * driver can handle that.
3662 * TODO: figure out a way to safely iterate the slaves
3663 * list, but without holding a lock around the actual
3664 * call to the base driver.
3667 bond_for_each_slave(bond
, slave
) {
3668 pr_debug("s %p s->p %p c_m %p\n",
3670 bond_prev_slave(bond
, slave
),
3671 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3673 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3676 /* If we failed to set the slave's mtu to the new value
3677 * we must abort the operation even in ACTIVE_BACKUP
3678 * mode, because if we allow the backup slaves to have
3679 * different mtu values than the active slave we'll
3680 * need to change their mtu when doing a failover. That
3681 * means changing their mtu from timer context, which
3682 * is probably not a good idea.
3684 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3689 bond_dev
->mtu
= new_mtu
;
3694 /* unwind from head to the slave that failed */
3695 bond_for_each_slave_continue_reverse(bond
, slave
) {
3698 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3700 pr_debug("unwind err %d dev %s\n",
3701 tmp_res
, slave
->dev
->name
);
3711 * Note that many devices must be down to change the HW address, and
3712 * downing the master releases all slaves. We can make bonds full of
3713 * bonding devices to test this, however.
3715 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3717 struct bonding
*bond
= netdev_priv(bond_dev
);
3718 struct sockaddr
*sa
= addr
, tmp_sa
;
3719 struct slave
*slave
;
3722 if (bond
->params
.mode
== BOND_MODE_ALB
)
3723 return bond_alb_set_mac_address(bond_dev
, addr
);
3726 pr_debug("bond=%p, name=%s\n",
3727 bond
, bond_dev
? bond_dev
->name
: "None");
3729 /* If fail_over_mac is enabled, do nothing and return success.
3730 * Returning an error causes ifenslave to fail.
3732 if (bond
->params
.fail_over_mac
)
3735 if (!is_valid_ether_addr(sa
->sa_data
))
3736 return -EADDRNOTAVAIL
;
3738 /* Can't hold bond->lock with bh disabled here since
3739 * some base drivers panic. On the other hand we can't
3740 * hold bond->lock without bh disabled because we'll
3741 * deadlock. The only solution is to rely on the fact
3742 * that we're under rtnl_lock here, and the slaves
3743 * list won't change. This doesn't solve the problem
3744 * of setting the slave's hw address while it is
3745 * transmitting, but the assumption is that the base
3746 * driver can handle that.
3748 * TODO: figure out a way to safely iterate the slaves
3749 * list, but without holding a lock around the actual
3750 * call to the base driver.
3753 bond_for_each_slave(bond
, slave
) {
3754 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3755 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3757 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3759 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3763 res
= dev_set_mac_address(slave
->dev
, addr
);
3765 /* TODO: consider downing the slave
3767 * User should expect communications
3768 * breakage anyway until ARP finish
3771 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3777 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3781 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3782 tmp_sa
.sa_family
= bond_dev
->type
;
3784 /* unwind from head to the slave that failed */
3785 bond_for_each_slave_continue_reverse(bond
, slave
) {
3788 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3790 pr_debug("unwind err %d dev %s\n",
3791 tmp_res
, slave
->dev
->name
);
3798 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3800 struct bonding
*bond
= netdev_priv(bond_dev
);
3801 struct slave
*slave
, *start_at
;
3802 int i
, slave_no
, res
= 1;
3803 struct iphdr
*iph
= ip_hdr(skb
);
3806 * Start with the curr_active_slave that joined the bond as the
3807 * default for sending IGMP traffic. For failover purposes one
3808 * needs to maintain some consistency for the interface that will
3809 * send the join/membership reports. The curr_active_slave found
3810 * will send all of this type of traffic.
