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 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
] = { 0, } ;
158 static int arp_ip_count
= 0;
159 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
160 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
161 static int lacp_fast
= 0;
164 const struct bond_parm_tbl bond_lacp_tbl
[] = {
165 { "slow", AD_LACP_SLOW
},
166 { "fast", AD_LACP_FAST
},
170 const struct bond_parm_tbl bond_mode_tbl
[] = {
171 { "balance-rr", BOND_MODE_ROUNDROBIN
},
172 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
173 { "balance-xor", BOND_MODE_XOR
},
174 { "broadcast", BOND_MODE_BROADCAST
},
175 { "802.3ad", BOND_MODE_8023AD
},
176 { "balance-tlb", BOND_MODE_TLB
},
177 { "balance-alb", BOND_MODE_ALB
},
181 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
182 { "layer2", BOND_XMIT_POLICY_LAYER2
},
183 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
184 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
188 const struct bond_parm_tbl arp_validate_tbl
[] = {
189 { "none", BOND_ARP_VALIDATE_NONE
},
190 { "active", BOND_ARP_VALIDATE_ACTIVE
},
191 { "backup", BOND_ARP_VALIDATE_BACKUP
},
192 { "all", BOND_ARP_VALIDATE_ALL
},
196 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
197 { "none", BOND_FOM_NONE
},
198 { "active", BOND_FOM_ACTIVE
},
199 { "follow", BOND_FOM_FOLLOW
},
203 struct bond_parm_tbl ad_select_tbl
[] = {
204 { "stable", BOND_AD_STABLE
},
205 { "bandwidth", BOND_AD_BANDWIDTH
},
206 { "count", BOND_AD_COUNT
},
210 /*-------------------------- Forward declarations ---------------------------*/
212 static void bond_send_gratuitous_arp(struct bonding
*bond
);
213 static void bond_deinit(struct net_device
*bond_dev
);
215 /*---------------------------- General routines -----------------------------*/
217 static const char *bond_mode_name(int mode
)
219 static const char *names
[] = {
220 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR
] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
224 [BOND_MODE_8023AD
]= "IEEE 802.3ad Dynamic link aggregation",
225 [BOND_MODE_TLB
] = "transmit load balancing",
226 [BOND_MODE_ALB
] = "adaptive load balancing",
229 if (mode
< 0 || mode
> BOND_MODE_ALB
)
235 /*---------------------------------- VLAN -----------------------------------*/
238 * bond_add_vlan - add a new vlan id on bond
239 * @bond: bond that got the notification
240 * @vlan_id: the vlan id to add
242 * Returns -ENOMEM if allocation failed.
244 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
246 struct vlan_entry
*vlan
;
248 pr_debug("bond: %s, vlan id %d\n",
249 (bond
? bond
->dev
->name
: "None"), vlan_id
);
251 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
256 INIT_LIST_HEAD(&vlan
->vlan_list
);
257 vlan
->vlan_id
= vlan_id
;
259 write_lock_bh(&bond
->lock
);
261 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
263 write_unlock_bh(&bond
->lock
);
265 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
271 * bond_del_vlan - delete a vlan id from bond
272 * @bond: bond that got the notification
273 * @vlan_id: the vlan id to delete
275 * returns -ENODEV if @vlan_id was not found in @bond.
277 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
279 struct vlan_entry
*vlan
;
282 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
284 write_lock_bh(&bond
->lock
);
286 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
287 if (vlan
->vlan_id
== vlan_id
) {
288 list_del(&vlan
->vlan_list
);
290 if (bond_is_lb(bond
))
291 bond_alb_clear_vlan(bond
, vlan_id
);
293 pr_debug("removed VLAN ID %d from bond %s\n", vlan_id
,
298 if (list_empty(&bond
->vlan_list
) &&
299 (bond
->slave_cnt
== 0)) {
300 /* Last VLAN removed and no slaves, so
301 * restore block on adding VLANs. This will
302 * be removed once new slaves that are not
303 * VLAN challenged will be added.
305 bond
->dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
313 pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id
,
317 write_unlock_bh(&bond
->lock
);
322 * bond_has_challenged_slaves
323 * @bond: the bond we're working on
325 * Searches the slave list. Returns 1 if a vlan challenged slave
326 * was found, 0 otherwise.
328 * Assumes bond->lock is held.
330 static int bond_has_challenged_slaves(struct bonding
*bond
)
335 bond_for_each_slave(bond
, slave
, i
) {
336 if (slave
->dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
337 pr_debug("found VLAN challenged slave - %s\n",
343 pr_debug("no VLAN challenged slaves found\n");
348 * bond_next_vlan - safely skip to the next item in the vlans list.
349 * @bond: the bond we're working on
350 * @curr: item we're advancing from
352 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
353 * or @curr->next otherwise (even if it is @curr itself again).
355 * Caller must hold bond->lock
357 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
359 struct vlan_entry
*next
, *last
;
361 if (list_empty(&bond
->vlan_list
)) {
366 next
= list_entry(bond
->vlan_list
.next
,
367 struct vlan_entry
, vlan_list
);
369 last
= list_entry(bond
->vlan_list
.prev
,
370 struct vlan_entry
, vlan_list
);
372 next
= list_entry(bond
->vlan_list
.next
,
373 struct vlan_entry
, vlan_list
);
375 next
= list_entry(curr
->vlan_list
.next
,
376 struct vlan_entry
, vlan_list
);
384 * bond_dev_queue_xmit - Prepare skb for xmit.
386 * @bond: bond device that got this skb for tx.
387 * @skb: hw accel VLAN tagged skb to transmit
388 * @slave_dev: slave that is supposed to xmit this skbuff
390 * When the bond gets an skb to transmit that is
391 * already hardware accelerated VLAN tagged, and it
392 * needs to relay this skb to a slave that is not
393 * hw accel capable, the skb needs to be "unaccelerated",
394 * i.e. strip the hwaccel tag and re-insert it as part
397 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
)
399 unsigned short uninitialized_var(vlan_id
);
401 if (!list_empty(&bond
->vlan_list
) &&
402 !(slave_dev
->features
& NETIF_F_HW_VLAN_TX
) &&
403 vlan_get_tag(skb
, &vlan_id
) == 0) {
404 skb
->dev
= slave_dev
;
405 skb
= vlan_put_tag(skb
, vlan_id
);
407 /* vlan_put_tag() frees the skb in case of error,
408 * so return success here so the calling functions
409 * won't attempt to free is again.
414 skb
->dev
= slave_dev
;
424 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
425 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
427 * a. This operation is performed in IOCTL context,
428 * b. The operation is protected by the RTNL semaphore in the 8021q code,
429 * c. Holding a lock with BH disabled while directly calling a base driver
430 * entry point is generally a BAD idea.
432 * The design of synchronization/protection for this operation in the 8021q
433 * module is good for one or more VLAN devices over a single physical device
434 * and cannot be extended for a teaming solution like bonding, so there is a
435 * potential race condition here where a net device from the vlan group might
436 * be referenced (either by a base driver or the 8021q code) while it is being
437 * removed from the system. However, it turns out we're not making matters
438 * worse, and if it works for regular VLAN usage it will work here too.
442 * bond_vlan_rx_register - Propagates registration to slaves
443 * @bond_dev: bonding net device that got called
444 * @grp: vlan group being registered
446 static void bond_vlan_rx_register(struct net_device
*bond_dev
, struct vlan_group
*grp
)
448 struct bonding
*bond
= netdev_priv(bond_dev
);
454 bond_for_each_slave(bond
, slave
, i
) {
455 struct net_device
*slave_dev
= slave
->dev
;
456 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
458 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
459 slave_ops
->ndo_vlan_rx_register
) {
460 slave_ops
->ndo_vlan_rx_register(slave_dev
, grp
);
466 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
467 * @bond_dev: bonding net device that got called
468 * @vid: vlan id being added
470 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
472 struct bonding
*bond
= netdev_priv(bond_dev
);
476 bond_for_each_slave(bond
, slave
, i
) {
477 struct net_device
*slave_dev
= slave
->dev
;
478 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
480 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
481 slave_ops
->ndo_vlan_rx_add_vid
) {
482 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vid
);
486 res
= bond_add_vlan(bond
, vid
);
488 printk(KERN_ERR DRV_NAME
489 ": %s: Error: Failed to add vlan id %d\n",
490 bond_dev
->name
, vid
);
495 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
496 * @bond_dev: bonding net device that got called
497 * @vid: vlan id being removed
499 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
501 struct bonding
*bond
= netdev_priv(bond_dev
);
503 struct net_device
*vlan_dev
;
506 bond_for_each_slave(bond
, slave
, i
) {
507 struct net_device
*slave_dev
= slave
->dev
;
508 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
510 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
511 slave_ops
->ndo_vlan_rx_kill_vid
) {
512 /* Save and then restore vlan_dev in the grp array,
513 * since the slave's driver might clear it.
515 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vid
);
516 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vid
);
517 vlan_group_set_device(bond
->vlgrp
, vid
, vlan_dev
);
521 res
= bond_del_vlan(bond
, vid
);
523 printk(KERN_ERR DRV_NAME
524 ": %s: Error: Failed to remove vlan id %d\n",
525 bond_dev
->name
, vid
);
529 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
531 struct vlan_entry
*vlan
;
532 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
534 write_lock_bh(&bond
->lock
);
536 if (list_empty(&bond
->vlan_list
))
539 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
540 slave_ops
->ndo_vlan_rx_register
)
541 slave_ops
->ndo_vlan_rx_register(slave_dev
, bond
->vlgrp
);
543 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
544 !(slave_ops
->ndo_vlan_rx_add_vid
))
547 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
)
548 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
551 write_unlock_bh(&bond
->lock
);
554 static void bond_del_vlans_from_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
556 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
557 struct vlan_entry
*vlan
;
558 struct net_device
*vlan_dev
;
560 write_lock_bh(&bond
->lock
);
562 if (list_empty(&bond
->vlan_list
))
565 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
566 !(slave_ops
->ndo_vlan_rx_kill_vid
))
569 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
570 /* Save and then restore vlan_dev in the grp array,
571 * since the slave's driver might clear it.
573 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
574 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
575 vlan_group_set_device(bond
->vlgrp
, vlan
->vlan_id
, vlan_dev
);
579 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
580 slave_ops
->ndo_vlan_rx_register
)
581 slave_ops
->ndo_vlan_rx_register(slave_dev
, NULL
);
584 write_unlock_bh(&bond
->lock
);
587 /*------------------------------- Link status -------------------------------*/
590 * Set the carrier state for the master according to the state of its
591 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
592 * do special 802.3ad magic.
594 * Returns zero if carrier state does not change, nonzero if it does.
596 static int bond_set_carrier(struct bonding
*bond
)
601 if (bond
->slave_cnt
== 0)
604 if (bond
->params
.mode
== BOND_MODE_8023AD
)
605 return bond_3ad_set_carrier(bond
);
607 bond_for_each_slave(bond
, slave
, i
) {
608 if (slave
->link
== BOND_LINK_UP
) {
609 if (!netif_carrier_ok(bond
->dev
)) {
610 netif_carrier_on(bond
->dev
);
618 if (netif_carrier_ok(bond
->dev
)) {
619 netif_carrier_off(bond
->dev
);
626 * Get link speed and duplex from the slave's base driver
627 * using ethtool. If for some reason the call fails or the
628 * values are invalid, fake speed and duplex to 100/Full
631 static int bond_update_speed_duplex(struct slave
*slave
)
633 struct net_device
*slave_dev
= slave
->dev
;
634 struct ethtool_cmd etool
;
637 /* Fake speed and duplex */
638 slave
->speed
= SPEED_100
;
639 slave
->duplex
= DUPLEX_FULL
;
641 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
644 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
648 switch (etool
.speed
) {
658 switch (etool
.duplex
) {
666 slave
->speed
= etool
.speed
;
667 slave
->duplex
= etool
.duplex
;
673 * if <dev> supports MII link status reporting, check its link status.
675 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
676 * depening upon the setting of the use_carrier parameter.
678 * Return either BMSR_LSTATUS, meaning that the link is up (or we
679 * can't tell and just pretend it is), or 0, meaning that the link is
682 * If reporting is non-zero, instead of faking link up, return -1 if
683 * both ETHTOOL and MII ioctls fail (meaning the device does not
684 * support them). If use_carrier is set, return whatever it says.
685 * It'd be nice if there was a good way to tell if a driver supports
686 * netif_carrier, but there really isn't.
688 static int bond_check_dev_link(struct bonding
*bond
, struct net_device
*slave_dev
, int reporting
)
690 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
691 static int (* ioctl
)(struct net_device
*, struct ifreq
*, int);
693 struct mii_ioctl_data
*mii
;
695 if (bond
->params
.use_carrier
)
696 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
698 ioctl
= slave_ops
->ndo_do_ioctl
;
700 /* TODO: set pointer to correct ioctl on a per team member */
701 /* bases to make this more efficient. that is, once */
702 /* we determine the correct ioctl, we will always */
703 /* call it and not the others for that team */
707 * We cannot assume that SIOCGMIIPHY will also read a
708 * register; not all network drivers (e.g., e100)
712 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
713 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
715 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
716 mii
->reg_num
= MII_BMSR
;
717 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0) {
718 return (mii
->val_out
& BMSR_LSTATUS
);
724 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
725 * attempt to get link status from it if the above MII ioctls fail.
727 if (slave_dev
->ethtool_ops
) {
728 if (slave_dev
->ethtool_ops
->get_link
) {
731 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
733 return link
? BMSR_LSTATUS
: 0;
738 * If reporting, report that either there's no dev->do_ioctl,
739 * or both SIOCGMIIREG and get_link failed (meaning that we
740 * cannot report link status). If not reporting, pretend
743 return (reporting
? -1 : BMSR_LSTATUS
);
746 /*----------------------------- Multicast list ------------------------------*/
749 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
751 static inline int bond_is_dmi_same(struct dev_mc_list
*dmi1
, struct dev_mc_list
*dmi2
)
753 return memcmp(dmi1
->dmi_addr
, dmi2
->dmi_addr
, dmi1
->dmi_addrlen
) == 0 &&
754 dmi1
->dmi_addrlen
== dmi2
->dmi_addrlen
;
758 * returns dmi entry if found, NULL otherwise
760 static struct dev_mc_list
*bond_mc_list_find_dmi(struct dev_mc_list
*dmi
, struct dev_mc_list
*mc_list
)
762 struct dev_mc_list
*idmi
;
764 for (idmi
= mc_list
; idmi
; idmi
= idmi
->next
) {
765 if (bond_is_dmi_same(dmi
, idmi
)) {
774 * Push the promiscuity flag down to appropriate slaves
776 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
779 if (USES_PRIMARY(bond
->params
.mode
)) {
780 /* write lock already acquired */
781 if (bond
->curr_active_slave
) {
782 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
788 bond_for_each_slave(bond
, slave
, i
) {
789 err
= dev_set_promiscuity(slave
->dev
, inc
);
798 * Push the allmulti flag down to all slaves
800 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
803 if (USES_PRIMARY(bond
->params
.mode
)) {
804 /* write lock already acquired */
805 if (bond
->curr_active_slave
) {
806 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
812 bond_for_each_slave(bond
, slave
, i
) {
813 err
= dev_set_allmulti(slave
->dev
, inc
);
822 * Add a Multicast address to slaves
825 static void bond_mc_add(struct bonding
*bond
, void *addr
, int alen
)
827 if (USES_PRIMARY(bond
->params
.mode
)) {
828 /* write lock already acquired */
829 if (bond
->curr_active_slave
) {
830 dev_mc_add(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
835 bond_for_each_slave(bond
, slave
, i
) {
836 dev_mc_add(slave
->dev
, addr
, alen
, 0);
842 * Remove a multicast address from slave
845 static void bond_mc_delete(struct bonding
*bond
, void *addr
, int alen
)
847 if (USES_PRIMARY(bond
->params
.mode
)) {
848 /* write lock already acquired */
849 if (bond
->curr_active_slave
) {
850 dev_mc_delete(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
855 bond_for_each_slave(bond
, slave
, i
) {
856 dev_mc_delete(slave
->dev
, addr
, alen
, 0);
863 * Retrieve the list of registered multicast addresses for the bonding
864 * device and retransmit an IGMP JOIN request to the current active
867 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
869 struct in_device
*in_dev
;
870 struct ip_mc_list
*im
;
873 in_dev
= __in_dev_get_rcu(bond
->dev
);
875 for (im
= in_dev
->mc_list
; im
; im
= im
->next
) {
876 ip_mc_rejoin_group(im
);
884 * Totally destroys the mc_list in bond
886 static void bond_mc_list_destroy(struct bonding
*bond
)
888 struct dev_mc_list
*dmi
;
892 bond
->mc_list
= dmi
->next
;
896 bond
->mc_list
= NULL
;
900 * Copy all the Multicast addresses from src to the bonding device dst
902 static int bond_mc_list_copy(struct dev_mc_list
*mc_list
, struct bonding
*bond
,
905 struct dev_mc_list
*dmi
, *new_dmi
;
907 for (dmi
= mc_list
; dmi
; dmi
= dmi
->next
) {
908 new_dmi
= kmalloc(sizeof(struct dev_mc_list
), gfp_flag
);
911 /* FIXME: Potential memory leak !!! */
915 new_dmi
->next
= bond
->mc_list
;
916 bond
->mc_list
= new_dmi
;
917 new_dmi
->dmi_addrlen
= dmi
->dmi_addrlen
;
918 memcpy(new_dmi
->dmi_addr
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
919 new_dmi
->dmi_users
= dmi
->dmi_users
;
920 new_dmi
->dmi_gusers
= dmi
->dmi_gusers
;
927 * flush all members of flush->mc_list from device dev->mc_list
929 static void bond_mc_list_flush(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
931 struct bonding
*bond
= netdev_priv(bond_dev
);
932 struct dev_mc_list
*dmi
;
934 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
935 dev_mc_delete(slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
938 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
939 /* del lacpdu mc addr from mc list */
940 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
942 dev_mc_delete(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
946 /*--------------------------- Active slave change ---------------------------*/
949 * Update the mc list and multicast-related flags for the new and
950 * old active slaves (if any) according to the multicast mode, and
951 * promiscuous flags unconditionally.
