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 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
54 #include <asm/system.h>
57 #include <asm/uaccess.h>
58 #include <linux/errno.h>
59 #include <linux/netdevice.h>
60 #include <linux/inetdevice.h>
61 #include <linux/igmp.h>
62 #include <linux/etherdevice.h>
63 #include <linux/skbuff.h>
65 #include <linux/rtnetlink.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
82 /*---------------------------- Module parameters ----------------------------*/
84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
85 #define BOND_LINK_MON_INTERV 0
86 #define BOND_LINK_ARP_INTERV 0
88 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
89 static int num_grat_arp
= 1;
90 static int num_unsol_na
= 1;
91 static int miimon
= BOND_LINK_MON_INTERV
;
92 static int updelay
= 0;
93 static int downdelay
= 0;
94 static int use_carrier
= 1;
95 static char *mode
= NULL
;
96 static char *primary
= NULL
;
97 static char *lacp_rate
= NULL
;
98 static char *ad_select
= NULL
;
99 static char *xmit_hash_policy
= NULL
;
100 static int arp_interval
= BOND_LINK_ARP_INTERV
;
101 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
] = { NULL
, };
102 static char *arp_validate
= NULL
;
103 static char *fail_over_mac
= NULL
;
104 struct bond_params bonding_defaults
;
106 module_param(max_bonds
, int, 0);
107 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
108 module_param(num_grat_arp
, int, 0644);
109 MODULE_PARM_DESC(num_grat_arp
, "Number of gratuitous ARP packets to send on failover event");
110 module_param(num_unsol_na
, int, 0644);
111 MODULE_PARM_DESC(num_unsol_na
, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
112 module_param(miimon
, int, 0);
113 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
114 module_param(updelay
, int, 0);
115 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
116 module_param(downdelay
, int, 0);
117 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
119 module_param(use_carrier
, int, 0);
120 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
121 "0 for off, 1 for on (default)");
122 module_param(mode
, charp
, 0);
123 MODULE_PARM_DESC(mode
, "Mode of operation : 0 for balance-rr, "
124 "1 for active-backup, 2 for balance-xor, "
125 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
126 "6 for balance-alb");
127 module_param(primary
, charp
, 0);
128 MODULE_PARM_DESC(primary
, "Primary network device to use");
129 module_param(lacp_rate
, charp
, 0);
130 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner "
132 module_param(ad_select
, charp
, 0);
133 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
134 module_param(xmit_hash_policy
, charp
, 0);
135 MODULE_PARM_DESC(xmit_hash_policy
, "XOR hashing method: 0 for layer 2 (default)"
136 ", 1 for layer 3+4");
137 module_param(arp_interval
, int, 0);
138 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
139 module_param_array(arp_ip_target
, charp
, NULL
, 0);
140 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
141 module_param(arp_validate
, charp
, 0);
142 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes: none (default), active, backup or all");
143 module_param(fail_over_mac
, charp
, 0);
144 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
146 /*----------------------------- Global variables ----------------------------*/
148 static const char * const version
=
149 DRV_DESCRIPTION
": v" DRV_VERSION
" (" DRV_RELDATE
")\n";
151 LIST_HEAD(bond_dev_list
);
153 #ifdef CONFIG_PROC_FS
154 static struct proc_dir_entry
*bond_proc_dir
= NULL
;
157 extern struct rw_semaphore bonding_rwsem
;
158 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
] = { 0, } ;
159 static int arp_ip_count
= 0;
160 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
161 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
162 static int lacp_fast
= 0;
165 const struct bond_parm_tbl bond_lacp_tbl
[] = {
166 { "slow", AD_LACP_SLOW
},
167 { "fast", AD_LACP_FAST
},
171 const struct bond_parm_tbl bond_mode_tbl
[] = {
172 { "balance-rr", BOND_MODE_ROUNDROBIN
},
173 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
174 { "balance-xor", BOND_MODE_XOR
},
175 { "broadcast", BOND_MODE_BROADCAST
},
176 { "802.3ad", BOND_MODE_8023AD
},
177 { "balance-tlb", BOND_MODE_TLB
},
178 { "balance-alb", BOND_MODE_ALB
},
182 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
183 { "layer2", BOND_XMIT_POLICY_LAYER2
},
184 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
185 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
189 const struct bond_parm_tbl arp_validate_tbl
[] = {
190 { "none", BOND_ARP_VALIDATE_NONE
},
191 { "active", BOND_ARP_VALIDATE_ACTIVE
},
192 { "backup", BOND_ARP_VALIDATE_BACKUP
},
193 { "all", BOND_ARP_VALIDATE_ALL
},
197 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
198 { "none", BOND_FOM_NONE
},
199 { "active", BOND_FOM_ACTIVE
},
200 { "follow", BOND_FOM_FOLLOW
},
204 struct bond_parm_tbl ad_select_tbl
[] = {
205 { "stable", BOND_AD_STABLE
},
206 { "bandwidth", BOND_AD_BANDWIDTH
},
207 { "count", BOND_AD_COUNT
},
211 /*-------------------------- Forward declarations ---------------------------*/
213 static void bond_send_gratuitous_arp(struct bonding
*bond
);
214 static void bond_deinit(struct net_device
*bond_dev
);
216 /*---------------------------- General routines -----------------------------*/
218 static const char *bond_mode_name(int mode
)
220 static const char *names
[] = {
221 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
222 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
223 [BOND_MODE_XOR
] = "load balancing (xor)",
224 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
225 [BOND_MODE_8023AD
]= "IEEE 802.3ad Dynamic link aggregation",
226 [BOND_MODE_TLB
] = "transmit load balancing",
227 [BOND_MODE_ALB
] = "adaptive load balancing",
230 if (mode
< 0 || mode
> BOND_MODE_ALB
)
236 /*---------------------------------- VLAN -----------------------------------*/
239 * bond_add_vlan - add a new vlan id on bond
240 * @bond: bond that got the notification
241 * @vlan_id: the vlan id to add
243 * Returns -ENOMEM if allocation failed.
245 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
247 struct vlan_entry
*vlan
;
249 pr_debug("bond: %s, vlan id %d\n",
250 (bond
? bond
->dev
->name
: "None"), vlan_id
);
252 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
257 INIT_LIST_HEAD(&vlan
->vlan_list
);
258 vlan
->vlan_id
= vlan_id
;
260 write_lock_bh(&bond
->lock
);
262 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
264 write_unlock_bh(&bond
->lock
);
266 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
272 * bond_del_vlan - delete a vlan id from bond
273 * @bond: bond that got the notification
274 * @vlan_id: the vlan id to delete
276 * returns -ENODEV if @vlan_id was not found in @bond.
278 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
280 struct vlan_entry
*vlan
;
283 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
285 write_lock_bh(&bond
->lock
);
287 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
288 if (vlan
->vlan_id
== vlan_id
) {
289 list_del(&vlan
->vlan_list
);
291 if (bond_is_lb(bond
))
292 bond_alb_clear_vlan(bond
, vlan_id
);
294 pr_debug("removed VLAN ID %d from bond %s\n", vlan_id
,
299 if (list_empty(&bond
->vlan_list
) &&
300 (bond
->slave_cnt
== 0)) {
301 /* Last VLAN removed and no slaves, so
302 * restore block on adding VLANs. This will
303 * be removed once new slaves that are not
304 * VLAN challenged will be added.
306 bond
->dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
314 pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id
,
318 write_unlock_bh(&bond
->lock
);
323 * bond_has_challenged_slaves
324 * @bond: the bond we're working on
326 * Searches the slave list. Returns 1 if a vlan challenged slave
327 * was found, 0 otherwise.
329 * Assumes bond->lock is held.
331 static int bond_has_challenged_slaves(struct bonding
*bond
)
336 bond_for_each_slave(bond
, slave
, i
) {
337 if (slave
->dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
338 pr_debug("found VLAN challenged slave - %s\n",
344 pr_debug("no VLAN challenged slaves found\n");
349 * bond_next_vlan - safely skip to the next item in the vlans list.
350 * @bond: the bond we're working on
351 * @curr: item we're advancing from
353 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
354 * or @curr->next otherwise (even if it is @curr itself again).
356 * Caller must hold bond->lock
358 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
360 struct vlan_entry
*next
, *last
;
362 if (list_empty(&bond
->vlan_list
)) {
367 next
= list_entry(bond
->vlan_list
.next
,
368 struct vlan_entry
, vlan_list
);
370 last
= list_entry(bond
->vlan_list
.prev
,
371 struct vlan_entry
, vlan_list
);
373 next
= list_entry(bond
->vlan_list
.next
,
374 struct vlan_entry
, vlan_list
);
376 next
= list_entry(curr
->vlan_list
.next
,
377 struct vlan_entry
, vlan_list
);
385 * bond_dev_queue_xmit - Prepare skb for xmit.
387 * @bond: bond device that got this skb for tx.
388 * @skb: hw accel VLAN tagged skb to transmit
389 * @slave_dev: slave that is supposed to xmit this skbuff
391 * When the bond gets an skb to transmit that is
392 * already hardware accelerated VLAN tagged, and it
393 * needs to relay this skb to a slave that is not
394 * hw accel capable, the skb needs to be "unaccelerated",
395 * i.e. strip the hwaccel tag and re-insert it as part
398 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
)
400 unsigned short uninitialized_var(vlan_id
);
402 if (!list_empty(&bond
->vlan_list
) &&
403 !(slave_dev
->features
& NETIF_F_HW_VLAN_TX
) &&
404 vlan_get_tag(skb
, &vlan_id
) == 0) {
405 skb
->dev
= slave_dev
;
406 skb
= vlan_put_tag(skb
, vlan_id
);
408 /* vlan_put_tag() frees the skb in case of error,
409 * so return success here so the calling functions
410 * won't attempt to free is again.
415 skb
->dev
= slave_dev
;
425 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
426 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
428 * a. This operation is performed in IOCTL context,
429 * b. The operation is protected by the RTNL semaphore in the 8021q code,
430 * c. Holding a lock with BH disabled while directly calling a base driver
431 * entry point is generally a BAD idea.
433 * The design of synchronization/protection for this operation in the 8021q
434 * module is good for one or more VLAN devices over a single physical device
435 * and cannot be extended for a teaming solution like bonding, so there is a
436 * potential race condition here where a net device from the vlan group might
437 * be referenced (either by a base driver or the 8021q code) while it is being
438 * removed from the system. However, it turns out we're not making matters
439 * worse, and if it works for regular VLAN usage it will work here too.
443 * bond_vlan_rx_register - Propagates registration to slaves
444 * @bond_dev: bonding net device that got called
445 * @grp: vlan group being registered
447 static void bond_vlan_rx_register(struct net_device
*bond_dev
, struct vlan_group
*grp
)
449 struct bonding
*bond
= netdev_priv(bond_dev
);
455 bond_for_each_slave(bond
, slave
, i
) {
456 struct net_device
*slave_dev
= slave
->dev
;
457 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
459 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
460 slave_ops
->ndo_vlan_rx_register
) {
461 slave_ops
->ndo_vlan_rx_register(slave_dev
, grp
);
467 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
468 * @bond_dev: bonding net device that got called
469 * @vid: vlan id being added
471 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
473 struct bonding
*bond
= netdev_priv(bond_dev
);
477 bond_for_each_slave(bond
, slave
, i
) {
478 struct net_device
*slave_dev
= slave
->dev
;
479 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
481 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
482 slave_ops
->ndo_vlan_rx_add_vid
) {
483 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vid
);
487 res
= bond_add_vlan(bond
, vid
);
489 printk(KERN_ERR DRV_NAME
490 ": %s: Error: Failed to add vlan id %d\n",
491 bond_dev
->name
, vid
);
496 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
497 * @bond_dev: bonding net device that got called
498 * @vid: vlan id being removed
500 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
502 struct bonding
*bond
= netdev_priv(bond_dev
);
504 struct net_device
*vlan_dev
;
507 bond_for_each_slave(bond
, slave
, i
) {
508 struct net_device
*slave_dev
= slave
->dev
;
509 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
511 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
512 slave_ops
->ndo_vlan_rx_kill_vid
) {
513 /* Save and then restore vlan_dev in the grp array,
514 * since the slave's driver might clear it.
516 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vid
);
517 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vid
);
518 vlan_group_set_device(bond
->vlgrp
, vid
, vlan_dev
);
522 res
= bond_del_vlan(bond
, vid
);
524 printk(KERN_ERR DRV_NAME
525 ": %s: Error: Failed to remove vlan id %d\n",
526 bond_dev
->name
, vid
);
530 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
532 struct vlan_entry
*vlan
;
533 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
535 write_lock_bh(&bond
->lock
);
537 if (list_empty(&bond
->vlan_list
))
540 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
541 slave_ops
->ndo_vlan_rx_register
)
542 slave_ops
->ndo_vlan_rx_register(slave_dev
, bond
->vlgrp
);
544 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
545 !(slave_ops
->ndo_vlan_rx_add_vid
))
548 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
)
549 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
552 write_unlock_bh(&bond
->lock
);
555 static void bond_del_vlans_from_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
557 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
558 struct vlan_entry
*vlan
;
559 struct net_device
*vlan_dev
;
561 write_lock_bh(&bond
->lock
);
563 if (list_empty(&bond
->vlan_list
))
566 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
567 !(slave_ops
->ndo_vlan_rx_kill_vid
))
570 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
571 /* Save and then restore vlan_dev in the grp array,
572 * since the slave's driver might clear it.
574 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
575 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
576 vlan_group_set_device(bond
->vlgrp
, vlan
->vlan_id
, vlan_dev
);
580 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
581 slave_ops
->ndo_vlan_rx_register
)
582 slave_ops
->ndo_vlan_rx_register(slave_dev
, NULL
);
585 write_unlock_bh(&bond
->lock
);
588 /*------------------------------- Link status -------------------------------*/
591 * Set the carrier state for the master according to the state of its
592 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
593 * do special 802.3ad magic.
595 * Returns zero if carrier state does not change, nonzero if it does.
597 static int bond_set_carrier(struct bonding
*bond
)
602 if (bond
->slave_cnt
== 0)
605 if (bond
->params
.mode
== BOND_MODE_8023AD
)
606 return bond_3ad_set_carrier(bond
);
608 bond_for_each_slave(bond
, slave
, i
) {
609 if (slave
->link
== BOND_LINK_UP
) {
610 if (!netif_carrier_ok(bond
->dev
)) {
611 netif_carrier_on(bond
->dev
);
619 if (netif_carrier_ok(bond
->dev
)) {
620 netif_carrier_off(bond
->dev
);
627 * Get link speed and duplex from the slave's base driver
628 * using ethtool. If for some reason the call fails or the
629 * values are invalid, fake speed and duplex to 100/Full
632 static int bond_update_speed_duplex(struct slave
*slave
)
634 struct net_device
*slave_dev
= slave
->dev
;
635 struct ethtool_cmd etool
;
638 /* Fake speed and duplex */
639 slave
->speed
= SPEED_100
;
640 slave
->duplex
= DUPLEX_FULL
;
642 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
645 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
649 switch (etool
.speed
) {
659 switch (etool
.duplex
) {
667 slave
->speed
= etool
.speed
;
668 slave
->duplex
= etool
.duplex
;
674 * if <dev> supports MII link status reporting, check its link status.
676 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
677 * depening upon the setting of the use_carrier parameter.
679 * Return either BMSR_LSTATUS, meaning that the link is up (or we
680 * can't tell and just pretend it is), or 0, meaning that the link is
683 * If reporting is non-zero, instead of faking link up, return -1 if
684 * both ETHTOOL and MII ioctls fail (meaning the device does not
685 * support them). If use_carrier is set, return whatever it says.
686 * It'd be nice if there was a good way to tell if a driver supports
687 * netif_carrier, but there really isn't.
689 static int bond_check_dev_link(struct bonding
*bond
, struct net_device
*slave_dev
, int reporting
)
691 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
692 static int (* ioctl
)(struct net_device
*, struct ifreq
*, int);
694 struct mii_ioctl_data
*mii
;
696 if (bond
->params
.use_carrier
)
697 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
699 ioctl
= slave_ops
->ndo_do_ioctl
;
701 /* TODO: set pointer to correct ioctl on a per team member */
702 /* bases to make this more efficient. that is, once */
703 /* we determine the correct ioctl, we will always */
704 /* call it and not the others for that team */
708 * We cannot assume that SIOCGMIIPHY will also read a
709 * register; not all network drivers (e.g., e100)
713 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
714 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
716 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
717 mii
->reg_num
= MII_BMSR
;
718 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0) {
719 return (mii
->val_out
& BMSR_LSTATUS
);
725 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
726 * attempt to get link status from it if the above MII ioctls fail.