3812 if ((iph
->protocol
== IPPROTO_IGMP
) &&
3813 (skb
->protocol
== htons(ETH_P_IP
))) {
3814 slave
= bond
->curr_active_slave
;
3819 * Concurrent TX may collide on rr_tx_counter; we accept
3820 * that as being rare enough not to justify using an
3823 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
3825 bond_for_each_slave(bond
, slave
) {
3833 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3834 if (IS_UP(slave
->dev
) &&
3835 (slave
->link
== BOND_LINK_UP
) &&
3836 bond_is_active_slave(slave
)) {
3837 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3844 /* no suitable interface, frame not sent */
3848 return NETDEV_TX_OK
;
3853 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3854 * the bond has a usable interface.
3856 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3858 struct bonding
*bond
= netdev_priv(bond_dev
);
3859 struct slave
*slave
;
3862 slave
= bond
->curr_active_slave
;
3864 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3867 /* no suitable interface, frame not sent */
3870 return NETDEV_TX_OK
;
3874 * In bond_xmit_xor() , we determine the output device by using a pre-
3875 * determined xmit_hash_policy(), If the selected device is not enabled,
3876 * find the next active slave.
3878 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3880 struct bonding
*bond
= netdev_priv(bond_dev
);
3881 struct slave
*slave
, *start_at
;
3886 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
3888 bond_for_each_slave(bond
, slave
) {
3896 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3897 if (IS_UP(slave
->dev
) &&
3898 (slave
->link
== BOND_LINK_UP
) &&
3899 bond_is_active_slave(slave
)) {
3900 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3906 /* no suitable interface, frame not sent */
3910 return NETDEV_TX_OK
;
3913 /* in broadcast mode, we send everything to all usable interfaces. */
3914 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3916 struct bonding
*bond
= netdev_priv(bond_dev
);
3917 struct slave
*slave
= NULL
;
3919 bond_for_each_slave(bond
, slave
) {
3920 if (bond_is_last_slave(bond
, slave
))
3922 if (IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
) {
3923 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
3926 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3930 /* bond_dev_queue_xmit always returns 0 */
3931 bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3934 if (slave
&& IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
)
3935 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3939 return NETDEV_TX_OK
;
3942 /*------------------------- Device initialization ---------------------------*/
3944 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
3946 switch (bond
->params
.xmit_policy
) {
3947 case BOND_XMIT_POLICY_LAYER23
:
3948 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
3950 case BOND_XMIT_POLICY_LAYER34
:
3951 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
3953 case BOND_XMIT_POLICY_LAYER2
:
3955 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
3961 * Lookup the slave that corresponds to a qid
3963 static inline int bond_slave_override(struct bonding
*bond
,
3964 struct sk_buff
*skb
)
3966 struct slave
*slave
= NULL
;
3967 struct slave
*check_slave
;
3970 if (!skb
->queue_mapping
)
3973 /* Find out if any slaves have the same mapping as this skb. */
3974 bond_for_each_slave(bond
, check_slave
) {
3975 if (check_slave
->queue_id
== skb
->queue_mapping
) {
3976 slave
= check_slave
;
3981 /* If the slave isn't UP, use default transmit policy. */
3982 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
3983 (slave
->link
== BOND_LINK_UP
)) {
3984 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3991 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
3994 * This helper function exists to help dev_pick_tx get the correct
3995 * destination queue. Using a helper function skips a call to
3996 * skb_tx_hash and will put the skbs in the queue we expect on their
3997 * way down to the bonding driver.
