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 *fail_over_mac
;
108 static int all_slaves_active
= 0;
109 static struct bond_params bonding_defaults
;
110 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
112 module_param(max_bonds
, int, 0);
113 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
114 module_param(tx_queues
, int, 0);
115 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
116 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
118 "failover event (alias of num_unsol_na)");
119 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
121 "failover event (alias of num_grat_arp)");
122 module_param(miimon
, int, 0);
123 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
124 module_param(updelay
, int, 0);
125 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
126 module_param(downdelay
, int, 0);
127 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
129 module_param(use_carrier
, int, 0);
130 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132 module_param(mode
, charp
, 0);
133 MODULE_PARM_DESC(mode
, "Mode of operation; 0 for balance-rr, "
134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
137 module_param(primary
, charp
, 0);
138 MODULE_PARM_DESC(primary
, "Primary network device to use");
139 module_param(primary_reselect
, charp
, 0);
140 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
145 "2 for only on active slave "
147 module_param(lacp_rate
, charp
, 0);
148 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner; "
149 "0 for slow, 1 for fast");
150 module_param(ad_select
, charp
, 0);
151 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
152 "0 for stable (default), 1 for bandwidth, "
154 module_param(min_links
, int, 0);
155 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
157 module_param(xmit_hash_policy
, charp
, 0);
158 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
159 "0 for layer 2 (default), 1 for layer 3+4, "
161 module_param(arp_interval
, int, 0);
162 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
163 module_param_array(arp_ip_target
, charp
, NULL
, 0);
164 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
165 module_param(arp_validate
, charp
, 0);
166 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
167 "0 for none (default), 1 for active, "
168 "2 for backup, 3 for all");
169 module_param(fail_over_mac
, charp
, 0);
170 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
171 "the same MAC; 0 for none (default), "
172 "1 for active, 2 for follow");
173 module_param(all_slaves_active
, int, 0);
174 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
175 "by setting active flag for all slaves; "
176 "0 for never (default), 1 for always.");
177 module_param(resend_igmp
, int, 0);
178 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
181 /*----------------------------- Global variables ----------------------------*/
183 #ifdef CONFIG_NET_POLL_CONTROLLER
184 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
187 int bond_net_id __read_mostly
;
189 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
190 static int arp_ip_count
;
191 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
192 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
193 static int lacp_fast
;
195 const struct bond_parm_tbl bond_lacp_tbl
[] = {
196 { "slow", AD_LACP_SLOW
},
197 { "fast", AD_LACP_FAST
},
201 const struct bond_parm_tbl bond_mode_tbl
[] = {
202 { "balance-rr", BOND_MODE_ROUNDROBIN
},
203 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
204 { "balance-xor", BOND_MODE_XOR
},
205 { "broadcast", BOND_MODE_BROADCAST
},
206 { "802.3ad", BOND_MODE_8023AD
},
207 { "balance-tlb", BOND_MODE_TLB
},
208 { "balance-alb", BOND_MODE_ALB
},
212 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
213 { "layer2", BOND_XMIT_POLICY_LAYER2
},
214 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
215 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
219 const struct bond_parm_tbl arp_validate_tbl
[] = {
220 { "none", BOND_ARP_VALIDATE_NONE
},
221 { "active", BOND_ARP_VALIDATE_ACTIVE
},
222 { "backup", BOND_ARP_VALIDATE_BACKUP
},
223 { "all", BOND_ARP_VALIDATE_ALL
},
227 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
228 { "none", BOND_FOM_NONE
},
229 { "active", BOND_FOM_ACTIVE
},
230 { "follow", BOND_FOM_FOLLOW
},
234 const struct bond_parm_tbl pri_reselect_tbl
[] = {
235 { "always", BOND_PRI_RESELECT_ALWAYS
},
236 { "better", BOND_PRI_RESELECT_BETTER
},
237 { "failure", BOND_PRI_RESELECT_FAILURE
},
241 struct bond_parm_tbl ad_select_tbl
[] = {
242 { "stable", BOND_AD_STABLE
},
243 { "bandwidth", BOND_AD_BANDWIDTH
},
244 { "count", BOND_AD_COUNT
},
248 /*-------------------------- Forward declarations ---------------------------*/
250 static int bond_init(struct net_device
*bond_dev
);
251 static void bond_uninit(struct net_device
*bond_dev
);
253 /*---------------------------- General routines -----------------------------*/
255 const char *bond_mode_name(int mode
)
257 static const char *names
[] = {
258 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
259 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
260 [BOND_MODE_XOR
] = "load balancing (xor)",
261 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
262 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
263 [BOND_MODE_TLB
] = "transmit load balancing",
264 [BOND_MODE_ALB
] = "adaptive load balancing",
267 if (mode
< 0 || mode
> BOND_MODE_ALB
)
273 /*---------------------------------- VLAN -----------------------------------*/
276 * bond_add_vlan - add a new vlan id on bond
277 * @bond: bond that got the notification
278 * @vlan_id: the vlan id to add
280 * Returns -ENOMEM if allocation failed.
282 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
284 struct vlan_entry
*vlan
;
286 pr_debug("bond: %s, vlan id %d\n",
287 (bond
? bond
->dev
->name
: "None"), vlan_id
);
289 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
293 INIT_LIST_HEAD(&vlan
->vlan_list
);
294 vlan
->vlan_id
= vlan_id
;
296 write_lock_bh(&bond
->lock
);
298 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
300 write_unlock_bh(&bond
->lock
);
302 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
308 * bond_del_vlan - delete a vlan id from bond
309 * @bond: bond that got the notification
310 * @vlan_id: the vlan id to delete
312 * returns -ENODEV if @vlan_id was not found in @bond.
314 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
316 struct vlan_entry
*vlan
;
319 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
322 write_lock_bh(&bond
->lock
);
324 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
325 if (vlan
->vlan_id
== vlan_id
) {
326 list_del(&vlan
->vlan_list
);
328 if (bond_is_lb(bond
))
329 bond_alb_clear_vlan(bond
, vlan_id
);
331 pr_debug("removed VLAN ID %d from bond %s\n",
332 vlan_id
, bond
->dev
->name
);
341 pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 vlan_id
, bond
->dev
->name
);
345 write_unlock_bh(&bond
->lock
);
346 unblock_netpoll_tx();
351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
358 * Caller must hold bond->lock
360 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
362 struct vlan_entry
*next
, *last
;
364 if (list_empty(&bond
->vlan_list
))
368 next
= list_entry(bond
->vlan_list
.next
,
369 struct vlan_entry
, vlan_list
);
371 last
= list_entry(bond
->vlan_list
.prev
,
372 struct vlan_entry
, vlan_list
);
374 next
= list_entry(bond
->vlan_list
.next
,
375 struct vlan_entry
, vlan_list
);
377 next
= list_entry(curr
->vlan_list
.next
,
378 struct vlan_entry
, vlan_list
);
386 * bond_dev_queue_xmit - Prepare skb for xmit.
388 * @bond: bond device that got this skb for tx.
389 * @skb: hw accel VLAN tagged skb to transmit
390 * @slave_dev: slave that is supposed to xmit this skbuff
392 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
393 struct net_device
*slave_dev
)
395 skb
->dev
= slave_dev
;
397 BUILD_BUG_ON(sizeof(skb
->queue_mapping
) !=
398 sizeof(qdisc_skb_cb(skb
)->slave_dev_queue_mapping
));
399 skb
->queue_mapping
= qdisc_skb_cb(skb
)->slave_dev_queue_mapping
;
401 if (unlikely(netpoll_tx_running(bond
->dev
)))
402 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
410 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411 * We don't protect the slave list iteration with a lock because:
412 * a. This operation is performed in IOCTL context,
413 * b. The operation is protected by the RTNL semaphore in the 8021q code,
414 * c. Holding a lock with BH disabled while directly calling a base driver
415 * entry point is generally a BAD idea.
417 * The design of synchronization/protection for this operation in the 8021q
418 * module is good for one or more VLAN devices over a single physical device
419 * and cannot be extended for a teaming solution like bonding, so there is a
420 * potential race condition here where a net device from the vlan group might
421 * be referenced (either by a base driver or the 8021q code) while it is being
422 * removed from the system. However, it turns out we're not making matters
423 * worse, and if it works for regular VLAN usage it will work here too.
427 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428 * @bond_dev: bonding net device that got called
429 * @vid: vlan id being added
431 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
433 struct bonding
*bond
= netdev_priv(bond_dev
);
434 struct slave
*slave
, *stop_at
;
437 bond_for_each_slave(bond
, slave
, i
) {
438 res
= vlan_vid_add(slave
->dev
, vid
);
443 res
= bond_add_vlan(bond
, vid
);
445 pr_err("%s: Error: Failed to add vlan id %d\n",
446 bond_dev
->name
, vid
);
453 /* unwind from head to the slave that failed */
455 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
)
456 vlan_vid_del(slave
->dev
, vid
);
462 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
463 * @bond_dev: bonding net device that got called
464 * @vid: vlan id being removed
466 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
468 struct bonding
*bond
= netdev_priv(bond_dev
);
472 bond_for_each_slave(bond
, slave
, i
)
473 vlan_vid_del(slave
->dev
, vid
);
475 res
= bond_del_vlan(bond
, vid
);
477 pr_err("%s: Error: Failed to remove vlan id %d\n",
478 bond_dev
->name
, vid
);
485 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
487 struct vlan_entry
*vlan
;
490 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
491 res
= vlan_vid_add(slave_dev
, vlan
->vlan_id
);
493 pr_warning("%s: Failed to add vlan id %d to device %s\n",
494 bond
->dev
->name
, vlan
->vlan_id
,
499 static void bond_del_vlans_from_slave(struct bonding
*bond
,
500 struct net_device
*slave_dev
)
502 struct vlan_entry
*vlan
;
504 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
507 vlan_vid_del(slave_dev
, vlan
->vlan_id
);
511 /*------------------------------- Link status -------------------------------*/
514 * Set the carrier state for the master according to the state of its
515 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
516 * do special 802.3ad magic.
518 * Returns zero if carrier state does not change, nonzero if it does.
520 static int bond_set_carrier(struct bonding
*bond
)
525 if (bond
->slave_cnt
== 0)
528 if (bond
->params
.mode
== BOND_MODE_8023AD
)
529 return bond_3ad_set_carrier(bond
);
531 bond_for_each_slave(bond
, slave
, i
) {
532 if (slave
->link
== BOND_LINK_UP
) {
533 if (!netif_carrier_ok(bond
->dev
)) {
534 netif_carrier_on(bond
->dev
);
542 if (netif_carrier_ok(bond
->dev
)) {
543 netif_carrier_off(bond
->dev
);
550 * Get link speed and duplex from the slave's base driver
551 * using ethtool. If for some reason the call fails or the
552 * values are invalid, set speed and duplex to -1,
555 static void bond_update_speed_duplex(struct slave
*slave
)
557 struct net_device
*slave_dev
= slave
->dev
;
558 struct ethtool_cmd ecmd
;
562 slave
->speed
= SPEED_UNKNOWN
;
563 slave
->duplex
= DUPLEX_UNKNOWN
;
565 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
569 slave_speed
= ethtool_cmd_speed(&ecmd
);
570 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
573 switch (ecmd
.duplex
) {
581 slave
->speed
= slave_speed
;
582 slave
->duplex
= ecmd
.duplex
;
588 * if <dev> supports MII link status reporting, check its link status.
590 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
591 * depending upon the setting of the use_carrier parameter.
593 * Return either BMSR_LSTATUS, meaning that the link is up (or we
594 * can't tell and just pretend it is), or 0, meaning that the link is
597 * If reporting is non-zero, instead of faking link up, return -1 if
598 * both ETHTOOL and MII ioctls fail (meaning the device does not
599 * support them). If use_carrier is set, return whatever it says.
600 * It'd be nice if there was a good way to tell if a driver supports
601 * netif_carrier, but there really isn't.
603 static int bond_check_dev_link(struct bonding
*bond
,
604 struct net_device
*slave_dev
, int reporting
)
606 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
607 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
609 struct mii_ioctl_data
*mii
;
611 if (!reporting
&& !netif_running(slave_dev
))
614 if (bond
->params
.use_carrier
)
615 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
617 /* Try to get link status using Ethtool first. */
618 if (slave_dev
->ethtool_ops
->get_link
)
619 return slave_dev
->ethtool_ops
->get_link(slave_dev
) ?