728 if (slave_dev
->ethtool_ops
) {
729 if (slave_dev
->ethtool_ops
->get_link
) {
732 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
734 return link
? BMSR_LSTATUS
: 0;
739 * If reporting, report that either there's no dev->do_ioctl,
740 * or both SIOCGMIIREG and get_link failed (meaning that we
741 * cannot report link status). If not reporting, pretend
744 return (reporting
? -1 : BMSR_LSTATUS
);
747 /*----------------------------- Multicast list ------------------------------*/
750 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
752 static inline int bond_is_dmi_same(struct dev_mc_list
*dmi1
, struct dev_mc_list
*dmi2
)
754 return memcmp(dmi1
->dmi_addr
, dmi2
->dmi_addr
, dmi1
->dmi_addrlen
) == 0 &&
755 dmi1
->dmi_addrlen
== dmi2
->dmi_addrlen
;
759 * returns dmi entry if found, NULL otherwise
761 static struct dev_mc_list
*bond_mc_list_find_dmi(struct dev_mc_list
*dmi
, struct dev_mc_list
*mc_list
)
763 struct dev_mc_list
*idmi
;
765 for (idmi
= mc_list
; idmi
; idmi
= idmi
->next
) {
766 if (bond_is_dmi_same(dmi
, idmi
)) {
775 * Push the promiscuity flag down to appropriate slaves
777 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
780 if (USES_PRIMARY(bond
->params
.mode
)) {
781 /* write lock already acquired */
782 if (bond
->curr_active_slave
) {
783 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
789 bond_for_each_slave(bond
, slave
, i
) {
790 err
= dev_set_promiscuity(slave
->dev
, inc
);
799 * Push the allmulti flag down to all slaves
801 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
804 if (USES_PRIMARY(bond
->params
.mode
)) {
805 /* write lock already acquired */
806 if (bond
->curr_active_slave
) {
807 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
813 bond_for_each_slave(bond
, slave
, i
) {
814 err
= dev_set_allmulti(slave
->dev
, inc
);
823 * Add a Multicast address to slaves
826 static void bond_mc_add(struct bonding
*bond
, void *addr
, int alen
)
828 if (USES_PRIMARY(bond
->params
.mode
)) {
829 /* write lock already acquired */
830 if (bond
->curr_active_slave
) {
831 dev_mc_add(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
836 bond_for_each_slave(bond
, slave
, i
) {
837 dev_mc_add(slave
->dev
, addr
, alen
, 0);
843 * Remove a multicast address from slave
846 static void bond_mc_delete(struct bonding
*bond
, void *addr
, int alen
)
848 if (USES_PRIMARY(bond
->params
.mode
)) {
849 /* write lock already acquired */
850 if (bond
->curr_active_slave
) {
851 dev_mc_delete(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
856 bond_for_each_slave(bond
, slave
, i
) {
857 dev_mc_delete(slave
->dev
, addr
, alen
, 0);
864 * Retrieve the list of registered multicast addresses for the bonding
865 * device and retransmit an IGMP JOIN request to the current active
868 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
870 struct in_device
*in_dev
;
871 struct ip_mc_list
*im
;
874 in_dev
= __in_dev_get_rcu(bond
->dev
);
876 for (im
= in_dev
->mc_list
; im
; im
= im
->next
) {
877 ip_mc_rejoin_group(im
);
885 * Totally destroys the mc_list in bond
887 static void bond_mc_list_destroy(struct bonding
*bond
)
889 struct dev_mc_list
*dmi
;
893 bond
->mc_list
= dmi
->next
;
897 bond
->mc_list
= NULL
;
901 * Copy all the Multicast addresses from src to the bonding device dst
903 static int bond_mc_list_copy(struct dev_mc_list
*mc_list
, struct bonding
*bond
,
906 struct dev_mc_list
*dmi
, *new_dmi
;
908 for (dmi
= mc_list
; dmi
; dmi
= dmi
->next
) {
909 new_dmi
= kmalloc(sizeof(struct dev_mc_list
), gfp_flag
);
912 /* FIXME: Potential memory leak !!! */
916 new_dmi
->next
= bond
->mc_list
;
917 bond
->mc_list
= new_dmi
;
918 new_dmi
->dmi_addrlen
= dmi
->dmi_addrlen
;
919 memcpy(new_dmi
->dmi_addr
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
920 new_dmi
->dmi_users
= dmi
->dmi_users
;
921 new_dmi
->dmi_gusers
= dmi
->dmi_gusers
;
928 * flush all members of flush->mc_list from device dev->mc_list
930 static void bond_mc_list_flush(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
932 struct bonding
*bond
= netdev_priv(bond_dev
);
933 struct dev_mc_list
*dmi
;
935 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
936 dev_mc_delete(slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
939 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
940 /* del lacpdu mc addr from mc list */
941 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
943 dev_mc_delete(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
947 /*--------------------------- Active slave change ---------------------------*/
950 * Update the mc list and multicast-related flags for the new and
951 * old active slaves (if any) according to the multicast mode, and
952 * promiscuous flags unconditionally.
954 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
, struct slave
*old_active
)
956 struct dev_mc_list
*dmi
;
958 if (!USES_PRIMARY(bond
->params
.mode
)) {
959 /* nothing to do - mc list is already up-to-date on
966 if (bond
->dev
->flags
& IFF_PROMISC
) {
967 dev_set_promiscuity(old_active
->dev
, -1);
970 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
971 dev_set_allmulti(old_active
->dev
, -1);
974 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
975 dev_mc_delete(old_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
980 /* FIXME: Signal errors upstream. */
981 if (bond
->dev
->flags
& IFF_PROMISC
) {
982 dev_set_promiscuity(new_active
->dev
, 1);
985 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
986 dev_set_allmulti(new_active
->dev
, 1);
989 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
990 dev_mc_add(new_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
992 bond_resend_igmp_join_requests(bond
);
997 * bond_do_fail_over_mac
999 * Perform special MAC address swapping for fail_over_mac settings
1001 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1003 static void bond_do_fail_over_mac(struct bonding
*bond
,
1004 struct slave
*new_active
,
1005 struct slave
*old_active
)
1007 u8 tmp_mac
[ETH_ALEN
];
1008 struct sockaddr saddr
;
1011 switch (bond
->params
.fail_over_mac
) {
1012 case BOND_FOM_ACTIVE
:
1014 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
1015 new_active
->dev
->addr_len
);
1017 case BOND_FOM_FOLLOW
:
1019 * if new_active && old_active, swap them
1020 * if just old_active, do nothing (going to no active slave)
1021 * if just new_active, set new_active to bond's MAC
1026 write_unlock_bh(&bond
->curr_slave_lock
);
1027 read_unlock(&bond
->lock
);
1030 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
1031 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
1033 saddr
.sa_family
= new_active
->dev
->type
;
1035 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
1036 saddr
.sa_family
= bond
->dev
->type
;
1039 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
1041 printk(KERN_ERR DRV_NAME
1042 ": %s: Error %d setting MAC of slave %s\n",
1043 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1050 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1051 saddr
.sa_family
= old_active
->dev
->type
;
1053 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
1055 printk(KERN_ERR DRV_NAME
1056 ": %s: Error %d setting MAC of slave %s\n",
1057 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1059 read_lock(&bond
->lock
);
1060 write_lock_bh(&bond
->curr_slave_lock
);
1063 printk(KERN_ERR DRV_NAME
1064 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1065 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1073 * find_best_interface - select the best available slave to be the active one
1074 * @bond: our bonding struct
1076 * Warning: Caller must hold curr_slave_lock for writing.
1078 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1080 struct slave
*new_active
, *old_active
;
1081 struct slave
*bestslave
= NULL
;
1082 int mintime
= bond
->params
.updelay
;
1085 new_active
= old_active
= bond
->curr_active_slave
;
1087 if (!new_active
) { /* there were no active slaves left */
1088 if (bond
->slave_cnt
> 0) { /* found one slave */
1089 new_active
= bond
->first_slave
;
1091 return NULL
; /* still no slave, return NULL */
1095 /* first try the primary link; if arping, a link must tx/rx traffic
1096 * before it can be considered the curr_active_slave - also, we would skip
1097 * slaves between the curr_active_slave and primary_slave that may be up
1100 if ((bond
->primary_slave
) &&
1101 (!bond
->params
.arp_interval
) &&
1102 (IS_UP(bond
->primary_slave
->dev
))) {
1103 new_active
= bond
->primary_slave
;
1106 /* remember where to stop iterating over the slaves */
1107 old_active
= new_active
;
1109 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1110 if (IS_UP(new_active
->dev
)) {
1111 if (new_active
->link
== BOND_LINK_UP
) {
1113 } else if (new_active
->link
== BOND_LINK_BACK
) {
1114 /* link up, but waiting for stabilization */
1115 if (new_active
->delay
< mintime
) {
1116 mintime
= new_active
->delay
;
1117 bestslave
= new_active
;
1127 * change_active_interface - change the active slave into the specified one
1128 * @bond: our bonding struct
1129 * @new: the new slave to make the active one
1131 * Set the new slave to the bond's settings and unset them on the old
1132 * curr_active_slave.
1133 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1135 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1136 * because it is apparently the best available slave we have, even though its
1137 * updelay hasn't timed out yet.
1139 * If new_active is not NULL, caller must hold bond->lock for read and
1140 * curr_slave_lock for write_bh.
1142 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1144 struct slave
*old_active
= bond
->curr_active_slave
;
1146 if (old_active
== new_active
) {
1151 new_active
->jiffies
= jiffies
;
1153 if (new_active
->link
== BOND_LINK_BACK
) {
1154 if (USES_PRIMARY(bond
->params
.mode
)) {
1155 printk(KERN_INFO DRV_NAME
1156 ": %s: making interface %s the new "
1157 "active one %d ms earlier.\n",
1158 bond
->dev
->name
, new_active
->dev
->name
,
1159 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1162 new_active
->delay
= 0;
1163 new_active
->link
= BOND_LINK_UP
;
1165 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1166 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1169 if (bond_is_lb(bond
))
1170 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1172 if (USES_PRIMARY(bond
->params
.mode
)) {
1173 printk(KERN_INFO DRV_NAME
1174 ": %s: making interface %s the new "
1176 bond
->dev
->name
, new_active
->dev
->name
);
1181 if (USES_PRIMARY(bond
->params
.mode
)) {
1182 bond_mc_swap(bond
, new_active
, old_active
);
1185 if (bond_is_lb(bond
)) {
1186 bond_alb_handle_active_change(bond
, new_active
);
1188 bond_set_slave_inactive_flags(old_active
);
1190 bond_set_slave_active_flags(new_active
);
1192 bond
->curr_active_slave
= new_active
;
1195 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1197 bond_set_slave_inactive_flags(old_active
);
1201 bond_set_slave_active_flags(new_active
);
1203 if (bond
->params
.fail_over_mac
)
1204 bond_do_fail_over_mac(bond
, new_active
,
1207 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1208 bond_send_gratuitous_arp(bond
);
1210 bond
->send_unsol_na
= bond
->params
.num_unsol_na
;
1211 bond_send_unsolicited_na(bond
);
1213 write_unlock_bh(&bond
->curr_slave_lock
);
1214 read_unlock(&bond
->lock
);
1216 netdev_bonding_change(bond
->dev
);
1218 read_lock(&bond
->lock
);
1219 write_lock_bh(&bond
->curr_slave_lock
);
1225 * bond_select_active_slave - select a new active slave, if needed
1226 * @bond: our bonding struct
1228 * This functions shoud be called when one of the following occurs:
1229 * - The old curr_active_slave has been released or lost its link.
1230 * - The primary_slave has got its link back.
1231 * - A slave has got its link back and there's no old curr_active_slave.
1233 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1235 void bond_select_active_slave(struct bonding
*bond
)
1237 struct slave
*best_slave
;
1240 best_slave
= bond_find_best_slave(bond
);
1241 if (best_slave
!= bond
->curr_active_slave
) {
1242 bond_change_active_slave(bond
, best_slave
);
1243 rv
= bond_set_carrier(bond
);
1247 if (netif_carrier_ok(bond
->dev
)) {
1248 printk(KERN_INFO DRV_NAME
1249 ": %s: first active interface up!\n",
1252 printk(KERN_INFO DRV_NAME
": %s: "
1253 "now running without any active interface !\n",
1259 /*--------------------------- slave list handling ---------------------------*/
1262 * This function attaches the slave to the end of list.
1264 * bond->lock held for writing by caller.
1266 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1268 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1269 new_slave
->next
= new_slave
;
1270 new_slave
->prev
= new_slave
;
1271 bond
->first_slave
= new_slave
;
1273 new_slave
->next
= bond
->first_slave
;
1274 new_slave
->prev
= bond
->first_slave
->prev
;
1275 new_slave
->next
->prev
= new_slave
;
1276 new_slave
->prev
->next
= new_slave
;
1283 * This function detaches the slave from the list.
1284 * WARNING: no check is made to verify if the slave effectively
1285 * belongs to <bond>.
1286 * Nothing is freed on return, structures are just unchained.
1287 * If any slave pointer in bond was pointing to <slave>,
1288 * it should be changed by the calling function.
1290 * bond->lock held for writing by caller.
1292 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1295 slave
->next
->prev
= slave
->prev
;
1299 slave
->prev
->next
= slave
->next
;
1302 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1303 if (bond
->slave_cnt
> 1) { /* there are more slave */
1304 bond
->first_slave
= slave
->next
;
1306 bond
->first_slave
= NULL
; /* slave was the last one */
1315 /*---------------------------------- IOCTL ----------------------------------*/
1317 static int bond_sethwaddr(struct net_device
*bond_dev
,
1318 struct net_device
*slave_dev
)
1320 pr_debug("bond_dev=%p\n", bond_dev
);
1321 pr_debug("slave_dev=%p\n", slave_dev
);
1322 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1323 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1327 #define BOND_VLAN_FEATURES \
1328 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1329 NETIF_F_HW_VLAN_FILTER)
1332 * Compute the common dev->feature set available to all slaves. Some
1333 * feature bits are managed elsewhere, so preserve those feature bits
1334 * on the master device.
1336 static int bond_compute_features(struct bonding
*bond
)
1338 struct slave
*slave
;
1339 struct net_device
*bond_dev
= bond
->dev
;
1340 unsigned long features
= bond_dev
->features
;
1341 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1342 bond_dev
->hard_header_len
);
1345 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1346 features
|= NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1348 if (!bond
->first_slave
)
1351 features
&= ~NETIF_F_ONE_FOR_ALL
;
1353 bond_for_each_slave(bond
, slave
, i
) {
1354 features
= netdev_increment_features(features
,
1355 slave
->dev
->features
,
1356 NETIF_F_ONE_FOR_ALL
);
1357 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1358 max_hard_header_len
= slave
->dev
->hard_header_len
;
1362 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1363 bond_dev
->features
= netdev_fix_features(features
, NULL
);
1364 bond_dev
->hard_header_len
= max_hard_header_len
;
1369 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1370 struct net_device
*slave_dev
)
1372 struct bonding
*bond
= netdev_priv(bond_dev
);
1374 bond_dev
->header_ops
= slave_dev
->header_ops
;
1376 bond_dev
->type
= slave_dev
->type
;
1377 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1378 bond_dev
->addr_len
= slave_dev
->addr_len
;
1380 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1381 slave_dev
->addr_len
);
1382 bond
->setup_by_slave
= 1;
1385 /* enslave device <slave> to bond device <master> */
1386 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1388 struct bonding
*bond
= netdev_priv(bond_dev
);
1389 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1390 struct slave
*new_slave
= NULL
;
1391 struct dev_mc_list
*dmi
;
1392 struct sockaddr addr
;
1394 int old_features
= bond_dev
->features
;
1397 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1398 slave_ops
->ndo_do_ioctl
== NULL
) {
1399 printk(KERN_WARNING DRV_NAME
1400 ": %s: Warning: no link monitoring support for %s\n",
1401 bond_dev
->name
, slave_dev
->name
);
1404 /* bond must be initialized by bond_open() before enslaving */
1405 if (!(bond_dev
->flags
& IFF_UP
)) {
1406 printk(KERN_WARNING DRV_NAME
1407 " %s: master_dev is not up in bond_enslave\n",
1411 /* already enslaved */
1412 if (slave_dev
->flags
& IFF_SLAVE
) {
1413 pr_debug("Error, Device was already enslaved\n");
1417 /* vlan challenged mutual exclusion */
1418 /* no need to lock since we're protected by rtnl_lock */
1419 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1420 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1421 if (!list_empty(&bond
->vlan_list
)) {
1422 printk(KERN_ERR DRV_NAME
1423 ": %s: Error: cannot enslave VLAN "
1424 "challenged slave %s on VLAN enabled "
1425 "bond %s\n", bond_dev
->name
, slave_dev
->name
,
1429 printk(KERN_WARNING DRV_NAME
1430 ": %s: Warning: enslaved VLAN challenged "
1431 "slave %s. Adding VLANs will be blocked as "
1432 "long as %s is part of bond %s\n",
1433 bond_dev
->name
, slave_dev
->name
, slave_dev
->name
,
1435 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1438 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1439 if (bond
->slave_cnt
== 0) {
1440 /* First slave, and it is not VLAN challenged,
1441 * so remove the block of adding VLANs over the bond.