953 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
, struct slave
*old_active
)
955 struct dev_mc_list
*dmi
;
957 if (!USES_PRIMARY(bond
->params
.mode
)) {
958 /* nothing to do - mc list is already up-to-date on
965 if (bond
->dev
->flags
& IFF_PROMISC
) {
966 dev_set_promiscuity(old_active
->dev
, -1);
969 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
970 dev_set_allmulti(old_active
->dev
, -1);
973 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
974 dev_mc_delete(old_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
979 /* FIXME: Signal errors upstream. */
980 if (bond
->dev
->flags
& IFF_PROMISC
) {
981 dev_set_promiscuity(new_active
->dev
, 1);
984 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
985 dev_set_allmulti(new_active
->dev
, 1);
988 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
989 dev_mc_add(new_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
991 bond_resend_igmp_join_requests(bond
);
996 * bond_do_fail_over_mac
998 * Perform special MAC address swapping for fail_over_mac settings
1000 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1002 static void bond_do_fail_over_mac(struct bonding
*bond
,
1003 struct slave
*new_active
,
1004 struct slave
*old_active
)
1006 u8 tmp_mac
[ETH_ALEN
];
1007 struct sockaddr saddr
;
1010 switch (bond
->params
.fail_over_mac
) {
1011 case BOND_FOM_ACTIVE
:
1013 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
1014 new_active
->dev
->addr_len
);
1016 case BOND_FOM_FOLLOW
:
1018 * if new_active && old_active, swap them
1019 * if just old_active, do nothing (going to no active slave)
1020 * if just new_active, set new_active to bond's MAC
1025 write_unlock_bh(&bond
->curr_slave_lock
);
1026 read_unlock(&bond
->lock
);
1029 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
1030 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
1032 saddr
.sa_family
= new_active
->dev
->type
;
1034 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
1035 saddr
.sa_family
= bond
->dev
->type
;
1038 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
1040 printk(KERN_ERR DRV_NAME
1041 ": %s: Error %d setting MAC of slave %s\n",
1042 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1049 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1050 saddr
.sa_family
= old_active
->dev
->type
;
1052 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
1054 printk(KERN_ERR DRV_NAME
1055 ": %s: Error %d setting MAC of slave %s\n",
1056 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1058 read_lock(&bond
->lock
);
1059 write_lock_bh(&bond
->curr_slave_lock
);
1062 printk(KERN_ERR DRV_NAME
1063 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1064 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1072 * find_best_interface - select the best available slave to be the active one
1073 * @bond: our bonding struct
1075 * Warning: Caller must hold curr_slave_lock for writing.
1077 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1079 struct slave
*new_active
, *old_active
;
1080 struct slave
*bestslave
= NULL
;
1081 int mintime
= bond
->params
.updelay
;
1084 new_active
= old_active
= bond
->curr_active_slave
;
1086 if (!new_active
) { /* there were no active slaves left */
1087 if (bond
->slave_cnt
> 0) { /* found one slave */
1088 new_active
= bond
->first_slave
;
1090 return NULL
; /* still no slave, return NULL */
1094 /* first try the primary link; if arping, a link must tx/rx traffic
1095 * before it can be considered the curr_active_slave - also, we would skip
1096 * slaves between the curr_active_slave and primary_slave that may be up
1099 if ((bond
->primary_slave
) &&
1100 (!bond
->params
.arp_interval
) &&
1101 (IS_UP(bond
->primary_slave
->dev
))) {
1102 new_active
= bond
->primary_slave
;
1105 /* remember where to stop iterating over the slaves */
1106 old_active
= new_active
;
1108 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1109 if (IS_UP(new_active
->dev
)) {
1110 if (new_active
->link
== BOND_LINK_UP
) {
1112 } else if (new_active
->link
== BOND_LINK_BACK
) {
1113 /* link up, but waiting for stabilization */
1114 if (new_active
->delay
< mintime
) {
1115 mintime
= new_active
->delay
;
1116 bestslave
= new_active
;
1126 * change_active_interface - change the active slave into the specified one
1127 * @bond: our bonding struct
1128 * @new: the new slave to make the active one
1130 * Set the new slave to the bond's settings and unset them on the old
1131 * curr_active_slave.
1132 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1134 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1135 * because it is apparently the best available slave we have, even though its
1136 * updelay hasn't timed out yet.
1138 * If new_active is not NULL, caller must hold bond->lock for read and
1139 * curr_slave_lock for write_bh.
1141 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1143 struct slave
*old_active
= bond
->curr_active_slave
;
1145 if (old_active
== new_active
) {
1150 new_active
->jiffies
= jiffies
;
1152 if (new_active
->link
== BOND_LINK_BACK
) {
1153 if (USES_PRIMARY(bond
->params
.mode
)) {
1154 printk(KERN_INFO DRV_NAME
1155 ": %s: making interface %s the new "
1156 "active one %d ms earlier.\n",
1157 bond
->dev
->name
, new_active
->dev
->name
,
1158 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1161 new_active
->delay
= 0;
1162 new_active
->link
= BOND_LINK_UP
;
1164 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1165 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1168 if (bond_is_lb(bond
))
1169 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1171 if (USES_PRIMARY(bond
->params
.mode
)) {
1172 printk(KERN_INFO DRV_NAME
1173 ": %s: making interface %s the new "
1175 bond
->dev
->name
, new_active
->dev
->name
);
1180 if (USES_PRIMARY(bond
->params
.mode
)) {
1181 bond_mc_swap(bond
, new_active
, old_active
);
1184 if (bond_is_lb(bond
)) {
1185 bond_alb_handle_active_change(bond
, new_active
);
1187 bond_set_slave_inactive_flags(old_active
);
1189 bond_set_slave_active_flags(new_active
);
1191 bond
->curr_active_slave
= new_active
;
1194 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1196 bond_set_slave_inactive_flags(old_active
);
1200 bond_set_slave_active_flags(new_active
);
1202 if (bond
->params
.fail_over_mac
)
1203 bond_do_fail_over_mac(bond
, new_active
,
1206 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1207 bond_send_gratuitous_arp(bond
);
1209 bond
->send_unsol_na
= bond
->params
.num_unsol_na
;
1210 bond_send_unsolicited_na(bond
);
1212 write_unlock_bh(&bond
->curr_slave_lock
);
1213 read_unlock(&bond
->lock
);
1215 netdev_bonding_change(bond
->dev
);
1217 read_lock(&bond
->lock
);
1218 write_lock_bh(&bond
->curr_slave_lock
);
1224 * bond_select_active_slave - select a new active slave, if needed
1225 * @bond: our bonding struct
1227 * This functions shoud be called when one of the following occurs:
1228 * - The old curr_active_slave has been released or lost its link.
1229 * - The primary_slave has got its link back.
1230 * - A slave has got its link back and there's no old curr_active_slave.
1232 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1234 void bond_select_active_slave(struct bonding
*bond
)
1236 struct slave
*best_slave
;
1239 best_slave
= bond_find_best_slave(bond
);
1240 if (best_slave
!= bond
->curr_active_slave
) {
1241 bond_change_active_slave(bond
, best_slave
);
1242 rv
= bond_set_carrier(bond
);
1246 if (netif_carrier_ok(bond
->dev
)) {
1247 printk(KERN_INFO DRV_NAME
1248 ": %s: first active interface up!\n",
1251 printk(KERN_INFO DRV_NAME
": %s: "
1252 "now running without any active interface !\n",
1258 /*--------------------------- slave list handling ---------------------------*/
1261 * This function attaches the slave to the end of list.
1263 * bond->lock held for writing by caller.
1265 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1267 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1268 new_slave
->next
= new_slave
;
1269 new_slave
->prev
= new_slave
;
1270 bond
->first_slave
= new_slave
;
1272 new_slave
->next
= bond
->first_slave
;
1273 new_slave
->prev
= bond
->first_slave
->prev
;
1274 new_slave
->next
->prev
= new_slave
;
1275 new_slave
->prev
->next
= new_slave
;
1282 * This function detaches the slave from the list.
1283 * WARNING: no check is made to verify if the slave effectively
1284 * belongs to <bond>.
1285 * Nothing is freed on return, structures are just unchained.
1286 * If any slave pointer in bond was pointing to <slave>,
1287 * it should be changed by the calling function.
1289 * bond->lock held for writing by caller.
1291 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1294 slave
->next
->prev
= slave
->prev
;
1298 slave
->prev
->next
= slave
->next
;
1301 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1302 if (bond
->slave_cnt
> 1) { /* there are more slave */
1303 bond
->first_slave
= slave
->next
;
1305 bond
->first_slave
= NULL
; /* slave was the last one */
1314 /*---------------------------------- IOCTL ----------------------------------*/
1316 static int bond_sethwaddr(struct net_device
*bond_dev
,
1317 struct net_device
*slave_dev
)
1319 pr_debug("bond_dev=%p\n", bond_dev
);
1320 pr_debug("slave_dev=%p\n", slave_dev
);
1321 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1322 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1326 #define BOND_VLAN_FEATURES \
1327 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1328 NETIF_F_HW_VLAN_FILTER)
1331 * Compute the common dev->feature set available to all slaves. Some
1332 * feature bits are managed elsewhere, so preserve those feature bits
1333 * on the master device.
1335 static int bond_compute_features(struct bonding
*bond
)
1337 struct slave
*slave
;
1338 struct net_device
*bond_dev
= bond
->dev
;
1339 unsigned long features
= bond_dev
->features
;
1340 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1341 bond_dev
->hard_header_len
);
1344 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1345 features
|= NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1347 if (!bond
->first_slave
)
1350 features
&= ~NETIF_F_ONE_FOR_ALL
;
1352 bond_for_each_slave(bond
, slave
, i
) {
1353 features
= netdev_increment_features(features
,
1354 slave
->dev
->features
,
1355 NETIF_F_ONE_FOR_ALL
);
1356 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1357 max_hard_header_len
= slave
->dev
->hard_header_len
;
1361 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1362 bond_dev
->features
= netdev_fix_features(features
, NULL
);
1363 bond_dev
->hard_header_len
= max_hard_header_len
;
1368 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1369 struct net_device
*slave_dev
)
1371 struct bonding
*bond
= netdev_priv(bond_dev
);
1373 bond_dev
->header_ops
= slave_dev
->header_ops
;
1375 bond_dev
->type
= slave_dev
->type
;
1376 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1377 bond_dev
->addr_len
= slave_dev
->addr_len
;
1379 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1380 slave_dev
->addr_len
);
1381 bond
->setup_by_slave
= 1;
1384 /* enslave device <slave> to bond device <master> */
1385 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1387 struct bonding
*bond
= netdev_priv(bond_dev
);
1388 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1389 struct slave
*new_slave
= NULL
;
1390 struct dev_mc_list
*dmi
;
1391 struct sockaddr addr
;
1393 int old_features
= bond_dev
->features
;
1396 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1397 slave_ops
->ndo_do_ioctl
== NULL
) {
1398 printk(KERN_WARNING DRV_NAME
1399 ": %s: Warning: no link monitoring support for %s\n",
1400 bond_dev
->name
, slave_dev
->name
);
1403 /* bond must be initialized by bond_open() before enslaving */
1404 if (!(bond_dev
->flags
& IFF_UP
)) {
1405 printk(KERN_WARNING DRV_NAME
1406 " %s: master_dev is not up in bond_enslave\n",
1410 /* already enslaved */
1411 if (slave_dev
->flags
& IFF_SLAVE
) {
1412 pr_debug("Error, Device was already enslaved\n");
1416 /* vlan challenged mutual exclusion */
1417 /* no need to lock since we're protected by rtnl_lock */
1418 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1419 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1420 if (!list_empty(&bond
->vlan_list
)) {
1421 printk(KERN_ERR DRV_NAME
1422 ": %s: Error: cannot enslave VLAN "
1423 "challenged slave %s on VLAN enabled "
1424 "bond %s\n", bond_dev
->name
, slave_dev
->name
,
1428 printk(KERN_WARNING DRV_NAME
1429 ": %s: Warning: enslaved VLAN challenged "
1430 "slave %s. Adding VLANs will be blocked as "
1431 "long as %s is part of bond %s\n",
1432 bond_dev
->name
, slave_dev
->name
, slave_dev
->name
,
1434 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1437 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1438 if (bond
->slave_cnt
== 0) {
1439 /* First slave, and it is not VLAN challenged,
1440 * so remove the block of adding VLANs over the bond.
1442 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1447 * Old ifenslave binaries are no longer supported. These can
1448 * be identified with moderate accurary by the state of the slave:
1449 * the current ifenslave will set the interface down prior to
1450 * enslaving it; the old ifenslave will not.
1452 if ((slave_dev
->flags
& IFF_UP
)) {
1453 printk(KERN_ERR DRV_NAME
": %s is up. "
1454 "This may be due to an out of date ifenslave.\n",
1457 goto err_undo_flags
;
1460 /* set bonding device ether type by slave - bonding netdevices are
1461 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1462 * there is a need to override some of the type dependent attribs/funcs.