622 /* Ethtool can't be used, fallback to MII ioctls. */
623 ioctl
= slave_ops
->ndo_do_ioctl
;
625 /* TODO: set pointer to correct ioctl on a per team member */
626 /* bases to make this more efficient. that is, once */
627 /* we determine the correct ioctl, we will always */
628 /* call it and not the others for that team */
632 * We cannot assume that SIOCGMIIPHY will also read a
633 * register; not all network drivers (e.g., e100)
637 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
638 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
640 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
641 mii
->reg_num
= MII_BMSR
;
642 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
643 return mii
->val_out
& BMSR_LSTATUS
;
648 * If reporting, report that either there's no dev->do_ioctl,
649 * or both SIOCGMIIREG and get_link failed (meaning that we
650 * cannot report link status). If not reporting, pretend
653 return reporting
? -1 : BMSR_LSTATUS
;
656 /*----------------------------- Multicast list ------------------------------*/
659 * Push the promiscuity flag down to appropriate slaves
661 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
664 if (USES_PRIMARY(bond
->params
.mode
)) {
665 /* write lock already acquired */
666 if (bond
->curr_active_slave
) {
667 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
673 bond_for_each_slave(bond
, slave
, i
) {
674 err
= dev_set_promiscuity(slave
->dev
, inc
);
683 * Push the allmulti flag down to all slaves
685 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
688 if (USES_PRIMARY(bond
->params
.mode
)) {
689 /* write lock already acquired */
690 if (bond
->curr_active_slave
) {
691 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
697 bond_for_each_slave(bond
, slave
, i
) {
698 err
= dev_set_allmulti(slave
->dev
, inc
);
707 * Add a Multicast address to slaves
710 static void bond_mc_add(struct bonding
*bond
, void *addr
)
712 if (USES_PRIMARY(bond
->params
.mode
)) {
713 /* write lock already acquired */
714 if (bond
->curr_active_slave
)
715 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
720 bond_for_each_slave(bond
, slave
, i
)
721 dev_mc_add(slave
->dev
, addr
);
726 * Remove a multicast address from slave
729 static void bond_mc_del(struct bonding
*bond
, void *addr
)
731 if (USES_PRIMARY(bond
->params
.mode
)) {
732 /* write lock already acquired */
733 if (bond
->curr_active_slave
)
734 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
738 bond_for_each_slave(bond
, slave
, i
) {
739 dev_mc_del(slave
->dev
, addr
);
745 static void __bond_resend_igmp_join_requests(struct net_device
*dev
)
747 struct in_device
*in_dev
;
749 in_dev
= __in_dev_get_rcu(dev
);
751 ip_mc_rejoin_groups(in_dev
);
755 * Retrieve the list of registered multicast addresses for the bonding
756 * device and retransmit an IGMP JOIN request to the current active
759 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
761 struct net_device
*bond_dev
, *vlan_dev
, *upper_dev
;
762 struct vlan_entry
*vlan
;
765 read_lock(&bond
->lock
);
767 bond_dev
= bond
->dev
;
769 /* rejoin all groups on bond device */
770 __bond_resend_igmp_join_requests(bond_dev
);
773 * if bond is enslaved to a bridge,
774 * then rejoin all groups on its master
776 upper_dev
= netdev_master_upper_dev_get_rcu(bond_dev
);
777 if (upper_dev
&& upper_dev
->priv_flags
& IFF_EBRIDGE
)
778 __bond_resend_igmp_join_requests(upper_dev
);
780 /* rejoin all groups on vlan devices */
781 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
782 vlan_dev
= __vlan_find_dev_deep(bond_dev
,
785 __bond_resend_igmp_join_requests(vlan_dev
);
788 if (--bond
->igmp_retrans
> 0)
789 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
791 read_unlock(&bond
->lock
);
795 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
797 struct bonding
*bond
= container_of(work
, struct bonding
,
800 bond_resend_igmp_join_requests(bond
);
805 * flush all members of flush->mc_list from device dev->mc_list
807 static void bond_mc_list_flush(struct net_device
*bond_dev
,
808 struct net_device
*slave_dev
)
810 struct bonding
*bond
= netdev_priv(bond_dev
);
811 struct netdev_hw_addr
*ha
;
813 netdev_for_each_mc_addr(ha
, bond_dev
)
814 dev_mc_del(slave_dev
, ha
->addr
);
816 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
817 /* del lacpdu mc addr from mc list */
818 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
820 dev_mc_del(slave_dev
, lacpdu_multicast
);
824 /*--------------------------- Active slave change ---------------------------*/
827 * Update the mc list and multicast-related flags for the new and
828 * old active slaves (if any) according to the multicast mode, and
829 * promiscuous flags unconditionally.
831 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
832 struct slave
*old_active
)
834 struct netdev_hw_addr
*ha
;
836 if (!USES_PRIMARY(bond
->params
.mode
))
837 /* nothing to do - mc list is already up-to-date on
843 if (bond
->dev
->flags
& IFF_PROMISC
)
844 dev_set_promiscuity(old_active
->dev
, -1);
846 if (bond
->dev
->flags
& IFF_ALLMULTI
)
847 dev_set_allmulti(old_active
->dev
, -1);
849 netdev_for_each_mc_addr(ha
, bond
->dev
)
850 dev_mc_del(old_active
->dev
, ha
->addr
);
854 /* FIXME: Signal errors upstream. */
855 if (bond
->dev
->flags
& IFF_PROMISC
)
856 dev_set_promiscuity(new_active
->dev
, 1);
858 if (bond
->dev
->flags
& IFF_ALLMULTI
)
859 dev_set_allmulti(new_active
->dev
, 1);
861 netdev_for_each_mc_addr(ha
, bond
->dev
)
862 dev_mc_add(new_active
->dev
, ha
->addr
);
867 * bond_do_fail_over_mac
869 * Perform special MAC address swapping for fail_over_mac settings
871 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
873 static void bond_do_fail_over_mac(struct bonding
*bond
,
874 struct slave
*new_active
,
875 struct slave
*old_active
)
876 __releases(&bond
->curr_slave_lock
)
877 __releases(&bond
->lock
)
878 __acquires(&bond
->lock
)
879 __acquires(&bond
->curr_slave_lock
)
881 u8 tmp_mac
[ETH_ALEN
];
882 struct sockaddr saddr
;
885 switch (bond
->params
.fail_over_mac
) {
886 case BOND_FOM_ACTIVE
:
888 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
889 new_active
->dev
->addr_len
);
890 write_unlock_bh(&bond
->curr_slave_lock
);
891 read_unlock(&bond
->lock
);
892 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
893 read_lock(&bond
->lock
);
894 write_lock_bh(&bond
->curr_slave_lock
);
897 case BOND_FOM_FOLLOW
:
899 * if new_active && old_active, swap them
900 * if just old_active, do nothing (going to no active slave)
901 * if just new_active, set new_active to bond's MAC
906 write_unlock_bh(&bond
->curr_slave_lock
);
907 read_unlock(&bond
->lock
);
910 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
911 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
913 saddr
.sa_family
= new_active
->dev
->type
;
915 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
916 saddr
.sa_family
= bond
->dev
->type
;
919 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
921 pr_err("%s: Error %d setting MAC of slave %s\n",
922 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
929 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
930 saddr
.sa_family
= old_active
->dev
->type
;
932 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
934 pr_err("%s: Error %d setting MAC of slave %s\n",
935 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
937 read_lock(&bond
->lock
);
938 write_lock_bh(&bond
->curr_slave_lock
);
941 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
942 bond
->dev
->name
, bond
->params
.fail_over_mac
);
948 static bool bond_should_change_active(struct bonding
*bond
)
950 struct slave
*prim
= bond
->primary_slave
;
951 struct slave
*curr
= bond
->curr_active_slave
;
953 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
955 if (bond
->force_primary
) {
956 bond
->force_primary
= false;
959 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
960 (prim
->speed
< curr
->speed
||
961 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
963 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
969 * find_best_interface - select the best available slave to be the active one
970 * @bond: our bonding struct
972 * Warning: Caller must hold curr_slave_lock for writing.
974 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
976 struct slave
*new_active
, *old_active
;
977 struct slave
*bestslave
= NULL
;
978 int mintime
= bond
->params
.updelay
;
981 new_active
= bond
->curr_active_slave
;
983 if (!new_active
) { /* there were no active slaves left */
984 if (bond
->slave_cnt
> 0) /* found one slave */
985 new_active
= bond
->first_slave
;
987 return NULL
; /* still no slave, return NULL */
990 if ((bond
->primary_slave
) &&
991 bond
->primary_slave
->link
== BOND_LINK_UP
&&
992 bond_should_change_active(bond
)) {
993 new_active
= bond
->primary_slave
;
996 /* remember where to stop iterating over the slaves */
997 old_active
= new_active
;
999 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1000 if (new_active
->link
== BOND_LINK_UP
) {
1002 } else if (new_active
->link
== BOND_LINK_BACK
&&
1003 IS_UP(new_active
->dev
)) {
1004 /* link up, but waiting for stabilization */
1005 if (new_active
->delay
< mintime
) {
1006 mintime
= new_active
->delay
;
1007 bestslave
= new_active
;
1015 static bool bond_should_notify_peers(struct bonding
*bond
)
1017 struct slave
*slave
= bond
->curr_active_slave
;
1019 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1020 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
1022 if (!slave
|| !bond
->send_peer_notif
||
1023 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
1026 bond
->send_peer_notif
--;
1031 * change_active_interface - change the active slave into the specified one
1032 * @bond: our bonding struct
1033 * @new: the new slave to make the active one
1035 * Set the new slave to the bond's settings and unset them on the old
1036 * curr_active_slave.
1037 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1039 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1040 * because it is apparently the best available slave we have, even though its
1041 * updelay hasn't timed out yet.
1043 * If new_active is not NULL, caller must hold bond->lock for read and
1044 * curr_slave_lock for write_bh.
1046 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1048 struct slave
*old_active
= bond
->curr_active_slave
;
1050 if (old_active
== new_active
)
1054 new_active
->jiffies
= jiffies
;
1056 if (new_active
->link
== BOND_LINK_BACK
) {
1057 if (USES_PRIMARY(bond
->params
.mode
)) {
1058 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1059 bond
->dev
->name
, new_active
->dev
->name
,
1060 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1063 new_active
->delay
= 0;
1064 new_active
->link
= BOND_LINK_UP
;
1066 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1067 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1069 if (bond_is_lb(bond
))
1070 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1072 if (USES_PRIMARY(bond
->params
.mode
)) {
1073 pr_info("%s: making interface %s the new active one.\n",
1074 bond
->dev
->name
, new_active
->dev
->name
);
1079 if (USES_PRIMARY(bond
->params
.mode
))
1080 bond_mc_swap(bond
, new_active
, old_active
);
1082 if (bond_is_lb(bond
)) {
1083 bond_alb_handle_active_change(bond
, new_active
);
1085 bond_set_slave_inactive_flags(old_active
);
1087 bond_set_slave_active_flags(new_active
);
1089 bond
->curr_active_slave
= new_active
;
1092 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1094 bond_set_slave_inactive_flags(old_active
);
1097 bool should_notify_peers
= false;
1099 bond_set_slave_active_flags(new_active
);
1101 if (bond
->params
.fail_over_mac
)
1102 bond_do_fail_over_mac(bond
, new_active
,
1105 if (netif_running(bond
->dev
)) {
1106 bond
->send_peer_notif
=
1107 bond
->params
.num_peer_notif
;
1108 should_notify_peers
=
1109 bond_should_notify_peers(bond
);
1112 write_unlock_bh(&bond
->curr_slave_lock
);
1113 read_unlock(&bond
->lock
);
1115 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
1116 if (should_notify_peers
)
1117 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
1120 read_lock(&bond
->lock
);
1121 write_lock_bh(&bond
->curr_slave_lock
);
1125 /* resend IGMP joins since active slave has changed or
1126 * all were sent on curr_active_slave.
1127 * resend only if bond is brought up with the affected
1128 * bonding modes and the retransmission is enabled */
1129 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
1130 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
1131 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
1132 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
1133 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 0);
1138 * bond_select_active_slave - select a new active slave, if needed
1139 * @bond: our bonding struct
1141 * This functions should be called when one of the following occurs:
1142 * - The old curr_active_slave has been released or lost its link.
1143 * - The primary_slave has got its link back.
1144 * - A slave has got its link back and there's no old curr_active_slave.
1146 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1148 void bond_select_active_slave(struct bonding
*bond
)
1150 struct slave
*best_slave
;
1153 best_slave
= bond_find_best_slave(bond
);
1154 if (best_slave
!= bond
->curr_active_slave
) {
1155 bond_change_active_slave(bond
, best_slave
);
1156 rv
= bond_set_carrier(bond
);
1160 if (netif_carrier_ok(bond
->dev
)) {
1161 pr_info("%s: first active interface up!\n",
1164 pr_info("%s: now running without any active interface !\n",
1170 /*--------------------------- slave list handling ---------------------------*/
1173 * This function attaches the slave to the end of list.
1175 * bond->lock held for writing by caller.
1177 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1179 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1180 new_slave
->next
= new_slave
;
1181 new_slave
->prev
= new_slave
;
1182 bond
->first_slave
= new_slave
;
1184 new_slave
->next
= bond
->first_slave
;
1185 new_slave
->prev
= bond
->first_slave
->prev
;
1186 new_slave
->next
->prev
= new_slave
;
1187 new_slave
->prev
->next
= new_slave
;
1194 * This function detaches the slave from the list.
1195 * WARNING: no check is made to verify if the slave effectively
1196 * belongs to <bond>.
1197 * Nothing is freed on return, structures are just unchained.
1198 * If any slave pointer in bond was pointing to <slave>,
1199 * it should be changed by the calling function.
1201 * bond->lock held for writing by caller.
1203 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1206 slave
->next
->prev
= slave
->prev
;
1209 slave
->prev
->next
= slave
->next
;
1211 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1212 if (bond
->slave_cnt
> 1) { /* there are more slave */
1213 bond
->first_slave
= slave
->next
;
1215 bond
->first_slave
= NULL
; /* slave was the last one */
1224 #ifdef CONFIG_NET_POLL_CONTROLLER
1225 static inline int slave_enable_netpoll(struct slave
*slave
)
1230 np
= kzalloc(sizeof(*np
), GFP_ATOMIC
);
1235 err
= __netpoll_setup(np
, slave
->dev
, GFP_ATOMIC
);
1244 static inline void slave_disable_netpoll(struct slave
*slave
)
1246 struct netpoll
*np
= slave
->np
;
1252 __netpoll_free_async(np
);
1254 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1256 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1258 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1263 static void bond_poll_controller(struct net_device
*bond_dev
)
1267 static void __bond_netpoll_cleanup(struct bonding
*bond
)
1269 struct slave
*slave
;
1272 bond_for_each_slave(bond
, slave
, i
)
1273 if (IS_UP(slave
->dev
))
1274 slave_disable_netpoll(slave
);
1276 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1278 struct bonding
*bond
= netdev_priv(bond_dev
);
1280 read_lock(&bond
->lock
);
1281 __bond_netpoll_cleanup(bond
);
1282 read_unlock(&bond
->lock
);
1285 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
, gfp_t gfp
)
1287 struct bonding
*bond
= netdev_priv(dev
);
1288 struct slave
*slave
;
1291 read_lock(&bond
->lock
);
1292 bond_for_each_slave(bond
, slave
, i
) {
1293 err
= slave_enable_netpoll(slave
);
1295 __bond_netpoll_cleanup(bond
);
1299 read_unlock(&bond
->lock
);
1303 static struct netpoll_info
*bond_netpoll_info(struct bonding
*bond
)
1305 return bond
->dev
->npinfo
;
1309 static inline int slave_enable_netpoll(struct slave
*slave
)
1313 static inline void slave_disable_netpoll(struct slave
*slave
)
1316 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1321 /*---------------------------------- IOCTL ----------------------------------*/
1323 static void bond_set_dev_addr(struct net_device
*bond_dev
,
1324 struct net_device
*slave_dev
)
1326 pr_debug("bond_dev=%p\n", bond_dev
);
1327 pr_debug("slave_dev=%p\n", slave_dev
);
1328 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1329 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1330 bond_dev
->addr_assign_type
= NET_ADDR_SET
;
1331 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
1334 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1335 netdev_features_t features
)
1337 struct slave
*slave
;
1338 struct bonding
*bond
= netdev_priv(dev
);
1339 netdev_features_t mask
;
1342 read_lock(&bond
->lock
);
1344 if (!bond
->first_slave
) {
1345 /* Disable adding VLANs to empty bond. But why? --mq */
1346 features
|= NETIF_F_VLAN_CHALLENGED
;
1351 features
&= ~NETIF_F_ONE_FOR_ALL
;
1352 features
|= NETIF_F_ALL_FOR_ALL
;
1354 bond_for_each_slave(bond
, slave
, i
) {
1355 features
= netdev_increment_features(features
,
1356 slave
->dev
->features
,
1361 read_unlock(&bond
->lock
);
1365 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1366 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1367 NETIF_F_HIGHDMA | NETIF_F_LRO)
1369 static void bond_compute_features(struct bonding
*bond
)
1371 struct slave
*slave
;
1372 struct net_device
*bond_dev
= bond
->dev
;
1373 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1374 unsigned short max_hard_header_len
= ETH_HLEN
;
1375 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1376 u16 gso_max_segs
= GSO_MAX_SEGS
;
1378 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1380 read_lock(&bond
->lock
);
1382 if (!bond
->first_slave
)
1385 bond_for_each_slave(bond
, slave
, i
) {
1386 vlan_features
= netdev_increment_features(vlan_features
,
1387 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1389 dst_release_flag
&= slave
->dev
->priv_flags
;
1390 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1391 max_hard_header_len
= slave
->dev
->hard_header_len
;
1393 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1394 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1398 bond_dev
->vlan_features
= vlan_features
;
1399 bond_dev
->hard_header_len
= max_hard_header_len
;
1400 bond_dev
->gso_max_segs
= gso_max_segs
;
1401 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1403 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1404 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1406 read_unlock(&bond
->lock
);
1408 netdev_change_features(bond_dev
);
1411 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1412 struct net_device
*slave_dev
)
1414 struct bonding
*bond
= netdev_priv(bond_dev
);
1416 bond_dev
->header_ops
= slave_dev
->header_ops
;
1418 bond_dev
->type
= slave_dev
->type
;
1419 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1420 bond_dev
->addr_len
= slave_dev
->addr_len
;
1422 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1423 slave_dev
->addr_len
);
1424 bond
->setup_by_slave
= 1;
1427 /* On bonding slaves other than the currently active slave, suppress
1428 * duplicates except for alb non-mcast/bcast.