1443 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1448 * Old ifenslave binaries are no longer supported. These can
1449 * be identified with moderate accurary by the state of the slave:
1450 * the current ifenslave will set the interface down prior to
1451 * enslaving it; the old ifenslave will not.
1453 if ((slave_dev
->flags
& IFF_UP
)) {
1454 printk(KERN_ERR DRV_NAME
": %s is up. "
1455 "This may be due to an out of date ifenslave.\n",
1458 goto err_undo_flags
;
1461 /* set bonding device ether type by slave - bonding netdevices are
1462 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1463 * there is a need to override some of the type dependent attribs/funcs.
1465 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1466 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1468 if (bond
->slave_cnt
== 0) {
1469 if (slave_dev
->type
!= ARPHRD_ETHER
)
1470 bond_setup_by_slave(bond_dev
, slave_dev
);
1471 } else if (bond_dev
->type
!= slave_dev
->type
) {
1472 printk(KERN_ERR DRV_NAME
": %s ether type (%d) is different "
1473 "from other slaves (%d), can not enslave it.\n",
1475 slave_dev
->type
, bond_dev
->type
);
1477 goto err_undo_flags
;
1480 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1481 if (bond
->slave_cnt
== 0) {
1482 printk(KERN_WARNING DRV_NAME
1483 ": %s: Warning: The first slave device "
1484 "specified does not support setting the MAC "
1485 "address. Setting fail_over_mac to active.",
1487 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1488 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1489 printk(KERN_ERR DRV_NAME
1490 ": %s: Error: The slave device specified "
1491 "does not support setting the MAC address, "
1492 "but fail_over_mac is not set to active.\n"
1495 goto err_undo_flags
;
1499 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1502 goto err_undo_flags
;
1505 /* save slave's original flags before calling
1506 * netdev_set_master and dev_open
1508 new_slave
->original_flags
= slave_dev
->flags
;
1511 * Save slave's original ("permanent") mac address for modes
1512 * that need it, and for restoring it upon release, and then
1513 * set it to the master's address
1515 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1517 if (!bond
->params
.fail_over_mac
) {
1519 * Set slave to master's mac address. The application already
1520 * set the master's mac address to that of the first slave
1522 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1523 addr
.sa_family
= slave_dev
->type
;
1524 res
= dev_set_mac_address(slave_dev
, &addr
);
1526 pr_debug("Error %d calling set_mac_address\n", res
);
1531 res
= netdev_set_master(slave_dev
, bond_dev
);
1533 pr_debug("Error %d calling netdev_set_master\n", res
);
1534 goto err_restore_mac
;
1536 /* open the slave since the application closed it */
1537 res
= dev_open(slave_dev
);
1539 pr_debug("Openning slave %s failed\n", slave_dev
->name
);
1540 goto err_unset_master
;
1543 new_slave
->dev
= slave_dev
;
1544 slave_dev
->priv_flags
|= IFF_BONDING
;
1546 if (bond_is_lb(bond
)) {
1547 /* bond_alb_init_slave() must be called before all other stages since
1548 * it might fail and we do not want to have to undo everything
1550 res
= bond_alb_init_slave(bond
, new_slave
);
1556 /* If the mode USES_PRIMARY, then the new slave gets the
1557 * master's promisc (and mc) settings only if it becomes the
1558 * curr_active_slave, and that is taken care of later when calling
1559 * bond_change_active()
1561 if (!USES_PRIMARY(bond
->params
.mode
)) {
1562 /* set promiscuity level to new slave */
1563 if (bond_dev
->flags
& IFF_PROMISC
) {
1564 res
= dev_set_promiscuity(slave_dev
, 1);
1569 /* set allmulti level to new slave */
1570 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1571 res
= dev_set_allmulti(slave_dev
, 1);
1576 netif_addr_lock_bh(bond_dev
);
1577 /* upload master's mc_list to new slave */
1578 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
1579 dev_mc_add (slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
1581 netif_addr_unlock_bh(bond_dev
);
1584 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1585 /* add lacpdu mc addr to mc list */
1586 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1588 dev_mc_add(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
1591 bond_add_vlans_on_slave(bond
, slave_dev
);
1593 write_lock_bh(&bond
->lock
);
1595 bond_attach_slave(bond
, new_slave
);
1597 new_slave
->delay
= 0;
1598 new_slave
->link_failure_count
= 0;
1600 bond_compute_features(bond
);
1602 write_unlock_bh(&bond
->lock
);
1604 read_lock(&bond
->lock
);
1606 new_slave
->last_arp_rx
= jiffies
;
1608 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1609 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1611 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1613 * miimon is set but a bonded network driver
1614 * does not support ETHTOOL/MII and
1615 * arp_interval is not set. Note: if
1616 * use_carrier is enabled, we will never go
1617 * here (because netif_carrier is always
1618 * supported); thus, we don't need to change
1619 * the messages for netif_carrier.
1621 printk(KERN_WARNING DRV_NAME
1622 ": %s: Warning: MII and ETHTOOL support not "
1623 "available for interface %s, and "
1624 "arp_interval/arp_ip_target module parameters "
1625 "not specified, thus bonding will not detect "
1626 "link failures! see bonding.txt for details.\n",
1627 bond_dev
->name
, slave_dev
->name
);
1628 } else if (link_reporting
== -1) {
1629 /* unable get link status using mii/ethtool */
1630 printk(KERN_WARNING DRV_NAME
1631 ": %s: Warning: can't get link status from "
1632 "interface %s; the network driver associated "
1633 "with this interface does not support MII or "
1634 "ETHTOOL link status reporting, thus miimon "
1635 "has no effect on this interface.\n",
1636 bond_dev
->name
, slave_dev
->name
);
1640 /* check for initial state */
1641 if (!bond
->params
.miimon
||
1642 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1643 if (bond
->params
.updelay
) {
1644 pr_debug("Initial state of slave_dev is "
1645 "BOND_LINK_BACK\n");
1646 new_slave
->link
= BOND_LINK_BACK
;
1647 new_slave
->delay
= bond
->params
.updelay
;
1649 pr_debug("Initial state of slave_dev is "
1651 new_slave
->link
= BOND_LINK_UP
;
1653 new_slave
->jiffies
= jiffies
;
1655 pr_debug("Initial state of slave_dev is "
1656 "BOND_LINK_DOWN\n");
1657 new_slave
->link
= BOND_LINK_DOWN
;
1660 if (bond_update_speed_duplex(new_slave
) &&
1661 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1662 printk(KERN_WARNING DRV_NAME
1663 ": %s: Warning: failed to get speed and duplex from %s, "
1664 "assumed to be 100Mb/sec and Full.\n",
1665 bond_dev
->name
, new_slave
->dev
->name
);
1667 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1668 printk(KERN_WARNING DRV_NAME
1669 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1670 "support in base driver for proper aggregator "
1671 "selection.\n", bond_dev
->name
);
1675 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1676 /* if there is a primary slave, remember it */
1677 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1678 bond
->primary_slave
= new_slave
;
1682 write_lock_bh(&bond
->curr_slave_lock
);
1684 switch (bond
->params
.mode
) {
1685 case BOND_MODE_ACTIVEBACKUP
:
1686 bond_set_slave_inactive_flags(new_slave
);
1687 bond_select_active_slave(bond
);
1689 case BOND_MODE_8023AD
:
1690 /* in 802.3ad mode, the internal mechanism
1691 * will activate the slaves in the selected
1694 bond_set_slave_inactive_flags(new_slave
);
1695 /* if this is the first slave */
1696 if (bond
->slave_cnt
== 1) {
1697 SLAVE_AD_INFO(new_slave
).id
= 1;
1698 /* Initialize AD with the number of times that the AD timer is called in 1 second
1699 * can be called only after the mac address of the bond is set
1701 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1702 bond
->params
.lacp_fast
);
1704 SLAVE_AD_INFO(new_slave
).id
=
1705 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1708 bond_3ad_bind_slave(new_slave
);
1712 new_slave
->state
= BOND_STATE_ACTIVE
;
1713 bond_set_slave_inactive_flags(new_slave
);
1716 pr_debug("This slave is always active in trunk mode\n");
1718 /* always active in trunk mode */
1719 new_slave
->state
= BOND_STATE_ACTIVE
;
1721 /* In trunking mode there is little meaning to curr_active_slave
1722 * anyway (it holds no special properties of the bond device),
1723 * so we can change it without calling change_active_interface()
1725 if (!bond
->curr_active_slave
) {
1726 bond
->curr_active_slave
= new_slave
;
1729 } /* switch(bond_mode) */
1731 write_unlock_bh(&bond
->curr_slave_lock
);
1733 bond_set_carrier(bond
);
1735 read_unlock(&bond
->lock
);
1737 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1741 printk(KERN_INFO DRV_NAME
1742 ": %s: enslaving %s as a%s interface with a%s link.\n",
1743 bond_dev
->name
, slave_dev
->name
,
1744 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1745 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1747 /* enslave is successful */
1750 /* Undo stages on error */
1752 dev_close(slave_dev
);
1755 netdev_set_master(slave_dev
, NULL
);
1758 if (!bond
->params
.fail_over_mac
) {
1759 /* XXX TODO - fom follow mode needs to change master's
1760 * MAC if this slave's MAC is in use by the bond, or at
1761 * least print a warning.
1763 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1764 addr
.sa_family
= slave_dev
->type
;
1765 dev_set_mac_address(slave_dev
, &addr
);
1772 bond_dev
->features
= old_features
;
1778 * Try to release the slave device <slave> from the bond device <master>
1779 * It is legal to access curr_active_slave without a lock because all the function
1782 * The rules for slave state should be:
1783 * for Active/Backup:
1784 * Active stays on all backups go down
1785 * for Bonded connections:
1786 * The first up interface should be left on and all others downed.
1788 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1790 struct bonding
*bond
= netdev_priv(bond_dev
);
1791 struct slave
*slave
, *oldcurrent
;
1792 struct sockaddr addr
;
1793 int mac_addr_differ
;
1795 /* slave is not a slave or master is not master of this slave */
1796 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1797 (slave_dev
->master
!= bond_dev
)) {
1798 printk(KERN_ERR DRV_NAME
1799 ": %s: Error: cannot release %s.\n",
1800 bond_dev
->name
, slave_dev
->name
);
1804 write_lock_bh(&bond
->lock
);
1806 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1808 /* not a slave of this bond */
1809 printk(KERN_INFO DRV_NAME
1810 ": %s: %s not enslaved\n",
1811 bond_dev
->name
, slave_dev
->name
);
1812 write_unlock_bh(&bond
->lock
);
1816 if (!bond
->params
.fail_over_mac
) {
1817 mac_addr_differ
= memcmp(bond_dev
->dev_addr
, slave
->perm_hwaddr
,
1819 if (!mac_addr_differ
&& (bond
->slave_cnt
> 1))
1820 printk(KERN_WARNING DRV_NAME
1821 ": %s: Warning: the permanent HWaddr of %s - "
1822 "%pM - is still in use by %s. "
1823 "Set the HWaddr of %s to a different address "
1824 "to avoid conflicts.\n",
1825 bond_dev
->name
, slave_dev
->name
,
1827 bond_dev
->name
, slave_dev
->name
);
1830 /* Inform AD package of unbinding of slave. */
1831 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1832 /* must be called before the slave is
1833 * detached from the list
1835 bond_3ad_unbind_slave(slave
);
1838 printk(KERN_INFO DRV_NAME
1839 ": %s: releasing %s interface %s\n",
1841 (slave
->state
== BOND_STATE_ACTIVE
)
1842 ? "active" : "backup",
1845 oldcurrent
= bond
->curr_active_slave
;
1847 bond
->current_arp_slave
= NULL
;
1849 /* release the slave from its bond */
1850 bond_detach_slave(bond
, slave
);
1852 bond_compute_features(bond
);
1854 if (bond
->primary_slave
== slave
) {
1855 bond
->primary_slave
= NULL
;
1858 if (oldcurrent
== slave
) {
1859 bond_change_active_slave(bond
, NULL
);
1862 if (bond_is_lb(bond
)) {
1863 /* Must be called only after the slave has been
1864 * detached from the list and the curr_active_slave
1865 * has been cleared (if our_slave == old_current),
1866 * but before a new active slave is selected.
1868 write_unlock_bh(&bond
->lock
);
1869 bond_alb_deinit_slave(bond
, slave
);
1870 write_lock_bh(&bond
->lock
);
1873 if (oldcurrent
== slave
) {
1875 * Note that we hold RTNL over this sequence, so there
1876 * is no concern that another slave add/remove event
1879 write_unlock_bh(&bond
->lock
);
1880 read_lock(&bond
->lock
);
1881 write_lock_bh(&bond
->curr_slave_lock
);
1883 bond_select_active_slave(bond
);
1885 write_unlock_bh(&bond
->curr_slave_lock
);
1886 read_unlock(&bond
->lock
);
1887 write_lock_bh(&bond
->lock
);
1890 if (bond
->slave_cnt
== 0) {
1891 bond_set_carrier(bond
);
1893 /* if the last slave was removed, zero the mac address
1894 * of the master so it will be set by the application
1895 * to the mac address of the first slave
1897 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1899 if (list_empty(&bond
->vlan_list
)) {
1900 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1902 printk(KERN_WARNING DRV_NAME
1903 ": %s: Warning: clearing HW address of %s while it "
1904 "still has VLANs.\n",
1905 bond_dev
->name
, bond_dev
->name
);
1906 printk(KERN_WARNING DRV_NAME
1907 ": %s: When re-adding slaves, make sure the bond's "
1908 "HW address matches its VLANs'.\n",
1911 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1912 !bond_has_challenged_slaves(bond
)) {
1913 printk(KERN_INFO DRV_NAME
1914 ": %s: last VLAN challenged slave %s "
1915 "left bond %s. VLAN blocking is removed\n",
1916 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1917 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1920 write_unlock_bh(&bond
->lock
);
1922 /* must do this from outside any spinlocks */
1923 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1925 bond_del_vlans_from_slave(bond
, slave_dev
);
1927 /* If the mode USES_PRIMARY, then we should only remove its
1928 * promisc and mc settings if it was the curr_active_slave, but that was
1929 * already taken care of above when we detached the slave
1931 if (!USES_PRIMARY(bond
->params
.mode
)) {
1932 /* unset promiscuity level from slave */
1933 if (bond_dev
->flags
& IFF_PROMISC
) {
1934 dev_set_promiscuity(slave_dev
, -1);
1937 /* unset allmulti level from slave */
1938 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1939 dev_set_allmulti(slave_dev
, -1);
1942 /* flush master's mc_list from slave */
1943 netif_addr_lock_bh(bond_dev
);
1944 bond_mc_list_flush(bond_dev
, slave_dev
);
1945 netif_addr_unlock_bh(bond_dev
);
1948 netdev_set_master(slave_dev
, NULL
);
1950 /* close slave before restoring its mac address */
1951 dev_close(slave_dev
);
1953 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1954 /* restore original ("permanent") mac address */
1955 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1956 addr
.sa_family
= slave_dev
->type
;
1957 dev_set_mac_address(slave_dev
, &addr
);
1960 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
1961 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
1966 return 0; /* deletion OK */
1970 * Destroy a bonding device.
1971 * Must be under rtnl_lock when this function is called.
1973 void bond_destroy(struct bonding
*bond
)
1975 bond_deinit(bond
->dev
);
1976 bond_destroy_sysfs_entry(bond
);
1977 unregister_netdevice(bond
->dev
);
1980 static void bond_destructor(struct net_device
*bond_dev
)
1982 struct bonding
*bond
= netdev_priv(bond_dev
);
1985 destroy_workqueue(bond
->wq
);
1987 netif_addr_lock_bh(bond_dev
);
1988 bond_mc_list_destroy(bond
);
1989 netif_addr_unlock_bh(bond_dev
);
1991 free_netdev(bond_dev
);
1995 * First release a slave and than destroy the bond if no more slaves iare left.