1464 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1465 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1467 if (bond
->slave_cnt
== 0) {
1468 if (slave_dev
->type
!= ARPHRD_ETHER
)
1469 bond_setup_by_slave(bond_dev
, slave_dev
);
1470 } else if (bond_dev
->type
!= slave_dev
->type
) {
1471 printk(KERN_ERR DRV_NAME
": %s ether type (%d) is different "
1472 "from other slaves (%d), can not enslave it.\n",
1474 slave_dev
->type
, bond_dev
->type
);
1476 goto err_undo_flags
;
1479 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1480 if (bond
->slave_cnt
== 0) {
1481 printk(KERN_WARNING DRV_NAME
1482 ": %s: Warning: The first slave device "
1483 "specified does not support setting the MAC "
1484 "address. Setting fail_over_mac to active.",
1486 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1487 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1488 printk(KERN_ERR DRV_NAME
1489 ": %s: Error: The slave device specified "
1490 "does not support setting the MAC address, "
1491 "but fail_over_mac is not set to active.\n"
1494 goto err_undo_flags
;
1498 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1501 goto err_undo_flags
;
1504 /* save slave's original flags before calling
1505 * netdev_set_master and dev_open
1507 new_slave
->original_flags
= slave_dev
->flags
;
1510 * Save slave's original ("permanent") mac address for modes
1511 * that need it, and for restoring it upon release, and then
1512 * set it to the master's address
1514 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1516 if (!bond
->params
.fail_over_mac
) {
1518 * Set slave to master's mac address. The application already
1519 * set the master's mac address to that of the first slave
1521 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1522 addr
.sa_family
= slave_dev
->type
;
1523 res
= dev_set_mac_address(slave_dev
, &addr
);
1525 pr_debug("Error %d calling set_mac_address\n", res
);
1530 res
= netdev_set_master(slave_dev
, bond_dev
);
1532 pr_debug("Error %d calling netdev_set_master\n", res
);
1533 goto err_restore_mac
;
1535 /* open the slave since the application closed it */
1536 res
= dev_open(slave_dev
);
1538 pr_debug("Openning slave %s failed\n", slave_dev
->name
);
1539 goto err_unset_master
;
1542 new_slave
->dev
= slave_dev
;
1543 slave_dev
->priv_flags
|= IFF_BONDING
;
1545 if (bond_is_lb(bond
)) {
1546 /* bond_alb_init_slave() must be called before all other stages since
1547 * it might fail and we do not want to have to undo everything
1549 res
= bond_alb_init_slave(bond
, new_slave
);
1555 /* If the mode USES_PRIMARY, then the new slave gets the
1556 * master's promisc (and mc) settings only if it becomes the
1557 * curr_active_slave, and that is taken care of later when calling
1558 * bond_change_active()
1560 if (!USES_PRIMARY(bond
->params
.mode
)) {
1561 /* set promiscuity level to new slave */
1562 if (bond_dev
->flags
& IFF_PROMISC
) {
1563 res
= dev_set_promiscuity(slave_dev
, 1);
1568 /* set allmulti level to new slave */
1569 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1570 res
= dev_set_allmulti(slave_dev
, 1);
1575 netif_addr_lock_bh(bond_dev
);
1576 /* upload master's mc_list to new slave */
1577 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
1578 dev_mc_add (slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
1580 netif_addr_unlock_bh(bond_dev
);
1583 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1584 /* add lacpdu mc addr to mc list */
1585 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1587 dev_mc_add(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
1590 bond_add_vlans_on_slave(bond
, slave_dev
);
1592 write_lock_bh(&bond
->lock
);
1594 bond_attach_slave(bond
, new_slave
);
1596 new_slave
->delay
= 0;
1597 new_slave
->link_failure_count
= 0;
1599 bond_compute_features(bond
);
1601 write_unlock_bh(&bond
->lock
);
1603 read_lock(&bond
->lock
);
1605 new_slave
->last_arp_rx
= jiffies
;
1607 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1608 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1610 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1612 * miimon is set but a bonded network driver
1613 * does not support ETHTOOL/MII and
1614 * arp_interval is not set. Note: if
1615 * use_carrier is enabled, we will never go
1616 * here (because netif_carrier is always
1617 * supported); thus, we don't need to change
1618 * the messages for netif_carrier.
1620 printk(KERN_WARNING DRV_NAME
1621 ": %s: Warning: MII and ETHTOOL support not "
1622 "available for interface %s, and "
1623 "arp_interval/arp_ip_target module parameters "
1624 "not specified, thus bonding will not detect "
1625 "link failures! see bonding.txt for details.\n",
1626 bond_dev
->name
, slave_dev
->name
);
1627 } else if (link_reporting
== -1) {
1628 /* unable get link status using mii/ethtool */
1629 printk(KERN_WARNING DRV_NAME
1630 ": %s: Warning: can't get link status from "
1631 "interface %s; the network driver associated "
1632 "with this interface does not support MII or "
1633 "ETHTOOL link status reporting, thus miimon "
1634 "has no effect on this interface.\n",
1635 bond_dev
->name
, slave_dev
->name
);
1639 /* check for initial state */
1640 if (!bond
->params
.miimon
||
1641 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1642 if (bond
->params
.updelay
) {
1643 pr_debug("Initial state of slave_dev is "
1644 "BOND_LINK_BACK\n");
1645 new_slave
->link
= BOND_LINK_BACK
;
1646 new_slave
->delay
= bond
->params
.updelay
;
1648 pr_debug("Initial state of slave_dev is "
1650 new_slave
->link
= BOND_LINK_UP
;
1652 new_slave
->jiffies
= jiffies
;
1654 pr_debug("Initial state of slave_dev is "
1655 "BOND_LINK_DOWN\n");
1656 new_slave
->link
= BOND_LINK_DOWN
;
1659 if (bond_update_speed_duplex(new_slave
) &&
1660 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1661 printk(KERN_WARNING DRV_NAME
1662 ": %s: Warning: failed to get speed and duplex from %s, "
1663 "assumed to be 100Mb/sec and Full.\n",
1664 bond_dev
->name
, new_slave
->dev
->name
);
1666 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1667 printk(KERN_WARNING DRV_NAME
1668 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1669 "support in base driver for proper aggregator "
1670 "selection.\n", bond_dev
->name
);
1674 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1675 /* if there is a primary slave, remember it */
1676 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1677 bond
->primary_slave
= new_slave
;
1681 write_lock_bh(&bond
->curr_slave_lock
);
1683 switch (bond
->params
.mode
) {
1684 case BOND_MODE_ACTIVEBACKUP
:
1685 bond_set_slave_inactive_flags(new_slave
);
1686 bond_select_active_slave(bond
);
1688 case BOND_MODE_8023AD
:
1689 /* in 802.3ad mode, the internal mechanism
1690 * will activate the slaves in the selected
1693 bond_set_slave_inactive_flags(new_slave
);
1694 /* if this is the first slave */
1695 if (bond
->slave_cnt
== 1) {
1696 SLAVE_AD_INFO(new_slave
).id
= 1;
1697 /* Initialize AD with the number of times that the AD timer is called in 1 second
1698 * can be called only after the mac address of the bond is set
1700 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1701 bond
->params
.lacp_fast
);
1703 SLAVE_AD_INFO(new_slave
).id
=
1704 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1707 bond_3ad_bind_slave(new_slave
);
1711 new_slave
->state
= BOND_STATE_ACTIVE
;
1712 bond_set_slave_inactive_flags(new_slave
);
1715 pr_debug("This slave is always active in trunk mode\n");
1717 /* always active in trunk mode */
1718 new_slave
->state
= BOND_STATE_ACTIVE
;
1720 /* In trunking mode there is little meaning to curr_active_slave
1721 * anyway (it holds no special properties of the bond device),
1722 * so we can change it without calling change_active_interface()
1724 if (!bond
->curr_active_slave
) {
1725 bond
->curr_active_slave
= new_slave
;
1728 } /* switch(bond_mode) */
1730 write_unlock_bh(&bond
->curr_slave_lock
);
1732 bond_set_carrier(bond
);
1734 read_unlock(&bond
->lock
);
1736 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1740 printk(KERN_INFO DRV_NAME
1741 ": %s: enslaving %s as a%s interface with a%s link.\n",
1742 bond_dev
->name
, slave_dev
->name
,
1743 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1744 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1746 /* enslave is successful */
1749 /* Undo stages on error */
1751 dev_close(slave_dev
);
1754 netdev_set_master(slave_dev
, NULL
);
1757 if (!bond
->params
.fail_over_mac
) {
1758 /* XXX TODO - fom follow mode needs to change master's
1759 * MAC if this slave's MAC is in use by the bond, or at
1760 * least print a warning.
1762 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1763 addr
.sa_family
= slave_dev
->type
;
1764 dev_set_mac_address(slave_dev
, &addr
);
1771 bond_dev
->features
= old_features
;
1777 * Try to release the slave device <slave> from the bond device <master>
1778 * It is legal to access curr_active_slave without a lock because all the function
1781 * The rules for slave state should be:
1782 * for Active/Backup:
1783 * Active stays on all backups go down
1784 * for Bonded connections:
1785 * The first up interface should be left on and all others downed.
1787 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1789 struct bonding
*bond
= netdev_priv(bond_dev
);
1790 struct slave
*slave
, *oldcurrent
;
1791 struct sockaddr addr
;
1792 int mac_addr_differ
;
1794 /* slave is not a slave or master is not master of this slave */
1795 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1796 (slave_dev
->master
!= bond_dev
)) {
1797 printk(KERN_ERR DRV_NAME
1798 ": %s: Error: cannot release %s.\n",
1799 bond_dev
->name
, slave_dev
->name
);
1803 write_lock_bh(&bond
->lock
);
1805 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1807 /* not a slave of this bond */
1808 printk(KERN_INFO DRV_NAME
1809 ": %s: %s not enslaved\n",
1810 bond_dev
->name
, slave_dev
->name
);
1811 write_unlock_bh(&bond
->lock
);
1815 if (!bond
->params
.fail_over_mac
) {
1816 mac_addr_differ
= memcmp(bond_dev
->dev_addr
, slave
->perm_hwaddr
,
1818 if (!mac_addr_differ
&& (bond
->slave_cnt
> 1))
1819 printk(KERN_WARNING DRV_NAME
1820 ": %s: Warning: the permanent HWaddr of %s - "
1821 "%pM - is still in use by %s. "
1822 "Set the HWaddr of %s to a different address "
1823 "to avoid conflicts.\n",
1824 bond_dev
->name
, slave_dev
->name
,
1826 bond_dev
->name
, slave_dev
->name
);
1829 /* Inform AD package of unbinding of slave. */
1830 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1831 /* must be called before the slave is
1832 * detached from the list
1834 bond_3ad_unbind_slave(slave
);
1837 printk(KERN_INFO DRV_NAME
1838 ": %s: releasing %s interface %s\n",
1840 (slave
->state
== BOND_STATE_ACTIVE
)
1841 ? "active" : "backup",
1844 oldcurrent
= bond
->curr_active_slave
;
1846 bond
->current_arp_slave
= NULL
;
1848 /* release the slave from its bond */
1849 bond_detach_slave(bond
, slave
);
1851 bond_compute_features(bond
);
1853 if (bond
->primary_slave
== slave
) {
1854 bond
->primary_slave
= NULL
;
1857 if (oldcurrent
== slave
) {
1858 bond_change_active_slave(bond
, NULL
);
1861 if (bond_is_lb(bond
)) {
1862 /* Must be called only after the slave has been
1863 * detached from the list and the curr_active_slave
1864 * has been cleared (if our_slave == old_current),
1865 * but before a new active slave is selected.
1867 write_unlock_bh(&bond
->lock
);
1868 bond_alb_deinit_slave(bond
, slave
);
1869 write_lock_bh(&bond
->lock
);
1872 if (oldcurrent
== slave
) {
1874 * Note that we hold RTNL over this sequence, so there
1875 * is no concern that another slave add/remove event
1878 write_unlock_bh(&bond
->lock
);
1879 read_lock(&bond
->lock
);
1880 write_lock_bh(&bond
->curr_slave_lock
);
1882 bond_select_active_slave(bond
);
1884 write_unlock_bh(&bond
->curr_slave_lock
);
1885 read_unlock(&bond
->lock
);
1886 write_lock_bh(&bond
->lock
);
1889 if (bond
->slave_cnt
== 0) {
1890 bond_set_carrier(bond
);
1892 /* if the last slave was removed, zero the mac address
1893 * of the master so it will be set by the application
1894 * to the mac address of the first slave
1896 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1898 if (list_empty(&bond
->vlan_list
)) {
1899 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1901 printk(KERN_WARNING DRV_NAME
1902 ": %s: Warning: clearing HW address of %s while it "
1903 "still has VLANs.\n",
1904 bond_dev
->name
, bond_dev
->name
);
1905 printk(KERN_WARNING DRV_NAME
1906 ": %s: When re-adding slaves, make sure the bond's "
1907 "HW address matches its VLANs'.\n",
1910 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1911 !bond_has_challenged_slaves(bond
)) {
1912 printk(KERN_INFO DRV_NAME
1913 ": %s: last VLAN challenged slave %s "
1914 "left bond %s. VLAN blocking is removed\n",
1915 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1916 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1919 write_unlock_bh(&bond
->lock
);
1921 /* must do this from outside any spinlocks */
1922 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1924 bond_del_vlans_from_slave(bond
, slave_dev
);
1926 /* If the mode USES_PRIMARY, then we should only remove its
1927 * promisc and mc settings if it was the curr_active_slave, but that was
1928 * already taken care of above when we detached the slave
1930 if (!USES_PRIMARY(bond
->params
.mode
)) {
1931 /* unset promiscuity level from slave */
1932 if (bond_dev
->flags
& IFF_PROMISC
) {
1933 dev_set_promiscuity(slave_dev
, -1);
1936 /* unset allmulti level from slave */
1937 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1938 dev_set_allmulti(slave_dev
, -1);
1941 /* flush master's mc_list from slave */
1942 netif_addr_lock_bh(bond_dev
);
1943 bond_mc_list_flush(bond_dev
, slave_dev
);
1944 netif_addr_unlock_bh(bond_dev
);
1947 netdev_set_master(slave_dev
, NULL
);
1949 /* close slave before restoring its mac address */
1950 dev_close(slave_dev
);
1952 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1953 /* restore original ("permanent") mac address */
1954 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1955 addr
.sa_family
= slave_dev
->type
;
1956 dev_set_mac_address(slave_dev
, &addr
);
1959 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
1960 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
1965 return 0; /* deletion OK */
1969 * Destroy a bonding device.
1970 * Must be under rtnl_lock when this function is called.
1972 void bond_destroy(struct bonding
*bond
)
1974 bond_deinit(bond
->dev
);
1975 bond_destroy_sysfs_entry(bond
);
1976 unregister_netdevice(bond
->dev
);
1979 static void bond_destructor(struct net_device
*bond_dev
)
1981 struct bonding
*bond
= netdev_priv(bond_dev
);
1984 destroy_workqueue(bond
->wq
);
1986 netif_addr_lock_bh(bond_dev
);
1987 bond_mc_list_destroy(bond
);
1988 netif_addr_unlock_bh(bond_dev
);
1990 free_netdev(bond_dev
);
1994 * First release a slave and than destroy the bond if no more slaves iare left.
1995 * Must be under rtnl_lock when this function is called.
1997 int bond_release_and_destroy(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1999 struct bonding
*bond
= netdev_priv(bond_dev
);
2002 ret
= bond_release(bond_dev
, slave_dev
);
2003 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2004 printk(KERN_INFO DRV_NAME
": %s: destroying bond %s.\n",
2005 bond_dev
->name
, bond_dev
->name
);
2012 * This function releases all slaves.
2014 static int bond_release_all(struct net_device
*bond_dev
)
2016 struct bonding
*bond
= netdev_priv(bond_dev
);
2017 struct slave
*slave
;
2018 struct net_device
*slave_dev
;
2019 struct sockaddr addr
;
2021 write_lock_bh(&bond
->lock
);
2023 netif_carrier_off(bond_dev
);
2025 if (bond
->slave_cnt
== 0) {
2029 bond
->current_arp_slave
= NULL
;
2030 bond
->primary_slave
= NULL
;
2031 bond_change_active_slave(bond
, NULL
);
2033 while ((slave
= bond
->first_slave
) != NULL
) {
2034 /* Inform AD package of unbinding of slave
2035 * before slave is detached from the list.