1430 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1431 struct slave
*slave
,
1432 struct bonding
*bond
)
1434 if (bond_is_slave_inactive(slave
)) {
1435 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1436 skb
->pkt_type
!= PACKET_BROADCAST
&&
1437 skb
->pkt_type
!= PACKET_MULTICAST
)
1444 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1446 struct sk_buff
*skb
= *pskb
;
1447 struct slave
*slave
;
1448 struct bonding
*bond
;
1449 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1451 int ret
= RX_HANDLER_ANOTHER
;
1453 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1455 return RX_HANDLER_CONSUMED
;
1459 slave
= bond_slave_get_rcu(skb
->dev
);
1462 if (bond
->params
.arp_interval
)
1463 slave
->dev
->last_rx
= jiffies
;
1465 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1467 ret
= recv_probe(skb
, bond
, slave
);
1468 if (ret
== RX_HANDLER_CONSUMED
) {
1474 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1475 return RX_HANDLER_EXACT
;
1478 skb
->dev
= bond
->dev
;
1480 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1481 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1482 skb
->pkt_type
== PACKET_HOST
) {
1484 if (unlikely(skb_cow_head(skb
,
1485 skb
->data
- skb_mac_header(skb
)))) {
1487 return RX_HANDLER_CONSUMED
;
1489 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1495 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1496 struct net_device
*slave_dev
)
1500 err
= netdev_master_upper_dev_link(slave_dev
, bond_dev
);
1503 slave_dev
->flags
|= IFF_SLAVE
;
1504 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
);
1508 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1509 struct net_device
*slave_dev
)
1511 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1512 slave_dev
->flags
&= ~IFF_SLAVE
;
1513 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
);
1516 /* enslave device <slave> to bond device <master> */
1517 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1519 struct bonding
*bond
= netdev_priv(bond_dev
);
1520 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1521 struct slave
*new_slave
= NULL
;
1522 struct netdev_hw_addr
*ha
;
1523 struct sockaddr addr
;
1527 if (!bond
->params
.use_carrier
&&
1528 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1529 slave_ops
->ndo_do_ioctl
== NULL
) {
1530 pr_warning("%s: Warning: no link monitoring support for %s\n",
1531 bond_dev
->name
, slave_dev
->name
);
1534 /* already enslaved */
1535 if (slave_dev
->flags
& IFF_SLAVE
) {
1536 pr_debug("Error, Device was already enslaved\n");
1540 /* vlan challenged mutual exclusion */
1541 /* no need to lock since we're protected by rtnl_lock */
1542 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1543 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1544 if (vlan_uses_dev(bond_dev
)) {
1545 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1546 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1549 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1550 bond_dev
->name
, slave_dev
->name
,
1551 slave_dev
->name
, bond_dev
->name
);
1554 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1558 * Old ifenslave binaries are no longer supported. These can
1559 * be identified with moderate accuracy by the state of the slave:
1560 * the current ifenslave will set the interface down prior to
1561 * enslaving it; the old ifenslave will not.
1563 if ((slave_dev
->flags
& IFF_UP
)) {
1564 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1567 goto err_undo_flags
;
1570 /* set bonding device ether type by slave - bonding netdevices are
1571 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1572 * there is a need to override some of the type dependent attribs/funcs.
1574 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1575 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1577 if (bond
->slave_cnt
== 0) {
1578 if (bond_dev
->type
!= slave_dev
->type
) {
1579 pr_debug("%s: change device type from %d to %d\n",
1581 bond_dev
->type
, slave_dev
->type
);
1583 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1585 res
= notifier_to_errno(res
);
1587 pr_err("%s: refused to change device type\n",
1590 goto err_undo_flags
;
1593 /* Flush unicast and multicast addresses */
1594 dev_uc_flush(bond_dev
);
1595 dev_mc_flush(bond_dev
);
1597 if (slave_dev
->type
!= ARPHRD_ETHER
)
1598 bond_setup_by_slave(bond_dev
, slave_dev
);
1600 ether_setup(bond_dev
);
1601 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1604 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1607 } else if (bond_dev
->type
!= slave_dev
->type
) {
1608 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1610 slave_dev
->type
, bond_dev
->type
);
1612 goto err_undo_flags
;
1615 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1616 if (bond
->slave_cnt
== 0) {
1617 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1619 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1620 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1621 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",
1624 goto err_undo_flags
;
1628 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1630 /* If this is the first slave, then we need to set the master's hardware
1631 * address to be the same as the slave's. */
1632 if (bond
->slave_cnt
== 0 && bond
->dev_addr_from_first
)
1633 bond_set_dev_addr(bond
->dev
, slave_dev
);
1635 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1638 goto err_undo_flags
;
1642 * Set the new_slave's queue_id to be zero. Queue ID mapping
1643 * is set via sysfs or module option if desired.
1645 new_slave
->queue_id
= 0;
1647 /* Save slave's original mtu and then set it to match the bond */
1648 new_slave
->original_mtu
= slave_dev
->mtu
;
1649 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1651 pr_debug("Error %d calling dev_set_mtu\n", res
);
1656 * Save slave's original ("permanent") mac address for modes
1657 * that need it, and for restoring it upon release, and then
1658 * set it to the master's address
1660 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1662 if (!bond
->params
.fail_over_mac
) {
1664 * Set slave to master's mac address. The application already
1665 * set the master's mac address to that of the first slave
1667 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1668 addr
.sa_family
= slave_dev
->type
;
1669 res
= dev_set_mac_address(slave_dev
, &addr
);
1671 pr_debug("Error %d calling set_mac_address\n", res
);
1672 goto err_restore_mtu
;
1676 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
);
1678 pr_debug("Error %d calling bond_master_upper_dev_link\n", res
);
1679 goto err_restore_mac
;
1682 /* open the slave since the application closed it */
1683 res
= dev_open(slave_dev
);
1685 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1686 goto err_unset_master
;
1689 new_slave
->bond
= bond
;
1690 new_slave
->dev
= slave_dev
;
1691 slave_dev
->priv_flags
|= IFF_BONDING
;
1693 if (bond_is_lb(bond
)) {
1694 /* bond_alb_init_slave() must be called before all other stages since
1695 * it might fail and we do not want to have to undo everything
1697 res
= bond_alb_init_slave(bond
, new_slave
);
1702 /* If the mode USES_PRIMARY, then the new slave gets the
1703 * master's promisc (and mc) settings only if it becomes the
1704 * curr_active_slave, and that is taken care of later when calling
1705 * bond_change_active()
1707 if (!USES_PRIMARY(bond
->params
.mode
)) {
1708 /* set promiscuity level to new slave */
1709 if (bond_dev
->flags
& IFF_PROMISC
) {
1710 res
= dev_set_promiscuity(slave_dev
, 1);
1715 /* set allmulti level to new slave */
1716 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1717 res
= dev_set_allmulti(slave_dev
, 1);
1722 netif_addr_lock_bh(bond_dev
);
1723 /* upload master's mc_list to new slave */
1724 netdev_for_each_mc_addr(ha
, bond_dev
)
1725 dev_mc_add(slave_dev
, ha
->addr
);
1726 netif_addr_unlock_bh(bond_dev
);
1729 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1730 /* add lacpdu mc addr to mc list */
1731 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1733 dev_mc_add(slave_dev
, lacpdu_multicast
);
1736 bond_add_vlans_on_slave(bond
, slave_dev
);
1738 write_lock_bh(&bond
->lock
);
1740 bond_attach_slave(bond
, new_slave
);
1742 new_slave
->delay
= 0;
1743 new_slave
->link_failure_count
= 0;
1745 write_unlock_bh(&bond
->lock
);
1747 bond_compute_features(bond
);
1749 bond_update_speed_duplex(new_slave
);
1751 read_lock(&bond
->lock
);
1753 new_slave
->last_arp_rx
= jiffies
-
1754 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1756 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1757 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1759 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1761 * miimon is set but a bonded network driver
1762 * does not support ETHTOOL/MII and
1763 * arp_interval is not set. Note: if
1764 * use_carrier is enabled, we will never go
1765 * here (because netif_carrier is always
1766 * supported); thus, we don't need to change
1767 * the messages for netif_carrier.
1769 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",
1770 bond_dev
->name
, slave_dev
->name
);
1771 } else if (link_reporting
== -1) {
1772 /* unable get link status using mii/ethtool */
1773 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",
1774 bond_dev
->name
, slave_dev
->name
);
1778 /* check for initial state */
1779 if (bond
->params
.miimon
) {
1780 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1781 if (bond
->params
.updelay
) {
1782 new_slave
->link
= BOND_LINK_BACK
;
1783 new_slave
->delay
= bond
->params
.updelay
;
1785 new_slave
->link
= BOND_LINK_UP
;
1788 new_slave
->link
= BOND_LINK_DOWN
;
1790 } else if (bond
->params
.arp_interval
) {
1791 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1792 BOND_LINK_UP
: BOND_LINK_DOWN
);
1794 new_slave
->link
= BOND_LINK_UP
;
1797 if (new_slave
->link
!= BOND_LINK_DOWN
)
1798 new_slave
->jiffies
= jiffies
;
1799 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1800 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1801 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1803 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1804 /* if there is a primary slave, remember it */
1805 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1806 bond
->primary_slave
= new_slave
;
1807 bond
->force_primary
= true;
1811 write_lock_bh(&bond
->curr_slave_lock
);
1813 switch (bond
->params
.mode
) {
1814 case BOND_MODE_ACTIVEBACKUP
:
1815 bond_set_slave_inactive_flags(new_slave
);
1816 bond_select_active_slave(bond
);
1818 case BOND_MODE_8023AD
:
1819 /* in 802.3ad mode, the internal mechanism
1820 * will activate the slaves in the selected
1823 bond_set_slave_inactive_flags(new_slave
);
1824 /* if this is the first slave */
1825 if (bond
->slave_cnt
== 1) {
1826 SLAVE_AD_INFO(new_slave
).id
= 1;
1827 /* Initialize AD with the number of times that the AD timer is called in 1 second
1828 * can be called only after the mac address of the bond is set
1830 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1832 SLAVE_AD_INFO(new_slave
).id
=
1833 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1836 bond_3ad_bind_slave(new_slave
);
1840 bond_set_active_slave(new_slave
);
1841 bond_set_slave_inactive_flags(new_slave
);
1842 bond_select_active_slave(bond
);
1845 pr_debug("This slave is always active in trunk mode\n");
1847 /* always active in trunk mode */
1848 bond_set_active_slave(new_slave
);
1850 /* In trunking mode there is little meaning to curr_active_slave
1851 * anyway (it holds no special properties of the bond device),
1852 * so we can change it without calling change_active_interface()
1854 if (!bond
->curr_active_slave
&& new_slave
->link
== BOND_LINK_UP
)
1855 bond
->curr_active_slave
= new_slave
;
1858 } /* switch(bond_mode) */
1860 write_unlock_bh(&bond
->curr_slave_lock
);
1862 bond_set_carrier(bond
);
1864 #ifdef CONFIG_NET_POLL_CONTROLLER
1865 slave_dev
->npinfo
= bond_netpoll_info(bond
);
1866 if (slave_dev
->npinfo
) {
1867 if (slave_enable_netpoll(new_slave
)) {
1868 read_unlock(&bond
->lock
);
1869 pr_info("Error, %s: master_dev is using netpoll, "
1870 "but new slave device does not support netpoll.\n",
1878 read_unlock(&bond
->lock
);
1880 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1884 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1887 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1888 goto err_dest_symlinks
;
1891 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1892 bond_dev
->name
, slave_dev
->name
,
1893 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1894 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1896 /* enslave is successful */
1899 /* Undo stages on error */
1901 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1904 write_lock_bh(&bond
->lock
);
1905 bond_detach_slave(bond
, new_slave
);
1906 write_unlock_bh(&bond
->lock
);
1909 dev_close(slave_dev
);
1912 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1915 if (!bond
->params
.fail_over_mac
) {
1916 /* XXX TODO - fom follow mode needs to change master's
1917 * MAC if this slave's MAC is in use by the bond, or at
1918 * least print a warning.
1920 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1921 addr
.sa_family
= slave_dev
->type
;
1922 dev_set_mac_address(slave_dev
, &addr
);
1926 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1932 bond_compute_features(bond
);
1938 * Try to release the slave device <slave> from the bond device <master>
1939 * It is legal to access curr_active_slave without a lock because all the function
1940 * is write-locked. If "all" is true it means that the function is being called
1941 * while destroying a bond interface and all slaves are being released.