1996 * Must be under rtnl_lock when this function is called.
1998 int bond_release_and_destroy(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2000 struct bonding
*bond
= netdev_priv(bond_dev
);
2003 ret
= bond_release(bond_dev
, slave_dev
);
2004 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2005 printk(KERN_INFO DRV_NAME
": %s: destroying bond %s.\n",
2006 bond_dev
->name
, bond_dev
->name
);
2013 * This function releases all slaves.
2015 static int bond_release_all(struct net_device
*bond_dev
)
2017 struct bonding
*bond
= netdev_priv(bond_dev
);
2018 struct slave
*slave
;
2019 struct net_device
*slave_dev
;
2020 struct sockaddr addr
;
2022 write_lock_bh(&bond
->lock
);
2024 netif_carrier_off(bond_dev
);
2026 if (bond
->slave_cnt
== 0) {
2030 bond
->current_arp_slave
= NULL
;
2031 bond
->primary_slave
= NULL
;
2032 bond_change_active_slave(bond
, NULL
);
2034 while ((slave
= bond
->first_slave
) != NULL
) {
2035 /* Inform AD package of unbinding of slave
2036 * before slave is detached from the list.
2038 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2039 bond_3ad_unbind_slave(slave
);
2042 slave_dev
= slave
->dev
;
2043 bond_detach_slave(bond
, slave
);
2045 /* now that the slave is detached, unlock and perform
2046 * all the undo steps that should not be called from
2049 write_unlock_bh(&bond
->lock
);
2051 if (bond_is_lb(bond
)) {
2052 /* must be called only after the slave
2053 * has been detached from the list
2055 bond_alb_deinit_slave(bond
, slave
);
2058 bond_compute_features(bond
);
2060 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2061 bond_del_vlans_from_slave(bond
, slave_dev
);
2063 /* If the mode USES_PRIMARY, then we should only remove its
2064 * promisc and mc settings if it was the curr_active_slave, but that was
2065 * already taken care of above when we detached the slave
2067 if (!USES_PRIMARY(bond
->params
.mode
)) {
2068 /* unset promiscuity level from slave */
2069 if (bond_dev
->flags
& IFF_PROMISC
) {
2070 dev_set_promiscuity(slave_dev
, -1);
2073 /* unset allmulti level from slave */
2074 if (bond_dev
->flags
& IFF_ALLMULTI
) {
2075 dev_set_allmulti(slave_dev
, -1);
2078 /* flush master's mc_list from slave */
2079 netif_addr_lock_bh(bond_dev
);
2080 bond_mc_list_flush(bond_dev
, slave_dev
);
2081 netif_addr_unlock_bh(bond_dev
);
2084 netdev_set_master(slave_dev
, NULL
);
2086 /* close slave before restoring its mac address */
2087 dev_close(slave_dev
);
2089 if (!bond
->params
.fail_over_mac
) {
2090 /* restore original ("permanent") mac address*/
2091 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2092 addr
.sa_family
= slave_dev
->type
;
2093 dev_set_mac_address(slave_dev
, &addr
);
2096 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2097 IFF_SLAVE_INACTIVE
);
2101 /* re-acquire the lock before getting the next slave */
2102 write_lock_bh(&bond
->lock
);
2105 /* zero the mac address of the master so it will be
2106 * set by the application to the mac address of the
2109 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2111 if (list_empty(&bond
->vlan_list
)) {
2112 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2114 printk(KERN_WARNING DRV_NAME
2115 ": %s: Warning: clearing HW address of %s while it "
2116 "still has VLANs.\n",
2117 bond_dev
->name
, bond_dev
->name
);
2118 printk(KERN_WARNING DRV_NAME
2119 ": %s: When re-adding slaves, make sure the bond's "
2120 "HW address matches its VLANs'.\n",
2124 printk(KERN_INFO DRV_NAME
2125 ": %s: released all slaves\n",
2129 write_unlock_bh(&bond
->lock
);
2135 * This function changes the active slave to slave <slave_dev>.
2136 * It returns -EINVAL in the following cases.
2137 * - <slave_dev> is not found in the list.
2138 * - There is not active slave now.
2139 * - <slave_dev> is already active.
2140 * - The link state of <slave_dev> is not BOND_LINK_UP.
2141 * - <slave_dev> is not running.
2142 * In these cases, this fuction does nothing.
2143 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2145 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2147 struct bonding
*bond
= netdev_priv(bond_dev
);
2148 struct slave
*old_active
= NULL
;
2149 struct slave
*new_active
= NULL
;
2152 if (!USES_PRIMARY(bond
->params
.mode
)) {
2156 /* Verify that master_dev is indeed the master of slave_dev */
2157 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
2158 (slave_dev
->master
!= bond_dev
)) {
2162 read_lock(&bond
->lock
);
2164 read_lock(&bond
->curr_slave_lock
);
2165 old_active
= bond
->curr_active_slave
;
2166 read_unlock(&bond
->curr_slave_lock
);
2168 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2171 * Changing to the current active: do nothing; return success.
2173 if (new_active
&& (new_active
== old_active
)) {
2174 read_unlock(&bond
->lock
);
2180 (new_active
->link
== BOND_LINK_UP
) &&
2181 IS_UP(new_active
->dev
)) {
2182 write_lock_bh(&bond
->curr_slave_lock
);
2183 bond_change_active_slave(bond
, new_active
);
2184 write_unlock_bh(&bond
->curr_slave_lock
);
2189 read_unlock(&bond
->lock
);
2194 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2196 struct bonding
*bond
= netdev_priv(bond_dev
);
2198 info
->bond_mode
= bond
->params
.mode
;
2199 info
->miimon
= bond
->params
.miimon
;
2201 read_lock(&bond
->lock
);
2202 info
->num_slaves
= bond
->slave_cnt
;
2203 read_unlock(&bond
->lock
);
2208 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2210 struct bonding
*bond
= netdev_priv(bond_dev
);
2211 struct slave
*slave
;
2214 if (info
->slave_id
< 0) {
2218 read_lock(&bond
->lock
);
2220 bond_for_each_slave(bond
, slave
, i
) {
2221 if (i
== (int)info
->slave_id
) {
2227 read_unlock(&bond
->lock
);
2230 strcpy(info
->slave_name
, slave
->dev
->name
);
2231 info
->link
= slave
->link
;
2232 info
->state
= slave
->state
;
2233 info
->link_failure_count
= slave
->link_failure_count
;
2241 /*-------------------------------- Monitoring -------------------------------*/
2244 static int bond_miimon_inspect(struct bonding
*bond
)
2246 struct slave
*slave
;
2247 int i
, link_state
, commit
= 0;
2249 bond_for_each_slave(bond
, slave
, i
) {
2250 slave
->new_link
= BOND_LINK_NOCHANGE
;
2252 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2254 switch (slave
->link
) {
2259 slave
->link
= BOND_LINK_FAIL
;
2260 slave
->delay
= bond
->params
.downdelay
;
2262 printk(KERN_INFO DRV_NAME
2263 ": %s: link status down for %s"
2264 "interface %s, disabling it in %d ms.\n",
2266 (bond
->params
.mode
==
2267 BOND_MODE_ACTIVEBACKUP
) ?
2268 ((slave
->state
== BOND_STATE_ACTIVE
) ?
2269 "active " : "backup ") : "",
2271 bond
->params
.downdelay
* bond
->params
.miimon
);
2274 case BOND_LINK_FAIL
:
2277 * recovered before downdelay expired
2279 slave
->link
= BOND_LINK_UP
;
2280 slave
->jiffies
= jiffies
;
2281 printk(KERN_INFO DRV_NAME
2282 ": %s: link status up again after %d "
2283 "ms for interface %s.\n",
2285 (bond
->params
.downdelay
- slave
->delay
) *
2286 bond
->params
.miimon
,
2291 if (slave
->delay
<= 0) {
2292 slave
->new_link
= BOND_LINK_DOWN
;
2300 case BOND_LINK_DOWN
:
2304 slave
->link
= BOND_LINK_BACK
;
2305 slave
->delay
= bond
->params
.updelay
;
2308 printk(KERN_INFO DRV_NAME
2309 ": %s: link status up for "
2310 "interface %s, enabling it in %d ms.\n",
2311 bond
->dev
->name
, slave
->dev
->name
,
2312 bond
->params
.updelay
*
2313 bond
->params
.miimon
);
2316 case BOND_LINK_BACK
:
2318 slave
->link
= BOND_LINK_DOWN
;
2319 printk(KERN_INFO DRV_NAME
2320 ": %s: link status down again after %d "
2321 "ms for interface %s.\n",
2323 (bond
->params
.updelay
- slave
->delay
) *
2324 bond
->params
.miimon
,
2330 if (slave
->delay
<= 0) {
2331 slave
->new_link
= BOND_LINK_UP
;
2344 static void bond_miimon_commit(struct bonding
*bond
)
2346 struct slave
*slave
;
2349 bond_for_each_slave(bond
, slave
, i
) {
2350 switch (slave
->new_link
) {
2351 case BOND_LINK_NOCHANGE
:
2355 slave
->link
= BOND_LINK_UP
;
2356 slave
->jiffies
= jiffies
;
2358 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2359 /* prevent it from being the active one */
2360 slave
->state
= BOND_STATE_BACKUP
;
2361 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2362 /* make it immediately active */
2363 slave
->state
= BOND_STATE_ACTIVE
;
2364 } else if (slave
!= bond
->primary_slave
) {
2365 /* prevent it from being the active one */
2366 slave
->state
= BOND_STATE_BACKUP
;
2369 printk(KERN_INFO DRV_NAME
2370 ": %s: link status definitely "
2371 "up for interface %s.\n",
2372 bond
->dev
->name
, slave
->dev
->name
);
2374 /* notify ad that the link status has changed */
2375 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2376 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2378 if (bond_is_lb(bond
))
2379 bond_alb_handle_link_change(bond
, slave
,
2382 if (!bond
->curr_active_slave
||
2383 (slave
== bond
->primary_slave
))
2388 case BOND_LINK_DOWN
:
2389 if (slave
->link_failure_count
< UINT_MAX
)
2390 slave
->link_failure_count
++;
2392 slave
->link
= BOND_LINK_DOWN
;
2394 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2395 bond
->params
.mode
== BOND_MODE_8023AD
)
2396 bond_set_slave_inactive_flags(slave
);
2398 printk(KERN_INFO DRV_NAME
2399 ": %s: link status definitely down for "
2400 "interface %s, disabling it\n",
2401 bond
->dev
->name
, slave
->dev
->name
);
2403 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2404 bond_3ad_handle_link_change(slave
,
2407 if (bond
->params
.mode
== BOND_MODE_TLB
||
2408 bond
->params
.mode
== BOND_MODE_ALB
)
2409 bond_alb_handle_link_change(bond
, slave
,
2412 if (slave
== bond
->curr_active_slave
)
2418 printk(KERN_ERR DRV_NAME
2419 ": %s: invalid new link %d on slave %s\n",
2420 bond
->dev
->name
, slave
->new_link
,
2422 slave
->new_link
= BOND_LINK_NOCHANGE
;
2429 write_lock_bh(&bond
->curr_slave_lock
);
2430 bond_select_active_slave(bond
);
2431 write_unlock_bh(&bond
->curr_slave_lock
);
2434 bond_set_carrier(bond
);
2440 * Really a wrapper that splits the mii monitor into two phases: an
2441 * inspection, then (if inspection indicates something needs to be done)
2442 * an acquisition of appropriate locks followed by a commit phase to
2443 * implement whatever link state changes are indicated.
2445 void bond_mii_monitor(struct work_struct
*work
)
2447 struct bonding
*bond
= container_of(work
, struct bonding
,
2450 read_lock(&bond
->lock
);
2451 if (bond
->kill_timers
)
2454 if (bond
->slave_cnt
== 0)
2457 if (bond
->send_grat_arp
) {
2458 read_lock(&bond
->curr_slave_lock
);
2459 bond_send_gratuitous_arp(bond
);
2460 read_unlock(&bond
->curr_slave_lock
);
2463 if (bond
->send_unsol_na
) {
2464 read_lock(&bond
->curr_slave_lock
);
2465 bond_send_unsolicited_na(bond
);
2466 read_unlock(&bond
->curr_slave_lock
);
2469 if (bond_miimon_inspect(bond
)) {
2470 read_unlock(&bond
->lock
);
2472 read_lock(&bond
->lock
);
2474 bond_miimon_commit(bond
);
2476 read_unlock(&bond
->lock
);
2477 rtnl_unlock(); /* might sleep, hold no other locks */
2478 read_lock(&bond
->lock
);
2482 if (bond
->params
.miimon
)
2483 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2484 msecs_to_jiffies(bond
->params
.miimon
));
2486 read_unlock(&bond
->lock
);
2489 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2491 struct in_device
*idev
;
2492 struct in_ifaddr
*ifa
;
2499 idev
= __in_dev_get_rcu(dev
);
2503 ifa
= idev
->ifa_list
;
2507 addr
= ifa
->ifa_local
;
2513 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2515 struct vlan_entry
*vlan
;
2517 if (ip
== bond
->master_ip
)
2520 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2521 if (ip
== vlan
->vlan_ip
)
2529 * We go to the (large) trouble of VLAN tagging ARP frames because
2530 * switches in VLAN mode (especially if ports are configured as
2531 * "native" to a VLAN) might not pass non-tagged frames.
2533 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2535 struct sk_buff
*skb
;
2537 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2538 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2540 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2541 NULL
, slave_dev
->dev_addr
, NULL
);
2544 printk(KERN_ERR DRV_NAME
": ARP packet allocation failed\n");
2548 skb
= vlan_put_tag(skb
, vlan_id
);
2550 printk(KERN_ERR DRV_NAME
": failed to insert VLAN tag\n");
2558 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2561 __be32
*targets
= bond
->params
.arp_targets
;
2562 struct vlan_entry
*vlan
;
2563 struct net_device
*vlan_dev
;
2567 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2570 pr_debug("basa: target %x\n", targets
[i
]);
2571 if (list_empty(&bond
->vlan_list
)) {
2572 pr_debug("basa: empty vlan: arp_send\n");
2573 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2574 bond
->master_ip
, 0);
2579 * If VLANs are configured, we do a route lookup to
2580 * determine which VLAN interface would be used, so we
2581 * can tag the ARP with the proper VLAN tag.
2583 memset(&fl
, 0, sizeof(fl
));
2584 fl
.fl4_dst
= targets
[i
];
2585 fl
.fl4_tos
= RTO_ONLINK
;
2587 rv
= ip_route_output_key(&init_net
, &rt
, &fl
);
2589 if (net_ratelimit()) {
2590 printk(KERN_WARNING DRV_NAME
2591 ": %s: no route to arp_ip_target %pI4\n",
2592 bond
->dev
->name
, &fl
.fl4_dst
);
2598 * This target is not on a VLAN
2600 if (rt
->u
.dst
.dev
== bond
->dev
) {
2602 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2603 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2604 bond
->master_ip
, 0);
2609 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2610 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2611 if (vlan_dev
== rt
->u
.dst
.dev
) {
2612 vlan_id
= vlan
->vlan_id
;
2613 pr_debug("basa: vlan match on %s %d\n",
2614 vlan_dev
->name
, vlan_id
);
2621 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2622 vlan
->vlan_ip
, vlan_id
);
2626 if (net_ratelimit()) {
2627 printk(KERN_WARNING DRV_NAME
2628 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2629 bond
->dev
->name
, &fl
.fl4_dst
,
2630 rt
->u
.dst
.dev
? rt
->u
.dst
.dev
->name
: "NULL");
2637 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2638 * for each VLAN above us.