2037 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2038 bond_3ad_unbind_slave(slave
);
2041 slave_dev
= slave
->dev
;
2042 bond_detach_slave(bond
, slave
);
2044 /* now that the slave is detached, unlock and perform
2045 * all the undo steps that should not be called from
2048 write_unlock_bh(&bond
->lock
);
2050 if (bond_is_lb(bond
)) {
2051 /* must be called only after the slave
2052 * has been detached from the list
2054 bond_alb_deinit_slave(bond
, slave
);
2057 bond_compute_features(bond
);
2059 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2060 bond_del_vlans_from_slave(bond
, slave_dev
);
2062 /* If the mode USES_PRIMARY, then we should only remove its
2063 * promisc and mc settings if it was the curr_active_slave, but that was
2064 * already taken care of above when we detached the slave
2066 if (!USES_PRIMARY(bond
->params
.mode
)) {
2067 /* unset promiscuity level from slave */
2068 if (bond_dev
->flags
& IFF_PROMISC
) {
2069 dev_set_promiscuity(slave_dev
, -1);
2072 /* unset allmulti level from slave */
2073 if (bond_dev
->flags
& IFF_ALLMULTI
) {
2074 dev_set_allmulti(slave_dev
, -1);
2077 /* flush master's mc_list from slave */
2078 netif_addr_lock_bh(bond_dev
);
2079 bond_mc_list_flush(bond_dev
, slave_dev
);
2080 netif_addr_unlock_bh(bond_dev
);
2083 netdev_set_master(slave_dev
, NULL
);
2085 /* close slave before restoring its mac address */
2086 dev_close(slave_dev
);
2088 if (!bond
->params
.fail_over_mac
) {
2089 /* restore original ("permanent") mac address*/
2090 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2091 addr
.sa_family
= slave_dev
->type
;
2092 dev_set_mac_address(slave_dev
, &addr
);
2095 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2096 IFF_SLAVE_INACTIVE
);
2100 /* re-acquire the lock before getting the next slave */
2101 write_lock_bh(&bond
->lock
);
2104 /* zero the mac address of the master so it will be
2105 * set by the application to the mac address of the
2108 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2110 if (list_empty(&bond
->vlan_list
)) {
2111 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2113 printk(KERN_WARNING DRV_NAME
2114 ": %s: Warning: clearing HW address of %s while it "
2115 "still has VLANs.\n",
2116 bond_dev
->name
, bond_dev
->name
);
2117 printk(KERN_WARNING DRV_NAME
2118 ": %s: When re-adding slaves, make sure the bond's "
2119 "HW address matches its VLANs'.\n",
2123 printk(KERN_INFO DRV_NAME
2124 ": %s: released all slaves\n",
2128 write_unlock_bh(&bond
->lock
);
2134 * This function changes the active slave to slave <slave_dev>.
2135 * It returns -EINVAL in the following cases.
2136 * - <slave_dev> is not found in the list.
2137 * - There is not active slave now.
2138 * - <slave_dev> is already active.
2139 * - The link state of <slave_dev> is not BOND_LINK_UP.
2140 * - <slave_dev> is not running.
2141 * In these cases, this fuction does nothing.
2142 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2144 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2146 struct bonding
*bond
= netdev_priv(bond_dev
);
2147 struct slave
*old_active
= NULL
;
2148 struct slave
*new_active
= NULL
;
2151 if (!USES_PRIMARY(bond
->params
.mode
)) {
2155 /* Verify that master_dev is indeed the master of slave_dev */
2156 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
2157 (slave_dev
->master
!= bond_dev
)) {
2161 read_lock(&bond
->lock
);
2163 read_lock(&bond
->curr_slave_lock
);
2164 old_active
= bond
->curr_active_slave
;
2165 read_unlock(&bond
->curr_slave_lock
);
2167 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2170 * Changing to the current active: do nothing; return success.
2172 if (new_active
&& (new_active
== old_active
)) {
2173 read_unlock(&bond
->lock
);
2179 (new_active
->link
== BOND_LINK_UP
) &&
2180 IS_UP(new_active
->dev
)) {
2181 write_lock_bh(&bond
->curr_slave_lock
);
2182 bond_change_active_slave(bond
, new_active
);
2183 write_unlock_bh(&bond
->curr_slave_lock
);
2188 read_unlock(&bond
->lock
);
2193 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2195 struct bonding
*bond
= netdev_priv(bond_dev
);
2197 info
->bond_mode
= bond
->params
.mode
;
2198 info
->miimon
= bond
->params
.miimon
;
2200 read_lock(&bond
->lock
);
2201 info
->num_slaves
= bond
->slave_cnt
;
2202 read_unlock(&bond
->lock
);
2207 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2209 struct bonding
*bond
= netdev_priv(bond_dev
);
2210 struct slave
*slave
;
2213 if (info
->slave_id
< 0) {
2217 read_lock(&bond
->lock
);
2219 bond_for_each_slave(bond
, slave
, i
) {
2220 if (i
== (int)info
->slave_id
) {
2226 read_unlock(&bond
->lock
);
2229 strcpy(info
->slave_name
, slave
->dev
->name
);
2230 info
->link
= slave
->link
;
2231 info
->state
= slave
->state
;
2232 info
->link_failure_count
= slave
->link_failure_count
;
2240 /*-------------------------------- Monitoring -------------------------------*/
2243 static int bond_miimon_inspect(struct bonding
*bond
)
2245 struct slave
*slave
;
2246 int i
, link_state
, commit
= 0;
2248 bond_for_each_slave(bond
, slave
, i
) {
2249 slave
->new_link
= BOND_LINK_NOCHANGE
;
2251 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2253 switch (slave
->link
) {
2258 slave
->link
= BOND_LINK_FAIL
;
2259 slave
->delay
= bond
->params
.downdelay
;
2261 printk(KERN_INFO DRV_NAME
2262 ": %s: link status down for %s"
2263 "interface %s, disabling it in %d ms.\n",
2265 (bond
->params
.mode
==
2266 BOND_MODE_ACTIVEBACKUP
) ?
2267 ((slave
->state
== BOND_STATE_ACTIVE
) ?
2268 "active " : "backup ") : "",
2270 bond
->params
.downdelay
* bond
->params
.miimon
);
2273 case BOND_LINK_FAIL
:
2276 * recovered before downdelay expired
2278 slave
->link
= BOND_LINK_UP
;
2279 slave
->jiffies
= jiffies
;
2280 printk(KERN_INFO DRV_NAME
2281 ": %s: link status up again after %d "
2282 "ms for interface %s.\n",
2284 (bond
->params
.downdelay
- slave
->delay
) *
2285 bond
->params
.miimon
,
2290 if (slave
->delay
<= 0) {
2291 slave
->new_link
= BOND_LINK_DOWN
;
2299 case BOND_LINK_DOWN
:
2303 slave
->link
= BOND_LINK_BACK
;
2304 slave
->delay
= bond
->params
.updelay
;
2307 printk(KERN_INFO DRV_NAME
2308 ": %s: link status up for "
2309 "interface %s, enabling it in %d ms.\n",
2310 bond
->dev
->name
, slave
->dev
->name
,
2311 bond
->params
.updelay
*
2312 bond
->params
.miimon
);
2315 case BOND_LINK_BACK
:
2317 slave
->link
= BOND_LINK_DOWN
;
2318 printk(KERN_INFO DRV_NAME
2319 ": %s: link status down again after %d "
2320 "ms for interface %s.\n",
2322 (bond
->params
.updelay
- slave
->delay
) *
2323 bond
->params
.miimon
,
2329 if (slave
->delay
<= 0) {
2330 slave
->new_link
= BOND_LINK_UP
;
2343 static void bond_miimon_commit(struct bonding
*bond
)
2345 struct slave
*slave
;
2348 bond_for_each_slave(bond
, slave
, i
) {
2349 switch (slave
->new_link
) {
2350 case BOND_LINK_NOCHANGE
:
2354 slave
->link
= BOND_LINK_UP
;
2355 slave
->jiffies
= jiffies
;
2357 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2358 /* prevent it from being the active one */
2359 slave
->state
= BOND_STATE_BACKUP
;
2360 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2361 /* make it immediately active */
2362 slave
->state
= BOND_STATE_ACTIVE
;
2363 } else if (slave
!= bond
->primary_slave
) {
2364 /* prevent it from being the active one */
2365 slave
->state
= BOND_STATE_BACKUP
;
2368 printk(KERN_INFO DRV_NAME
2369 ": %s: link status definitely "
2370 "up for interface %s.\n",
2371 bond
->dev
->name
, slave
->dev
->name
);
2373 /* notify ad that the link status has changed */
2374 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2375 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2377 if (bond_is_lb(bond
))
2378 bond_alb_handle_link_change(bond
, slave
,
2381 if (!bond
->curr_active_slave
||
2382 (slave
== bond
->primary_slave
))
2387 case BOND_LINK_DOWN
:
2388 if (slave
->link_failure_count
< UINT_MAX
)
2389 slave
->link_failure_count
++;
2391 slave
->link
= BOND_LINK_DOWN
;
2393 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2394 bond
->params
.mode
== BOND_MODE_8023AD
)
2395 bond_set_slave_inactive_flags(slave
);
2397 printk(KERN_INFO DRV_NAME
2398 ": %s: link status definitely down for "
2399 "interface %s, disabling it\n",
2400 bond
->dev
->name
, slave
->dev
->name
);
2402 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2403 bond_3ad_handle_link_change(slave
,
2406 if (bond
->params
.mode
== BOND_MODE_TLB
||
2407 bond
->params
.mode
== BOND_MODE_ALB
)
2408 bond_alb_handle_link_change(bond
, slave
,
2411 if (slave
== bond
->curr_active_slave
)
2417 printk(KERN_ERR DRV_NAME
2418 ": %s: invalid new link %d on slave %s\n",
2419 bond
->dev
->name
, slave
->new_link
,
2421 slave
->new_link
= BOND_LINK_NOCHANGE
;
2428 write_lock_bh(&bond
->curr_slave_lock
);
2429 bond_select_active_slave(bond
);
2430 write_unlock_bh(&bond
->curr_slave_lock
);
2433 bond_set_carrier(bond
);
2439 * Really a wrapper that splits the mii monitor into two phases: an
2440 * inspection, then (if inspection indicates something needs to be done)
2441 * an acquisition of appropriate locks followed by a commit phase to
2442 * implement whatever link state changes are indicated.
2444 void bond_mii_monitor(struct work_struct
*work
)
2446 struct bonding
*bond
= container_of(work
, struct bonding
,
2449 read_lock(&bond
->lock
);
2450 if (bond
->kill_timers
)
2453 if (bond
->slave_cnt
== 0)
2456 if (bond
->send_grat_arp
) {
2457 read_lock(&bond
->curr_slave_lock
);
2458 bond_send_gratuitous_arp(bond
);
2459 read_unlock(&bond
->curr_slave_lock
);
2462 if (bond
->send_unsol_na
) {
2463 read_lock(&bond
->curr_slave_lock
);
2464 bond_send_unsolicited_na(bond
);
2465 read_unlock(&bond
->curr_slave_lock
);
2468 if (bond_miimon_inspect(bond
)) {
2469 read_unlock(&bond
->lock
);
2471 read_lock(&bond
->lock
);
2473 bond_miimon_commit(bond
);
2475 read_unlock(&bond
->lock
);
2476 rtnl_unlock(); /* might sleep, hold no other locks */
2477 read_lock(&bond
->lock
);
2481 if (bond
->params
.miimon
)
2482 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2483 msecs_to_jiffies(bond
->params
.miimon
));
2485 read_unlock(&bond
->lock
);
2488 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2490 struct in_device
*idev
;
2491 struct in_ifaddr
*ifa
;
2498 idev
= __in_dev_get_rcu(dev
);
2502 ifa
= idev
->ifa_list
;
2506 addr
= ifa
->ifa_local
;
2512 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2514 struct vlan_entry
*vlan
;
2516 if (ip
== bond
->master_ip
)
2519 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2520 if (ip
== vlan
->vlan_ip
)
2528 * We go to the (large) trouble of VLAN tagging ARP frames because
2529 * switches in VLAN mode (especially if ports are configured as
2530 * "native" to a VLAN) might not pass non-tagged frames.
2532 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2534 struct sk_buff
*skb
;
2536 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2537 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2539 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2540 NULL
, slave_dev
->dev_addr
, NULL
);
2543 printk(KERN_ERR DRV_NAME
": ARP packet allocation failed\n");
2547 skb
= vlan_put_tag(skb
, vlan_id
);
2549 printk(KERN_ERR DRV_NAME
": failed to insert VLAN tag\n");
2557 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2560 __be32
*targets
= bond
->params
.arp_targets
;
2561 struct vlan_entry
*vlan
;
2562 struct net_device
*vlan_dev
;
2566 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2569 pr_debug("basa: target %x\n", targets
[i
]);
2570 if (list_empty(&bond
->vlan_list
)) {
2571 pr_debug("basa: empty vlan: arp_send\n");
2572 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2573 bond
->master_ip
, 0);
2578 * If VLANs are configured, we do a route lookup to
2579 * determine which VLAN interface would be used, so we
2580 * can tag the ARP with the proper VLAN tag.
2582 memset(&fl
, 0, sizeof(fl
));
2583 fl
.fl4_dst
= targets
[i
];
2584 fl
.fl4_tos
= RTO_ONLINK
;
2586 rv
= ip_route_output_key(&init_net
, &rt
, &fl
);
2588 if (net_ratelimit()) {
2589 printk(KERN_WARNING DRV_NAME
2590 ": %s: no route to arp_ip_target %pI4\n",
2591 bond
->dev
->name
, &fl
.fl4_dst
);
2597 * This target is not on a VLAN
2599 if (rt
->u
.dst
.dev
== bond
->dev
) {
2601 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2602 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2603 bond
->master_ip
, 0);
2608 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2609 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2610 if (vlan_dev
== rt
->u
.dst
.dev
) {
2611 vlan_id
= vlan
->vlan_id
;
2612 pr_debug("basa: vlan match on %s %d\n",
2613 vlan_dev
->name
, vlan_id
);
2620 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2621 vlan
->vlan_ip
, vlan_id
);
2625 if (net_ratelimit()) {
2626 printk(KERN_WARNING DRV_NAME
2627 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2628 bond
->dev
->name
, &fl
.fl4_dst
,
2629 rt
->u
.dst
.dev
? rt
->u
.dst
.dev
->name
: "NULL");
2636 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2637 * for each VLAN above us.
2639 * Caller must hold curr_slave_lock for read or better
2641 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2643 struct slave
*slave
= bond
->curr_active_slave
;
2644 struct vlan_entry
*vlan
;
2645 struct net_device
*vlan_dev
;
2647 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond
->dev
->name
,
2648 slave
? slave
->dev
->name
: "NULL");
2650 if (!slave
|| !bond
->send_grat_arp
||
2651 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2654 bond
->send_grat_arp
--;
2656 if (bond
->master_ip
) {
2657 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2658 bond
->master_ip
, 0);
2661 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2662 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2663 if (vlan
->vlan_ip
) {
2664 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2665 vlan
->vlan_ip
, vlan
->vlan_id
);
2670 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2673 __be32
*targets
= bond
->params
.arp_targets
;
2675 targets
= bond
->params
.arp_targets
;
2676 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2677 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2678 &sip
, &tip
, i
, &targets
[i
], bond_has_this_ip(bond
, tip
));
2679 if (sip
== targets
[i
]) {
2680 if (bond_has_this_ip(bond
, tip
))
2681 slave
->last_arp_rx
= jiffies
;
2687 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2690 struct slave
*slave
;
2691 struct bonding
*bond
;
2692 unsigned char *arp_ptr
;
2695 if (dev_net(dev
) != &init_net
)
2698 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2701 bond
= netdev_priv(dev
);
2702 read_lock(&bond
->lock
);
2704 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2705 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2706 orig_dev
? orig_dev
->name
: "NULL");
2708 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2709 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2712 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2716 if (arp
->ar_hln
!= dev
->addr_len
||
2717 skb
->pkt_type
== PACKET_OTHERHOST
||
2718 skb
->pkt_type
== PACKET_LOOPBACK
||
2719 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2720 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2724 arp_ptr
= (unsigned char *)(arp
+ 1);
2725 arp_ptr
+= dev
->addr_len
;
2726 memcpy(&sip
, arp_ptr
, 4);
2727 arp_ptr
+= 4 + dev
->addr_len
;
2728 memcpy(&tip
, arp_ptr
, 4);
2730 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2731 bond
->dev
->name
, slave
->dev
->name
, slave
->state
,
2732 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2736 * Backup slaves won't see the ARP reply, but do come through
2737 * here for each ARP probe (so we swap the sip/tip to validate
2738 * the probe). In a "redundant switch, common router" type of
2739 * configuration, the ARP probe will (hopefully) travel from
2740 * the active, through one switch, the router, then the other
2741 * switch before reaching the backup.