1943 * The rules for slave state should be:
1944 * for Active/Backup:
1945 * Active stays on all backups go down
1946 * for Bonded connections:
1947 * The first up interface should be left on and all others downed.
1949 static int __bond_release_one(struct net_device
*bond_dev
,
1950 struct net_device
*slave_dev
,
1953 struct bonding
*bond
= netdev_priv(bond_dev
);
1954 struct slave
*slave
, *oldcurrent
;
1955 struct sockaddr addr
;
1956 netdev_features_t old_features
= bond_dev
->features
;
1958 /* slave is not a slave or master is not master of this slave */
1959 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1960 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1961 pr_err("%s: Error: cannot release %s.\n",
1962 bond_dev
->name
, slave_dev
->name
);
1967 write_lock_bh(&bond
->lock
);
1969 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1971 /* not a slave of this bond */
1972 pr_info("%s: %s not enslaved\n",
1973 bond_dev
->name
, slave_dev
->name
);
1974 write_unlock_bh(&bond
->lock
);
1975 unblock_netpoll_tx();
1979 write_unlock_bh(&bond
->lock
);
1980 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1981 * for this slave anymore.
1983 netdev_rx_handler_unregister(slave_dev
);
1984 write_lock_bh(&bond
->lock
);
1986 if (!all
&& !bond
->params
.fail_over_mac
) {
1987 if (ether_addr_equal(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1988 bond
->slave_cnt
> 1)
1989 pr_warning("%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",
1990 bond_dev
->name
, slave_dev
->name
,
1992 bond_dev
->name
, slave_dev
->name
);
1995 /* Inform AD package of unbinding of slave. */
1996 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1997 /* must be called before the slave is
1998 * detached from the list
2000 bond_3ad_unbind_slave(slave
);
2003 pr_info("%s: releasing %s interface %s\n",
2005 bond_is_active_slave(slave
) ? "active" : "backup",
2008 oldcurrent
= bond
->curr_active_slave
;
2010 bond
->current_arp_slave
= NULL
;
2012 /* release the slave from its bond */
2013 bond_detach_slave(bond
, slave
);
2015 if (bond
->primary_slave
== slave
)
2016 bond
->primary_slave
= NULL
;
2018 if (oldcurrent
== slave
)
2019 bond_change_active_slave(bond
, NULL
);
2021 if (bond_is_lb(bond
)) {
2022 /* Must be called only after the slave has been
2023 * detached from the list and the curr_active_slave
2024 * has been cleared (if our_slave == old_current),
2025 * but before a new active slave is selected.
2027 write_unlock_bh(&bond
->lock
);
2028 bond_alb_deinit_slave(bond
, slave
);
2029 write_lock_bh(&bond
->lock
);
2033 bond
->curr_active_slave
= NULL
;
2034 } else if (oldcurrent
== slave
) {
2036 * Note that we hold RTNL over this sequence, so there
2037 * is no concern that another slave add/remove event
2040 write_unlock_bh(&bond
->lock
);
2041 read_lock(&bond
->lock
);
2042 write_lock_bh(&bond
->curr_slave_lock
);
2044 bond_select_active_slave(bond
);
2046 write_unlock_bh(&bond
->curr_slave_lock
);
2047 read_unlock(&bond
->lock
);
2048 write_lock_bh(&bond
->lock
);
2051 if (bond
->slave_cnt
== 0) {
2052 bond_set_carrier(bond
);
2053 eth_hw_addr_random(bond_dev
);
2054 bond
->dev_addr_from_first
= true;
2056 if (bond_vlan_used(bond
)) {
2057 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2058 bond_dev
->name
, bond_dev
->name
);
2059 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2064 write_unlock_bh(&bond
->lock
);
2065 unblock_netpoll_tx();
2067 if (bond
->slave_cnt
== 0) {
2068 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
2069 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
2072 bond_compute_features(bond
);
2073 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
2074 (old_features
& NETIF_F_VLAN_CHALLENGED
))
2075 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2076 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
2078 /* must do this from outside any spinlocks */
2079 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2081 bond_del_vlans_from_slave(bond
, slave_dev
);
2083 /* If the mode USES_PRIMARY, then we should only remove its
2084 * promisc and mc settings if it was the curr_active_slave, but that was
2085 * already taken care of above when we detached the slave
2087 if (!USES_PRIMARY(bond
->params
.mode
)) {
2088 /* unset promiscuity level from slave */
2089 if (bond_dev
->flags
& IFF_PROMISC
)
2090 dev_set_promiscuity(slave_dev
, -1);
2092 /* unset allmulti level from slave */
2093 if (bond_dev
->flags
& IFF_ALLMULTI
)
2094 dev_set_allmulti(slave_dev
, -1);
2096 /* flush master's mc_list from slave */
2097 netif_addr_lock_bh(bond_dev
);
2098 bond_mc_list_flush(bond_dev
, slave_dev
);
2099 netif_addr_unlock_bh(bond_dev
);
2102 bond_upper_dev_unlink(bond_dev
, slave_dev
);
2104 slave_disable_netpoll(slave
);
2106 /* close slave before restoring its mac address */
2107 dev_close(slave_dev
);
2109 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2110 /* restore original ("permanent") mac address */
2111 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2112 addr
.sa_family
= slave_dev
->type
;
2113 dev_set_mac_address(slave_dev
, &addr
);
2116 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2118 slave_dev
->priv_flags
&= ~IFF_BONDING
;
2122 return 0; /* deletion OK */
2125 /* A wrapper used because of ndo_del_link */
2126 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2128 return __bond_release_one(bond_dev
, slave_dev
, false);
2132 * First release a slave and then destroy the bond if no more slaves are left.
2133 * Must be under rtnl_lock when this function is called.
2135 static int bond_release_and_destroy(struct net_device
*bond_dev
,
2136 struct net_device
*slave_dev
)
2138 struct bonding
*bond
= netdev_priv(bond_dev
);
2141 ret
= bond_release(bond_dev
, slave_dev
);
2142 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2143 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
2144 pr_info("%s: destroying bond %s.\n",
2145 bond_dev
->name
, bond_dev
->name
);
2146 unregister_netdevice(bond_dev
);
2152 * This function changes the active slave to slave <slave_dev>.
2153 * It returns -EINVAL in the following cases.
2154 * - <slave_dev> is not found in the list.
2155 * - There is not active slave now.
2156 * - <slave_dev> is already active.
2157 * - The link state of <slave_dev> is not BOND_LINK_UP.
2158 * - <slave_dev> is not running.
2159 * In these cases, this function does nothing.
2160 * In the other cases, current_slave pointer is changed and 0 is returned.
2162 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2164 struct bonding
*bond
= netdev_priv(bond_dev
);
2165 struct slave
*old_active
= NULL
;
2166 struct slave
*new_active
= NULL
;
2169 if (!USES_PRIMARY(bond
->params
.mode
))
2172 /* Verify that bond_dev is indeed the master of slave_dev */
2173 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
2174 !netdev_has_upper_dev(slave_dev
, bond_dev
))
2177 read_lock(&bond
->lock
);
2179 read_lock(&bond
->curr_slave_lock
);
2180 old_active
= bond
->curr_active_slave
;
2181 read_unlock(&bond
->curr_slave_lock
);
2183 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2186 * Changing to the current active: do nothing; return success.
2188 if (new_active
&& (new_active
== old_active
)) {
2189 read_unlock(&bond
->lock
);
2195 (new_active
->link
== BOND_LINK_UP
) &&
2196 IS_UP(new_active
->dev
)) {
2198 write_lock_bh(&bond
->curr_slave_lock
);
2199 bond_change_active_slave(bond
, new_active
);
2200 write_unlock_bh(&bond
->curr_slave_lock
);
2201 unblock_netpoll_tx();
2205 read_unlock(&bond
->lock
);
2210 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2212 struct bonding
*bond
= netdev_priv(bond_dev
);
2214 info
->bond_mode
= bond
->params
.mode
;
2215 info
->miimon
= bond
->params
.miimon
;
2217 read_lock(&bond
->lock
);
2218 info
->num_slaves
= bond
->slave_cnt
;
2219 read_unlock(&bond
->lock
);
2224 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2226 struct bonding
*bond
= netdev_priv(bond_dev
);
2227 struct slave
*slave
;
2228 int i
, res
= -ENODEV
;
2230 read_lock(&bond
->lock
);
2232 bond_for_each_slave(bond
, slave
, i
) {
2233 if (i
== (int)info
->slave_id
) {
2235 strcpy(info
->slave_name
, slave
->dev
->name
);
2236 info
->link
= slave
->link
;
2237 info
->state
= bond_slave_state(slave
);
2238 info
->link_failure_count
= slave
->link_failure_count
;
2243 read_unlock(&bond
->lock
);
2248 /*-------------------------------- Monitoring -------------------------------*/
2251 static int bond_miimon_inspect(struct bonding
*bond
)
2253 struct slave
*slave
;
2254 int i
, link_state
, commit
= 0;
2255 bool ignore_updelay
;
2257 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2259 bond_for_each_slave(bond
, slave
, i
) {
2260 slave
->new_link
= BOND_LINK_NOCHANGE
;
2262 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2264 switch (slave
->link
) {
2269 slave
->link
= BOND_LINK_FAIL
;
2270 slave
->delay
= bond
->params
.downdelay
;
2272 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2274 (bond
->params
.mode
==
2275 BOND_MODE_ACTIVEBACKUP
) ?
2276 (bond_is_active_slave(slave
) ?
2277 "active " : "backup ") : "",
2279 bond
->params
.downdelay
* bond
->params
.miimon
);
2282 case BOND_LINK_FAIL
:
2285 * recovered before downdelay expired
2287 slave
->link
= BOND_LINK_UP
;
2288 slave
->jiffies
= jiffies
;
2289 pr_info("%s: link status up again after %d ms for interface %s.\n",
2291 (bond
->params
.downdelay
- slave
->delay
) *
2292 bond
->params
.miimon
,
2297 if (slave
->delay
<= 0) {
2298 slave
->new_link
= BOND_LINK_DOWN
;
2306 case BOND_LINK_DOWN
:
2310 slave
->link
= BOND_LINK_BACK
;
2311 slave
->delay
= bond
->params
.updelay
;
2314 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2315 bond
->dev
->name
, slave
->dev
->name
,
2316 ignore_updelay
? 0 :
2317 bond
->params
.updelay
*
2318 bond
->params
.miimon
);
2321 case BOND_LINK_BACK
:
2323 slave
->link
= BOND_LINK_DOWN
;
2324 pr_info("%s: link status down again after %d ms for interface %s.\n",
2326 (bond
->params
.updelay
- slave
->delay
) *
2327 bond
->params
.miimon
,
2336 if (slave
->delay
<= 0) {
2337 slave
->new_link
= BOND_LINK_UP
;
2339 ignore_updelay
= false;
2351 static void bond_miimon_commit(struct bonding
*bond
)
2353 struct slave
*slave
;
2356 bond_for_each_slave(bond
, slave
, i
) {
2357 switch (slave
->new_link
) {
2358 case BOND_LINK_NOCHANGE
:
2362 slave
->link
= BOND_LINK_UP
;
2363 slave
->jiffies
= jiffies
;
2365 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2366 /* prevent it from being the active one */
2367 bond_set_backup_slave(slave
);
2368 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2369 /* make it immediately active */
2370 bond_set_active_slave(slave
);
2371 } else if (slave
!= bond
->primary_slave
) {
2372 /* prevent it from being the active one */
2373 bond_set_backup_slave(slave
);
2376 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2377 bond
->dev
->name
, slave
->dev
->name
,
2378 slave
->speed
, slave
->duplex
? "full" : "half");
2380 /* notify ad that the link status has changed */
2381 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2382 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2384 if (bond_is_lb(bond
))
2385 bond_alb_handle_link_change(bond
, slave
,
2388 if (!bond
->curr_active_slave
||
2389 (slave
== bond
->primary_slave
))
2394 case BOND_LINK_DOWN
:
2395 if (slave
->link_failure_count
< UINT_MAX
)
2396 slave
->link_failure_count
++;
2398 slave
->link
= BOND_LINK_DOWN
;
2400 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2401 bond
->params
.mode
== BOND_MODE_8023AD
)
2402 bond_set_slave_inactive_flags(slave
);
2404 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2405 bond
->dev
->name
, slave
->dev
->name
);
2407 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2408 bond_3ad_handle_link_change(slave
,
2411 if (bond_is_lb(bond
))
2412 bond_alb_handle_link_change(bond
, slave
,
2415 if (slave
== bond
->curr_active_slave
)
2421 pr_err("%s: invalid new link %d on slave %s\n",
2422 bond
->dev
->name
, slave
->new_link
,
2424 slave
->new_link
= BOND_LINK_NOCHANGE
;
2432 write_lock_bh(&bond
->curr_slave_lock
);
2433 bond_select_active_slave(bond
);
2434 write_unlock_bh(&bond
->curr_slave_lock
);
2435 unblock_netpoll_tx();
2438 bond_set_carrier(bond
);
2444 * Really a wrapper that splits the mii monitor into two phases: an
2445 * inspection, then (if inspection indicates something needs to be done)
2446 * an acquisition of appropriate locks followed by a commit phase to
2447 * implement whatever link state changes are indicated.
2449 void bond_mii_monitor(struct work_struct
*work
)
2451 struct bonding
*bond
= container_of(work
, struct bonding
,
2453 bool should_notify_peers
= false;
2454 unsigned long delay
;
2456 read_lock(&bond
->lock
);
2458 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2460 if (bond
->slave_cnt
== 0)
2463 should_notify_peers
= bond_should_notify_peers(bond
);
2465 if (bond_miimon_inspect(bond
)) {
2466 read_unlock(&bond
->lock
);
2468 /* Race avoidance with bond_close cancel of workqueue */
2469 if (!rtnl_trylock()) {
2470 read_lock(&bond
->lock
);
2472 should_notify_peers
= false;
2476 read_lock(&bond
->lock
);
2478 bond_miimon_commit(bond
);
2480 read_unlock(&bond
->lock
);
2481 rtnl_unlock(); /* might sleep, hold no other locks */
2482 read_lock(&bond
->lock
);
2486 if (bond
->params
.miimon
)
2487 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2489 read_unlock(&bond
->lock
);
2491 if (should_notify_peers
) {
2492 if (!rtnl_trylock()) {
2493 read_lock(&bond
->lock
);
2494 bond
->send_peer_notif
++;
2495 read_unlock(&bond
->lock
);
2498 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2503 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2505 struct vlan_entry
*vlan
;
2506 struct net_device
*vlan_dev
;
2508 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2511 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2513 vlan_dev
= __vlan_find_dev_deep(bond
->dev
, vlan
->vlan_id
);
2515 if (vlan_dev
&& ip
== bond_confirm_addr(vlan_dev
, 0, ip
))
2523 * We go to the (large) trouble of VLAN tagging ARP frames because
2524 * switches in VLAN mode (especially if ports are configured as
2525 * "native" to a VLAN) might not pass non-tagged frames.