2640 * Caller must hold curr_slave_lock for read or better
2642 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2644 struct slave
*slave
= bond
->curr_active_slave
;
2645 struct vlan_entry
*vlan
;
2646 struct net_device
*vlan_dev
;
2648 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond
->dev
->name
,
2649 slave
? slave
->dev
->name
: "NULL");
2651 if (!slave
|| !bond
->send_grat_arp
||
2652 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2655 bond
->send_grat_arp
--;
2657 if (bond
->master_ip
) {
2658 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2659 bond
->master_ip
, 0);
2662 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2663 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2664 if (vlan
->vlan_ip
) {
2665 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2666 vlan
->vlan_ip
, vlan
->vlan_id
);
2671 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2674 __be32
*targets
= bond
->params
.arp_targets
;
2676 targets
= bond
->params
.arp_targets
;
2677 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2678 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2679 &sip
, &tip
, i
, &targets
[i
], bond_has_this_ip(bond
, tip
));
2680 if (sip
== targets
[i
]) {
2681 if (bond_has_this_ip(bond
, tip
))
2682 slave
->last_arp_rx
= jiffies
;
2688 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2691 struct slave
*slave
;
2692 struct bonding
*bond
;
2693 unsigned char *arp_ptr
;
2696 if (dev_net(dev
) != &init_net
)
2699 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2702 bond
= netdev_priv(dev
);
2703 read_lock(&bond
->lock
);
2705 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2706 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2707 orig_dev
? orig_dev
->name
: "NULL");
2709 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2710 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2713 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2717 if (arp
->ar_hln
!= dev
->addr_len
||
2718 skb
->pkt_type
== PACKET_OTHERHOST
||
2719 skb
->pkt_type
== PACKET_LOOPBACK
||
2720 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2721 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2725 arp_ptr
= (unsigned char *)(arp
+ 1);
2726 arp_ptr
+= dev
->addr_len
;
2727 memcpy(&sip
, arp_ptr
, 4);
2728 arp_ptr
+= 4 + dev
->addr_len
;
2729 memcpy(&tip
, arp_ptr
, 4);
2731 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2732 bond
->dev
->name
, slave
->dev
->name
, slave
->state
,
2733 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2737 * Backup slaves won't see the ARP reply, but do come through
2738 * here for each ARP probe (so we swap the sip/tip to validate
2739 * the probe). In a "redundant switch, common router" type of
2740 * configuration, the ARP probe will (hopefully) travel from
2741 * the active, through one switch, the router, then the other
2742 * switch before reaching the backup.
2744 if (slave
->state
== BOND_STATE_ACTIVE
)
2745 bond_validate_arp(bond
, slave
, sip
, tip
);
2747 bond_validate_arp(bond
, slave
, tip
, sip
);
2750 read_unlock(&bond
->lock
);
2753 return NET_RX_SUCCESS
;
2757 * this function is called regularly to monitor each slave's link
2758 * ensuring that traffic is being sent and received when arp monitoring
2759 * is used in load-balancing mode. if the adapter has been dormant, then an
2760 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2761 * arp monitoring in active backup mode.
2763 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2765 struct bonding
*bond
= container_of(work
, struct bonding
,
2767 struct slave
*slave
, *oldcurrent
;
2768 int do_failover
= 0;
2772 read_lock(&bond
->lock
);
2774 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2776 if (bond
->kill_timers
) {
2780 if (bond
->slave_cnt
== 0) {
2784 read_lock(&bond
->curr_slave_lock
);
2785 oldcurrent
= bond
->curr_active_slave
;
2786 read_unlock(&bond
->curr_slave_lock
);
2788 /* see if any of the previous devices are up now (i.e. they have
2789 * xmt and rcv traffic). the curr_active_slave does not come into
2790 * the picture unless it is null. also, slave->jiffies is not needed
2791 * here because we send an arp on each slave and give a slave as
2792 * long as it needs to get the tx/rx within the delta.
2793 * TODO: what about up/down delay in arp mode? it wasn't here before
2796 bond_for_each_slave(bond
, slave
, i
) {
2797 if (slave
->link
!= BOND_LINK_UP
) {
2798 if (time_before_eq(jiffies
, slave
->dev
->trans_start
+ delta_in_ticks
) &&
2799 time_before_eq(jiffies
, slave
->dev
->last_rx
+ delta_in_ticks
)) {
2801 slave
->link
= BOND_LINK_UP
;
2802 slave
->state
= BOND_STATE_ACTIVE
;
2804 /* primary_slave has no meaning in round-robin
2805 * mode. the window of a slave being up and
2806 * curr_active_slave being null after enslaving
2810 printk(KERN_INFO DRV_NAME
2811 ": %s: link status definitely "
2812 "up for interface %s, ",
2817 printk(KERN_INFO DRV_NAME
2818 ": %s: interface %s is now up\n",
2824 /* slave->link == BOND_LINK_UP */
2826 /* not all switches will respond to an arp request
2827 * when the source ip is 0, so don't take the link down
2828 * if we don't know our ip yet
2830 if (time_after_eq(jiffies
, slave
->dev
->trans_start
+ 2*delta_in_ticks
) ||
2831 (time_after_eq(jiffies
, slave
->dev
->last_rx
+ 2*delta_in_ticks
))) {
2833 slave
->link
= BOND_LINK_DOWN
;
2834 slave
->state
= BOND_STATE_BACKUP
;
2836 if (slave
->link_failure_count
< UINT_MAX
) {
2837 slave
->link_failure_count
++;
2840 printk(KERN_INFO DRV_NAME
2841 ": %s: interface %s is now down.\n",
2845 if (slave
== oldcurrent
) {
2851 /* note: if switch is in round-robin mode, all links
2852 * must tx arp to ensure all links rx an arp - otherwise
2853 * links may oscillate or not come up at all; if switch is
2854 * in something like xor mode, there is nothing we can
2855 * do - all replies will be rx'ed on same link causing slaves
2856 * to be unstable during low/no traffic periods
2858 if (IS_UP(slave
->dev
)) {
2859 bond_arp_send_all(bond
, slave
);
2864 write_lock_bh(&bond
->curr_slave_lock
);
2866 bond_select_active_slave(bond
);
2868 write_unlock_bh(&bond
->curr_slave_lock
);
2872 if (bond
->params
.arp_interval
)
2873 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2875 read_unlock(&bond
->lock
);
2879 * Called to inspect slaves for active-backup mode ARP monitor link state
2880 * changes. Sets new_link in slaves to specify what action should take
2881 * place for the slave. Returns 0 if no changes are found, >0 if changes
2882 * to link states must be committed.
2884 * Called with bond->lock held for read.
2886 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2888 struct slave
*slave
;
2891 bond_for_each_slave(bond
, slave
, i
) {
2892 slave
->new_link
= BOND_LINK_NOCHANGE
;
2894 if (slave
->link
!= BOND_LINK_UP
) {
2895 if (time_before_eq(jiffies
, slave_last_rx(bond
, slave
) +
2897 slave
->new_link
= BOND_LINK_UP
;
2905 * Give slaves 2*delta after being enslaved or made
2906 * active. This avoids bouncing, as the last receive
2907 * times need a full ARP monitor cycle to be updated.
2909 if (!time_after_eq(jiffies
, slave
->jiffies
+
2910 2 * delta_in_ticks
))
2914 * Backup slave is down if:
2915 * - No current_arp_slave AND
2916 * - more than 3*delta since last receive AND
2917 * - the bond has an IP address
2919 * Note: a non-null current_arp_slave indicates
2920 * the curr_active_slave went down and we are
2921 * searching for a new one; under this condition
2922 * we only take the curr_active_slave down - this
2923 * gives each slave a chance to tx/rx traffic
2924 * before being taken out
2926 if (slave
->state
== BOND_STATE_BACKUP
&&
2927 !bond
->current_arp_slave
&&
2928 time_after(jiffies
, slave_last_rx(bond
, slave
) +
2929 3 * delta_in_ticks
)) {
2930 slave
->new_link
= BOND_LINK_DOWN
;
2935 * Active slave is down if:
2936 * - more than 2*delta since transmitting OR
2937 * - (more than 2*delta since receive AND
2938 * the bond has an IP address)
2940 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
2941 (time_after_eq(jiffies
, slave
->dev
->trans_start
+
2942 2 * delta_in_ticks
) ||
2943 (time_after_eq(jiffies
, slave_last_rx(bond
, slave
)
2944 + 2 * delta_in_ticks
)))) {
2945 slave
->new_link
= BOND_LINK_DOWN
;
2950 read_lock(&bond
->curr_slave_lock
);
2953 * Trigger a commit if the primary option setting has changed.
2955 if (bond
->primary_slave
&&
2956 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
2957 (bond
->primary_slave
->link
== BOND_LINK_UP
))
2960 read_unlock(&bond
->curr_slave_lock
);
2966 * Called to commit link state changes noted by inspection step of
2967 * active-backup mode ARP monitor.
2969 * Called with RTNL and bond->lock for read.
2971 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2973 struct slave
*slave
;
2976 bond_for_each_slave(bond
, slave
, i
) {
2977 switch (slave
->new_link
) {
2978 case BOND_LINK_NOCHANGE
:
2982 write_lock_bh(&bond
->curr_slave_lock
);
2984 if (!bond
->curr_active_slave
&&
2985 time_before_eq(jiffies
, slave
->dev
->trans_start
+
2987 slave
->link
= BOND_LINK_UP
;
2988 bond_change_active_slave(bond
, slave
);
2989 bond
->current_arp_slave
= NULL
;
2991 printk(KERN_INFO DRV_NAME
2992 ": %s: %s is up and now the "
2993 "active interface\n",
2994 bond
->dev
->name
, slave
->dev
->name
);
2996 } else if (bond
->curr_active_slave
!= slave
) {
2997 /* this slave has just come up but we
2998 * already have a current slave; this can
2999 * also happen if bond_enslave adds a new
3000 * slave that is up while we are searching
3003 slave
->link
= BOND_LINK_UP
;
3004 bond_set_slave_inactive_flags(slave
);
3005 bond
->current_arp_slave
= NULL
;
3007 printk(KERN_INFO DRV_NAME
3008 ": %s: backup interface %s is now up\n",
3009 bond
->dev
->name
, slave
->dev
->name
);
3012 write_unlock_bh(&bond
->curr_slave_lock
);
3016 case BOND_LINK_DOWN
:
3017 if (slave
->link_failure_count
< UINT_MAX
)
3018 slave
->link_failure_count
++;
3020 slave
->link
= BOND_LINK_DOWN
;
3022 if (slave
== bond
->curr_active_slave
) {
3023 printk(KERN_INFO DRV_NAME
3024 ": %s: link status down for active "
3025 "interface %s, disabling it\n",
3026 bond
->dev
->name
, slave
->dev
->name
);
3028 bond_set_slave_inactive_flags(slave
);
3030 write_lock_bh(&bond
->curr_slave_lock
);
3032 bond_select_active_slave(bond
);
3033 if (bond
->curr_active_slave
)
3034 bond
->curr_active_slave
->jiffies
=
3037 write_unlock_bh(&bond
->curr_slave_lock
);
3039 bond
->current_arp_slave
= NULL
;
3041 } else if (slave
->state
== BOND_STATE_BACKUP
) {
3042 printk(KERN_INFO DRV_NAME
3043 ": %s: backup interface %s is now down\n",
3044 bond
->dev
->name
, slave
->dev
->name
);
3046 bond_set_slave_inactive_flags(slave
);
3051 printk(KERN_ERR DRV_NAME
3052 ": %s: impossible: new_link %d on slave %s\n",
3053 bond
->dev
->name
, slave
->new_link
,
3059 * No race with changes to primary via sysfs, as we hold rtnl.
3061 if (bond
->primary_slave
&&
3062 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
3063 (bond
->primary_slave
->link
== BOND_LINK_UP
)) {
3064 write_lock_bh(&bond
->curr_slave_lock
);
3065 bond_change_active_slave(bond
, bond
->primary_slave
);
3066 write_unlock_bh(&bond
->curr_slave_lock
);
3069 bond_set_carrier(bond
);
3073 * Send ARP probes for active-backup mode ARP monitor.
3075 * Called with bond->lock held for read.
3077 static void bond_ab_arp_probe(struct bonding
*bond
)
3079 struct slave
*slave
;
3082 read_lock(&bond
->curr_slave_lock
);
3084 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3085 printk("PROBE: c_arp %s && cas %s BAD\n",
3086 bond
->current_arp_slave
->dev
->name
,
3087 bond
->curr_active_slave
->dev
->name
);
3089 if (bond
->curr_active_slave
) {
3090 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3091 read_unlock(&bond
->curr_slave_lock
);
3095 read_unlock(&bond
->curr_slave_lock
);
3097 /* if we don't have a curr_active_slave, search for the next available
3098 * backup slave from the current_arp_slave and make it the candidate
3099 * for becoming the curr_active_slave
3102 if (!bond
->current_arp_slave
) {
3103 bond
->current_arp_slave
= bond
->first_slave
;
3104 if (!bond
->current_arp_slave
)
3108 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3110 /* search for next candidate */
3111 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3112 if (IS_UP(slave
->dev
)) {
3113 slave
->link
= BOND_LINK_BACK
;
3114 bond_set_slave_active_flags(slave
);
3115 bond_arp_send_all(bond
, slave
);
3116 slave
->jiffies
= jiffies
;
3117 bond
->current_arp_slave
= slave
;
3121 /* if the link state is up at this point, we
3122 * mark it down - this can happen if we have
3123 * simultaneous link failures and
3124 * reselect_active_interface doesn't make this
3125 * one the current slave so it is still marked
3126 * up when it is actually down
3128 if (slave
->link
== BOND_LINK_UP
) {
3129 slave
->link
= BOND_LINK_DOWN
;
3130 if (slave
->link_failure_count
< UINT_MAX
)
3131 slave
->link_failure_count
++;
3133 bond_set_slave_inactive_flags(slave
);
3135 printk(KERN_INFO DRV_NAME
3136 ": %s: backup interface %s is now down.\n",
3137 bond
->dev
->name
, slave
->dev
->name
);
3142 void bond_activebackup_arp_mon(struct work_struct
*work
)
3144 struct bonding
*bond
= container_of(work
, struct bonding
,
3148 read_lock(&bond
->lock
);
3150 if (bond
->kill_timers
)
3153 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3155 if (bond
->slave_cnt
== 0)
3158 if (bond
->send_grat_arp
) {
3159 read_lock(&bond
->curr_slave_lock
);
3160 bond_send_gratuitous_arp(bond
);
3161 read_unlock(&bond
->curr_slave_lock
);
3164 if (bond
->send_unsol_na
) {
3165 read_lock(&bond
->curr_slave_lock
);
3166 bond_send_unsolicited_na(bond
);
3167 read_unlock(&bond
->curr_slave_lock
);
3170 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3171 read_unlock(&bond
->lock
);
3173 read_lock(&bond
->lock
);
3175 bond_ab_arp_commit(bond
, delta_in_ticks
);
3177 read_unlock(&bond
->lock
);
3179 read_lock(&bond
->lock
);
3182 bond_ab_arp_probe(bond
);
3185 if (bond
->params
.arp_interval
) {
3186 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3189 read_unlock(&bond
->lock
);
3192 /*------------------------------ proc/seq_file-------------------------------*/
3194 #ifdef CONFIG_PROC_FS
3196 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3198 struct bonding
*bond
= seq
->private;
3200 struct slave
*slave
;
3203 /* make sure the bond won't be taken away */
3204 read_lock(&dev_base_lock
);
3205 read_lock(&bond
->lock
);
3208 return SEQ_START_TOKEN
;
3211 bond_for_each_slave(bond
, slave
, i
) {
3212 if (++off
== *pos
) {
3220 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3222 struct bonding
*bond
= seq
->private;
3223 struct slave
*slave
= v
;
3226 if (v
== SEQ_START_TOKEN
) {
3227 return bond
->first_slave
;
3230 slave
= slave
->next
;
3232 return (slave
== bond
->first_slave
) ? NULL
: slave
;
3235 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
3237 struct bonding
*bond
= seq
->private;
3239 read_unlock(&bond
->lock
);
3240 read_unlock(&dev_base_lock
);
3243 static void bond_info_show_master(struct seq_file
*seq
)
3245 struct bonding
*bond
= seq
->private;
3249 read_lock(&bond
->curr_slave_lock
);
3250 curr
= bond
->curr_active_slave
;
3251 read_unlock(&bond
->curr_slave_lock
);
3253 seq_printf(seq
, "Bonding Mode: %s",
3254 bond_mode_name(bond
->params
.mode
));
3256 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
&&
3257 bond
->params
.fail_over_mac
)
3258 seq_printf(seq
, " (fail_over_mac %s)",
3259 fail_over_mac_tbl
[bond
->params
.fail_over_mac
].modename
);
3261 seq_printf(seq
, "\n");
3263 if (bond
->params
.mode
== BOND_MODE_XOR
||
3264 bond
->params
.mode
== BOND_MODE_8023AD
) {
3265 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
3266 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
3267 bond
->params
.xmit_policy
);
3270 if (USES_PRIMARY(bond
->params
.mode
)) {
3271 seq_printf(seq
, "Primary Slave: %s\n",
3272 (bond
->primary_slave
) ?