2743 if (slave
->state
== BOND_STATE_ACTIVE
)
2744 bond_validate_arp(bond
, slave
, sip
, tip
);
2746 bond_validate_arp(bond
, slave
, tip
, sip
);
2749 read_unlock(&bond
->lock
);
2752 return NET_RX_SUCCESS
;
2756 * this function is called regularly to monitor each slave's link
2757 * ensuring that traffic is being sent and received when arp monitoring
2758 * is used in load-balancing mode. if the adapter has been dormant, then an
2759 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2760 * arp monitoring in active backup mode.
2762 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2764 struct bonding
*bond
= container_of(work
, struct bonding
,
2766 struct slave
*slave
, *oldcurrent
;
2767 int do_failover
= 0;
2771 read_lock(&bond
->lock
);
2773 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2775 if (bond
->kill_timers
) {
2779 if (bond
->slave_cnt
== 0) {
2783 read_lock(&bond
->curr_slave_lock
);
2784 oldcurrent
= bond
->curr_active_slave
;
2785 read_unlock(&bond
->curr_slave_lock
);
2787 /* see if any of the previous devices are up now (i.e. they have
2788 * xmt and rcv traffic). the curr_active_slave does not come into
2789 * the picture unless it is null. also, slave->jiffies is not needed
2790 * here because we send an arp on each slave and give a slave as
2791 * long as it needs to get the tx/rx within the delta.
2792 * TODO: what about up/down delay in arp mode? it wasn't here before
2795 bond_for_each_slave(bond
, slave
, i
) {
2796 if (slave
->link
!= BOND_LINK_UP
) {
2797 if (time_before_eq(jiffies
, slave
->dev
->trans_start
+ delta_in_ticks
) &&
2798 time_before_eq(jiffies
, slave
->dev
->last_rx
+ delta_in_ticks
)) {
2800 slave
->link
= BOND_LINK_UP
;
2801 slave
->state
= BOND_STATE_ACTIVE
;
2803 /* primary_slave has no meaning in round-robin
2804 * mode. the window of a slave being up and
2805 * curr_active_slave being null after enslaving
2809 printk(KERN_INFO DRV_NAME
2810 ": %s: link status definitely "
2811 "up for interface %s, ",
2816 printk(KERN_INFO DRV_NAME
2817 ": %s: interface %s is now up\n",
2823 /* slave->link == BOND_LINK_UP */
2825 /* not all switches will respond to an arp request
2826 * when the source ip is 0, so don't take the link down
2827 * if we don't know our ip yet
2829 if (time_after_eq(jiffies
, slave
->dev
->trans_start
+ 2*delta_in_ticks
) ||
2830 (time_after_eq(jiffies
, slave
->dev
->last_rx
+ 2*delta_in_ticks
))) {
2832 slave
->link
= BOND_LINK_DOWN
;
2833 slave
->state
= BOND_STATE_BACKUP
;
2835 if (slave
->link_failure_count
< UINT_MAX
) {
2836 slave
->link_failure_count
++;
2839 printk(KERN_INFO DRV_NAME
2840 ": %s: interface %s is now down.\n",
2844 if (slave
== oldcurrent
) {
2850 /* note: if switch is in round-robin mode, all links
2851 * must tx arp to ensure all links rx an arp - otherwise
2852 * links may oscillate or not come up at all; if switch is
2853 * in something like xor mode, there is nothing we can
2854 * do - all replies will be rx'ed on same link causing slaves
2855 * to be unstable during low/no traffic periods
2857 if (IS_UP(slave
->dev
)) {
2858 bond_arp_send_all(bond
, slave
);
2863 write_lock_bh(&bond
->curr_slave_lock
);
2865 bond_select_active_slave(bond
);
2867 write_unlock_bh(&bond
->curr_slave_lock
);
2871 if (bond
->params
.arp_interval
)
2872 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2874 read_unlock(&bond
->lock
);
2878 * Called to inspect slaves for active-backup mode ARP monitor link state
2879 * changes. Sets new_link in slaves to specify what action should take
2880 * place for the slave. Returns 0 if no changes are found, >0 if changes
2881 * to link states must be committed.
2883 * Called with bond->lock held for read.
2885 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2887 struct slave
*slave
;
2890 bond_for_each_slave(bond
, slave
, i
) {
2891 slave
->new_link
= BOND_LINK_NOCHANGE
;
2893 if (slave
->link
!= BOND_LINK_UP
) {
2894 if (time_before_eq(jiffies
, slave_last_rx(bond
, slave
) +
2896 slave
->new_link
= BOND_LINK_UP
;
2904 * Give slaves 2*delta after being enslaved or made
2905 * active. This avoids bouncing, as the last receive
2906 * times need a full ARP monitor cycle to be updated.
2908 if (!time_after_eq(jiffies
, slave
->jiffies
+
2909 2 * delta_in_ticks
))
2913 * Backup slave is down if:
2914 * - No current_arp_slave AND
2915 * - more than 3*delta since last receive AND
2916 * - the bond has an IP address
2918 * Note: a non-null current_arp_slave indicates
2919 * the curr_active_slave went down and we are
2920 * searching for a new one; under this condition
2921 * we only take the curr_active_slave down - this
2922 * gives each slave a chance to tx/rx traffic
2923 * before being taken out
2925 if (slave
->state
== BOND_STATE_BACKUP
&&
2926 !bond
->current_arp_slave
&&
2927 time_after(jiffies
, slave_last_rx(bond
, slave
) +
2928 3 * delta_in_ticks
)) {
2929 slave
->new_link
= BOND_LINK_DOWN
;
2934 * Active slave is down if:
2935 * - more than 2*delta since transmitting OR
2936 * - (more than 2*delta since receive AND
2937 * the bond has an IP address)
2939 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
2940 (time_after_eq(jiffies
, slave
->dev
->trans_start
+
2941 2 * delta_in_ticks
) ||
2942 (time_after_eq(jiffies
, slave_last_rx(bond
, slave
)
2943 + 2 * delta_in_ticks
)))) {
2944 slave
->new_link
= BOND_LINK_DOWN
;
2949 read_lock(&bond
->curr_slave_lock
);
2952 * Trigger a commit if the primary option setting has changed.
2954 if (bond
->primary_slave
&&
2955 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
2956 (bond
->primary_slave
->link
== BOND_LINK_UP
))
2959 read_unlock(&bond
->curr_slave_lock
);
2965 * Called to commit link state changes noted by inspection step of
2966 * active-backup mode ARP monitor.
2968 * Called with RTNL and bond->lock for read.
2970 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2972 struct slave
*slave
;
2975 bond_for_each_slave(bond
, slave
, i
) {
2976 switch (slave
->new_link
) {
2977 case BOND_LINK_NOCHANGE
:
2981 write_lock_bh(&bond
->curr_slave_lock
);
2983 if (!bond
->curr_active_slave
&&
2984 time_before_eq(jiffies
, slave
->dev
->trans_start
+
2986 slave
->link
= BOND_LINK_UP
;
2987 bond_change_active_slave(bond
, slave
);
2988 bond
->current_arp_slave
= NULL
;
2990 printk(KERN_INFO DRV_NAME
2991 ": %s: %s is up and now the "
2992 "active interface\n",
2993 bond
->dev
->name
, slave
->dev
->name
);
2995 } else if (bond
->curr_active_slave
!= slave
) {
2996 /* this slave has just come up but we
2997 * already have a current slave; this can
2998 * also happen if bond_enslave adds a new
2999 * slave that is up while we are searching
3002 slave
->link
= BOND_LINK_UP
;
3003 bond_set_slave_inactive_flags(slave
);
3004 bond
->current_arp_slave
= NULL
;
3006 printk(KERN_INFO DRV_NAME
3007 ": %s: backup interface %s is now up\n",
3008 bond
->dev
->name
, slave
->dev
->name
);
3011 write_unlock_bh(&bond
->curr_slave_lock
);
3015 case BOND_LINK_DOWN
:
3016 if (slave
->link_failure_count
< UINT_MAX
)
3017 slave
->link_failure_count
++;
3019 slave
->link
= BOND_LINK_DOWN
;
3021 if (slave
== bond
->curr_active_slave
) {
3022 printk(KERN_INFO DRV_NAME
3023 ": %s: link status down for active "
3024 "interface %s, disabling it\n",
3025 bond
->dev
->name
, slave
->dev
->name
);
3027 bond_set_slave_inactive_flags(slave
);
3029 write_lock_bh(&bond
->curr_slave_lock
);
3031 bond_select_active_slave(bond
);
3032 if (bond
->curr_active_slave
)
3033 bond
->curr_active_slave
->jiffies
=
3036 write_unlock_bh(&bond
->curr_slave_lock
);
3038 bond
->current_arp_slave
= NULL
;
3040 } else if (slave
->state
== BOND_STATE_BACKUP
) {
3041 printk(KERN_INFO DRV_NAME
3042 ": %s: backup interface %s is now down\n",
3043 bond
->dev
->name
, slave
->dev
->name
);
3045 bond_set_slave_inactive_flags(slave
);
3050 printk(KERN_ERR DRV_NAME
3051 ": %s: impossible: new_link %d on slave %s\n",
3052 bond
->dev
->name
, slave
->new_link
,
3058 * No race with changes to primary via sysfs, as we hold rtnl.
3060 if (bond
->primary_slave
&&
3061 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
3062 (bond
->primary_slave
->link
== BOND_LINK_UP
)) {
3063 write_lock_bh(&bond
->curr_slave_lock
);
3064 bond_change_active_slave(bond
, bond
->primary_slave
);
3065 write_unlock_bh(&bond
->curr_slave_lock
);
3068 bond_set_carrier(bond
);
3072 * Send ARP probes for active-backup mode ARP monitor.
3074 * Called with bond->lock held for read.
3076 static void bond_ab_arp_probe(struct bonding
*bond
)
3078 struct slave
*slave
;
3081 read_lock(&bond
->curr_slave_lock
);
3083 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3084 printk("PROBE: c_arp %s && cas %s BAD\n",
3085 bond
->current_arp_slave
->dev
->name
,
3086 bond
->curr_active_slave
->dev
->name
);
3088 if (bond
->curr_active_slave
) {
3089 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3090 read_unlock(&bond
->curr_slave_lock
);
3094 read_unlock(&bond
->curr_slave_lock
);
3096 /* if we don't have a curr_active_slave, search for the next available
3097 * backup slave from the current_arp_slave and make it the candidate
3098 * for becoming the curr_active_slave
3101 if (!bond
->current_arp_slave
) {
3102 bond
->current_arp_slave
= bond
->first_slave
;
3103 if (!bond
->current_arp_slave
)
3107 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3109 /* search for next candidate */
3110 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3111 if (IS_UP(slave
->dev
)) {
3112 slave
->link
= BOND_LINK_BACK
;
3113 bond_set_slave_active_flags(slave
);
3114 bond_arp_send_all(bond
, slave
);
3115 slave
->jiffies
= jiffies
;
3116 bond
->current_arp_slave
= slave
;
3120 /* if the link state is up at this point, we
3121 * mark it down - this can happen if we have
3122 * simultaneous link failures and
3123 * reselect_active_interface doesn't make this
3124 * one the current slave so it is still marked
3125 * up when it is actually down
3127 if (slave
->link
== BOND_LINK_UP
) {
3128 slave
->link
= BOND_LINK_DOWN
;
3129 if (slave
->link_failure_count
< UINT_MAX
)
3130 slave
->link_failure_count
++;
3132 bond_set_slave_inactive_flags(slave
);
3134 printk(KERN_INFO DRV_NAME
3135 ": %s: backup interface %s is now down.\n",
3136 bond
->dev
->name
, slave
->dev
->name
);
3141 void bond_activebackup_arp_mon(struct work_struct
*work
)
3143 struct bonding
*bond
= container_of(work
, struct bonding
,
3147 read_lock(&bond
->lock
);
3149 if (bond
->kill_timers
)
3152 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3154 if (bond
->slave_cnt
== 0)
3157 if (bond
->send_grat_arp
) {
3158 read_lock(&bond
->curr_slave_lock
);
3159 bond_send_gratuitous_arp(bond
);
3160 read_unlock(&bond
->curr_slave_lock
);
3163 if (bond
->send_unsol_na
) {
3164 read_lock(&bond
->curr_slave_lock
);
3165 bond_send_unsolicited_na(bond
);
3166 read_unlock(&bond
->curr_slave_lock
);
3169 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3170 read_unlock(&bond
->lock
);
3172 read_lock(&bond
->lock
);
3174 bond_ab_arp_commit(bond
, delta_in_ticks
);
3176 read_unlock(&bond
->lock
);
3178 read_lock(&bond
->lock
);
3181 bond_ab_arp_probe(bond
);
3184 if (bond
->params
.arp_interval
) {
3185 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3188 read_unlock(&bond
->lock
);
3191 /*------------------------------ proc/seq_file-------------------------------*/
3193 #ifdef CONFIG_PROC_FS
3195 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3197 struct bonding
*bond
= seq
->private;
3199 struct slave
*slave
;
3202 /* make sure the bond won't be taken away */
3203 read_lock(&dev_base_lock
);
3204 read_lock(&bond
->lock
);
3207 return SEQ_START_TOKEN
;
3210 bond_for_each_slave(bond
, slave
, i
) {
3211 if (++off
== *pos
) {
3219 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3221 struct bonding
*bond
= seq
->private;
3222 struct slave
*slave
= v
;
3225 if (v
== SEQ_START_TOKEN
) {
3226 return bond
->first_slave
;
3229 slave
= slave
->next
;
3231 return (slave
== bond
->first_slave
) ? NULL
: slave
;
3234 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
3236 struct bonding
*bond
= seq
->private;
3238 read_unlock(&bond
->lock
);
3239 read_unlock(&dev_base_lock
);
3242 static void bond_info_show_master(struct seq_file
*seq
)
3244 struct bonding
*bond
= seq
->private;
3248 read_lock(&bond
->curr_slave_lock
);
3249 curr
= bond
->curr_active_slave
;
3250 read_unlock(&bond
->curr_slave_lock
);
3252 seq_printf(seq
, "Bonding Mode: %s",
3253 bond_mode_name(bond
->params
.mode
));
3255 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
&&
3256 bond
->params
.fail_over_mac
)
3257 seq_printf(seq
, " (fail_over_mac %s)",
3258 fail_over_mac_tbl
[bond
->params
.fail_over_mac
].modename
);
3260 seq_printf(seq
, "\n");
3262 if (bond
->params
.mode
== BOND_MODE_XOR
||
3263 bond
->params
.mode
== BOND_MODE_8023AD
) {
3264 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
3265 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
3266 bond
->params
.xmit_policy
);
3269 if (USES_PRIMARY(bond
->params
.mode
)) {
3270 seq_printf(seq
, "Primary Slave: %s\n",
3271 (bond
->primary_slave
) ?