3999 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
4002 * Save the original txq to restore before passing to the driver
4004 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
4006 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
4008 txq
-= dev
->real_num_tx_queues
;
4009 } while (txq
>= dev
->real_num_tx_queues
);
4014 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4016 struct bonding
*bond
= netdev_priv(dev
);
4018 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4019 if (!bond_slave_override(bond
, skb
))
4020 return NETDEV_TX_OK
;
4023 switch (bond
->params
.mode
) {
4024 case BOND_MODE_ROUNDROBIN
:
4025 return bond_xmit_roundrobin(skb
, dev
);
4026 case BOND_MODE_ACTIVEBACKUP
:
4027 return bond_xmit_activebackup(skb
, dev
);
4029 return bond_xmit_xor(skb
, dev
);
4030 case BOND_MODE_BROADCAST
:
4031 return bond_xmit_broadcast(skb
, dev
);
4032 case BOND_MODE_8023AD
:
4033 return bond_3ad_xmit_xor(skb
, dev
);
4036 return bond_alb_xmit(skb
, dev
);
4038 /* Should never happen, mode already checked */
4039 pr_err("%s: Error: Unknown bonding mode %d\n",
4040 dev
->name
, bond
->params
.mode
);
4043 return NETDEV_TX_OK
;
4047 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4049 struct bonding
*bond
= netdev_priv(dev
);
4050 netdev_tx_t ret
= NETDEV_TX_OK
;
4053 * If we risk deadlock from transmitting this in the
4054 * netpoll path, tell netpoll to queue the frame for later tx
4056 if (is_netpoll_tx_blocked(dev
))
4057 return NETDEV_TX_BUSY
;
4059 read_lock(&bond
->lock
);
4061 if (!list_empty(&bond
->slave_list
))
4062 ret
= __bond_start_xmit(skb
, dev
);
4066 read_unlock(&bond
->lock
);
4072 * set bond mode specific net device operations
4074 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4076 struct net_device
*bond_dev
= bond
->dev
;
4079 case BOND_MODE_ROUNDROBIN
:
4081 case BOND_MODE_ACTIVEBACKUP
:
4084 bond_set_xmit_hash_policy(bond
);
4086 case BOND_MODE_BROADCAST
:
4088 case BOND_MODE_8023AD
:
4089 bond_set_xmit_hash_policy(bond
);
4096 /* Should never happen, mode already checked */
4097 pr_err("%s: Error: Unknown bonding mode %d\n",
4098 bond_dev
->name
, mode
);
4103 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
4104 struct ethtool_cmd
*ecmd
)
4106 struct bonding
*bond
= netdev_priv(bond_dev
);
4107 unsigned long speed
= 0;
4108 struct slave
*slave
;
4110 ecmd
->duplex
= DUPLEX_UNKNOWN
;
4111 ecmd
->port
= PORT_OTHER
;
4113 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4114 * do not need to check mode. Though link speed might not represent
4115 * the true receive or transmit bandwidth (not all modes are symmetric)
4116 * this is an accurate maximum.
4118 read_lock(&bond
->lock
);
4119 bond_for_each_slave(bond
, slave
) {
4120 if (SLAVE_IS_OK(slave
)) {
4121 if (slave
->speed
!= SPEED_UNKNOWN
)
4122 speed
+= slave
->speed
;
4123 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
4124 slave
->duplex
!= DUPLEX_UNKNOWN
)
4125 ecmd
->duplex
= slave
->duplex
;
4128 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
4129 read_unlock(&bond
->lock
);
4134 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4135 struct ethtool_drvinfo
*drvinfo
)
4137 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
4138 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
4139 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
4143 static const struct ethtool_ops bond_ethtool_ops
= {
4144 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4145 .get_settings
= bond_ethtool_get_settings
,
4146 .get_link
= ethtool_op_get_link
,
4149 static const struct net_device_ops bond_netdev_ops
= {
4150 .ndo_init
= bond_init
,
4151 .ndo_uninit
= bond_uninit
,
4152 .ndo_open
= bond_open
,
4153 .ndo_stop
= bond_close
,
4154 .ndo_start_xmit
= bond_start_xmit
,
4155 .ndo_select_queue
= bond_select_queue
,
4156 .ndo_get_stats64
= bond_get_stats
,
4157 .ndo_do_ioctl
= bond_do_ioctl
,
4158 .ndo_change_rx_flags
= bond_change_rx_flags
,
4159 .ndo_set_rx_mode
= bond_set_rx_mode
,
4160 .ndo_change_mtu
= bond_change_mtu
,
4161 .ndo_set_mac_address
= bond_set_mac_address
,
4162 .ndo_neigh_setup
= bond_neigh_setup
,