3273 bond
->primary_slave
->dev
->name
: "None");
3275 seq_printf(seq
, "Currently Active Slave: %s\n",
3276 (curr
) ? curr
->dev
->name
: "None");
3279 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
3281 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
3282 seq_printf(seq
, "Up Delay (ms): %d\n",
3283 bond
->params
.updelay
* bond
->params
.miimon
);
3284 seq_printf(seq
, "Down Delay (ms): %d\n",
3285 bond
->params
.downdelay
* bond
->params
.miimon
);
3288 /* ARP information */
3289 if(bond
->params
.arp_interval
> 0) {
3291 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
3292 bond
->params
.arp_interval
);
3294 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
3296 for(i
= 0; (i
< BOND_MAX_ARP_TARGETS
) ;i
++) {
3297 if (!bond
->params
.arp_targets
[i
])
3300 seq_printf(seq
, ",");
3301 seq_printf(seq
, " %pI4", &bond
->params
.arp_targets
[i
]);
3304 seq_printf(seq
, "\n");
3307 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3308 struct ad_info ad_info
;
3310 seq_puts(seq
, "\n802.3ad info\n");
3311 seq_printf(seq
, "LACP rate: %s\n",
3312 (bond
->params
.lacp_fast
) ? "fast" : "slow");
3313 seq_printf(seq
, "Aggregator selection policy (ad_select): %s\n",
3314 ad_select_tbl
[bond
->params
.ad_select
].modename
);
3316 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3317 seq_printf(seq
, "bond %s has no active aggregator\n",
3320 seq_printf(seq
, "Active Aggregator Info:\n");
3322 seq_printf(seq
, "\tAggregator ID: %d\n",
3323 ad_info
.aggregator_id
);
3324 seq_printf(seq
, "\tNumber of ports: %d\n",
3326 seq_printf(seq
, "\tActor Key: %d\n",
3328 seq_printf(seq
, "\tPartner Key: %d\n",
3329 ad_info
.partner_key
);
3330 seq_printf(seq
, "\tPartner Mac Address: %pM\n",
3331 ad_info
.partner_system
);
3336 static void bond_info_show_slave(struct seq_file
*seq
, const struct slave
*slave
)
3338 struct bonding
*bond
= seq
->private;
3340 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
3341 seq_printf(seq
, "MII Status: %s\n",
3342 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
3343 seq_printf(seq
, "Link Failure Count: %u\n",
3344 slave
->link_failure_count
);
3346 seq_printf(seq
, "Permanent HW addr: %pM\n", slave
->perm_hwaddr
);
3348 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3349 const struct aggregator
*agg
3350 = SLAVE_AD_INFO(slave
).port
.aggregator
;
3353 seq_printf(seq
, "Aggregator ID: %d\n",
3354 agg
->aggregator_identifier
);
3356 seq_puts(seq
, "Aggregator ID: N/A\n");
3361 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
3363 if (v
== SEQ_START_TOKEN
) {
3364 seq_printf(seq
, "%s\n", version
);
3365 bond_info_show_master(seq
);
3367 bond_info_show_slave(seq
, v
);
3373 static struct seq_operations bond_info_seq_ops
= {
3374 .start
= bond_info_seq_start
,
3375 .next
= bond_info_seq_next
,
3376 .stop
= bond_info_seq_stop
,
3377 .show
= bond_info_seq_show
,
3380 static int bond_info_open(struct inode
*inode
, struct file
*file
)
3382 struct seq_file
*seq
;
3383 struct proc_dir_entry
*proc
;
3386 res
= seq_open(file
, &bond_info_seq_ops
);
3388 /* recover the pointer buried in proc_dir_entry data */
3389 seq
= file
->private_data
;
3391 seq
->private = proc
->data
;
3397 static const struct file_operations bond_info_fops
= {
3398 .owner
= THIS_MODULE
,
3399 .open
= bond_info_open
,
3401 .llseek
= seq_lseek
,
3402 .release
= seq_release
,
3405 static int bond_create_proc_entry(struct bonding
*bond
)
3407 struct net_device
*bond_dev
= bond
->dev
;
3409 if (bond_proc_dir
) {
3410 bond
->proc_entry
= proc_create_data(bond_dev
->name
,
3411 S_IRUGO
, bond_proc_dir
,
3412 &bond_info_fops
, bond
);
3413 if (bond
->proc_entry
== NULL
) {
3414 printk(KERN_WARNING DRV_NAME
3415 ": Warning: Cannot create /proc/net/%s/%s\n",
3416 DRV_NAME
, bond_dev
->name
);
3418 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3425 static void bond_remove_proc_entry(struct bonding
*bond
)
3427 if (bond_proc_dir
&& bond
->proc_entry
) {
3428 remove_proc_entry(bond
->proc_file_name
, bond_proc_dir
);
3429 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3430 bond
->proc_entry
= NULL
;
3434 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3435 * Caller must hold rtnl_lock.
3437 static void bond_create_proc_dir(void)
3439 int len
= strlen(DRV_NAME
);
3441 for (bond_proc_dir
= init_net
.proc_net
->subdir
; bond_proc_dir
;
3442 bond_proc_dir
= bond_proc_dir
->next
) {
3443 if ((bond_proc_dir
->namelen
== len
) &&
3444 !memcmp(bond_proc_dir
->name
, DRV_NAME
, len
)) {
3449 if (!bond_proc_dir
) {
3450 bond_proc_dir
= proc_mkdir(DRV_NAME
, init_net
.proc_net
);
3451 if (bond_proc_dir
) {
3452 bond_proc_dir
->owner
= THIS_MODULE
;
3454 printk(KERN_WARNING DRV_NAME
3455 ": Warning: cannot create /proc/net/%s\n",
3461 /* Destroy the bonding directory under /proc/net, if empty.
3462 * Caller must hold rtnl_lock.
3464 static void bond_destroy_proc_dir(void)
3466 struct proc_dir_entry
*de
;
3468 if (!bond_proc_dir
) {
3472 /* verify that the /proc dir is empty */
3473 for (de
= bond_proc_dir
->subdir
; de
; de
= de
->next
) {
3474 /* ignore . and .. */
3475 if (*(de
->name
) != '.') {
3481 if (bond_proc_dir
->owner
== THIS_MODULE
) {
3482 bond_proc_dir
->owner
= NULL
;
3485 remove_proc_entry(DRV_NAME
, init_net
.proc_net
);
3486 bond_proc_dir
= NULL
;
3489 #endif /* CONFIG_PROC_FS */
3491 /*-------------------------- netdev event handling --------------------------*/
3494 * Change device name
3496 static int bond_event_changename(struct bonding
*bond
)
3498 #ifdef CONFIG_PROC_FS
3499 bond_remove_proc_entry(bond
);
3500 bond_create_proc_entry(bond
);
3502 down_write(&(bonding_rwsem
));
3503 bond_destroy_sysfs_entry(bond
);
3504 bond_create_sysfs_entry(bond
);
3505 up_write(&(bonding_rwsem
));
3509 static int bond_master_netdev_event(unsigned long event
, struct net_device
*bond_dev
)
3511 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3514 case NETDEV_CHANGENAME
:
3515 return bond_event_changename(event_bond
);
3516 case NETDEV_UNREGISTER
:
3517 bond_release_all(event_bond
->dev
);
3526 static int bond_slave_netdev_event(unsigned long event
, struct net_device
*slave_dev
)
3528 struct net_device
*bond_dev
= slave_dev
->master
;
3529 struct bonding
*bond
= netdev_priv(bond_dev
);
3532 case NETDEV_UNREGISTER
:
3534 if (bond
->setup_by_slave
)
3535 bond_release_and_destroy(bond_dev
, slave_dev
);
3537 bond_release(bond_dev
, slave_dev
);
3542 * TODO: is this what we get if somebody
3543 * sets up a hierarchical bond, then rmmod's
3544 * one of the slave bonding devices?
3549 * ... Or is it this?
3552 case NETDEV_CHANGEMTU
:
3554 * TODO: Should slaves be allowed to
3555 * independently alter their MTU? For
3556 * an active-backup bond, slaves need
3557 * not be the same type of device, so
3558 * MTUs may vary. For other modes,
3559 * slaves arguably should have the
3560 * same MTUs. To do this, we'd need to
3561 * take over the slave's change_mtu
3562 * function for the duration of their
3566 case NETDEV_CHANGENAME
:
3568 * TODO: handle changing the primary's name
3571 case NETDEV_FEAT_CHANGE
:
3572 bond_compute_features(bond
);
3582 * bond_netdev_event: handle netdev notifier chain events.
3584 * This function receives events for the netdev chain. The caller (an
3585 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3586 * locks for us to safely manipulate the slave devices (RTNL lock,
3589 static int bond_netdev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3591 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3593 if (dev_net(event_dev
) != &init_net
)
3596 pr_debug("event_dev: %s, event: %lx\n",
3597 (event_dev
? event_dev
->name
: "None"),
3600 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3603 if (event_dev
->flags
& IFF_MASTER
) {
3604 pr_debug("IFF_MASTER\n");
3605 return bond_master_netdev_event(event
, event_dev
);
3608 if (event_dev
->flags
& IFF_SLAVE
) {
3609 pr_debug("IFF_SLAVE\n");
3610 return bond_slave_netdev_event(event
, event_dev
);
3617 * bond_inetaddr_event: handle inetaddr notifier chain events.
3619 * We keep track of device IPs primarily to use as source addresses in
3620 * ARP monitor probes (rather than spewing out broadcasts all the time).
3622 * We track one IP for the main device (if it has one), plus one per VLAN.
3624 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3626 struct in_ifaddr
*ifa
= ptr
;
3627 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3628 struct bonding
*bond
;
3629 struct vlan_entry
*vlan
;
3631 if (dev_net(ifa
->ifa_dev
->dev
) != &init_net
)
3634 list_for_each_entry(bond
, &bond_dev_list
, bond_list
) {
3635 if (bond
->dev
== event_dev
) {
3638 bond
->master_ip
= ifa
->ifa_local
;
3641 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3648 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3649 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3650 if (vlan_dev
== event_dev
) {
3653 vlan
->vlan_ip
= ifa
->ifa_local
;
3657 bond_glean_dev_ip(vlan_dev
);
3668 static struct notifier_block bond_netdev_notifier
= {
3669 .notifier_call
= bond_netdev_event
,
3672 static struct notifier_block bond_inetaddr_notifier
= {
3673 .notifier_call
= bond_inetaddr_event
,
3676 /*-------------------------- Packet type handling ---------------------------*/
3678 /* register to receive lacpdus on a bond */
3679 static void bond_register_lacpdu(struct bonding
*bond
)
3681 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3683 /* initialize packet type */
3684 pk_type
->type
= PKT_TYPE_LACPDU
;
3685 pk_type
->dev
= bond
->dev
;
3686 pk_type
->func
= bond_3ad_lacpdu_recv
;
3688 dev_add_pack(pk_type
);
3691 /* unregister to receive lacpdus on a bond */
3692 static void bond_unregister_lacpdu(struct bonding
*bond
)
3694 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3697 void bond_register_arp(struct bonding
*bond
)
3699 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3704 pt
->type
= htons(ETH_P_ARP
);
3705 pt
->dev
= bond
->dev
;
3706 pt
->func
= bond_arp_rcv
;
3710 void bond_unregister_arp(struct bonding
*bond
)
3712 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3714 dev_remove_pack(pt
);
3718 /*---------------------------- Hashing Policies -----------------------------*/
3721 * Hash for the output device based upon layer 2 and layer 3 data. If
3722 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3724 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
,
3725 struct net_device
*bond_dev
, int count
)
3727 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3728 struct iphdr
*iph
= ip_hdr(skb
);
3730 if (skb
->protocol
== htons(ETH_P_IP
)) {
3731 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3732 (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5])) % count
;
3735 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3739 * Hash for the output device based upon layer 3 and layer 4 data. If
3740 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3741 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3743 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
,
3744 struct net_device
*bond_dev
, int count
)
3746 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3747 struct iphdr
*iph
= ip_hdr(skb
);
3748 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3751 if (skb
->protocol
== htons(ETH_P_IP
)) {
3752 if (!(iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) &&
3753 (iph
->protocol
== IPPROTO_TCP
||
3754 iph
->protocol
== IPPROTO_UDP
)) {
3755 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3757 return (layer4_xor
^
3758 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3762 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3766 * Hash for the output device based upon layer 2 data
3768 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
,
3769 struct net_device
*bond_dev
, int count
)
3771 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3773 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3776 /*-------------------------- Device entry points ----------------------------*/
3778 static int bond_open(struct net_device
*bond_dev
)
3780 struct bonding
*bond
= netdev_priv(bond_dev
);
3782 bond
->kill_timers
= 0;
3784 if (bond_is_lb(bond
)) {
3785 /* bond_alb_initialize must be called before the timer
3788 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3789 /* something went wrong - fail the open operation */
3793 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3794 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3797 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3798 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3799 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3802 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3803 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3804 INIT_DELAYED_WORK(&bond
->arp_work
,
3805 bond_activebackup_arp_mon
);
3807 INIT_DELAYED_WORK(&bond
->arp_work
,
3808 bond_loadbalance_arp_mon
);
3810 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3811 if (bond
->params
.arp_validate
)
3812 bond_register_arp(bond
);
3815 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3816 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3817 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3818 /* register to receive LACPDUs */
3819 bond_register_lacpdu(bond
);
3820 bond_3ad_initiate_agg_selection(bond
, 1);
3826 static int bond_close(struct net_device
*bond_dev
)
3828 struct bonding
*bond
= netdev_priv(bond_dev
);
3830 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3831 /* Unregister the receive of LACPDUs */
3832 bond_unregister_lacpdu(bond
);
3835 if (bond
->params
.arp_validate
)
3836 bond_unregister_arp(bond
);
3838 write_lock_bh(&bond
->lock
);
3840 bond
->send_grat_arp
= 0;
3841 bond
->send_unsol_na
= 0;
3843 /* signal timers not to re-arm */
3844 bond
->kill_timers
= 1;
3846 write_unlock_bh(&bond
->lock
);
3848 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3849 cancel_delayed_work(&bond
->mii_work
);
3852 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3853 cancel_delayed_work(&bond
->arp_work
);
3856 switch (bond
->params
.mode
) {
3857 case BOND_MODE_8023AD
:
3858 cancel_delayed_work(&bond
->ad_work
);
3862 cancel_delayed_work(&bond
->alb_work
);
3869 if (bond_is_lb(bond
)) {
3870 /* Must be called only after all
3871 * slaves have been released
3873 bond_alb_deinitialize(bond
);
3879 static struct net_device_stats
*bond_get_stats(struct net_device
*bond_dev
)
3881 struct bonding
*bond
= netdev_priv(bond_dev
);
3882 struct net_device_stats
*stats
= &bond
->stats
;
3883 struct net_device_stats local_stats
;
3884 struct slave
*slave
;
3887 memset(&local_stats
, 0, sizeof(struct net_device_stats
));
3889 read_lock_bh(&bond
->lock
);
3891 bond_for_each_slave(bond
, slave
, i
) {
3892 const struct net_device_stats
*sstats
= dev_get_stats(slave
->dev
);
3894 local_stats
.rx_packets
+= sstats
->rx_packets
;
3895 local_stats
.rx_bytes
+= sstats
->rx_bytes
;
3896 local_stats
.rx_errors
+= sstats
->rx_errors
;
3897 local_stats
.rx_dropped
+= sstats
->rx_dropped
;
3899 local_stats
.tx_packets
+= sstats
->tx_packets
;
3900 local_stats
.tx_bytes
+= sstats
->tx_bytes
;
3901 local_stats
.tx_errors
+= sstats
->tx_errors
;
3902 local_stats
.tx_dropped
+= sstats
->tx_dropped
;
3904 local_stats
.multicast
+= sstats
->multicast
;
3905 local_stats
.collisions
+= sstats
->collisions
;
3907 local_stats
.rx_length_errors
+= sstats
->rx_length_errors
;
3908 local_stats
.rx_over_errors
+= sstats
->rx_over_errors
;
3909 local_stats
.rx_crc_errors
+= sstats
->rx_crc_errors
;
3910 local_stats
.rx_frame_errors
+= sstats
->rx_frame_errors
;
3911 local_stats
.rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3912 local_stats
.rx_missed_errors
+= sstats
->rx_missed_errors
;
3914 local_stats
.tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3915 local_stats
.tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3916 local_stats
.tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3917 local_stats
.tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3918 local_stats
.tx_window_errors
+= sstats
->tx_window_errors
;
3921 memcpy(stats
, &local_stats
, sizeof(struct net_device_stats
));
3923 read_unlock_bh(&bond
->lock
);
3928 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3930 struct net_device
*slave_dev
= NULL
;
3931 struct ifbond k_binfo
;
3932 struct ifbond __user
*u_binfo
= NULL
;
3933 struct ifslave k_sinfo
;
3934 struct ifslave __user
*u_sinfo
= NULL
;
3935 struct mii_ioctl_data
*mii
= NULL
;
3938 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3939 bond_dev
->name
, cmd
);
3951 * We do this again just in case we were called by SIOCGMIIREG
3952 * instead of SIOCGMIIPHY.