3272 bond
->primary_slave
->dev
->name
: "None");
3274 seq_printf(seq
, "Currently Active Slave: %s\n",
3275 (curr
) ? curr
->dev
->name
: "None");
3278 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
3280 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
3281 seq_printf(seq
, "Up Delay (ms): %d\n",
3282 bond
->params
.updelay
* bond
->params
.miimon
);
3283 seq_printf(seq
, "Down Delay (ms): %d\n",
3284 bond
->params
.downdelay
* bond
->params
.miimon
);
3287 /* ARP information */
3288 if(bond
->params
.arp_interval
> 0) {
3290 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
3291 bond
->params
.arp_interval
);
3293 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
3295 for(i
= 0; (i
< BOND_MAX_ARP_TARGETS
) ;i
++) {
3296 if (!bond
->params
.arp_targets
[i
])
3299 seq_printf(seq
, ",");
3300 seq_printf(seq
, " %pI4", &bond
->params
.arp_targets
[i
]);
3303 seq_printf(seq
, "\n");
3306 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3307 struct ad_info ad_info
;
3309 seq_puts(seq
, "\n802.3ad info\n");
3310 seq_printf(seq
, "LACP rate: %s\n",
3311 (bond
->params
.lacp_fast
) ? "fast" : "slow");
3312 seq_printf(seq
, "Aggregator selection policy (ad_select): %s\n",
3313 ad_select_tbl
[bond
->params
.ad_select
].modename
);
3315 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3316 seq_printf(seq
, "bond %s has no active aggregator\n",
3319 seq_printf(seq
, "Active Aggregator Info:\n");
3321 seq_printf(seq
, "\tAggregator ID: %d\n",
3322 ad_info
.aggregator_id
);
3323 seq_printf(seq
, "\tNumber of ports: %d\n",
3325 seq_printf(seq
, "\tActor Key: %d\n",
3327 seq_printf(seq
, "\tPartner Key: %d\n",
3328 ad_info
.partner_key
);
3329 seq_printf(seq
, "\tPartner Mac Address: %pM\n",
3330 ad_info
.partner_system
);
3335 static void bond_info_show_slave(struct seq_file
*seq
, const struct slave
*slave
)
3337 struct bonding
*bond
= seq
->private;
3339 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
3340 seq_printf(seq
, "MII Status: %s\n",
3341 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
3342 seq_printf(seq
, "Link Failure Count: %u\n",
3343 slave
->link_failure_count
);
3345 seq_printf(seq
, "Permanent HW addr: %pM\n", slave
->perm_hwaddr
);
3347 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3348 const struct aggregator
*agg
3349 = SLAVE_AD_INFO(slave
).port
.aggregator
;
3352 seq_printf(seq
, "Aggregator ID: %d\n",
3353 agg
->aggregator_identifier
);
3355 seq_puts(seq
, "Aggregator ID: N/A\n");
3360 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
3362 if (v
== SEQ_START_TOKEN
) {
3363 seq_printf(seq
, "%s\n", version
);
3364 bond_info_show_master(seq
);
3366 bond_info_show_slave(seq
, v
);
3372 static struct seq_operations bond_info_seq_ops
= {
3373 .start
= bond_info_seq_start
,
3374 .next
= bond_info_seq_next
,
3375 .stop
= bond_info_seq_stop
,
3376 .show
= bond_info_seq_show
,
3379 static int bond_info_open(struct inode
*inode
, struct file
*file
)
3381 struct seq_file
*seq
;
3382 struct proc_dir_entry
*proc
;
3385 res
= seq_open(file
, &bond_info_seq_ops
);
3387 /* recover the pointer buried in proc_dir_entry data */
3388 seq
= file
->private_data
;
3390 seq
->private = proc
->data
;
3396 static const struct file_operations bond_info_fops
= {
3397 .owner
= THIS_MODULE
,
3398 .open
= bond_info_open
,
3400 .llseek
= seq_lseek
,
3401 .release
= seq_release
,
3404 static int bond_create_proc_entry(struct bonding
*bond
)
3406 struct net_device
*bond_dev
= bond
->dev
;
3408 if (bond_proc_dir
) {
3409 bond
->proc_entry
= proc_create_data(bond_dev
->name
,
3410 S_IRUGO
, bond_proc_dir
,
3411 &bond_info_fops
, bond
);
3412 if (bond
->proc_entry
== NULL
) {
3413 printk(KERN_WARNING DRV_NAME
3414 ": Warning: Cannot create /proc/net/%s/%s\n",
3415 DRV_NAME
, bond_dev
->name
);
3417 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3424 static void bond_remove_proc_entry(struct bonding
*bond
)
3426 if (bond_proc_dir
&& bond
->proc_entry
) {
3427 remove_proc_entry(bond
->proc_file_name
, bond_proc_dir
);
3428 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3429 bond
->proc_entry
= NULL
;
3433 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3434 * Caller must hold rtnl_lock.
3436 static void bond_create_proc_dir(void)
3438 int len
= strlen(DRV_NAME
);
3440 for (bond_proc_dir
= init_net
.proc_net
->subdir
; bond_proc_dir
;
3441 bond_proc_dir
= bond_proc_dir
->next
) {
3442 if ((bond_proc_dir
->namelen
== len
) &&
3443 !memcmp(bond_proc_dir
->name
, DRV_NAME
, len
)) {
3448 if (!bond_proc_dir
) {
3449 bond_proc_dir
= proc_mkdir(DRV_NAME
, init_net
.proc_net
);
3450 if (bond_proc_dir
) {
3451 bond_proc_dir
->owner
= THIS_MODULE
;
3453 printk(KERN_WARNING DRV_NAME
3454 ": Warning: cannot create /proc/net/%s\n",
3460 /* Destroy the bonding directory under /proc/net, if empty.
3461 * Caller must hold rtnl_lock.
3463 static void bond_destroy_proc_dir(void)
3465 struct proc_dir_entry
*de
;
3467 if (!bond_proc_dir
) {
3471 /* verify that the /proc dir is empty */
3472 for (de
= bond_proc_dir
->subdir
; de
; de
= de
->next
) {
3473 /* ignore . and .. */
3474 if (*(de
->name
) != '.') {
3480 if (bond_proc_dir
->owner
== THIS_MODULE
) {
3481 bond_proc_dir
->owner
= NULL
;
3484 remove_proc_entry(DRV_NAME
, init_net
.proc_net
);
3485 bond_proc_dir
= NULL
;
3488 #endif /* CONFIG_PROC_FS */
3490 /*-------------------------- netdev event handling --------------------------*/
3493 * Change device name
3495 static int bond_event_changename(struct bonding
*bond
)
3497 #ifdef CONFIG_PROC_FS
3498 bond_remove_proc_entry(bond
);
3499 bond_create_proc_entry(bond
);
3501 down_write(&(bonding_rwsem
));
3502 bond_destroy_sysfs_entry(bond
);
3503 bond_create_sysfs_entry(bond
);
3504 up_write(&(bonding_rwsem
));
3508 static int bond_master_netdev_event(unsigned long event
, struct net_device
*bond_dev
)
3510 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3513 case NETDEV_CHANGENAME
:
3514 return bond_event_changename(event_bond
);
3515 case NETDEV_UNREGISTER
:
3516 bond_release_all(event_bond
->dev
);
3525 static int bond_slave_netdev_event(unsigned long event
, struct net_device
*slave_dev
)
3527 struct net_device
*bond_dev
= slave_dev
->master
;
3528 struct bonding
*bond
= netdev_priv(bond_dev
);
3531 case NETDEV_UNREGISTER
:
3533 if (bond
->setup_by_slave
)
3534 bond_release_and_destroy(bond_dev
, slave_dev
);
3536 bond_release(bond_dev
, slave_dev
);
3541 * TODO: is this what we get if somebody
3542 * sets up a hierarchical bond, then rmmod's
3543 * one of the slave bonding devices?
3548 * ... Or is it this?
3551 case NETDEV_CHANGEMTU
:
3553 * TODO: Should slaves be allowed to
3554 * independently alter their MTU? For
3555 * an active-backup bond, slaves need
3556 * not be the same type of device, so
3557 * MTUs may vary. For other modes,
3558 * slaves arguably should have the
3559 * same MTUs. To do this, we'd need to
3560 * take over the slave's change_mtu
3561 * function for the duration of their
3565 case NETDEV_CHANGENAME
:
3567 * TODO: handle changing the primary's name
3570 case NETDEV_FEAT_CHANGE
:
3571 bond_compute_features(bond
);
3581 * bond_netdev_event: handle netdev notifier chain events.
3583 * This function receives events for the netdev chain. The caller (an
3584 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3585 * locks for us to safely manipulate the slave devices (RTNL lock,
3588 static int bond_netdev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3590 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3592 if (dev_net(event_dev
) != &init_net
)
3595 pr_debug("event_dev: %s, event: %lx\n",
3596 (event_dev
? event_dev
->name
: "None"),
3599 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3602 if (event_dev
->flags
& IFF_MASTER
) {
3603 pr_debug("IFF_MASTER\n");
3604 return bond_master_netdev_event(event
, event_dev
);
3607 if (event_dev
->flags
& IFF_SLAVE
) {
3608 pr_debug("IFF_SLAVE\n");
3609 return bond_slave_netdev_event(event
, event_dev
);
3616 * bond_inetaddr_event: handle inetaddr notifier chain events.
3618 * We keep track of device IPs primarily to use as source addresses in
3619 * ARP monitor probes (rather than spewing out broadcasts all the time).
3621 * We track one IP for the main device (if it has one), plus one per VLAN.
3623 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3625 struct in_ifaddr
*ifa
= ptr
;
3626 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3627 struct bonding
*bond
;
3628 struct vlan_entry
*vlan
;
3630 if (dev_net(ifa
->ifa_dev
->dev
) != &init_net
)
3633 list_for_each_entry(bond
, &bond_dev_list
, bond_list
) {
3634 if (bond
->dev
== event_dev
) {
3637 bond
->master_ip
= ifa
->ifa_local
;
3640 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3647 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3648 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3649 if (vlan_dev
== event_dev
) {
3652 vlan
->vlan_ip
= ifa
->ifa_local
;
3656 bond_glean_dev_ip(vlan_dev
);
3667 static struct notifier_block bond_netdev_notifier
= {
3668 .notifier_call
= bond_netdev_event
,
3671 static struct notifier_block bond_inetaddr_notifier
= {
3672 .notifier_call
= bond_inetaddr_event
,
3675 /*-------------------------- Packet type handling ---------------------------*/
3677 /* register to receive lacpdus on a bond */
3678 static void bond_register_lacpdu(struct bonding
*bond
)
3680 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3682 /* initialize packet type */
3683 pk_type
->type
= PKT_TYPE_LACPDU
;
3684 pk_type
->dev
= bond
->dev
;
3685 pk_type
->func
= bond_3ad_lacpdu_recv
;
3687 dev_add_pack(pk_type
);
3690 /* unregister to receive lacpdus on a bond */
3691 static void bond_unregister_lacpdu(struct bonding
*bond
)
3693 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3696 void bond_register_arp(struct bonding
*bond
)
3698 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3703 pt
->type
= htons(ETH_P_ARP
);
3704 pt
->dev
= bond
->dev
;
3705 pt
->func
= bond_arp_rcv
;
3709 void bond_unregister_arp(struct bonding
*bond
)
3711 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3713 dev_remove_pack(pt
);
3717 /*---------------------------- Hashing Policies -----------------------------*/
3720 * Hash for the output device based upon layer 2 and layer 3 data. If
3721 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3723 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
,
3724 struct net_device
*bond_dev
, int count
)
3726 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3727 struct iphdr
*iph
= ip_hdr(skb
);
3729 if (skb
->protocol
== htons(ETH_P_IP
)) {
3730 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3731 (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5])) % count
;
3734 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3738 * Hash for the output device based upon layer 3 and layer 4 data. If
3739 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3740 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3742 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
,
3743 struct net_device
*bond_dev
, int count
)
3745 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3746 struct iphdr
*iph
= ip_hdr(skb
);
3747 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3750 if (skb
->protocol
== htons(ETH_P_IP
)) {
3751 if (!(iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) &&
3752 (iph
->protocol
== IPPROTO_TCP
||
3753 iph
->protocol
== IPPROTO_UDP
)) {
3754 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3756 return (layer4_xor
^
3757 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3761 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3765 * Hash for the output device based upon layer 2 data
3767 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
,
3768 struct net_device
*bond_dev
, int count
)
3770 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3772 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3775 /*-------------------------- Device entry points ----------------------------*/
3777 static int bond_open(struct net_device
*bond_dev
)
3779 struct bonding
*bond
= netdev_priv(bond_dev
);
3781 bond
->kill_timers
= 0;
3783 if (bond_is_lb(bond
)) {
3784 /* bond_alb_initialize must be called before the timer
3787 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3788 /* something went wrong - fail the open operation */
3792 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3793 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3796 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3797 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3798 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3801 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3802 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3803 INIT_DELAYED_WORK(&bond
->arp_work
,
3804 bond_activebackup_arp_mon
);
3806 INIT_DELAYED_WORK(&bond
->arp_work
,
3807 bond_loadbalance_arp_mon
);
3809 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3810 if (bond
->params
.arp_validate
)
3811 bond_register_arp(bond
);
3814 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3815 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3816 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3817 /* register to receive LACPDUs */
3818 bond_register_lacpdu(bond
);
3819 bond_3ad_initiate_agg_selection(bond
, 1);
3825 static int bond_close(struct net_device
*bond_dev
)
3827 struct bonding
*bond
= netdev_priv(bond_dev
);
3829 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3830 /* Unregister the receive of LACPDUs */
3831 bond_unregister_lacpdu(bond
);
3834 if (bond
->params
.arp_validate
)
3835 bond_unregister_arp(bond
);
3837 write_lock_bh(&bond
->lock
);
3839 bond
->send_grat_arp
= 0;
3840 bond
->send_unsol_na
= 0;
3842 /* signal timers not to re-arm */
3843 bond
->kill_timers
= 1;
3845 write_unlock_bh(&bond
->lock
);
3847 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3848 cancel_delayed_work(&bond
->mii_work
);
3851 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3852 cancel_delayed_work(&bond
->arp_work
);
3855 switch (bond
->params
.mode
) {
3856 case BOND_MODE_8023AD
:
3857 cancel_delayed_work(&bond
->ad_work
);
3861 cancel_delayed_work(&bond
->alb_work
);
3868 if (bond_is_lb(bond
)) {
3869 /* Must be called only after all
3870 * slaves have been released
3872 bond_alb_deinitialize(bond
);
3878 static struct net_device_stats
*bond_get_stats(struct net_device
*bond_dev
)
3880 struct bonding
*bond
= netdev_priv(bond_dev
);
3881 struct net_device_stats
*stats
= &bond
->stats
;
3882 struct net_device_stats local_stats
;
3883 struct slave
*slave
;
3886 memset(&local_stats
, 0, sizeof(struct net_device_stats
));
3888 read_lock_bh(&bond
->lock
);
3890 bond_for_each_slave(bond
, slave
, i
) {
3891 const struct net_device_stats
*sstats
= dev_get_stats(slave
->dev
);
3893 local_stats
.rx_packets
+= sstats
->rx_packets
;
3894 local_stats
.rx_bytes
+= sstats
->rx_bytes
;
3895 local_stats
.rx_errors
+= sstats
->rx_errors
;
3896 local_stats
.rx_dropped
+= sstats
->rx_dropped
;
3898 local_stats
.tx_packets
+= sstats
->tx_packets
;
3899 local_stats
.tx_bytes
+= sstats
->tx_bytes
;
3900 local_stats
.tx_errors
+= sstats
->tx_errors
;
3901 local_stats
.tx_dropped
+= sstats
->tx_dropped
;
3903 local_stats
.multicast
+= sstats
->multicast
;
3904 local_stats
.collisions
+= sstats
->collisions
;
3906 local_stats
.rx_length_errors
+= sstats
->rx_length_errors
;
3907 local_stats
.rx_over_errors
+= sstats
->rx_over_errors
;
3908 local_stats
.rx_crc_errors
+= sstats
->rx_crc_errors
;
3909 local_stats
.rx_frame_errors
+= sstats
->rx_frame_errors
;
3910 local_stats
.rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3911 local_stats
.rx_missed_errors
+= sstats
->rx_missed_errors
;
3913 local_stats
.tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3914 local_stats
.tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3915 local_stats
.tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3916 local_stats
.tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3917 local_stats
.tx_window_errors
+= sstats
->tx_window_errors
;
3920 memcpy(stats
, &local_stats
, sizeof(struct net_device_stats
));
3922 read_unlock_bh(&bond
->lock
);
3927 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3929 struct net_device
*slave_dev
= NULL
;
3930 struct ifbond k_binfo
;
3931 struct ifbond __user
*u_binfo
= NULL
;
3932 struct ifslave k_sinfo
;
3933 struct ifslave __user
*u_sinfo
= NULL
;
3934 struct mii_ioctl_data
*mii
= NULL
;
3937 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3938 bond_dev
->name
, cmd
);
3950 * We do this again just in case we were called by SIOCGMIIREG
3951 * instead of SIOCGMIIPHY.