4163 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4164 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4165 #ifdef CONFIG_NET_POLL_CONTROLLER
4166 .ndo_netpoll_setup
= bond_netpoll_setup
,
4167 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4168 .ndo_poll_controller
= bond_poll_controller
,
4170 .ndo_add_slave
= bond_enslave
,
4171 .ndo_del_slave
= bond_release
,
4172 .ndo_fix_features
= bond_fix_features
,
4175 static const struct device_type bond_type
= {
4179 static void bond_destructor(struct net_device
*bond_dev
)
4181 struct bonding
*bond
= netdev_priv(bond_dev
);
4183 destroy_workqueue(bond
->wq
);
4184 free_netdev(bond_dev
);
4187 static void bond_setup(struct net_device
*bond_dev
)
4189 struct bonding
*bond
= netdev_priv(bond_dev
);
4191 /* initialize rwlocks */
4192 rwlock_init(&bond
->lock
);
4193 rwlock_init(&bond
->curr_slave_lock
);
4194 INIT_LIST_HEAD(&bond
->slave_list
);
4195 bond
->params
= bonding_defaults
;
4197 /* Initialize pointers */
4198 bond
->dev
= bond_dev
;
4199 INIT_LIST_HEAD(&bond
->vlan_list
);
4201 /* Initialize the device entry points */
4202 ether_setup(bond_dev
);
4203 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4204 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4205 bond_set_mode_ops(bond
, bond
->params
.mode
);
4207 bond_dev
->destructor
= bond_destructor
;
4209 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
4211 /* Initialize the device options */
4212 bond_dev
->tx_queue_len
= 0;
4213 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4214 bond_dev
->priv_flags
|= IFF_BONDING
;
4215 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
4217 /* At first, we block adding VLANs. That's the only way to
4218 * prevent problems that occur when adding VLANs over an
4219 * empty bond. The block will be removed once non-challenged
4220 * slaves are enslaved.
4222 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4224 /* don't acquire bond device's netif_tx_lock when
4226 bond_dev
->features
|= NETIF_F_LLTX
;
4228 /* By default, we declare the bond to be fully
4229 * VLAN hardware accelerated capable. Special
4230 * care is taken in the various xmit functions
4231 * when there are slaves that are not hw accel
4235 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4236 NETIF_F_HW_VLAN_CTAG_TX
|
4237 NETIF_F_HW_VLAN_CTAG_RX
|
4238 NETIF_F_HW_VLAN_CTAG_FILTER
;
4240 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
4241 bond_dev
->features
|= bond_dev
->hw_features
;
4245 * Destroy a bonding device.
4246 * Must be under rtnl_lock when this function is called.
4248 static void bond_uninit(struct net_device
*bond_dev
)
4250 struct bonding
*bond
= netdev_priv(bond_dev
);
4251 struct slave
*slave
, *tmp_slave
;
4252 struct vlan_entry
*vlan
, *tmp
;
4254 bond_netpoll_cleanup(bond_dev
);
4256 /* Release the bonded slaves */
4257 list_for_each_entry_safe(slave
, tmp_slave
, &bond
->slave_list
, list
)
4258 __bond_release_one(bond_dev
, slave
->dev
, true);
4259 pr_info("%s: released all slaves\n", bond_dev
->name
);
4261 list_del(&bond
->bond_list
);
4263 bond_debug_unregister(bond
);
4265 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4266 list_del(&vlan
->vlan_list
);
4271 /*------------------------- Module initialization ---------------------------*/
4274 * Convert string input module parms. Accept either the
4275 * number of the mode or its string name. A bit complicated because
4276 * some mode names are substrings of other names, and calls from sysfs
4277 * may have whitespace in the name (trailing newlines, for example).
4279 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4281 int modeint
= -1, i
, rv
;
4282 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4284 for (p
= (char *)buf
; *p
; p
++)
4285 if (!(isdigit(*p
) || isspace(*p
)))
4289 rv
= sscanf(buf
, "%20s", modestr
);
4291 rv
= sscanf(buf
, "%d", &modeint
);
4296 for (i
= 0; tbl
[i
].modename
; i
++) {
4297 if (modeint
== tbl
[i
].mode
)
4299 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4306 static int bond_check_params(struct bond_params
*params
)
4308 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
4309 int arp_all_targets_value
;
4312 * Convert string parameters.