3959 if (mii
->reg_num
== 1) {
3960 struct bonding
*bond
= netdev_priv(bond_dev
);
3962 read_lock(&bond
->lock
);
3963 read_lock(&bond
->curr_slave_lock
);
3964 if (netif_carrier_ok(bond
->dev
)) {
3965 mii
->val_out
= BMSR_LSTATUS
;
3967 read_unlock(&bond
->curr_slave_lock
);
3968 read_unlock(&bond
->lock
);
3972 case BOND_INFO_QUERY_OLD
:
3973 case SIOCBONDINFOQUERY
:
3974 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3976 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
))) {
3980 res
= bond_info_query(bond_dev
, &k_binfo
);
3982 if (copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
))) {
3988 case BOND_SLAVE_INFO_QUERY_OLD
:
3989 case SIOCBONDSLAVEINFOQUERY
:
3990 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3992 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
))) {
3996 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3998 if (copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
))) {
4009 if (!capable(CAP_NET_ADMIN
)) {
4013 down_write(&(bonding_rwsem
));
4014 slave_dev
= dev_get_by_name(&init_net
, ifr
->ifr_slave
);
4016 pr_debug("slave_dev=%p: \n", slave_dev
);
4021 pr_debug("slave_dev->name=%s: \n", slave_dev
->name
);
4023 case BOND_ENSLAVE_OLD
:
4024 case SIOCBONDENSLAVE
:
4025 res
= bond_enslave(bond_dev
, slave_dev
);
4027 case BOND_RELEASE_OLD
:
4028 case SIOCBONDRELEASE
:
4029 res
= bond_release(bond_dev
, slave_dev
);
4031 case BOND_SETHWADDR_OLD
:
4032 case SIOCBONDSETHWADDR
:
4033 res
= bond_sethwaddr(bond_dev
, slave_dev
);
4035 case BOND_CHANGE_ACTIVE_OLD
:
4036 case SIOCBONDCHANGEACTIVE
:
4037 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
4046 up_write(&(bonding_rwsem
));
4050 static void bond_set_multicast_list(struct net_device
*bond_dev
)
4052 struct bonding
*bond
= netdev_priv(bond_dev
);
4053 struct dev_mc_list
*dmi
;
4056 * Do promisc before checking multicast_mode
4058 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
)) {
4060 * FIXME: Need to handle the error when one of the multi-slaves
4063 bond_set_promiscuity(bond
, 1);
4066 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
)) {
4067 bond_set_promiscuity(bond
, -1);
4070 /* set allmulti flag to slaves */
4071 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
)) {
4073 * FIXME: Need to handle the error when one of the multi-slaves
4076 bond_set_allmulti(bond
, 1);
4079 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
)) {
4080 bond_set_allmulti(bond
, -1);
4083 read_lock(&bond
->lock
);
4085 bond
->flags
= bond_dev
->flags
;
4087 /* looking for addresses to add to slaves' mc list */
4088 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
4089 if (!bond_mc_list_find_dmi(dmi
, bond
->mc_list
)) {
4090 bond_mc_add(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4094 /* looking for addresses to delete from slaves' list */
4095 for (dmi
= bond
->mc_list
; dmi
; dmi
= dmi
->next
) {
4096 if (!bond_mc_list_find_dmi(dmi
, bond_dev
->mc_list
)) {
4097 bond_mc_delete(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4101 /* save master's multicast list */
4102 bond_mc_list_destroy(bond
);
4103 bond_mc_list_copy(bond_dev
->mc_list
, bond
, GFP_ATOMIC
);
4105 read_unlock(&bond
->lock
);
4108 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
4110 struct bonding
*bond
= netdev_priv(dev
);
4111 struct slave
*slave
= bond
->first_slave
;
4114 const struct net_device_ops
*slave_ops
4115 = slave
->dev
->netdev_ops
;
4116 if (slave_ops
->ndo_neigh_setup
)
4117 return slave_ops
->ndo_neigh_setup(dev
, parms
);
4123 * Change the MTU of all of a master's slaves to match the master
4125 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
4127 struct bonding
*bond
= netdev_priv(bond_dev
);
4128 struct slave
*slave
, *stop_at
;
4132 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
4133 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
4135 /* Can't hold bond->lock with bh disabled here since
4136 * some base drivers panic. On the other hand we can't
4137 * hold bond->lock without bh disabled because we'll
4138 * deadlock. The only solution is to rely on the fact
4139 * that we're under rtnl_lock here, and the slaves
4140 * list won't change. This doesn't solve the problem
4141 * of setting the slave's MTU while it is
4142 * transmitting, but the assumption is that the base
4143 * driver can handle that.
4145 * TODO: figure out a way to safely iterate the slaves
4146 * list, but without holding a lock around the actual
4147 * call to the base driver.
4150 bond_for_each_slave(bond
, slave
, i
) {
4151 pr_debug("s %p s->p %p c_m %p\n", slave
,
4152 slave
->prev
, slave
->dev
->change_mtu
);
4154 res
= dev_set_mtu(slave
->dev
, new_mtu
);
4157 /* If we failed to set the slave's mtu to the new value
4158 * we must abort the operation even in ACTIVE_BACKUP
4159 * mode, because if we allow the backup slaves to have
4160 * different mtu values than the active slave we'll
4161 * need to change their mtu when doing a failover. That
4162 * means changing their mtu from timer context, which
4163 * is probably not a good idea.
4165 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4170 bond_dev
->mtu
= new_mtu
;
4175 /* unwind from head to the slave that failed */
4177 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4180 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
4182 pr_debug("unwind err %d dev %s\n", tmp_res
,
4193 * Note that many devices must be down to change the HW address, and
4194 * downing the master releases all slaves. We can make bonds full of
4195 * bonding devices to test this, however.
4197 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4199 struct bonding
*bond
= netdev_priv(bond_dev
);
4200 struct sockaddr
*sa
= addr
, tmp_sa
;
4201 struct slave
*slave
, *stop_at
;
4205 if (bond
->params
.mode
== BOND_MODE_ALB
)
4206 return bond_alb_set_mac_address(bond_dev
, addr
);
4209 pr_debug("bond=%p, name=%s\n", bond
, (bond_dev
? bond_dev
->name
: "None"));
4212 * If fail_over_mac is set to active, do nothing and return
4213 * success. Returning an error causes ifenslave to fail.
4215 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4218 if (!is_valid_ether_addr(sa
->sa_data
)) {
4219 return -EADDRNOTAVAIL
;
4222 /* Can't hold bond->lock with bh disabled here since
4223 * some base drivers panic. On the other hand we can't
4224 * hold bond->lock without bh disabled because we'll
4225 * deadlock. The only solution is to rely on the fact
4226 * that we're under rtnl_lock here, and the slaves
4227 * list won't change. This doesn't solve the problem
4228 * of setting the slave's hw address while it is
4229 * transmitting, but the assumption is that the base
4230 * driver can handle that.
4232 * TODO: figure out a way to safely iterate the slaves
4233 * list, but without holding a lock around the actual
4234 * call to the base driver.
4237 bond_for_each_slave(bond
, slave
, i
) {
4238 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
4239 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
4241 if (slave_ops
->ndo_set_mac_address
== NULL
) {
4243 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
4247 res
= dev_set_mac_address(slave
->dev
, addr
);
4249 /* TODO: consider downing the slave
4251 * User should expect communications
4252 * breakage anyway until ARP finish
4255 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4261 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4265 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4266 tmp_sa
.sa_family
= bond_dev
->type
;
4268 /* unwind from head to the slave that failed */
4270 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4273 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4275 pr_debug("unwind err %d dev %s\n", tmp_res
,
4283 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4285 struct bonding
*bond
= netdev_priv(bond_dev
);
4286 struct slave
*slave
, *start_at
;
4287 int i
, slave_no
, res
= 1;
4289 read_lock(&bond
->lock
);
4291 if (!BOND_IS_OK(bond
)) {
4296 * Concurrent TX may collide on rr_tx_counter; we accept that
4297 * as being rare enough not to justify using an atomic op here
4299 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4301 bond_for_each_slave(bond
, slave
, i
) {
4309 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4310 if (IS_UP(slave
->dev
) &&
4311 (slave
->link
== BOND_LINK_UP
) &&
4312 (slave
->state
== BOND_STATE_ACTIVE
)) {
4313 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4320 /* no suitable interface, frame not sent */
4323 read_unlock(&bond
->lock
);
4329 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4330 * the bond has a usable interface.
4332 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4334 struct bonding
*bond
= netdev_priv(bond_dev
);
4337 read_lock(&bond
->lock
);
4338 read_lock(&bond
->curr_slave_lock
);
4340 if (!BOND_IS_OK(bond
)) {
4344 if (!bond
->curr_active_slave
)
4347 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4351 /* no suitable interface, frame not sent */
4354 read_unlock(&bond
->curr_slave_lock
);
4355 read_unlock(&bond
->lock
);
4360 * In bond_xmit_xor() , we determine the output device by using a pre-
4361 * determined xmit_hash_policy(), If the selected device is not enabled,
4362 * find the next active slave.
4364 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4366 struct bonding
*bond
= netdev_priv(bond_dev
);
4367 struct slave
*slave
, *start_at
;
4372 read_lock(&bond
->lock
);
4374 if (!BOND_IS_OK(bond
)) {
4378 slave_no
= bond
->xmit_hash_policy(skb
, bond_dev
, bond
->slave_cnt
);
4380 bond_for_each_slave(bond
, slave
, i
) {
4389 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4390 if (IS_UP(slave
->dev
) &&
4391 (slave
->link
== BOND_LINK_UP
) &&
4392 (slave
->state
== BOND_STATE_ACTIVE
)) {
4393 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4400 /* no suitable interface, frame not sent */
4403 read_unlock(&bond
->lock
);
4408 * in broadcast mode, we send everything to all usable interfaces.
4410 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4412 struct bonding
*bond
= netdev_priv(bond_dev
);
4413 struct slave
*slave
, *start_at
;
4414 struct net_device
*tx_dev
= NULL
;
4418 read_lock(&bond
->lock
);
4420 if (!BOND_IS_OK(bond
)) {
4424 read_lock(&bond
->curr_slave_lock
);
4425 start_at
= bond
->curr_active_slave
;
4426 read_unlock(&bond
->curr_slave_lock
);
4432 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4433 if (IS_UP(slave
->dev
) &&
4434 (slave
->link
== BOND_LINK_UP
) &&
4435 (slave
->state
== BOND_STATE_ACTIVE
)) {
4437 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4439 printk(KERN_ERR DRV_NAME
4440 ": %s: Error: bond_xmit_broadcast(): "
4441 "skb_clone() failed\n",
4446 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4448 dev_kfree_skb(skb2
);
4452 tx_dev
= slave
->dev
;
4457 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4462 /* no suitable interface, frame not sent */
4465 /* frame sent to all suitable interfaces */
4466 read_unlock(&bond
->lock
);
4470 /*------------------------- Device initialization ---------------------------*/
4472 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4474 switch (bond
->params
.xmit_policy
) {
4475 case BOND_XMIT_POLICY_LAYER23
:
4476 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4478 case BOND_XMIT_POLICY_LAYER34
:
4479 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4481 case BOND_XMIT_POLICY_LAYER2
:
4483 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4488 static int bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4490 const struct bonding
*bond
= netdev_priv(dev
);
4492 switch (bond
->params
.mode
) {
4493 case BOND_MODE_ROUNDROBIN
:
4494 return bond_xmit_roundrobin(skb
, dev
);
4495 case BOND_MODE_ACTIVEBACKUP
:
4496 return bond_xmit_activebackup(skb
, dev
);
4498 return bond_xmit_xor(skb
, dev
);
4499 case BOND_MODE_BROADCAST
:
4500 return bond_xmit_broadcast(skb
, dev
);
4501 case BOND_MODE_8023AD
:
4502 return bond_3ad_xmit_xor(skb
, dev
);
4505 return bond_alb_xmit(skb
, dev
);
4507 /* Should never happen, mode already checked */
4508 printk(KERN_ERR DRV_NAME
": %s: Error: Unknown bonding mode %d\n",
4509 dev
->name
, bond
->params
.mode
);
4512 return NETDEV_TX_OK
;
4518 * set bond mode specific net device operations
4520 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4522 struct net_device
*bond_dev
= bond
->dev
;
4525 case BOND_MODE_ROUNDROBIN
:
4527 case BOND_MODE_ACTIVEBACKUP
:
4530 bond_set_xmit_hash_policy(bond
);
4532 case BOND_MODE_BROADCAST
:
4534 case BOND_MODE_8023AD
:
4535 bond_set_master_3ad_flags(bond
);
4536 bond_set_xmit_hash_policy(bond
);
4539 bond_set_master_alb_flags(bond
);
4544 /* Should never happen, mode already checked */
4545 printk(KERN_ERR DRV_NAME
4546 ": %s: Error: Unknown bonding mode %d\n",
4553 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4554 struct ethtool_drvinfo
*drvinfo
)
4556 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4557 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4558 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4561 static const struct ethtool_ops bond_ethtool_ops
= {
4562 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4563 .get_link
= ethtool_op_get_link
,
4564 .get_tx_csum
= ethtool_op_get_tx_csum
,
4565 .get_sg
= ethtool_op_get_sg
,
4566 .get_tso
= ethtool_op_get_tso
,
4567 .get_ufo
= ethtool_op_get_ufo
,
4568 .get_flags
= ethtool_op_get_flags
,
4571 static const struct net_device_ops bond_netdev_ops
= {
4572 .ndo_open
= bond_open
,
4573 .ndo_stop
= bond_close
,
4574 .ndo_start_xmit
= bond_start_xmit
,
4575 .ndo_get_stats
= bond_get_stats
,
4576 .ndo_do_ioctl
= bond_do_ioctl
,
4577 .ndo_set_multicast_list
= bond_set_multicast_list
,
4578 .ndo_change_mtu
= bond_change_mtu
,
4579 .ndo_set_mac_address
= bond_set_mac_address
,
4580 .ndo_neigh_setup
= bond_neigh_setup
,
4581 .ndo_vlan_rx_register
= bond_vlan_rx_register
,
4582 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4583 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4587 * Does not allocate but creates a /proc entry.
4590 static int bond_init(struct net_device
*bond_dev
, struct bond_params
*params
)
4592 struct bonding
*bond
= netdev_priv(bond_dev
);
4594 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4596 /* initialize rwlocks */
4597 rwlock_init(&bond
->lock
);
4598 rwlock_init(&bond
->curr_slave_lock
);
4600 bond
->params
= *params
; /* copy params struct */
4602 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4606 /* Initialize pointers */
4607 bond
->first_slave
= NULL
;
4608 bond
->curr_active_slave
= NULL
;
4609 bond
->current_arp_slave
= NULL
;
4610 bond
->primary_slave
= NULL
;
4611 bond
->dev
= bond_dev
;
4612 bond
->send_grat_arp
= 0;
4613 bond
->send_unsol_na
= 0;
4614 bond
->setup_by_slave
= 0;
4615 INIT_LIST_HEAD(&bond
->vlan_list
);
4617 /* Initialize the device entry points */
4618 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4619 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4620 bond_set_mode_ops(bond
, bond
->params
.mode
);
4622 bond_dev
->destructor
= bond_destructor
;
4624 /* Initialize the device options */
4625 bond_dev
->tx_queue_len
= 0;
4626 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4627 bond_dev
->priv_flags
|= IFF_BONDING
;
4628 if (bond
->params
.arp_interval
)
4629 bond_dev
->priv_flags
|= IFF_MASTER_ARPMON
;
4631 /* At first, we block adding VLANs. That's the only way to
4632 * prevent problems that occur when adding VLANs over an
4633 * empty bond. The block will be removed once non-challenged
4634 * slaves are enslaved.