3958 if (mii
->reg_num
== 1) {
3959 struct bonding
*bond
= netdev_priv(bond_dev
);
3961 read_lock(&bond
->lock
);
3962 read_lock(&bond
->curr_slave_lock
);
3963 if (netif_carrier_ok(bond
->dev
)) {
3964 mii
->val_out
= BMSR_LSTATUS
;
3966 read_unlock(&bond
->curr_slave_lock
);
3967 read_unlock(&bond
->lock
);
3971 case BOND_INFO_QUERY_OLD
:
3972 case SIOCBONDINFOQUERY
:
3973 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3975 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
))) {
3979 res
= bond_info_query(bond_dev
, &k_binfo
);
3981 if (copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
))) {
3987 case BOND_SLAVE_INFO_QUERY_OLD
:
3988 case SIOCBONDSLAVEINFOQUERY
:
3989 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3991 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
))) {
3995 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3997 if (copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
))) {
4008 if (!capable(CAP_NET_ADMIN
)) {
4012 down_write(&(bonding_rwsem
));
4013 slave_dev
= dev_get_by_name(&init_net
, ifr
->ifr_slave
);
4015 pr_debug("slave_dev=%p: \n", slave_dev
);
4020 pr_debug("slave_dev->name=%s: \n", slave_dev
->name
);
4022 case BOND_ENSLAVE_OLD
:
4023 case SIOCBONDENSLAVE
:
4024 res
= bond_enslave(bond_dev
, slave_dev
);
4026 case BOND_RELEASE_OLD
:
4027 case SIOCBONDRELEASE
:
4028 res
= bond_release(bond_dev
, slave_dev
);
4030 case BOND_SETHWADDR_OLD
:
4031 case SIOCBONDSETHWADDR
:
4032 res
= bond_sethwaddr(bond_dev
, slave_dev
);
4034 case BOND_CHANGE_ACTIVE_OLD
:
4035 case SIOCBONDCHANGEACTIVE
:
4036 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
4045 up_write(&(bonding_rwsem
));
4049 static void bond_set_multicast_list(struct net_device
*bond_dev
)
4051 struct bonding
*bond
= netdev_priv(bond_dev
);
4052 struct dev_mc_list
*dmi
;
4055 * Do promisc before checking multicast_mode
4057 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
)) {
4059 * FIXME: Need to handle the error when one of the multi-slaves
4062 bond_set_promiscuity(bond
, 1);
4065 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
)) {
4066 bond_set_promiscuity(bond
, -1);
4069 /* set allmulti flag to slaves */
4070 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
)) {
4072 * FIXME: Need to handle the error when one of the multi-slaves
4075 bond_set_allmulti(bond
, 1);
4078 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
)) {
4079 bond_set_allmulti(bond
, -1);
4082 read_lock(&bond
->lock
);
4084 bond
->flags
= bond_dev
->flags
;
4086 /* looking for addresses to add to slaves' mc list */
4087 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
4088 if (!bond_mc_list_find_dmi(dmi
, bond
->mc_list
)) {
4089 bond_mc_add(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4093 /* looking for addresses to delete from slaves' list */
4094 for (dmi
= bond
->mc_list
; dmi
; dmi
= dmi
->next
) {
4095 if (!bond_mc_list_find_dmi(dmi
, bond_dev
->mc_list
)) {
4096 bond_mc_delete(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4100 /* save master's multicast list */
4101 bond_mc_list_destroy(bond
);
4102 bond_mc_list_copy(bond_dev
->mc_list
, bond
, GFP_ATOMIC
);
4104 read_unlock(&bond
->lock
);
4107 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
4109 struct bonding
*bond
= netdev_priv(dev
);
4110 struct slave
*slave
= bond
->first_slave
;
4113 const struct net_device_ops
*slave_ops
4114 = slave
->dev
->netdev_ops
;
4115 if (slave_ops
->ndo_neigh_setup
)
4116 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
4122 * Change the MTU of all of a master's slaves to match the master
4124 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
4126 struct bonding
*bond
= netdev_priv(bond_dev
);
4127 struct slave
*slave
, *stop_at
;
4131 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
4132 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
4134 /* Can't hold bond->lock with bh disabled here since
4135 * some base drivers panic. On the other hand we can't
4136 * hold bond->lock without bh disabled because we'll
4137 * deadlock. The only solution is to rely on the fact
4138 * that we're under rtnl_lock here, and the slaves
4139 * list won't change. This doesn't solve the problem
4140 * of setting the slave's MTU while it is
4141 * transmitting, but the assumption is that the base
4142 * driver can handle that.
4144 * TODO: figure out a way to safely iterate the slaves
4145 * list, but without holding a lock around the actual
4146 * call to the base driver.
4149 bond_for_each_slave(bond
, slave
, i
) {
4150 pr_debug("s %p s->p %p c_m %p\n", slave
,
4151 slave
->prev
, slave
->dev
->netdev_ops
->ndo_change_mtu
);
4153 res
= dev_set_mtu(slave
->dev
, new_mtu
);
4156 /* If we failed to set the slave's mtu to the new value
4157 * we must abort the operation even in ACTIVE_BACKUP
4158 * mode, because if we allow the backup slaves to have
4159 * different mtu values than the active slave we'll
4160 * need to change their mtu when doing a failover. That
4161 * means changing their mtu from timer context, which
4162 * is probably not a good idea.
4164 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4169 bond_dev
->mtu
= new_mtu
;
4174 /* unwind from head to the slave that failed */
4176 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4179 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
4181 pr_debug("unwind err %d dev %s\n", tmp_res
,
4192 * Note that many devices must be down to change the HW address, and
4193 * downing the master releases all slaves. We can make bonds full of
4194 * bonding devices to test this, however.
4196 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4198 struct bonding
*bond
= netdev_priv(bond_dev
);
4199 struct sockaddr
*sa
= addr
, tmp_sa
;
4200 struct slave
*slave
, *stop_at
;
4204 if (bond
->params
.mode
== BOND_MODE_ALB
)
4205 return bond_alb_set_mac_address(bond_dev
, addr
);
4208 pr_debug("bond=%p, name=%s\n", bond
, (bond_dev
? bond_dev
->name
: "None"));
4211 * If fail_over_mac is set to active, do nothing and return
4212 * success. Returning an error causes ifenslave to fail.
4214 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4217 if (!is_valid_ether_addr(sa
->sa_data
)) {
4218 return -EADDRNOTAVAIL
;
4221 /* Can't hold bond->lock with bh disabled here since
4222 * some base drivers panic. On the other hand we can't
4223 * hold bond->lock without bh disabled because we'll
4224 * deadlock. The only solution is to rely on the fact
4225 * that we're under rtnl_lock here, and the slaves
4226 * list won't change. This doesn't solve the problem
4227 * of setting the slave's hw address while it is
4228 * transmitting, but the assumption is that the base
4229 * driver can handle that.
4231 * TODO: figure out a way to safely iterate the slaves
4232 * list, but without holding a lock around the actual
4233 * call to the base driver.
4236 bond_for_each_slave(bond
, slave
, i
) {
4237 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
4238 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
4240 if (slave_ops
->ndo_set_mac_address
== NULL
) {
4242 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
4246 res
= dev_set_mac_address(slave
->dev
, addr
);
4248 /* TODO: consider downing the slave
4250 * User should expect communications
4251 * breakage anyway until ARP finish
4254 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4260 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4264 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4265 tmp_sa
.sa_family
= bond_dev
->type
;
4267 /* unwind from head to the slave that failed */
4269 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4272 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4274 pr_debug("unwind err %d dev %s\n", tmp_res
,
4282 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4284 struct bonding
*bond
= netdev_priv(bond_dev
);
4285 struct slave
*slave
, *start_at
;
4286 int i
, slave_no
, res
= 1;
4288 read_lock(&bond
->lock
);
4290 if (!BOND_IS_OK(bond
)) {
4295 * Concurrent TX may collide on rr_tx_counter; we accept that
4296 * as being rare enough not to justify using an atomic op here
4298 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4300 bond_for_each_slave(bond
, slave
, i
) {
4308 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4309 if (IS_UP(slave
->dev
) &&
4310 (slave
->link
== BOND_LINK_UP
) &&
4311 (slave
->state
== BOND_STATE_ACTIVE
)) {
4312 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4319 /* no suitable interface, frame not sent */
4322 read_unlock(&bond
->lock
);
4328 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4329 * the bond has a usable interface.
4331 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4333 struct bonding
*bond
= netdev_priv(bond_dev
);
4336 read_lock(&bond
->lock
);
4337 read_lock(&bond
->curr_slave_lock
);
4339 if (!BOND_IS_OK(bond
)) {
4343 if (!bond
->curr_active_slave
)
4346 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4350 /* no suitable interface, frame not sent */
4353 read_unlock(&bond
->curr_slave_lock
);
4354 read_unlock(&bond
->lock
);
4359 * In bond_xmit_xor() , we determine the output device by using a pre-
4360 * determined xmit_hash_policy(), If the selected device is not enabled,
4361 * find the next active slave.
4363 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4365 struct bonding
*bond
= netdev_priv(bond_dev
);
4366 struct slave
*slave
, *start_at
;
4371 read_lock(&bond
->lock
);
4373 if (!BOND_IS_OK(bond
)) {
4377 slave_no
= bond
->xmit_hash_policy(skb
, bond_dev
, bond
->slave_cnt
);
4379 bond_for_each_slave(bond
, slave
, i
) {
4388 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4389 if (IS_UP(slave
->dev
) &&
4390 (slave
->link
== BOND_LINK_UP
) &&
4391 (slave
->state
== BOND_STATE_ACTIVE
)) {
4392 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4399 /* no suitable interface, frame not sent */
4402 read_unlock(&bond
->lock
);
4407 * in broadcast mode, we send everything to all usable interfaces.
4409 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4411 struct bonding
*bond
= netdev_priv(bond_dev
);
4412 struct slave
*slave
, *start_at
;
4413 struct net_device
*tx_dev
= NULL
;
4417 read_lock(&bond
->lock
);
4419 if (!BOND_IS_OK(bond
)) {
4423 read_lock(&bond
->curr_slave_lock
);
4424 start_at
= bond
->curr_active_slave
;
4425 read_unlock(&bond
->curr_slave_lock
);
4431 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4432 if (IS_UP(slave
->dev
) &&
4433 (slave
->link
== BOND_LINK_UP
) &&
4434 (slave
->state
== BOND_STATE_ACTIVE
)) {
4436 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4438 printk(KERN_ERR DRV_NAME
4439 ": %s: Error: bond_xmit_broadcast(): "
4440 "skb_clone() failed\n",
4445 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4447 dev_kfree_skb(skb2
);
4451 tx_dev
= slave
->dev
;
4456 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4461 /* no suitable interface, frame not sent */
4464 /* frame sent to all suitable interfaces */
4465 read_unlock(&bond
->lock
);
4469 /*------------------------- Device initialization ---------------------------*/
4471 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4473 switch (bond
->params
.xmit_policy
) {
4474 case BOND_XMIT_POLICY_LAYER23
:
4475 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4477 case BOND_XMIT_POLICY_LAYER34
:
4478 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4480 case BOND_XMIT_POLICY_LAYER2
:
4482 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4487 static int bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4489 const struct bonding
*bond
= netdev_priv(dev
);
4491 switch (bond
->params
.mode
) {
4492 case BOND_MODE_ROUNDROBIN
:
4493 return bond_xmit_roundrobin(skb
, dev
);
4494 case BOND_MODE_ACTIVEBACKUP
:
4495 return bond_xmit_activebackup(skb
, dev
);
4497 return bond_xmit_xor(skb
, dev
);
4498 case BOND_MODE_BROADCAST
:
4499 return bond_xmit_broadcast(skb
, dev
);
4500 case BOND_MODE_8023AD
:
4501 return bond_3ad_xmit_xor(skb
, dev
);
4504 return bond_alb_xmit(skb
, dev
);
4506 /* Should never happen, mode already checked */
4507 printk(KERN_ERR DRV_NAME
": %s: Error: Unknown bonding mode %d\n",
4508 dev
->name
, bond
->params
.mode
);
4511 return NETDEV_TX_OK
;
4517 * set bond mode specific net device operations
4519 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4521 struct net_device
*bond_dev
= bond
->dev
;
4524 case BOND_MODE_ROUNDROBIN
:
4526 case BOND_MODE_ACTIVEBACKUP
:
4529 bond_set_xmit_hash_policy(bond
);
4531 case BOND_MODE_BROADCAST
:
4533 case BOND_MODE_8023AD
:
4534 bond_set_master_3ad_flags(bond
);
4535 bond_set_xmit_hash_policy(bond
);
4538 bond_set_master_alb_flags(bond
);
4543 /* Should never happen, mode already checked */
4544 printk(KERN_ERR DRV_NAME
4545 ": %s: Error: Unknown bonding mode %d\n",
4552 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4553 struct ethtool_drvinfo
*drvinfo
)
4555 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4556 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4557 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4560 static const struct ethtool_ops bond_ethtool_ops
= {
4561 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4562 .get_link
= ethtool_op_get_link
,
4563 .get_tx_csum
= ethtool_op_get_tx_csum
,
4564 .get_sg
= ethtool_op_get_sg
,
4565 .get_tso
= ethtool_op_get_tso
,
4566 .get_ufo
= ethtool_op_get_ufo
,
4567 .get_flags
= ethtool_op_get_flags
,
4570 static const struct net_device_ops bond_netdev_ops
= {
4571 .ndo_open
= bond_open
,
4572 .ndo_stop
= bond_close
,
4573 .ndo_start_xmit
= bond_start_xmit
,
4574 .ndo_get_stats
= bond_get_stats
,
4575 .ndo_do_ioctl
= bond_do_ioctl
,
4576 .ndo_set_multicast_list
= bond_set_multicast_list
,
4577 .ndo_change_mtu
= bond_change_mtu
,
4578 .ndo_set_mac_address
= bond_set_mac_address
,
4579 .ndo_neigh_setup
= bond_neigh_setup
,
4580 .ndo_vlan_rx_register
= bond_vlan_rx_register
,
4581 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4582 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4586 * Does not allocate but creates a /proc entry.
4589 static int bond_init(struct net_device
*bond_dev
, struct bond_params
*params
)
4591 struct bonding
*bond
= netdev_priv(bond_dev
);
4593 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4595 /* initialize rwlocks */
4596 rwlock_init(&bond
->lock
);
4597 rwlock_init(&bond
->curr_slave_lock
);
4599 bond
->params
= *params
; /* copy params struct */
4601 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4605 /* Initialize pointers */
4606 bond
->first_slave
= NULL
;
4607 bond
->curr_active_slave
= NULL
;
4608 bond
->current_arp_slave
= NULL
;
4609 bond
->primary_slave
= NULL
;
4610 bond
->dev
= bond_dev
;
4611 bond
->send_grat_arp
= 0;
4612 bond
->send_unsol_na
= 0;
4613 bond
->setup_by_slave
= 0;
4614 INIT_LIST_HEAD(&bond
->vlan_list
);
4616 /* Initialize the device entry points */
4617 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4618 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4619 bond_set_mode_ops(bond
, bond
->params
.mode
);
4621 bond_dev
->destructor
= bond_destructor
;
4623 /* Initialize the device options */
4624 bond_dev
->tx_queue_len
= 0;
4625 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4626 bond_dev
->priv_flags
|= IFF_BONDING
;
4627 if (bond
->params
.arp_interval
)
4628 bond_dev
->priv_flags
|= IFF_MASTER_ARPMON
;
4630 /* At first, we block adding VLANs. That's the only way to
4631 * prevent problems that occur when adding VLANs over an
4632 * empty bond. The block will be removed once non-challenged
4633 * slaves are enslaved.