4315 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4316 if (bond_mode
== -1) {
4317 pr_err("Error: Invalid bonding mode \"%s\"\n",
4318 mode
== NULL
? "NULL" : mode
);
4323 if (xmit_hash_policy
) {
4324 if ((bond_mode
!= BOND_MODE_XOR
) &&
4325 (bond_mode
!= BOND_MODE_8023AD
)) {
4326 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4327 bond_mode_name(bond_mode
));
4329 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4331 if (xmit_hashtype
== -1) {
4332 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4333 xmit_hash_policy
== NULL
? "NULL" :
4341 if (bond_mode
!= BOND_MODE_8023AD
) {
4342 pr_info("lacp_rate param is irrelevant in mode %s\n",
4343 bond_mode_name(bond_mode
));
4345 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4346 if (lacp_fast
== -1) {
4347 pr_err("Error: Invalid lacp rate \"%s\"\n",
4348 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4355 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4356 if (params
->ad_select
== -1) {
4357 pr_err("Error: Invalid ad_select \"%s\"\n",
4358 ad_select
== NULL
? "NULL" : ad_select
);
4362 if (bond_mode
!= BOND_MODE_8023AD
) {
4363 pr_warning("ad_select param only affects 802.3ad mode\n");
4366 params
->ad_select
= BOND_AD_STABLE
;
4369 if (max_bonds
< 0) {
4370 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4371 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4372 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4376 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4377 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4378 miimon
= BOND_LINK_MON_INTERV
;
4382 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4387 if (downdelay
< 0) {
4388 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4389 downdelay
, INT_MAX
);
4393 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4394 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4399 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4400 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4405 /* reset values for 802.3ad */
4406 if (bond_mode
== BOND_MODE_8023AD
) {
4408 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");
4409 pr_warning("Forcing miimon to 100msec\n");
4414 if (tx_queues
< 1 || tx_queues
> 255) {
4415 pr_warning("Warning: tx_queues (%d) should be between "
4416 "1 and 255, resetting to %d\n",
4417 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4418 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4421 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4422 pr_warning("Warning: all_slaves_active module parameter (%d), "
4423 "not of valid value (0/1), so it was set to "
4424 "0\n", all_slaves_active
);
4425 all_slaves_active
= 0;
4428 if (resend_igmp
< 0 || resend_igmp
> 255) {
4429 pr_warning("Warning: resend_igmp (%d) should be between "
4430 "0 and 255, resetting to %d\n",
4431 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4432 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4435 /* reset values for TLB/ALB */
4436 if ((bond_mode
== BOND_MODE_TLB
) ||
4437 (bond_mode
== BOND_MODE_ALB
)) {
4439 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");
4440 pr_warning("Forcing miimon to 100msec\n");
4445 if (bond_mode
== BOND_MODE_ALB
) {
4446 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",
4451 if (updelay
|| downdelay
) {
4452 /* just warn the user the up/down delay will have
4453 * no effect since miimon is zero...