4636 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4638 /* don't acquire bond device's netif_tx_lock when
4640 bond_dev
->features
|= NETIF_F_LLTX
;
4642 /* By default, we declare the bond to be fully
4643 * VLAN hardware accelerated capable. Special
4644 * care is taken in the various xmit functions
4645 * when there are slaves that are not hw accel
4648 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4649 NETIF_F_HW_VLAN_RX
|
4650 NETIF_F_HW_VLAN_FILTER
);
4652 #ifdef CONFIG_PROC_FS
4653 bond_create_proc_entry(bond
);
4655 list_add_tail(&bond
->bond_list
, &bond_dev_list
);
4660 static void bond_work_cancel_all(struct bonding
*bond
)
4662 write_lock_bh(&bond
->lock
);
4663 bond
->kill_timers
= 1;
4664 write_unlock_bh(&bond
->lock
);
4666 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4667 cancel_delayed_work(&bond
->mii_work
);
4669 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4670 cancel_delayed_work(&bond
->arp_work
);
4672 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4673 delayed_work_pending(&bond
->alb_work
))
4674 cancel_delayed_work(&bond
->alb_work
);
4676 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4677 delayed_work_pending(&bond
->ad_work
))
4678 cancel_delayed_work(&bond
->ad_work
);
4681 /* De-initialize device specific data.
4682 * Caller must hold rtnl_lock.
4684 static void bond_deinit(struct net_device
*bond_dev
)
4686 struct bonding
*bond
= netdev_priv(bond_dev
);
4688 list_del(&bond
->bond_list
);
4690 bond_work_cancel_all(bond
);
4692 #ifdef CONFIG_PROC_FS
4693 bond_remove_proc_entry(bond
);
4697 /* Unregister and free all bond devices.
4698 * Caller must hold rtnl_lock.
4700 static void bond_free_all(void)
4702 struct bonding
*bond
, *nxt
;
4704 list_for_each_entry_safe(bond
, nxt
, &bond_dev_list
, bond_list
) {
4705 struct net_device
*bond_dev
= bond
->dev
;
4707 bond_work_cancel_all(bond
);
4708 /* Release the bonded slaves */
4709 bond_release_all(bond_dev
);
4713 #ifdef CONFIG_PROC_FS
4714 bond_destroy_proc_dir();
4718 /*------------------------- Module initialization ---------------------------*/
4721 * Convert string input module parms. Accept either the
4722 * number of the mode or its string name. A bit complicated because
4723 * some mode names are substrings of other names, and calls from sysfs
4724 * may have whitespace in the name (trailing newlines, for example).
4726 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4728 int mode
= -1, i
, rv
;
4729 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4731 for (p
= (char *)buf
; *p
; p
++)
4732 if (!(isdigit(*p
) || isspace(*p
)))
4736 rv
= sscanf(buf
, "%20s", modestr
);
4738 rv
= sscanf(buf
, "%d", &mode
);
4743 for (i
= 0; tbl
[i
].modename
; i
++) {
4744 if (mode
== tbl
[i
].mode
)
4746 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4753 static int bond_check_params(struct bond_params
*params
)
4755 int arp_validate_value
, fail_over_mac_value
;
4758 * Convert string parameters.
4761 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4762 if (bond_mode
== -1) {
4763 printk(KERN_ERR DRV_NAME
4764 ": Error: Invalid bonding mode \"%s\"\n",
4765 mode
== NULL
? "NULL" : mode
);
4770 if (xmit_hash_policy
) {
4771 if ((bond_mode
!= BOND_MODE_XOR
) &&
4772 (bond_mode
!= BOND_MODE_8023AD
)) {
4773 printk(KERN_INFO DRV_NAME
4774 ": xor_mode param is irrelevant in mode %s\n",
4775 bond_mode_name(bond_mode
));
4777 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4779 if (xmit_hashtype
== -1) {
4780 printk(KERN_ERR DRV_NAME
4781 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4782 xmit_hash_policy
== NULL
? "NULL" :
4790 if (bond_mode
!= BOND_MODE_8023AD
) {
4791 printk(KERN_INFO DRV_NAME
4792 ": lacp_rate param is irrelevant in mode %s\n",
4793 bond_mode_name(bond_mode
));
4795 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4796 if (lacp_fast
== -1) {
4797 printk(KERN_ERR DRV_NAME
4798 ": Error: Invalid lacp rate \"%s\"\n",
4799 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4806 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4807 if (params
->ad_select
== -1) {
4808 printk(KERN_ERR DRV_NAME
4809 ": Error: Invalid ad_select \"%s\"\n",
4810 ad_select
== NULL
? "NULL" : ad_select
);
4814 if (bond_mode
!= BOND_MODE_8023AD
) {
4815 printk(KERN_WARNING DRV_NAME
4816 ": ad_select param only affects 802.3ad mode\n");
4819 params
->ad_select
= BOND_AD_STABLE
;
4822 if (max_bonds
< 0 || max_bonds
> INT_MAX
) {
4823 printk(KERN_WARNING DRV_NAME
4824 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4825 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4826 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4827 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4831 printk(KERN_WARNING DRV_NAME
4832 ": Warning: miimon module parameter (%d), "
4833 "not in range 0-%d, so it was reset to %d\n",
4834 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4835 miimon
= BOND_LINK_MON_INTERV
;
4839 printk(KERN_WARNING DRV_NAME
4840 ": Warning: updelay module parameter (%d), "
4841 "not in range 0-%d, so it was reset to 0\n",
4846 if (downdelay
< 0) {
4847 printk(KERN_WARNING DRV_NAME
4848 ": Warning: downdelay module parameter (%d), "
4849 "not in range 0-%d, so it was reset to 0\n",
4850 downdelay
, INT_MAX
);
4854 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4855 printk(KERN_WARNING DRV_NAME
4856 ": Warning: use_carrier module parameter (%d), "
4857 "not of valid value (0/1), so it was set to 1\n",
4862 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
4863 printk(KERN_WARNING DRV_NAME
4864 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4865 "was reset to 1 \n", num_grat_arp
);
4869 if (num_unsol_na
< 0 || num_unsol_na
> 255) {
4870 printk(KERN_WARNING DRV_NAME
4871 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4872 "was reset to 1 \n", num_unsol_na
);
4876 /* reset values for 802.3ad */
4877 if (bond_mode
== BOND_MODE_8023AD
) {
4879 printk(KERN_WARNING DRV_NAME
4880 ": Warning: miimon must be specified, "
4881 "otherwise bonding will not detect link "
4882 "failure, speed and duplex which are "
4883 "essential for 802.3ad operation\n");
4884 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4889 /* reset values for TLB/ALB */
4890 if ((bond_mode
== BOND_MODE_TLB
) ||
4891 (bond_mode
== BOND_MODE_ALB
)) {
4893 printk(KERN_WARNING DRV_NAME
4894 ": Warning: miimon must be specified, "
4895 "otherwise bonding will not detect link "
4896 "failure and link speed which are essential "
4897 "for TLB/ALB load balancing\n");
4898 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4903 if (bond_mode
== BOND_MODE_ALB
) {
4904 printk(KERN_NOTICE DRV_NAME
4905 ": In ALB mode you might experience client "
4906 "disconnections upon reconnection of a link if the "
4907 "bonding module updelay parameter (%d msec) is "
4908 "incompatible with the forwarding delay time of the "
4914 if (updelay
|| downdelay
) {
4915 /* just warn the user the up/down delay will have
4916 * no effect since miimon is zero...
4918 printk(KERN_WARNING DRV_NAME
4919 ": Warning: miimon module parameter not set "
4920 "and updelay (%d) or downdelay (%d) module "
4921 "parameter is set; updelay and downdelay have "
4922 "no effect unless miimon is set\n",
4923 updelay
, downdelay
);
4926 /* don't allow arp monitoring */
4928 printk(KERN_WARNING DRV_NAME
4929 ": Warning: miimon (%d) and arp_interval (%d) "
4930 "can't be used simultaneously, disabling ARP "
4932 miimon
, arp_interval
);
4936 if ((updelay
% miimon
) != 0) {
4937 printk(KERN_WARNING DRV_NAME
4938 ": Warning: updelay (%d) is not a multiple "
4939 "of miimon (%d), updelay rounded to %d ms\n",
4940 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4945 if ((downdelay
% miimon
) != 0) {
4946 printk(KERN_WARNING DRV_NAME
4947 ": Warning: downdelay (%d) is not a multiple "
4948 "of miimon (%d), downdelay rounded to %d ms\n",
4950 (downdelay
/ miimon
) * miimon
);
4953 downdelay
/= miimon
;
4956 if (arp_interval
< 0) {
4957 printk(KERN_WARNING DRV_NAME
4958 ": Warning: arp_interval module parameter (%d) "
4959 ", not in range 0-%d, so it was reset to %d\n",
4960 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4961 arp_interval
= BOND_LINK_ARP_INTERV
;
4964 for (arp_ip_count
= 0;
4965 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4967 /* not complete check, but should be good enough to
4969 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4970 printk(KERN_WARNING DRV_NAME
4971 ": Warning: bad arp_ip_target module parameter "
4972 "(%s), ARP monitoring will not be performed\n",
4973 arp_ip_target
[arp_ip_count
]);
4976 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4977 arp_target
[arp_ip_count
] = ip
;
4981 if (arp_interval
&& !arp_ip_count
) {
4982 /* don't allow arping if no arp_ip_target given... */
4983 printk(KERN_WARNING DRV_NAME
4984 ": Warning: arp_interval module parameter (%d) "
4985 "specified without providing an arp_ip_target "
4986 "parameter, arp_interval was reset to 0\n",
4992 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4993 printk(KERN_ERR DRV_NAME
4994 ": arp_validate only supported in active-backup mode\n");
4997 if (!arp_interval
) {
4998 printk(KERN_ERR DRV_NAME
4999 ": arp_validate requires arp_interval\n");
5003 arp_validate_value
= bond_parse_parm(arp_validate
,
5005 if (arp_validate_value
== -1) {
5006 printk(KERN_ERR DRV_NAME
5007 ": Error: invalid arp_validate \"%s\"\n",
5008 arp_validate
== NULL
? "NULL" : arp_validate
);
5012 arp_validate_value
= 0;
5015 printk(KERN_INFO DRV_NAME
5016 ": MII link monitoring set to %d ms\n",
5018 } else if (arp_interval
) {
5021 printk(KERN_INFO DRV_NAME
5022 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5024 arp_validate_tbl
[arp_validate_value
].modename
,
5027 for (i
= 0; i
< arp_ip_count
; i
++)
5028 printk (" %s", arp_ip_target
[i
]);
5032 } else if (max_bonds
) {
5033 /* miimon and arp_interval not set, we need one so things
5034 * work as expected, see bonding.txt for details
5036 printk(KERN_WARNING DRV_NAME
5037 ": Warning: either miimon or arp_interval and "
5038 "arp_ip_target module parameters must be specified, "
5039 "otherwise bonding will not detect link failures! see "
5040 "bonding.txt for details.\n");
5043 if (primary
&& !USES_PRIMARY(bond_mode
)) {
5044 /* currently, using a primary only makes sense
5045 * in active backup, TLB or ALB modes
5047 printk(KERN_WARNING DRV_NAME
5048 ": Warning: %s primary device specified but has no "
5049 "effect in %s mode\n",
5050 primary
, bond_mode_name(bond_mode
));
5054 if (fail_over_mac
) {
5055 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
5057 if (fail_over_mac_value
== -1) {
5058 printk(KERN_ERR DRV_NAME
5059 ": Error: invalid fail_over_mac \"%s\"\n",
5060 arp_validate
== NULL
? "NULL" : arp_validate
);
5064 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
5065 printk(KERN_WARNING DRV_NAME
5066 ": Warning: fail_over_mac only affects "
5067 "active-backup mode.\n");
5069 fail_over_mac_value
= BOND_FOM_NONE
;
5072 /* fill params struct with the proper values */
5073 params
->mode
= bond_mode
;
5074 params
->xmit_policy
= xmit_hashtype
;
5075 params
->miimon
= miimon
;
5076 params
->num_grat_arp
= num_grat_arp
;
5077 params
->num_unsol_na
= num_unsol_na
;
5078 params
->arp_interval
= arp_interval
;
5079 params
->arp_validate
= arp_validate_value
;
5080 params
->updelay
= updelay
;
5081 params
->downdelay
= downdelay
;
5082 params
->use_carrier
= use_carrier
;
5083 params
->lacp_fast
= lacp_fast
;
5084 params
->primary
[0] = 0;
5085 params
->fail_over_mac
= fail_over_mac_value
;
5088 strncpy(params
->primary
, primary
, IFNAMSIZ
);
5089 params
->primary
[IFNAMSIZ
- 1] = 0;
5092 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
5097 static struct lock_class_key bonding_netdev_xmit_lock_key
;
5098 static struct lock_class_key bonding_netdev_addr_lock_key
;
5100 static void bond_set_lockdep_class_one(struct net_device
*dev
,
5101 struct netdev_queue
*txq
,
5104 lockdep_set_class(&txq
->_xmit_lock
,
5105 &bonding_netdev_xmit_lock_key
);
5108 static void bond_set_lockdep_class(struct net_device
*dev
)
5110 lockdep_set_class(&dev
->addr_list_lock
,
5111 &bonding_netdev_addr_lock_key
);
5112 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
5115 /* Create a new bond based on the specified name and bonding parameters.
5116 * If name is NULL, obtain a suitable "bond%d" name for us.
5117 * Caller must NOT hold rtnl_lock; we need to release it here before we
5118 * set up our sysfs entries.
5120 int bond_create(char *name
, struct bond_params
*params
)
5122 struct net_device
*bond_dev
;
5123 struct bonding
*bond
;
5127 down_write(&bonding_rwsem
);
5129 /* Check to see if the bond already exists. */
5131 list_for_each_entry(bond
, &bond_dev_list
, bond_list
)
5132 if (strnicmp(bond
->dev
->name
, name
, IFNAMSIZ
) == 0) {
5133 printk(KERN_ERR DRV_NAME
5134 ": cannot add bond %s; it already exists\n",
5141 bond_dev
= alloc_netdev(sizeof(struct bonding
), name
? name
: "",
5144 printk(KERN_ERR DRV_NAME
5145 ": %s: eek! can't alloc netdev!\n",
5152 res
= dev_alloc_name(bond_dev
, "bond%d");
5157 /* bond_init() must be called after dev_alloc_name() (for the
5158 * /proc files), but before register_netdevice(), because we
5159 * need to set function pointers.
5162 res
= bond_init(bond_dev
, params
);
5167 res
= register_netdevice(bond_dev
);
5172 bond_set_lockdep_class(bond_dev
);
5174 netif_carrier_off(bond_dev
);
5176 up_write(&bonding_rwsem
);
5177 rtnl_unlock(); /* allows sysfs registration of net device */
5178 res
= bond_create_sysfs_entry(netdev_priv(bond_dev
));
5181 down_write(&bonding_rwsem
);
5182 bond_deinit(bond_dev
);
5183 unregister_netdevice(bond_dev
);
5190 bond_deinit(bond_dev
);
5192 free_netdev(bond_dev
);
5194 up_write(&bonding_rwsem
);
5199 static int __init
bonding_init(void)
5203 struct bonding
*bond
;
5205 printk(KERN_INFO
"%s", version
);
5207 res
= bond_check_params(&bonding_defaults
);
5212 #ifdef CONFIG_PROC_FS
5213 bond_create_proc_dir();
5216 init_rwsem(&bonding_rwsem
);
5218 for (i
= 0; i
< max_bonds
; i
++) {
5219 res
= bond_create(NULL
, &bonding_defaults
);
5224 res
= bond_create_sysfs();
5228 register_netdevice_notifier(&bond_netdev_notifier
);
5229 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5230 bond_register_ipv6_notifier();
5234 list_for_each_entry(bond
, &bond_dev_list
, bond_list
) {
5235 bond_work_cancel_all(bond
);
5236 destroy_workqueue(bond
->wq
);
5239 bond_destroy_sysfs();
5249 static void __exit
bonding_exit(void)
5251 unregister_netdevice_notifier(&bond_netdev_notifier
);
5252 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5253 bond_unregister_ipv6_notifier();
5255 bond_destroy_sysfs();
5262 module_init(bonding_init
);
5263 module_exit(bonding_exit
);
5264 MODULE_LICENSE("GPL");
5265 MODULE_VERSION(DRV_VERSION
);
5266 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5267 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5268 MODULE_SUPPORTED_DEVICE("most ethernet devices");