2527 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2529 struct sk_buff
*skb
;
2531 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2532 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2534 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2535 NULL
, slave_dev
->dev_addr
, NULL
);
2538 pr_err("ARP packet allocation failed\n");
2542 skb
= vlan_put_tag(skb
, vlan_id
);
2544 pr_err("failed to insert VLAN tag\n");
2552 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2555 __be32
*targets
= bond
->params
.arp_targets
;
2556 struct vlan_entry
*vlan
;
2557 struct net_device
*vlan_dev
= NULL
;
2560 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2564 pr_debug("basa: target %x\n", targets
[i
]);
2565 if (!bond_vlan_used(bond
)) {
2566 pr_debug("basa: empty vlan: arp_send\n");
2567 addr
= bond_confirm_addr(bond
->dev
, targets
[i
], 0);
2568 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2574 * If VLANs are configured, we do a route lookup to
2575 * determine which VLAN interface would be used, so we
2576 * can tag the ARP with the proper VLAN tag.
2578 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2581 if (net_ratelimit()) {
2582 pr_warning("%s: no route to arp_ip_target %pI4\n",
2583 bond
->dev
->name
, &targets
[i
]);
2589 * This target is not on a VLAN
2591 if (rt
->dst
.dev
== bond
->dev
) {
2593 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2594 addr
= bond_confirm_addr(bond
->dev
, targets
[i
], 0);
2595 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2601 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2603 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
2606 if (vlan_dev
== rt
->dst
.dev
) {
2607 vlan_id
= vlan
->vlan_id
;
2608 pr_debug("basa: vlan match on %s %d\n",
2609 vlan_dev
->name
, vlan_id
);
2614 if (vlan_id
&& vlan_dev
) {
2616 addr
= bond_confirm_addr(vlan_dev
, targets
[i
], 0);
2617 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2622 if (net_ratelimit()) {
2623 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2624 bond
->dev
->name
, &targets
[i
],
2625 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2631 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2634 __be32
*targets
= bond
->params
.arp_targets
;
2636 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2637 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2638 &sip
, &tip
, i
, &targets
[i
],
2639 bond_has_this_ip(bond
, tip
));
2640 if (sip
== targets
[i
]) {
2641 if (bond_has_this_ip(bond
, tip
))
2642 slave
->last_arp_rx
= jiffies
;
2648 static int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2649 struct slave
*slave
)
2651 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2652 unsigned char *arp_ptr
;
2656 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2657 return RX_HANDLER_ANOTHER
;
2659 read_lock(&bond
->lock
);
2660 alen
= arp_hdr_len(bond
->dev
);
2662 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2663 bond
->dev
->name
, skb
->dev
->name
);
2665 if (alen
> skb_headlen(skb
)) {
2666 arp
= kmalloc(alen
, GFP_ATOMIC
);
2669 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2673 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2674 skb
->pkt_type
== PACKET_OTHERHOST
||
2675 skb
->pkt_type
== PACKET_LOOPBACK
||
2676 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2677 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2681 arp_ptr
= (unsigned char *)(arp
+ 1);
2682 arp_ptr
+= bond
->dev
->addr_len
;
2683 memcpy(&sip
, arp_ptr
, 4);
2684 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2685 memcpy(&tip
, arp_ptr
, 4);
2687 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2688 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2689 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2693 * Backup slaves won't see the ARP reply, but do come through
2694 * here for each ARP probe (so we swap the sip/tip to validate
2695 * the probe). In a "redundant switch, common router" type of
2696 * configuration, the ARP probe will (hopefully) travel from
2697 * the active, through one switch, the router, then the other
2698 * switch before reaching the backup.
2700 if (bond_is_active_slave(slave
))
2701 bond_validate_arp(bond
, slave
, sip
, tip
);
2703 bond_validate_arp(bond
, slave
, tip
, sip
);
2706 read_unlock(&bond
->lock
);
2707 if (arp
!= (struct arphdr
*)skb
->data
)
2709 return RX_HANDLER_ANOTHER
;
2713 * this function is called regularly to monitor each slave's link
2714 * ensuring that traffic is being sent and received when arp monitoring
2715 * is used in load-balancing mode. if the adapter has been dormant, then an
2716 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2717 * arp monitoring in active backup mode.
2719 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2721 struct bonding
*bond
= container_of(work
, struct bonding
,
2723 struct slave
*slave
, *oldcurrent
;
2724 int do_failover
= 0;
2725 int delta_in_ticks
, extra_ticks
;
2728 read_lock(&bond
->lock
);
2730 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2731 extra_ticks
= delta_in_ticks
/ 2;
2733 if (bond
->slave_cnt
== 0)
2736 read_lock(&bond
->curr_slave_lock
);
2737 oldcurrent
= bond
->curr_active_slave
;
2738 read_unlock(&bond
->curr_slave_lock
);
2740 /* see if any of the previous devices are up now (i.e. they have
2741 * xmt and rcv traffic). the curr_active_slave does not come into
2742 * the picture unless it is null. also, slave->jiffies is not needed
2743 * here because we send an arp on each slave and give a slave as
2744 * long as it needs to get the tx/rx within the delta.
2745 * TODO: what about up/down delay in arp mode? it wasn't here before
2748 bond_for_each_slave(bond
, slave
, i
) {
2749 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2751 if (slave
->link
!= BOND_LINK_UP
) {
2752 if (time_in_range(jiffies
,
2753 trans_start
- delta_in_ticks
,
2754 trans_start
+ delta_in_ticks
+ extra_ticks
) &&
2755 time_in_range(jiffies
,
2756 slave
->dev
->last_rx
- delta_in_ticks
,
2757 slave
->dev
->last_rx
+ delta_in_ticks
+ extra_ticks
)) {
2759 slave
->link
= BOND_LINK_UP
;
2760 bond_set_active_slave(slave
);
2762 /* primary_slave has no meaning in round-robin
2763 * mode. the window of a slave being up and
2764 * curr_active_slave being null after enslaving
2768 pr_info("%s: link status definitely up for interface %s, ",
2773 pr_info("%s: interface %s is now up\n",
2779 /* slave->link == BOND_LINK_UP */
2781 /* not all switches will respond to an arp request
2782 * when the source ip is 0, so don't take the link down
2783 * if we don't know our ip yet
2785 if (!time_in_range(jiffies
,
2786 trans_start
- delta_in_ticks
,
2787 trans_start
+ 2 * delta_in_ticks
+ extra_ticks
) ||
2788 !time_in_range(jiffies
,
2789 slave
->dev
->last_rx
- delta_in_ticks
,
2790 slave
->dev
->last_rx
+ 2 * delta_in_ticks
+ extra_ticks
)) {
2792 slave
->link
= BOND_LINK_DOWN
;
2793 bond_set_backup_slave(slave
);
2795 if (slave
->link_failure_count
< UINT_MAX
)
2796 slave
->link_failure_count
++;
2798 pr_info("%s: interface %s is now down.\n",
2802 if (slave
== oldcurrent
)
2807 /* note: if switch is in round-robin mode, all links
2808 * must tx arp to ensure all links rx an arp - otherwise
2809 * links may oscillate or not come up at all; if switch is
2810 * in something like xor mode, there is nothing we can
2811 * do - all replies will be rx'ed on same link causing slaves
2812 * to be unstable during low/no traffic periods
2814 if (IS_UP(slave
->dev
))
2815 bond_arp_send_all(bond
, slave
);
2820 write_lock_bh(&bond
->curr_slave_lock
);
2822 bond_select_active_slave(bond
);
2824 write_unlock_bh(&bond
->curr_slave_lock
);
2825 unblock_netpoll_tx();
2829 if (bond
->params
.arp_interval
)
2830 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2832 read_unlock(&bond
->lock
);
2836 * Called to inspect slaves for active-backup mode ARP monitor link state
2837 * changes. Sets new_link in slaves to specify what action should take
2838 * place for the slave. Returns 0 if no changes are found, >0 if changes
2839 * to link states must be committed.
2841 * Called with bond->lock held for read.
2843 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2845 struct slave
*slave
;
2847 unsigned long trans_start
;
2850 /* All the time comparisons below need some extra time. Otherwise, on
2851 * fast networks the ARP probe/reply may arrive within the same jiffy
2852 * as it was sent. Then, the next time the ARP monitor is run, one
2853 * arp_interval will already have passed in the comparisons.
2855 extra_ticks
= delta_in_ticks
/ 2;
2857 bond_for_each_slave(bond
, slave
, i
) {
2858 slave
->new_link
= BOND_LINK_NOCHANGE
;
2860 if (slave
->link
!= BOND_LINK_UP
) {
2861 if (time_in_range(jiffies
,
2862 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2863 slave_last_rx(bond
, slave
) + delta_in_ticks
+ extra_ticks
)) {
2865 slave
->new_link
= BOND_LINK_UP
;
2873 * Give slaves 2*delta after being enslaved or made
2874 * active. This avoids bouncing, as the last receive
2875 * times need a full ARP monitor cycle to be updated.
2877 if (time_in_range(jiffies
,
2878 slave
->jiffies
- delta_in_ticks
,
2879 slave
->jiffies
+ 2 * delta_in_ticks
+ extra_ticks
))
2883 * Backup slave is down if:
2884 * - No current_arp_slave AND
2885 * - more than 3*delta since last receive AND
2886 * - the bond has an IP address
2888 * Note: a non-null current_arp_slave indicates
2889 * the curr_active_slave went down and we are
2890 * searching for a new one; under this condition
2891 * we only take the curr_active_slave down - this
2892 * gives each slave a chance to tx/rx traffic
2893 * before being taken out
2895 if (!bond_is_active_slave(slave
) &&
2896 !bond
->current_arp_slave
&&
2897 !time_in_range(jiffies
,
2898 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2899 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
+ extra_ticks
)) {
2901 slave
->new_link
= BOND_LINK_DOWN
;
2906 * Active slave is down if:
2907 * - more than 2*delta since transmitting OR
2908 * - (more than 2*delta since receive AND
2909 * the bond has an IP address)
2911 trans_start
= dev_trans_start(slave
->dev
);
2912 if (bond_is_active_slave(slave
) &&
2913 (!time_in_range(jiffies
,
2914 trans_start
- delta_in_ticks
,
2915 trans_start
+ 2 * delta_in_ticks
+ extra_ticks
) ||
2916 !time_in_range(jiffies
,
2917 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2918 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
+ extra_ticks
))) {
2920 slave
->new_link
= BOND_LINK_DOWN
;
2929 * Called to commit link state changes noted by inspection step of
2930 * active-backup mode ARP monitor.
2932 * Called with RTNL and bond->lock for read.
2934 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2936 struct slave
*slave
;
2938 unsigned long trans_start
;
2940 bond_for_each_slave(bond
, slave
, i
) {
2941 switch (slave
->new_link
) {
2942 case BOND_LINK_NOCHANGE
:
2946 trans_start
= dev_trans_start(slave
->dev
);
2947 if ((!bond
->curr_active_slave
&&
2948 time_in_range(jiffies
,
2949 trans_start
- delta_in_ticks
,
2950 trans_start
+ delta_in_ticks
+ delta_in_ticks
/ 2)) ||
2951 bond
->curr_active_slave
!= slave
) {
2952 slave
->link
= BOND_LINK_UP
;
2953 if (bond
->current_arp_slave
) {
2954 bond_set_slave_inactive_flags(
2955 bond
->current_arp_slave
);
2956 bond
->current_arp_slave
= NULL
;
2959 pr_info("%s: link status definitely up for interface %s.\n",
2960 bond
->dev
->name
, slave
->dev
->name
);
2962 if (!bond
->curr_active_slave
||
2963 (slave
== bond
->primary_slave
))
2970 case BOND_LINK_DOWN
:
2971 if (slave
->link_failure_count
< UINT_MAX
)
2972 slave
->link_failure_count
++;
2974 slave
->link
= BOND_LINK_DOWN
;
2975 bond_set_slave_inactive_flags(slave
);
2977 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2978 bond
->dev
->name
, slave
->dev
->name
);
2980 if (slave
== bond
->curr_active_slave
) {
2981 bond
->current_arp_slave
= NULL
;
2988 pr_err("%s: impossible: new_link %d on slave %s\n",
2989 bond
->dev
->name
, slave
->new_link
,
2997 write_lock_bh(&bond
->curr_slave_lock
);
2998 bond_select_active_slave(bond
);
2999 write_unlock_bh(&bond
->curr_slave_lock
);
3000 unblock_netpoll_tx();
3003 bond_set_carrier(bond
);
3007 * Send ARP probes for active-backup mode ARP monitor.
3009 * Called with bond->lock held for read.