4635 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4637 /* don't acquire bond device's netif_tx_lock when
4639 bond_dev
->features
|= NETIF_F_LLTX
;
4641 /* By default, we declare the bond to be fully
4642 * VLAN hardware accelerated capable. Special
4643 * care is taken in the various xmit functions
4644 * when there are slaves that are not hw accel
4647 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4648 NETIF_F_HW_VLAN_RX
|
4649 NETIF_F_HW_VLAN_FILTER
);
4651 #ifdef CONFIG_PROC_FS
4652 bond_create_proc_entry(bond
);
4654 list_add_tail(&bond
->bond_list
, &bond_dev_list
);
4659 static void bond_work_cancel_all(struct bonding
*bond
)
4661 write_lock_bh(&bond
->lock
);
4662 bond
->kill_timers
= 1;
4663 write_unlock_bh(&bond
->lock
);
4665 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4666 cancel_delayed_work(&bond
->mii_work
);
4668 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4669 cancel_delayed_work(&bond
->arp_work
);
4671 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4672 delayed_work_pending(&bond
->alb_work
))
4673 cancel_delayed_work(&bond
->alb_work
);
4675 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4676 delayed_work_pending(&bond
->ad_work
))
4677 cancel_delayed_work(&bond
->ad_work
);
4680 /* De-initialize device specific data.
4681 * Caller must hold rtnl_lock.
4683 static void bond_deinit(struct net_device
*bond_dev
)
4685 struct bonding
*bond
= netdev_priv(bond_dev
);
4687 list_del(&bond
->bond_list
);
4689 bond_work_cancel_all(bond
);
4691 #ifdef CONFIG_PROC_FS
4692 bond_remove_proc_entry(bond
);
4696 /* Unregister and free all bond devices.
4697 * Caller must hold rtnl_lock.
4699 static void bond_free_all(void)
4701 struct bonding
*bond
, *nxt
;
4703 list_for_each_entry_safe(bond
, nxt
, &bond_dev_list
, bond_list
) {
4704 struct net_device
*bond_dev
= bond
->dev
;
4706 bond_work_cancel_all(bond
);
4707 /* Release the bonded slaves */
4708 bond_release_all(bond_dev
);
4712 #ifdef CONFIG_PROC_FS
4713 bond_destroy_proc_dir();
4717 /*------------------------- Module initialization ---------------------------*/
4720 * Convert string input module parms. Accept either the
4721 * number of the mode or its string name. A bit complicated because
4722 * some mode names are substrings of other names, and calls from sysfs
4723 * may have whitespace in the name (trailing newlines, for example).
4725 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4727 int mode
= -1, i
, rv
;
4728 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4730 for (p
= (char *)buf
; *p
; p
++)
4731 if (!(isdigit(*p
) || isspace(*p
)))
4735 rv
= sscanf(buf
, "%20s", modestr
);
4737 rv
= sscanf(buf
, "%d", &mode
);
4742 for (i
= 0; tbl
[i
].modename
; i
++) {
4743 if (mode
== tbl
[i
].mode
)
4745 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4752 static int bond_check_params(struct bond_params
*params
)
4754 int arp_validate_value
, fail_over_mac_value
;
4757 * Convert string parameters.
4760 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4761 if (bond_mode
== -1) {
4762 printk(KERN_ERR DRV_NAME
4763 ": Error: Invalid bonding mode \"%s\"\n",
4764 mode
== NULL
? "NULL" : mode
);
4769 if (xmit_hash_policy
) {
4770 if ((bond_mode
!= BOND_MODE_XOR
) &&
4771 (bond_mode
!= BOND_MODE_8023AD
)) {
4772 printk(KERN_INFO DRV_NAME
4773 ": xor_mode param is irrelevant in mode %s\n",
4774 bond_mode_name(bond_mode
));
4776 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4778 if (xmit_hashtype
== -1) {
4779 printk(KERN_ERR DRV_NAME
4780 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4781 xmit_hash_policy
== NULL
? "NULL" :
4789 if (bond_mode
!= BOND_MODE_8023AD
) {
4790 printk(KERN_INFO DRV_NAME
4791 ": lacp_rate param is irrelevant in mode %s\n",
4792 bond_mode_name(bond_mode
));
4794 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4795 if (lacp_fast
== -1) {
4796 printk(KERN_ERR DRV_NAME
4797 ": Error: Invalid lacp rate \"%s\"\n",
4798 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4805 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4806 if (params
->ad_select
== -1) {
4807 printk(KERN_ERR DRV_NAME
4808 ": Error: Invalid ad_select \"%s\"\n",
4809 ad_select
== NULL
? "NULL" : ad_select
);
4813 if (bond_mode
!= BOND_MODE_8023AD
) {
4814 printk(KERN_WARNING DRV_NAME
4815 ": ad_select param only affects 802.3ad mode\n");
4818 params
->ad_select
= BOND_AD_STABLE
;
4821 if (max_bonds
< 0 || max_bonds
> INT_MAX
) {
4822 printk(KERN_WARNING DRV_NAME
4823 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4824 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4825 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4826 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4830 printk(KERN_WARNING DRV_NAME
4831 ": Warning: miimon module parameter (%d), "
4832 "not in range 0-%d, so it was reset to %d\n",
4833 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4834 miimon
= BOND_LINK_MON_INTERV
;
4838 printk(KERN_WARNING DRV_NAME
4839 ": Warning: updelay module parameter (%d), "
4840 "not in range 0-%d, so it was reset to 0\n",
4845 if (downdelay
< 0) {
4846 printk(KERN_WARNING DRV_NAME
4847 ": Warning: downdelay module parameter (%d), "
4848 "not in range 0-%d, so it was reset to 0\n",
4849 downdelay
, INT_MAX
);
4853 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4854 printk(KERN_WARNING DRV_NAME
4855 ": Warning: use_carrier module parameter (%d), "
4856 "not of valid value (0/1), so it was set to 1\n",
4861 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
4862 printk(KERN_WARNING DRV_NAME
4863 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4864 "was reset to 1 \n", num_grat_arp
);
4868 if (num_unsol_na
< 0 || num_unsol_na
> 255) {
4869 printk(KERN_WARNING DRV_NAME
4870 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4871 "was reset to 1 \n", num_unsol_na
);
4875 /* reset values for 802.3ad */
4876 if (bond_mode
== BOND_MODE_8023AD
) {
4878 printk(KERN_WARNING DRV_NAME
4879 ": Warning: miimon must be specified, "
4880 "otherwise bonding will not detect link "
4881 "failure, speed and duplex which are "
4882 "essential for 802.3ad operation\n");
4883 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4888 /* reset values for TLB/ALB */
4889 if ((bond_mode
== BOND_MODE_TLB
) ||
4890 (bond_mode
== BOND_MODE_ALB
)) {
4892 printk(KERN_WARNING DRV_NAME
4893 ": Warning: miimon must be specified, "
4894 "otherwise bonding will not detect link "
4895 "failure and link speed which are essential "
4896 "for TLB/ALB load balancing\n");
4897 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4902 if (bond_mode
== BOND_MODE_ALB
) {
4903 printk(KERN_NOTICE DRV_NAME
4904 ": In ALB mode you might experience client "
4905 "disconnections upon reconnection of a link if the "
4906 "bonding module updelay parameter (%d msec) is "
4907 "incompatible with the forwarding delay time of the "
4913 if (updelay
|| downdelay
) {
4914 /* just warn the user the up/down delay will have
4915 * no effect since miimon is zero...
4917 printk(KERN_WARNING DRV_NAME
4918 ": Warning: miimon module parameter not set "
4919 "and updelay (%d) or downdelay (%d) module "
4920 "parameter is set; updelay and downdelay have "
4921 "no effect unless miimon is set\n",
4922 updelay
, downdelay
);
4925 /* don't allow arp monitoring */
4927 printk(KERN_WARNING DRV_NAME
4928 ": Warning: miimon (%d) and arp_interval (%d) "
4929 "can't be used simultaneously, disabling ARP "
4931 miimon
, arp_interval
);
4935 if ((updelay
% miimon
) != 0) {
4936 printk(KERN_WARNING DRV_NAME
4937 ": Warning: updelay (%d) is not a multiple "
4938 "of miimon (%d), updelay rounded to %d ms\n",
4939 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4944 if ((downdelay
% miimon
) != 0) {
4945 printk(KERN_WARNING DRV_NAME
4946 ": Warning: downdelay (%d) is not a multiple "
4947 "of miimon (%d), downdelay rounded to %d ms\n",
4949 (downdelay
/ miimon
) * miimon
);
4952 downdelay
/= miimon
;
4955 if (arp_interval
< 0) {
4956 printk(KERN_WARNING DRV_NAME
4957 ": Warning: arp_interval module parameter (%d) "
4958 ", not in range 0-%d, so it was reset to %d\n",
4959 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4960 arp_interval
= BOND_LINK_ARP_INTERV
;
4963 for (arp_ip_count
= 0;
4964 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4966 /* not complete check, but should be good enough to
4968 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4969 printk(KERN_WARNING DRV_NAME
4970 ": Warning: bad arp_ip_target module parameter "
4971 "(%s), ARP monitoring will not be performed\n",
4972 arp_ip_target
[arp_ip_count
]);
4975 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4976 arp_target
[arp_ip_count
] = ip
;
4980 if (arp_interval
&& !arp_ip_count
) {
4981 /* don't allow arping if no arp_ip_target given... */
4982 printk(KERN_WARNING DRV_NAME
4983 ": Warning: arp_interval module parameter (%d) "
4984 "specified without providing an arp_ip_target "
4985 "parameter, arp_interval was reset to 0\n",
4991 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4992 printk(KERN_ERR DRV_NAME
4993 ": arp_validate only supported in active-backup mode\n");
4996 if (!arp_interval
) {
4997 printk(KERN_ERR DRV_NAME
4998 ": arp_validate requires arp_interval\n");
5002 arp_validate_value
= bond_parse_parm(arp_validate
,
5004 if (arp_validate_value
== -1) {
5005 printk(KERN_ERR DRV_NAME
5006 ": Error: invalid arp_validate \"%s\"\n",
5007 arp_validate
== NULL
? "NULL" : arp_validate
);
5011 arp_validate_value
= 0;
5014 printk(KERN_INFO DRV_NAME
5015 ": MII link monitoring set to %d ms\n",
5017 } else if (arp_interval
) {
5020 printk(KERN_INFO DRV_NAME
5021 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5023 arp_validate_tbl
[arp_validate_value
].modename
,
5026 for (i
= 0; i
< arp_ip_count
; i
++)
5027 printk (" %s", arp_ip_target
[i
]);
5031 } else if (max_bonds
) {
5032 /* miimon and arp_interval not set, we need one so things
5033 * work as expected, see bonding.txt for details
5035 printk(KERN_WARNING DRV_NAME
5036 ": Warning: either miimon or arp_interval and "
5037 "arp_ip_target module parameters must be specified, "
5038 "otherwise bonding will not detect link failures! see "
5039 "bonding.txt for details.\n");
5042 if (primary
&& !USES_PRIMARY(bond_mode
)) {
5043 /* currently, using a primary only makes sense
5044 * in active backup, TLB or ALB modes
5046 printk(KERN_WARNING DRV_NAME
5047 ": Warning: %s primary device specified but has no "
5048 "effect in %s mode\n",
5049 primary
, bond_mode_name(bond_mode
));
5053 if (fail_over_mac
) {
5054 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
5056 if (fail_over_mac_value
== -1) {
5057 printk(KERN_ERR DRV_NAME
5058 ": Error: invalid fail_over_mac \"%s\"\n",
5059 arp_validate
== NULL
? "NULL" : arp_validate
);
5063 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
5064 printk(KERN_WARNING DRV_NAME
5065 ": Warning: fail_over_mac only affects "
5066 "active-backup mode.\n");
5068 fail_over_mac_value
= BOND_FOM_NONE
;
5071 /* fill params struct with the proper values */
5072 params
->mode
= bond_mode
;
5073 params
->xmit_policy
= xmit_hashtype
;
5074 params
->miimon
= miimon
;
5075 params
->num_grat_arp
= num_grat_arp
;
5076 params
->num_unsol_na
= num_unsol_na
;
5077 params
->arp_interval
= arp_interval
;
5078 params
->arp_validate
= arp_validate_value
;
5079 params
->updelay
= updelay
;
5080 params
->downdelay
= downdelay
;
5081 params
->use_carrier
= use_carrier
;
5082 params
->lacp_fast
= lacp_fast
;
5083 params
->primary
[0] = 0;
5084 params
->fail_over_mac
= fail_over_mac_value
;
5087 strncpy(params
->primary
, primary
, IFNAMSIZ
);
5088 params
->primary
[IFNAMSIZ
- 1] = 0;
5091 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
5096 static struct lock_class_key bonding_netdev_xmit_lock_key
;
5097 static struct lock_class_key bonding_netdev_addr_lock_key
;
5099 static void bond_set_lockdep_class_one(struct net_device
*dev
,
5100 struct netdev_queue
*txq
,
5103 lockdep_set_class(&txq
->_xmit_lock
,
5104 &bonding_netdev_xmit_lock_key
);
5107 static void bond_set_lockdep_class(struct net_device
*dev
)
5109 lockdep_set_class(&dev
->addr_list_lock
,
5110 &bonding_netdev_addr_lock_key
);
5111 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
5114 /* Create a new bond based on the specified name and bonding parameters.
5115 * If name is NULL, obtain a suitable "bond%d" name for us.
5116 * Caller must NOT hold rtnl_lock; we need to release it here before we
5117 * set up our sysfs entries.
5119 int bond_create(char *name
, struct bond_params
*params
)
5121 struct net_device
*bond_dev
;
5122 struct bonding
*bond
;
5126 down_write(&bonding_rwsem
);
5128 /* Check to see if the bond already exists. */
5130 list_for_each_entry(bond
, &bond_dev_list
, bond_list
)
5131 if (strnicmp(bond
->dev
->name
, name
, IFNAMSIZ
) == 0) {
5132 printk(KERN_ERR DRV_NAME
5133 ": cannot add bond %s; it already exists\n",
5140 bond_dev
= alloc_netdev(sizeof(struct bonding
), name
? name
: "",
5143 printk(KERN_ERR DRV_NAME
5144 ": %s: eek! can't alloc netdev!\n",
5151 res
= dev_alloc_name(bond_dev
, "bond%d");
5156 /* bond_init() must be called after dev_alloc_name() (for the
5157 * /proc files), but before register_netdevice(), because we
5158 * need to set function pointers.
5161 res
= bond_init(bond_dev
, params
);
5166 res
= register_netdevice(bond_dev
);
5171 bond_set_lockdep_class(bond_dev
);
5173 netif_carrier_off(bond_dev
);
5175 up_write(&bonding_rwsem
);
5176 rtnl_unlock(); /* allows sysfs registration of net device */
5177 res
= bond_create_sysfs_entry(netdev_priv(bond_dev
));
5180 down_write(&bonding_rwsem
);
5181 bond_deinit(bond_dev
);
5182 unregister_netdevice(bond_dev
);
5189 bond_deinit(bond_dev
);
5191 free_netdev(bond_dev
);
5193 up_write(&bonding_rwsem
);
5198 static int __init
bonding_init(void)
5202 struct bonding
*bond
;
5204 printk(KERN_INFO
"%s", version
);
5206 res
= bond_check_params(&bonding_defaults
);
5211 #ifdef CONFIG_PROC_FS
5212 bond_create_proc_dir();
5215 init_rwsem(&bonding_rwsem
);
5217 for (i
= 0; i
< max_bonds
; i
++) {
5218 res
= bond_create(NULL
, &bonding_defaults
);
5223 res
= bond_create_sysfs();
5227 register_netdevice_notifier(&bond_netdev_notifier
);
5228 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5229 bond_register_ipv6_notifier();
5233 list_for_each_entry(bond
, &bond_dev_list
, bond_list
) {
5234 bond_work_cancel_all(bond
);
5235 destroy_workqueue(bond
->wq
);
5238 bond_destroy_sysfs();
5248 static void __exit
bonding_exit(void)
5250 unregister_netdevice_notifier(&bond_netdev_notifier
);
5251 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5252 bond_unregister_ipv6_notifier();
5254 bond_destroy_sysfs();
5261 module_init(bonding_init
);
5262 module_exit(bonding_exit
);
5263 MODULE_LICENSE("GPL");
5264 MODULE_VERSION(DRV_VERSION
);
5265 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5266 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5267 MODULE_SUPPORTED_DEVICE("most ethernet devices");