4455 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",
4456 updelay
, downdelay
);
4459 /* don't allow arp monitoring */
4461 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4462 miimon
, arp_interval
);
4466 if ((updelay
% miimon
) != 0) {
4467 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4469 (updelay
/ miimon
) * miimon
);
4474 if ((downdelay
% miimon
) != 0) {
4475 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4477 (downdelay
/ miimon
) * miimon
);
4480 downdelay
/= miimon
;
4483 if (arp_interval
< 0) {
4484 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4485 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4486 arp_interval
= BOND_LINK_ARP_INTERV
;
4489 for (arp_ip_count
= 0, i
= 0;
4490 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4491 /* not complete check, but should be good enough to
4493 __be32 ip
= in_aton(arp_ip_target
[i
]);
4494 if (!isdigit(arp_ip_target
[i
][0]) || ip
== 0 ||
4495 ip
== htonl(INADDR_BROADCAST
)) {
4496 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4500 if (bond_get_targets_ip(arp_target
, ip
) == -1)
4501 arp_target
[arp_ip_count
++] = ip
;
4503 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4508 if (arp_interval
&& !arp_ip_count
) {
4509 /* don't allow arping if no arp_ip_target given... */
4510 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4516 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4517 pr_err("arp_validate only supported in active-backup mode\n");
4520 if (!arp_interval
) {
4521 pr_err("arp_validate requires arp_interval\n");
4525 arp_validate_value
= bond_parse_parm(arp_validate
,
4527 if (arp_validate_value
== -1) {
4528 pr_err("Error: invalid arp_validate \"%s\"\n",
4529 arp_validate
== NULL
? "NULL" : arp_validate
);
4533 arp_validate_value
= 0;
4535 arp_all_targets_value
= 0;
4536 if (arp_all_targets
) {
4537 arp_all_targets_value
= bond_parse_parm(arp_all_targets
,
4538 arp_all_targets_tbl
);
4540 if (arp_all_targets_value
== -1) {
4541 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4543 arp_all_targets_value
= 0;
4548 pr_info("MII link monitoring set to %d ms\n", miimon
);
4549 } else if (arp_interval
) {
4550 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4552 arp_validate_tbl
[arp_validate_value
].modename
,
4555 for (i
= 0; i
< arp_ip_count
; i
++)
4556 pr_info(" %s", arp_ip_target
[i
]);
4560 } else if (max_bonds
) {
4561 /* miimon and arp_interval not set, we need one so things
4562 * work as expected, see bonding.txt for details
4564 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");
4567 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4568 /* currently, using a primary only makes sense
4569 * in active backup, TLB or ALB modes
4571 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4572 primary
, bond_mode_name(bond_mode
));
4576 if (primary
&& primary_reselect
) {
4577 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4579 if (primary_reselect_value
== -1) {
4580 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4582 NULL
? "NULL" : primary_reselect
);
4586 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4589 if (fail_over_mac
) {
4590 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4592 if (fail_over_mac_value
== -1) {
4593 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4594 arp_validate
== NULL
? "NULL" : arp_validate
);
4598 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4599 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4601 fail_over_mac_value
= BOND_FOM_NONE
;
4604 /* fill params struct with the proper values */
4605 params
->mode
= bond_mode
;
4606 params
->xmit_policy
= xmit_hashtype
;
4607 params
->miimon
= miimon
;
4608 params
->num_peer_notif
= num_peer_notif
;
4609 params
->arp_interval
= arp_interval
;
4610 params
->arp_validate
= arp_validate_value
;
4611 params
->arp_all_targets
= arp_all_targets_value
;
4612 params
->updelay
= updelay
;
4613 params
->downdelay
= downdelay
;
4614 params
->use_carrier
= use_carrier
;
4615 params
->lacp_fast
= lacp_fast
;
4616 params
->primary
[0] = 0;
4617 params
->primary_reselect
= primary_reselect_value
;
4618 params
->fail_over_mac
= fail_over_mac_value
;
4619 params
->tx_queues
= tx_queues
;
4620 params
->all_slaves_active
= all_slaves_active
;
4621 params
->resend_igmp
= resend_igmp
;
4622 params
->min_links
= min_links
;
4625 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4626 params
->primary
[IFNAMSIZ
- 1] = 0;
4629 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4634 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4635 static struct lock_class_key bonding_netdev_addr_lock_key
;
4636 static struct lock_class_key bonding_tx_busylock_key
;
4638 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4639 struct netdev_queue
*txq
,
4642 lockdep_set_class(&txq
->_xmit_lock
,
4643 &bonding_netdev_xmit_lock_key
);
4646 static void bond_set_lockdep_class(struct net_device
*dev
)
4648 lockdep_set_class(&dev
->addr_list_lock
,
4649 &bonding_netdev_addr_lock_key
);
4650 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4651 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4655 * Called from registration process
4657 static int bond_init(struct net_device
*bond_dev
)
4659 struct bonding
*bond
= netdev_priv(bond_dev
);
4660 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4661 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4663 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4666 * Initialize locks that may be required during
4667 * en/deslave operations. All of the bond_open work
4668 * (of which this is part) should really be moved to
4669 * a phase prior to dev_open
4671 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4672 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4674 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4678 bond_set_lockdep_class(bond_dev
);
4680 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4682 bond_prepare_sysfs_group(bond
);
4684 bond_debug_register(bond
);
4686 /* Ensure valid dev_addr */
4687 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4688 bond_dev
->addr_assign_type
== NET_ADDR_PERM
)
4689 eth_hw_addr_random(bond_dev
);
4694 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4696 if (tb
[IFLA_ADDRESS
]) {
4697 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4699 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4700 return -EADDRNOTAVAIL
;
4705 static unsigned int bond_get_num_tx_queues(void)
4710 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
4712 .priv_size
= sizeof(struct bonding
),
4713 .setup
= bond_setup
,
4714 .validate
= bond_validate
,
4715 .get_num_tx_queues
= bond_get_num_tx_queues
,
4716 .get_num_rx_queues
= bond_get_num_tx_queues
, /* Use the same number
4720 /* Create a new bond based on the specified name and bonding parameters.