3011 static void bond_ab_arp_probe(struct bonding
*bond
)
3013 struct slave
*slave
;
3016 read_lock(&bond
->curr_slave_lock
);
3018 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3019 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3020 bond
->current_arp_slave
->dev
->name
,
3021 bond
->curr_active_slave
->dev
->name
);
3023 if (bond
->curr_active_slave
) {
3024 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3025 read_unlock(&bond
->curr_slave_lock
);
3029 read_unlock(&bond
->curr_slave_lock
);
3031 /* if we don't have a curr_active_slave, search for the next available
3032 * backup slave from the current_arp_slave and make it the candidate
3033 * for becoming the curr_active_slave
3036 if (!bond
->current_arp_slave
) {
3037 bond
->current_arp_slave
= bond
->first_slave
;
3038 if (!bond
->current_arp_slave
)
3042 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3044 /* search for next candidate */
3045 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3046 if (IS_UP(slave
->dev
)) {
3047 slave
->link
= BOND_LINK_BACK
;
3048 bond_set_slave_active_flags(slave
);
3049 bond_arp_send_all(bond
, slave
);
3050 slave
->jiffies
= jiffies
;
3051 bond
->current_arp_slave
= slave
;
3055 /* if the link state is up at this point, we
3056 * mark it down - this can happen if we have
3057 * simultaneous link failures and
3058 * reselect_active_interface doesn't make this
3059 * one the current slave so it is still marked
3060 * up when it is actually down
3062 if (slave
->link
== BOND_LINK_UP
) {
3063 slave
->link
= BOND_LINK_DOWN
;
3064 if (slave
->link_failure_count
< UINT_MAX
)
3065 slave
->link_failure_count
++;
3067 bond_set_slave_inactive_flags(slave
);
3069 pr_info("%s: backup interface %s is now down.\n",
3070 bond
->dev
->name
, slave
->dev
->name
);
3075 void bond_activebackup_arp_mon(struct work_struct
*work
)
3077 struct bonding
*bond
= container_of(work
, struct bonding
,
3079 bool should_notify_peers
= false;
3082 read_lock(&bond
->lock
);
3084 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3086 if (bond
->slave_cnt
== 0)
3089 should_notify_peers
= bond_should_notify_peers(bond
);
3091 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3092 read_unlock(&bond
->lock
);
3094 /* Race avoidance with bond_close flush of workqueue */
3095 if (!rtnl_trylock()) {
3096 read_lock(&bond
->lock
);
3098 should_notify_peers
= false;
3102 read_lock(&bond
->lock
);
3104 bond_ab_arp_commit(bond
, delta_in_ticks
);
3106 read_unlock(&bond
->lock
);
3108 read_lock(&bond
->lock
);
3111 bond_ab_arp_probe(bond
);
3114 if (bond
->params
.arp_interval
)
3115 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3117 read_unlock(&bond
->lock
);
3119 if (should_notify_peers
) {
3120 if (!rtnl_trylock()) {
3121 read_lock(&bond
->lock
);
3122 bond
->send_peer_notif
++;
3123 read_unlock(&bond
->lock
);
3126 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
3131 /*-------------------------- netdev event handling --------------------------*/
3134 * Change device name
3136 static int bond_event_changename(struct bonding
*bond
)
3138 bond_remove_proc_entry(bond
);
3139 bond_create_proc_entry(bond
);
3141 bond_debug_reregister(bond
);
3146 static int bond_master_netdev_event(unsigned long event
,
3147 struct net_device
*bond_dev
)
3149 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3152 case NETDEV_CHANGENAME
:
3153 return bond_event_changename(event_bond
);
3154 case NETDEV_UNREGISTER
:
3155 bond_remove_proc_entry(event_bond
);
3157 case NETDEV_REGISTER
:
3158 bond_create_proc_entry(event_bond
);
3167 static int bond_slave_netdev_event(unsigned long event
,
3168 struct net_device
*slave_dev
)
3170 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
);
3171 struct bonding
*bond
= slave
->bond
;
3172 struct net_device
*bond_dev
= slave
->bond
->dev
;
3177 case NETDEV_UNREGISTER
:
3178 if (bond
->setup_by_slave
)
3179 bond_release_and_destroy(bond_dev
, slave_dev
);
3181 bond_release(bond_dev
, slave_dev
);
3185 old_speed
= slave
->speed
;
3186 old_duplex
= slave
->duplex
;
3188 bond_update_speed_duplex(slave
);
3190 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3191 if (old_speed
!= slave
->speed
)
3192 bond_3ad_adapter_speed_changed(slave
);
3193 if (old_duplex
!= slave
->duplex
)
3194 bond_3ad_adapter_duplex_changed(slave
);
3199 * ... Or is it this?
3202 case NETDEV_CHANGEMTU
:
3204 * TODO: Should slaves be allowed to
3205 * independently alter their MTU? For
3206 * an active-backup bond, slaves need
3207 * not be the same type of device, so
3208 * MTUs may vary. For other modes,
3209 * slaves arguably should have the
3210 * same MTUs. To do this, we'd need to
3211 * take over the slave's change_mtu
3212 * function for the duration of their
3216 case NETDEV_CHANGENAME
:
3218 * TODO: handle changing the primary's name
3221 case NETDEV_FEAT_CHANGE
:
3222 bond_compute_features(bond
);
3232 * bond_netdev_event: handle netdev notifier chain events.
3234 * This function receives events for the netdev chain. The caller (an
3235 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3236 * locks for us to safely manipulate the slave devices (RTNL lock,
3239 static int bond_netdev_event(struct notifier_block
*this,
3240 unsigned long event
, void *ptr
)
3242 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3244 pr_debug("event_dev: %s, event: %lx\n",
3245 event_dev
? event_dev
->name
: "None",
3248 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3251 if (event_dev
->flags
& IFF_MASTER
) {
3252 pr_debug("IFF_MASTER\n");
3253 return bond_master_netdev_event(event
, event_dev
);
3256 if (event_dev
->flags
& IFF_SLAVE
) {
3257 pr_debug("IFF_SLAVE\n");
3258 return bond_slave_netdev_event(event
, event_dev
);
3264 static struct notifier_block bond_netdev_notifier
= {
3265 .notifier_call
= bond_netdev_event
,
3268 /*---------------------------- Hashing Policies -----------------------------*/
3271 * Hash for the output device based upon layer 2 data
3273 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3275 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3277 if (skb_headlen(skb
) >= offsetof(struct ethhdr
, h_proto
))
3278 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3284 * Hash for the output device based upon layer 2 and layer 3 data. If
3285 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3287 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3289 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3291 struct ipv6hdr
*ipv6h
;
3295 if (skb
->protocol
== htons(ETH_P_IP
) &&
3296 skb_network_header_len(skb
) >= sizeof(*iph
)) {
3298 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3299 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3300 } else if (skb
->protocol
== htons(ETH_P_IPV6
) &&
3301 skb_network_header_len(skb
) >= sizeof(*ipv6h
)) {
3302 ipv6h
= ipv6_hdr(skb
);
3303 s
= &ipv6h
->saddr
.s6_addr32
[0];
3304 d
= &ipv6h
->daddr
.s6_addr32
[0];
3305 v6hash
= (s
[1] ^ d
[1]) ^ (s
[2] ^ d
[2]) ^ (s
[3] ^ d
[3]);
3306 v6hash
^= (v6hash
>> 24) ^ (v6hash
>> 16) ^ (v6hash
>> 8);
3307 return (v6hash
^ data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3310 return bond_xmit_hash_policy_l2(skb
, count
);
3314 * Hash for the output device based upon layer 3 and layer 4 data. If
3315 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3316 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3318 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3322 struct ipv6hdr
*ipv6h
;
3326 if (skb
->protocol
== htons(ETH_P_IP
) &&
3327 skb_network_header_len(skb
) >= sizeof(*iph
)) {
3329 if (!ip_is_fragment(iph
) &&
3330 (iph
->protocol
== IPPROTO_TCP
||
3331 iph
->protocol
== IPPROTO_UDP
) &&
3332 (skb_headlen(skb
) - skb_network_offset(skb
) >=
3333 iph
->ihl
* sizeof(u32
) + sizeof(*layer4hdr
) * 2)) {
3334 layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3335 layer4_xor
= ntohs(*layer4hdr
^ *(layer4hdr
+ 1));
3337 return (layer4_xor
^
3338 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3339 } else if (skb
->protocol
== htons(ETH_P_IPV6
) &&
3340 skb_network_header_len(skb
) >= sizeof(*ipv6h
)) {
3341 ipv6h
= ipv6_hdr(skb
);
3342 if ((ipv6h
->nexthdr
== IPPROTO_TCP
||
3343 ipv6h
->nexthdr
== IPPROTO_UDP
) &&
3344 (skb_headlen(skb
) - skb_network_offset(skb
) >=
3345 sizeof(*ipv6h
) + sizeof(*layer4hdr
) * 2)) {
3346 layer4hdr
= (__be16
*)(ipv6h
+ 1);
3347 layer4_xor
= ntohs(*layer4hdr
^ *(layer4hdr
+ 1));
3349 s
= &ipv6h
->saddr
.s6_addr32
[0];
3350 d
= &ipv6h
->daddr
.s6_addr32
[0];
3351 layer4_xor
^= (s
[1] ^ d
[1]) ^ (s
[2] ^ d
[2]) ^ (s
[3] ^ d
[3]);
3352 layer4_xor
^= (layer4_xor
>> 24) ^ (layer4_xor
>> 16) ^
3354 return layer4_xor
% count
;
3357 return bond_xmit_hash_policy_l2(skb
, count
);
3360 /*-------------------------- Device entry points ----------------------------*/
3362 static void bond_work_init_all(struct bonding
*bond
)
3364 INIT_DELAYED_WORK(&bond
->mcast_work
,
3365 bond_resend_igmp_join_requests_delayed
);
3366 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3367 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3368 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3369 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3371 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3372 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3375 static void bond_work_cancel_all(struct bonding
*bond
)
3377 cancel_delayed_work_sync(&bond
->mii_work
);
3378 cancel_delayed_work_sync(&bond
->arp_work
);
3379 cancel_delayed_work_sync(&bond
->alb_work
);
3380 cancel_delayed_work_sync(&bond
->ad_work
);
3381 cancel_delayed_work_sync(&bond
->mcast_work
);
3384 static int bond_open(struct net_device
*bond_dev
)
3386 struct bonding
*bond
= netdev_priv(bond_dev
);
3387 struct slave
*slave
;
3390 /* reset slave->backup and slave->inactive */
3391 read_lock(&bond
->lock
);
3392 if (bond
->slave_cnt
> 0) {
3393 read_lock(&bond
->curr_slave_lock
);
3394 bond_for_each_slave(bond
, slave
, i
) {
3395 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3396 && (slave
!= bond
->curr_active_slave
)) {
3397 bond_set_slave_inactive_flags(slave
);
3399 bond_set_slave_active_flags(slave
);
3402 read_unlock(&bond
->curr_slave_lock
);
3404 read_unlock(&bond
->lock
);
3406 bond_work_init_all(bond
);
3408 if (bond_is_lb(bond
)) {
3409 /* bond_alb_initialize must be called before the timer
3412 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
)))
3414 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3417 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3418 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3420 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3421 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3422 if (bond
->params
.arp_validate
)
3423 bond
->recv_probe
= bond_arp_rcv
;
3426 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3427 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3428 /* register to receive LACPDUs */
3429 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3430 bond_3ad_initiate_agg_selection(bond
, 1);
3436 static int bond_close(struct net_device
*bond_dev
)
3438 struct bonding
*bond
= netdev_priv(bond_dev
);
3440 write_lock_bh(&bond
->lock
);
3441 bond
->send_peer_notif
= 0;
3442 write_unlock_bh(&bond
->lock
);
3444 bond_work_cancel_all(bond
);
3445 if (bond_is_lb(bond
)) {
3446 /* Must be called only after all
3447 * slaves have been released
3449 bond_alb_deinitialize(bond
);
3451 bond
->recv_probe
= NULL
;
3456 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3457 struct rtnl_link_stats64
*stats
)
3459 struct bonding
*bond
= netdev_priv(bond_dev
);
3460 struct rtnl_link_stats64 temp
;
3461 struct slave
*slave
;
3464 memset(stats
, 0, sizeof(*stats
));
3466 read_lock_bh(&bond
->lock
);
3468 bond_for_each_slave(bond
, slave
, i
) {
3469 const struct rtnl_link_stats64
*sstats
=
3470 dev_get_stats(slave
->dev
, &temp
);
3472 stats
->rx_packets
+= sstats
->rx_packets
;
3473 stats
->rx_bytes
+= sstats
->rx_bytes
;
3474 stats
->rx_errors
+= sstats
->rx_errors
;
3475 stats
->rx_dropped
+= sstats
->rx_dropped
;
3477 stats
->tx_packets
+= sstats
->tx_packets
;
3478 stats
->tx_bytes
+= sstats
->tx_bytes
;
3479 stats
->tx_errors
+= sstats
->tx_errors
;
3480 stats
->tx_dropped
+= sstats
->tx_dropped
;
3482 stats
->multicast
+= sstats
->multicast
;
3483 stats
->collisions
+= sstats
->collisions
;
3485 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3486 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3487 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3488 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3489 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3490 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3492 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3493 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3494 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3495 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3496 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3499 read_unlock_bh(&bond
->lock
);
3504 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3506 struct net_device
*slave_dev
= NULL
;
3507 struct ifbond k_binfo
;
3508 struct ifbond __user
*u_binfo
= NULL
;
3509 struct ifslave k_sinfo
;
3510 struct ifslave __user
*u_sinfo
= NULL
;
3511 struct mii_ioctl_data
*mii
= NULL
;
3515 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3527 * We do this again just in case we were called by SIOCGMIIREG
3528 * instead of SIOCGMIIPHY.
3535 if (mii
->reg_num
== 1) {
3536 struct bonding
*bond
= netdev_priv(bond_dev
);
3538 read_lock(&bond
->lock
);
3539 read_lock(&bond
->curr_slave_lock
);
3540 if (netif_carrier_ok(bond
->dev
))
3541 mii
->val_out
= BMSR_LSTATUS
;
3543 read_unlock(&bond
->curr_slave_lock
);
3544 read_unlock(&bond
->lock
);
3548 case BOND_INFO_QUERY_OLD
:
3549 case SIOCBONDINFOQUERY
:
3550 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3552 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3555 res
= bond_info_query(bond_dev
, &k_binfo
);
3557 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3561 case BOND_SLAVE_INFO_QUERY_OLD
:
3562 case SIOCBONDSLAVEINFOQUERY
:
3563 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3565 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3568 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3570 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3579 net
= dev_net(bond_dev
);
3581 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3584 slave_dev
= dev_get_by_name(net
, ifr
->ifr_slave
);
3586 pr_debug("slave_dev=%p:\n", slave_dev
);
3591 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3593 case BOND_ENSLAVE_OLD
:
3594 case SIOCBONDENSLAVE
:
3595 res
= bond_enslave(bond_dev
, slave_dev
);
3597 case BOND_RELEASE_OLD
:
3598 case SIOCBONDRELEASE
:
3599 res
= bond_release(bond_dev
, slave_dev
);
3601 case BOND_SETHWADDR_OLD
:
3602 case SIOCBONDSETHWADDR
:
3603 bond_set_dev_addr(bond_dev
, slave_dev
);
3606 case BOND_CHANGE_ACTIVE_OLD
:
3607 case SIOCBONDCHANGEACTIVE
:
3608 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3620 static bool bond_addr_in_mc_list(unsigned char *addr
,
3621 struct netdev_hw_addr_list
*list
,
3624 struct netdev_hw_addr
*ha
;
3626 netdev_hw_addr_list_for_each(ha
, list
)
3627 if (!memcmp(ha
->addr
, addr
, addrlen
))
3633 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3635 struct bonding
*bond
= netdev_priv(bond_dev
);
3637 if (change
& IFF_PROMISC
)
3638 bond_set_promiscuity(bond
,
3639 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3641 if (change
& IFF_ALLMULTI
)
3642 bond_set_allmulti(bond
,
3643 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3646 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3648 struct bonding
*bond
= netdev_priv(bond_dev
);
3649 struct netdev_hw_addr
*ha
;
3652 read_lock(&bond
->lock
);
3654 /* looking for addresses to add to slaves' mc list */
3655 netdev_for_each_mc_addr(ha
, bond_dev
) {
3656 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
3657 bond_dev
->addr_len
);
3659 bond_mc_add(bond
, ha
->addr
);
3662 /* looking for addresses to delete from slaves' list */
3663 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
3664 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
3665 bond_dev
->addr_len
);
3667 bond_mc_del(bond
, ha
->addr
);
3670 /* save master's multicast list */
3671 __hw_addr_flush(&bond
->mc_list
);
3672 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
3673 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
3675 read_unlock(&bond
->lock
);
3678 static int bond_neigh_init(struct neighbour
*n
)
3680 struct bonding
*bond
= netdev_priv(n
->dev
);
3681 struct slave
*slave
= bond
->first_slave
;
3682 const struct net_device_ops
*slave_ops
;
3683 struct neigh_parms parms
;
3689 slave_ops
= slave
->dev
->netdev_ops
;
3691 if (!slave_ops
->ndo_neigh_setup
)
3694 parms
.neigh_setup
= NULL
;
3695 parms
.neigh_cleanup
= NULL
;
3696 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3701 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3702 * after the last slave has been detached. Assumes that all slaves
3703 * utilize the same neigh_cleanup (true at this writing as only user
3706 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3708 if (!parms
.neigh_setup
)
3711 return parms
.neigh_setup(n
);
3715 * The bonding ndo_neigh_setup is called at init time beofre any
3716 * slave exists. So we must declare proxy setup function which will
3717 * be used at run time to resolve the actual slave neigh param setup.
3719 static int bond_neigh_setup(struct net_device
*dev
,
3720 struct neigh_parms
*parms
)
3722 parms
->neigh_setup
= bond_neigh_init
;
3728 * Change the MTU of all of a master's slaves to match the master
3730 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3732 struct bonding
*bond
= netdev_priv(bond_dev
);
3733 struct slave
*slave
, *stop_at
;
3737 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3738 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3740 /* Can't hold bond->lock with bh disabled here since
3741 * some base drivers panic. On the other hand we can't
3742 * hold bond->lock without bh disabled because we'll
3743 * deadlock. The only solution is to rely on the fact
3744 * that we're under rtnl_lock here, and the slaves
3745 * list won't change. This doesn't solve the problem
3746 * of setting the slave's MTU while it is
3747 * transmitting, but the assumption is that the base
3748 * driver can handle that.
3750 * TODO: figure out a way to safely iterate the slaves
3751 * list, but without holding a lock around the actual
3752 * call to the base driver.
3755 bond_for_each_slave(bond
, slave
, i
) {
3756 pr_debug("s %p s->p %p c_m %p\n",
3759 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3761 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3764 /* If we failed to set the slave's mtu to the new value
3765 * we must abort the operation even in ACTIVE_BACKUP
3766 * mode, because if we allow the backup slaves to have
3767 * different mtu values than the active slave we'll
3768 * need to change their mtu when doing a failover. That
3769 * means changing their mtu from timer context, which
3770 * is probably not a good idea.
3772 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3777 bond_dev
->mtu
= new_mtu
;
3782 /* unwind from head to the slave that failed */
3784 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3787 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3789 pr_debug("unwind err %d dev %s\n",
3790 tmp_res
, slave
->dev
->name
);
3800 * Note that many devices must be down to change the HW address, and
3801 * downing the master releases all slaves. We can make bonds full of
3802 * bonding devices to test this, however.
3804 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3806 struct bonding
*bond
= netdev_priv(bond_dev
);
3807 struct sockaddr
*sa
= addr
, tmp_sa
;
3808 struct slave
*slave
, *stop_at
;
3812 if (bond
->params
.mode
== BOND_MODE_ALB
)
3813 return bond_alb_set_mac_address(bond_dev
, addr
);
3816 pr_debug("bond=%p, name=%s\n",
3817 bond
, bond_dev
? bond_dev
->name
: "None");
3820 * If fail_over_mac is set to active, do nothing and return
3821 * success. Returning an error causes ifenslave to fail.
3823 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
3826 if (!is_valid_ether_addr(sa
->sa_data
))
3827 return -EADDRNOTAVAIL
;
3829 /* Can't hold bond->lock with bh disabled here since
3830 * some base drivers panic. On the other hand we can't
3831 * hold bond->lock without bh disabled because we'll
3832 * deadlock. The only solution is to rely on the fact
3833 * that we're under rtnl_lock here, and the slaves
3834 * list won't change. This doesn't solve the problem
3835 * of setting the slave's hw address while it is
3836 * transmitting, but the assumption is that the base
3837 * driver can handle that.
3839 * TODO: figure out a way to safely iterate the slaves
3840 * list, but without holding a lock around the actual
3841 * call to the base driver.
3844 bond_for_each_slave(bond
, slave
, i
) {
3845 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3846 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3848 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3850 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3854 res
= dev_set_mac_address(slave
->dev
, addr
);
3856 /* TODO: consider downing the slave
3858 * User should expect communications
3859 * breakage anyway until ARP finish
3862 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3868 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3872 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3873 tmp_sa
.sa_family
= bond_dev
->type
;
3875 /* unwind from head to the slave that failed */
3877 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3880 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3882 pr_debug("unwind err %d dev %s\n",
3883 tmp_res
, slave
->dev
->name
);
3890 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3892 struct bonding
*bond
= netdev_priv(bond_dev
);
3893 struct slave
*slave
, *start_at
;
3894 int i
, slave_no
, res
= 1;
3895 struct iphdr
*iph
= ip_hdr(skb
);
3898 * Start with the curr_active_slave that joined the bond as the
3899 * default for sending IGMP traffic. For failover purposes one
3900 * needs to maintain some consistency for the interface that will
3901 * send the join/membership reports. The curr_active_slave found
3902 * will send all of this type of traffic.
3904 if ((iph
->protocol
== IPPROTO_IGMP
) &&
3905 (skb
->protocol
== htons(ETH_P_IP
))) {
3907 read_lock(&bond
->curr_slave_lock
);
3908 slave
= bond
->curr_active_slave
;
3909 read_unlock(&bond
->curr_slave_lock
);
3915 * Concurrent TX may collide on rr_tx_counter; we accept
3916 * that as being rare enough not to justify using an
3919 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
3921 bond_for_each_slave(bond
, slave
, i
) {
3929 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3930 if (IS_UP(slave
->dev
) &&
3931 (slave
->link
== BOND_LINK_UP
) &&
3932 bond_is_active_slave(slave
)) {
3933 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3940 /* no suitable interface, frame not sent */
3944 return NETDEV_TX_OK
;
3949 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3950 * the bond has a usable interface.
3952 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3954 struct bonding
*bond
= netdev_priv(bond_dev
);
3957 read_lock(&bond
->curr_slave_lock
);
3959 if (bond
->curr_active_slave
)
3960 res
= bond_dev_queue_xmit(bond
, skb
,
3961 bond
->curr_active_slave
->dev
);
3963 read_unlock(&bond
->curr_slave_lock
);
3966 /* no suitable interface, frame not sent */
3969 return NETDEV_TX_OK
;
3973 * In bond_xmit_xor() , we determine the output device by using a pre-
3974 * determined xmit_hash_policy(), If the selected device is not enabled,
3975 * find the next active slave.
3977 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3979 struct bonding
*bond
= netdev_priv(bond_dev
);
3980 struct slave
*slave
, *start_at
;
3985 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
3987 bond_for_each_slave(bond
, slave
, i
) {
3995 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3996 if (IS_UP(slave
->dev
) &&
3997 (slave
->link
== BOND_LINK_UP
) &&
3998 bond_is_active_slave(slave
)) {
3999 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4005 /* no suitable interface, frame not sent */
4009 return NETDEV_TX_OK
;
4013 * in broadcast mode, we send everything to all usable interfaces.
4015 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4017 struct bonding
*bond
= netdev_priv(bond_dev
);
4018 struct slave
*slave
, *start_at
;
4019 struct net_device
*tx_dev
= NULL
;
4023 read_lock(&bond
->curr_slave_lock
);
4024 start_at
= bond
->curr_active_slave
;
4025 read_unlock(&bond
->curr_slave_lock
);
4030 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4031 if (IS_UP(slave
->dev
) &&
4032 (slave
->link
== BOND_LINK_UP
) &&
4033 bond_is_active_slave(slave
)) {
4035 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4037 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4042 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4048 tx_dev
= slave
->dev
;
4053 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4057 /* no suitable interface, frame not sent */
4060 /* frame sent to all suitable interfaces */
4061 return NETDEV_TX_OK
;
4064 /*------------------------- Device initialization ---------------------------*/
4066 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4068 switch (bond
->params
.xmit_policy
) {
4069 case BOND_XMIT_POLICY_LAYER23
:
4070 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4072 case BOND_XMIT_POLICY_LAYER34
:
4073 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4075 case BOND_XMIT_POLICY_LAYER2
:
4077 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4083 * Lookup the slave that corresponds to a qid
4085 static inline int bond_slave_override(struct bonding
*bond
,
4086 struct sk_buff
*skb
)
4089 struct slave
*slave
= NULL
;
4090 struct slave
*check_slave
;
4092 if (!skb
->queue_mapping
)
4095 /* Find out if any slaves have the same mapping as this skb. */
4096 bond_for_each_slave(bond
, check_slave
, i
) {
4097 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4098 slave
= check_slave
;
4103 /* If the slave isn't UP, use default transmit policy. */
4104 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4105 (slave
->link
== BOND_LINK_UP
)) {
4106 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4113 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4116 * This helper function exists to help dev_pick_tx get the correct
4117 * destination queue. Using a helper function skips a call to
4118 * skb_tx_hash and will put the skbs in the queue we expect on their
4119 * way down to the bonding driver.
4121 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
4124 * Save the original txq to restore before passing to the driver
4126 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
4128 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
4130 txq
-= dev
->real_num_tx_queues
;
4131 } while (txq
>= dev
->real_num_tx_queues
);
4136 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4138 struct bonding
*bond
= netdev_priv(dev
);
4140 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4141 if (!bond_slave_override(bond
, skb
))
4142 return NETDEV_TX_OK
;
4145 switch (bond
->params
.mode
) {
4146 case BOND_MODE_ROUNDROBIN
:
4147 return bond_xmit_roundrobin(skb
, dev
);
4148 case BOND_MODE_ACTIVEBACKUP
:
4149 return bond_xmit_activebackup(skb
, dev
);
4151 return bond_xmit_xor(skb
, dev
);
4152 case BOND_MODE_BROADCAST
:
4153 return bond_xmit_broadcast(skb
, dev
);
4154 case BOND_MODE_8023AD
:
4155 return bond_3ad_xmit_xor(skb
, dev
);
4158 return bond_alb_xmit(skb
, dev
);
4160 /* Should never happen, mode already checked */
4161 pr_err("%s: Error: Unknown bonding mode %d\n",
4162 dev
->name
, bond
->params
.mode
);
4165 return NETDEV_TX_OK
;
4169 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4171 struct bonding
*bond
= netdev_priv(dev
);
4172 netdev_tx_t ret
= NETDEV_TX_OK
;
4175 * If we risk deadlock from transmitting this in the
4176 * netpoll path, tell netpoll to queue the frame for later tx
4178 if (is_netpoll_tx_blocked(dev
))
4179 return NETDEV_TX_BUSY
;
4181 read_lock(&bond
->lock
);
4183 if (bond
->slave_cnt
)
4184 ret
= __bond_start_xmit(skb
, dev
);
4188 read_unlock(&bond
->lock
);
4194 * set bond mode specific net device operations
4196 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4198 struct net_device
*bond_dev
= bond
->dev
;
4201 case BOND_MODE_ROUNDROBIN
:
4203 case BOND_MODE_ACTIVEBACKUP
:
4206 bond_set_xmit_hash_policy(bond
);
4208 case BOND_MODE_BROADCAST
:
4210 case BOND_MODE_8023AD
:
4211 bond_set_xmit_hash_policy(bond
);
4218 /* Should never happen, mode already checked */
4219 pr_err("%s: Error: Unknown bonding mode %d\n",
4220 bond_dev
->name
, mode
);
4225 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4226 struct ethtool_drvinfo
*drvinfo
)
4228 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
4229 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
4230 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
4234 static const struct ethtool_ops bond_ethtool_ops
= {
4235 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4236 .get_link
= ethtool_op_get_link
,
4239 static const struct net_device_ops bond_netdev_ops
= {
4240 .ndo_init
= bond_init
,
4241 .ndo_uninit
= bond_uninit
,
4242 .ndo_open
= bond_open
,
4243 .ndo_stop
= bond_close
,
4244 .ndo_start_xmit
= bond_start_xmit
,
4245 .ndo_select_queue
= bond_select_queue
,
4246 .ndo_get_stats64
= bond_get_stats
,
4247 .ndo_do_ioctl
= bond_do_ioctl
,
4248 .ndo_change_rx_flags
= bond_change_rx_flags
,
4249 .ndo_set_rx_mode
= bond_set_multicast_list
,
4250 .ndo_change_mtu
= bond_change_mtu
,
4251 .ndo_set_mac_address
= bond_set_mac_address
,
4252 .ndo_neigh_setup
= bond_neigh_setup
,
4253 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4254 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4255 #ifdef CONFIG_NET_POLL_CONTROLLER
4256 .ndo_netpoll_setup
= bond_netpoll_setup
,
4257 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4258 .ndo_poll_controller
= bond_poll_controller
,
4260 .ndo_add_slave
= bond_enslave
,
4261 .ndo_del_slave
= bond_release
,
4262 .ndo_fix_features
= bond_fix_features
,
4265 static const struct device_type bond_type
= {
4269 static void bond_destructor(struct net_device
*bond_dev
)
4271 struct bonding
*bond
= netdev_priv(bond_dev
);
4273 destroy_workqueue(bond
->wq
);
4274 free_netdev(bond_dev
);
4277 static void bond_setup(struct net_device
*bond_dev
)
4279 struct bonding
*bond
= netdev_priv(bond_dev
);
4281 /* initialize rwlocks */
4282 rwlock_init(&bond
->lock
);
4283 rwlock_init(&bond
->curr_slave_lock
);
4285 bond
->params
= bonding_defaults
;
4287 /* Initialize pointers */
4288 bond
->dev
= bond_dev
;
4289 INIT_LIST_HEAD(&bond
->vlan_list
);
4291 /* Initialize the device entry points */
4292 ether_setup(bond_dev
);
4293 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4294 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4295 bond_set_mode_ops(bond
, bond
->params
.mode
);
4297 bond_dev
->destructor
= bond_destructor
;
4299 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
4301 /* Initialize the device options */
4302 bond_dev
->tx_queue_len
= 0;
4303 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4304 bond_dev
->priv_flags
|= IFF_BONDING
;
4305 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
4307 /* At first, we block adding VLANs. That's the only way to
4308 * prevent problems that occur when adding VLANs over an
4309 * empty bond. The block will be removed once non-challenged
4310 * slaves are enslaved.