4721 * If name is NULL, obtain a suitable "bond%d" name for us.
4722 * Caller must NOT hold rtnl_lock; we need to release it here before we
4723 * set up our sysfs entries.
4725 int bond_create(struct net
*net
, const char *name
)
4727 struct net_device
*bond_dev
;
4732 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4733 name
? name
: "bond%d",
4734 bond_setup
, tx_queues
);
4736 pr_err("%s: eek! can't alloc netdev!\n", name
);
4741 dev_net_set(bond_dev
, net
);
4742 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4744 res
= register_netdevice(bond_dev
);
4746 netif_carrier_off(bond_dev
);
4750 bond_destructor(bond_dev
);
4754 static int __net_init
bond_net_init(struct net
*net
)
4756 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4759 INIT_LIST_HEAD(&bn
->dev_list
);
4761 bond_create_proc_dir(bn
);
4762 bond_create_sysfs(bn
);
4767 static void __net_exit
bond_net_exit(struct net
*net
)
4769 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4770 struct bonding
*bond
, *tmp_bond
;
4773 bond_destroy_sysfs(bn
);
4774 bond_destroy_proc_dir(bn
);
4776 /* Kill off any bonds created after unregistering bond rtnl ops */
4778 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4779 unregister_netdevice_queue(bond
->dev
, &list
);
4780 unregister_netdevice_many(&list
);
4784 static struct pernet_operations bond_net_ops
= {
4785 .init
= bond_net_init
,
4786 .exit
= bond_net_exit
,
4788 .size
= sizeof(struct bond_net
),
4791 static int __init
bonding_init(void)
4796 pr_info("%s", bond_version
);
4798 res
= bond_check_params(&bonding_defaults
);
4802 res
= register_pernet_subsys(&bond_net_ops
);
4806 res
= rtnl_link_register(&bond_link_ops
);
4810 bond_create_debugfs();
4812 for (i
= 0; i
< max_bonds
; i
++) {
4813 res
= bond_create(&init_net
, NULL
);
4818 register_netdevice_notifier(&bond_netdev_notifier
);
4822 rtnl_link_unregister(&bond_link_ops
);
4824 unregister_pernet_subsys(&bond_net_ops
);
4829 static void __exit
bonding_exit(void)
4831 unregister_netdevice_notifier(&bond_netdev_notifier
);
4833 bond_destroy_debugfs();
4835 rtnl_link_unregister(&bond_link_ops
);
4836 unregister_pernet_subsys(&bond_net_ops
);
4838 #ifdef CONFIG_NET_POLL_CONTROLLER
4840 * Make sure we don't have an imbalance on our netpoll blocking
4842 WARN_ON(atomic_read(&netpoll_block_tx
));
4846 module_init(bonding_init
);
4847 module_exit(bonding_exit
);
4848 MODULE_LICENSE("GPL");
4849 MODULE_VERSION(DRV_VERSION
);
4850 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4851 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4852 MODULE_ALIAS_RTNL_LINK("bond");