4312 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4314 /* don't acquire bond device's netif_tx_lock when
4316 bond_dev
->features
|= NETIF_F_LLTX
;
4318 /* By default, we declare the bond to be fully
4319 * VLAN hardware accelerated capable. Special
4320 * care is taken in the various xmit functions
4321 * when there are slaves that are not hw accel
4325 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4326 NETIF_F_HW_VLAN_TX
|
4327 NETIF_F_HW_VLAN_RX
|
4328 NETIF_F_HW_VLAN_FILTER
;
4330 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
4331 bond_dev
->features
|= bond_dev
->hw_features
;
4335 * Destroy a bonding device.
4336 * Must be under rtnl_lock when this function is called.
4338 static void bond_uninit(struct net_device
*bond_dev
)
4340 struct bonding
*bond
= netdev_priv(bond_dev
);
4341 struct vlan_entry
*vlan
, *tmp
;
4343 bond_netpoll_cleanup(bond_dev
);
4345 /* Release the bonded slaves */
4346 while (bond
->first_slave
!= NULL
)
4347 __bond_release_one(bond_dev
, bond
->first_slave
->dev
, true);
4348 pr_info("%s: released all slaves\n", bond_dev
->name
);
4350 list_del(&bond
->bond_list
);
4352 bond_debug_unregister(bond
);
4354 __hw_addr_flush(&bond
->mc_list
);
4356 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4357 list_del(&vlan
->vlan_list
);
4362 /*------------------------- Module initialization ---------------------------*/
4365 * Convert string input module parms. Accept either the
4366 * number of the mode or its string name. A bit complicated because
4367 * some mode names are substrings of other names, and calls from sysfs
4368 * may have whitespace in the name (trailing newlines, for example).
4370 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4372 int modeint
= -1, i
, rv
;
4373 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4375 for (p
= (char *)buf
; *p
; p
++)
4376 if (!(isdigit(*p
) || isspace(*p
)))
4380 rv
= sscanf(buf
, "%20s", modestr
);
4382 rv
= sscanf(buf
, "%d", &modeint
);
4387 for (i
= 0; tbl
[i
].modename
; i
++) {
4388 if (modeint
== tbl
[i
].mode
)
4390 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4397 static int bond_check_params(struct bond_params
*params
)
4399 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4402 * Convert string parameters.
4405 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4406 if (bond_mode
== -1) {
4407 pr_err("Error: Invalid bonding mode \"%s\"\n",
4408 mode
== NULL
? "NULL" : mode
);
4413 if (xmit_hash_policy
) {
4414 if ((bond_mode
!= BOND_MODE_XOR
) &&
4415 (bond_mode
!= BOND_MODE_8023AD
)) {
4416 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4417 bond_mode_name(bond_mode
));
4419 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4421 if (xmit_hashtype
== -1) {
4422 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4423 xmit_hash_policy
== NULL
? "NULL" :
4431 if (bond_mode
!= BOND_MODE_8023AD
) {
4432 pr_info("lacp_rate param is irrelevant in mode %s\n",
4433 bond_mode_name(bond_mode
));
4435 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4436 if (lacp_fast
== -1) {
4437 pr_err("Error: Invalid lacp rate \"%s\"\n",
4438 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4445 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4446 if (params
->ad_select
== -1) {
4447 pr_err("Error: Invalid ad_select \"%s\"\n",
4448 ad_select
== NULL
? "NULL" : ad_select
);
4452 if (bond_mode
!= BOND_MODE_8023AD
) {
4453 pr_warning("ad_select param only affects 802.3ad mode\n");
4456 params
->ad_select
= BOND_AD_STABLE
;
4459 if (max_bonds
< 0) {
4460 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4461 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4462 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4466 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4467 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4468 miimon
= BOND_LINK_MON_INTERV
;
4472 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4477 if (downdelay
< 0) {
4478 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4479 downdelay
, INT_MAX
);
4483 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4484 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4489 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4490 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4495 /* reset values for 802.3ad */
4496 if (bond_mode
== BOND_MODE_8023AD
) {
4498 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");
4499 pr_warning("Forcing miimon to 100msec\n");
4504 if (tx_queues
< 1 || tx_queues
> 255) {
4505 pr_warning("Warning: tx_queues (%d) should be between "
4506 "1 and 255, resetting to %d\n",
4507 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4508 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4511 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4512 pr_warning("Warning: all_slaves_active module parameter (%d), "
4513 "not of valid value (0/1), so it was set to "
4514 "0\n", all_slaves_active
);
4515 all_slaves_active
= 0;
4518 if (resend_igmp
< 0 || resend_igmp
> 255) {
4519 pr_warning("Warning: resend_igmp (%d) should be between "
4520 "0 and 255, resetting to %d\n",
4521 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4522 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4525 /* reset values for TLB/ALB */
4526 if ((bond_mode
== BOND_MODE_TLB
) ||
4527 (bond_mode
== BOND_MODE_ALB
)) {
4529 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");
4530 pr_warning("Forcing miimon to 100msec\n");
4535 if (bond_mode
== BOND_MODE_ALB
) {
4536 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",
4541 if (updelay
|| downdelay
) {
4542 /* just warn the user the up/down delay will have
4543 * no effect since miimon is zero...
4545 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",
4546 updelay
, downdelay
);
4549 /* don't allow arp monitoring */
4551 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4552 miimon
, arp_interval
);
4556 if ((updelay
% miimon
) != 0) {
4557 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4559 (updelay
/ miimon
) * miimon
);
4564 if ((downdelay
% miimon
) != 0) {
4565 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4567 (downdelay
/ miimon
) * miimon
);
4570 downdelay
/= miimon
;
4573 if (arp_interval
< 0) {
4574 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4575 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4576 arp_interval
= BOND_LINK_ARP_INTERV
;
4579 for (arp_ip_count
= 0;
4580 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4582 /* not complete check, but should be good enough to
4584 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4585 if (!isdigit(arp_ip_target
[arp_ip_count
][0]) ||
4586 ip
== 0 || ip
== htonl(INADDR_BROADCAST
)) {
4587 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4588 arp_ip_target
[arp_ip_count
]);
4591 arp_target
[arp_ip_count
] = ip
;
4595 if (arp_interval
&& !arp_ip_count
) {
4596 /* don't allow arping if no arp_ip_target given... */
4597 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4603 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4604 pr_err("arp_validate only supported in active-backup mode\n");
4607 if (!arp_interval
) {
4608 pr_err("arp_validate requires arp_interval\n");
4612 arp_validate_value
= bond_parse_parm(arp_validate
,
4614 if (arp_validate_value
== -1) {
4615 pr_err("Error: invalid arp_validate \"%s\"\n",
4616 arp_validate
== NULL
? "NULL" : arp_validate
);
4620 arp_validate_value
= 0;
4623 pr_info("MII link monitoring set to %d ms\n", miimon
);
4624 } else if (arp_interval
) {
4627 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4629 arp_validate_tbl
[arp_validate_value
].modename
,
4632 for (i
= 0; i
< arp_ip_count
; i
++)
4633 pr_info(" %s", arp_ip_target
[i
]);
4637 } else if (max_bonds
) {
4638 /* miimon and arp_interval not set, we need one so things
4639 * work as expected, see bonding.txt for details
4641 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");
4644 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4645 /* currently, using a primary only makes sense
4646 * in active backup, TLB or ALB modes
4648 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4649 primary
, bond_mode_name(bond_mode
));
4653 if (primary
&& primary_reselect
) {
4654 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4656 if (primary_reselect_value
== -1) {
4657 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4659 NULL
? "NULL" : primary_reselect
);
4663 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4666 if (fail_over_mac
) {
4667 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4669 if (fail_over_mac_value
== -1) {
4670 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4671 arp_validate
== NULL
? "NULL" : arp_validate
);
4675 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4676 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4678 fail_over_mac_value
= BOND_FOM_NONE
;
4681 /* fill params struct with the proper values */
4682 params
->mode
= bond_mode
;
4683 params
->xmit_policy
= xmit_hashtype
;
4684 params
->miimon
= miimon
;
4685 params
->num_peer_notif
= num_peer_notif
;
4686 params
->arp_interval
= arp_interval
;
4687 params
->arp_validate
= arp_validate_value
;
4688 params
->updelay
= updelay
;
4689 params
->downdelay
= downdelay
;
4690 params
->use_carrier
= use_carrier
;
4691 params
->lacp_fast
= lacp_fast
;
4692 params
->primary
[0] = 0;
4693 params
->primary_reselect
= primary_reselect_value
;
4694 params
->fail_over_mac
= fail_over_mac_value
;
4695 params
->tx_queues
= tx_queues
;
4696 params
->all_slaves_active
= all_slaves_active
;
4697 params
->resend_igmp
= resend_igmp
;
4698 params
->min_links
= min_links
;
4701 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4702 params
->primary
[IFNAMSIZ
- 1] = 0;
4705 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4710 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4711 static struct lock_class_key bonding_netdev_addr_lock_key
;
4712 static struct lock_class_key bonding_tx_busylock_key
;
4714 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4715 struct netdev_queue
*txq
,
4718 lockdep_set_class(&txq
->_xmit_lock
,
4719 &bonding_netdev_xmit_lock_key
);
4722 static void bond_set_lockdep_class(struct net_device
*dev
)
4724 lockdep_set_class(&dev
->addr_list_lock
,
4725 &bonding_netdev_addr_lock_key
);
4726 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4727 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4731 * Called from registration process
4733 static int bond_init(struct net_device
*bond_dev
)
4735 struct bonding
*bond
= netdev_priv(bond_dev
);
4736 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4737 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4739 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4742 * Initialize locks that may be required during
4743 * en/deslave operations. All of the bond_open work
4744 * (of which this is part) should really be moved to
4745 * a phase prior to dev_open
4747 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4748 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4750 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4754 bond_set_lockdep_class(bond_dev
);
4756 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4758 bond_prepare_sysfs_group(bond
);
4760 bond_debug_register(bond
);
4762 /* Ensure valid dev_addr */
4763 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4764 bond_dev
->addr_assign_type
== NET_ADDR_PERM
) {
4765 eth_hw_addr_random(bond_dev
);
4766 bond
->dev_addr_from_first
= true;
4769 __hw_addr_init(&bond
->mc_list
);
4773 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4775 if (tb
[IFLA_ADDRESS
]) {
4776 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4778 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4779 return -EADDRNOTAVAIL
;
4784 static unsigned int bond_get_num_tx_queues(void)
4789 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
4791 .priv_size
= sizeof(struct bonding
),
4792 .setup
= bond_setup
,
4793 .validate
= bond_validate
,
4794 .get_num_tx_queues
= bond_get_num_tx_queues
,
4795 .get_num_rx_queues
= bond_get_num_tx_queues
, /* Use the same number
4799 /* Create a new bond based on the specified name and bonding parameters.
4800 * If name is NULL, obtain a suitable "bond%d" name for us.
4801 * Caller must NOT hold rtnl_lock; we need to release it here before we
4802 * set up our sysfs entries.
4804 int bond_create(struct net
*net
, const char *name
)
4806 struct net_device
*bond_dev
;
4811 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4812 name
? name
: "bond%d",
4813 bond_setup
, tx_queues
);
4815 pr_err("%s: eek! can't alloc netdev!\n", name
);
4820 dev_net_set(bond_dev
, net
);
4821 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4823 res
= register_netdevice(bond_dev
);
4825 netif_carrier_off(bond_dev
);
4829 bond_destructor(bond_dev
);
4833 static int __net_init
bond_net_init(struct net
*net
)
4835 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4838 INIT_LIST_HEAD(&bn
->dev_list
);
4840 bond_create_proc_dir(bn
);
4841 bond_create_sysfs(bn
);
4846 static void __net_exit
bond_net_exit(struct net
*net
)
4848 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4850 bond_destroy_sysfs(bn
);
4851 bond_destroy_proc_dir(bn
);
4854 static struct pernet_operations bond_net_ops
= {
4855 .init
= bond_net_init
,
4856 .exit
= bond_net_exit
,
4858 .size
= sizeof(struct bond_net
),
4861 static int __init
bonding_init(void)
4866 pr_info("%s", bond_version
);
4868 res
= bond_check_params(&bonding_defaults
);
4872 res
= register_pernet_subsys(&bond_net_ops
);
4876 res
= rtnl_link_register(&bond_link_ops
);
4880 bond_create_debugfs();
4882 for (i
= 0; i
< max_bonds
; i
++) {
4883 res
= bond_create(&init_net
, NULL
);
4888 register_netdevice_notifier(&bond_netdev_notifier
);
4892 rtnl_link_unregister(&bond_link_ops
);
4894 unregister_pernet_subsys(&bond_net_ops
);
4899 static void __exit
bonding_exit(void)
4901 unregister_netdevice_notifier(&bond_netdev_notifier
);
4903 bond_destroy_debugfs();
4905 rtnl_link_unregister(&bond_link_ops
);
4906 unregister_pernet_subsys(&bond_net_ops
);
4908 #ifdef CONFIG_NET_POLL_CONTROLLER
4910 * Make sure we don't have an imbalance on our netpoll blocking
4912 WARN_ON(atomic_read(&netpoll_block_tx
));
4916 module_init(bonding_init
);
4917 module_exit(bonding_exit
);
4918 MODULE_LICENSE("GPL");
4919 MODULE_VERSION(DRV_VERSION
);
4920 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4921 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4922 MODULE_ALIAS_RTNL_LINK("bond");