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 /* Try to get link status using Ethtool first. */
699 if (slave_dev
->ethtool_ops
) {
700 if (slave_dev
->ethtool_ops
->get_link
) {
703 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
705 return link
? BMSR_LSTATUS
: 0;
709 /* Ethtool can't be used, fallback to MII ioclts. */
710 ioctl
= slave_ops
->ndo_do_ioctl
;
712 /* TODO: set pointer to correct ioctl on a per team member */
713 /* bases to make this more efficient. that is, once */
714 /* we determine the correct ioctl, we will always */
715 /* call it and not the others for that team */
719 * We cannot assume that SIOCGMIIPHY will also read a
720 * register; not all network drivers (e.g., e100)
724 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
725 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
727 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
728 mii
->reg_num
= MII_BMSR
;
729 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0) {
730 return (mii
->val_out
& BMSR_LSTATUS
);
736 * If reporting, report that either there's no dev->do_ioctl,
737 * or both SIOCGMIIREG and get_link failed (meaning that we
738 * cannot report link status). If not reporting, pretend
741 return (reporting
? -1 : BMSR_LSTATUS
);
744 /*----------------------------- Multicast list ------------------------------*/
747 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
749 static inline int bond_is_dmi_same(struct dev_mc_list
*dmi1
, struct dev_mc_list
*dmi2
)
751 return memcmp(dmi1
->dmi_addr
, dmi2
->dmi_addr
, dmi1
->dmi_addrlen
) == 0 &&
752 dmi1
->dmi_addrlen
== dmi2
->dmi_addrlen
;
756 * returns dmi entry if found, NULL otherwise
758 static struct dev_mc_list
*bond_mc_list_find_dmi(struct dev_mc_list
*dmi
, struct dev_mc_list
*mc_list
)
760 struct dev_mc_list
*idmi
;
762 for (idmi
= mc_list
; idmi
; idmi
= idmi
->next
) {
763 if (bond_is_dmi_same(dmi
, idmi
)) {
772 * Push the promiscuity flag down to appropriate slaves
774 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
777 if (USES_PRIMARY(bond
->params
.mode
)) {
778 /* write lock already acquired */
779 if (bond
->curr_active_slave
) {
780 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
786 bond_for_each_slave(bond
, slave
, i
) {
787 err
= dev_set_promiscuity(slave
->dev
, inc
);
796 * Push the allmulti flag down to all slaves
798 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
801 if (USES_PRIMARY(bond
->params
.mode
)) {
802 /* write lock already acquired */
803 if (bond
->curr_active_slave
) {
804 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
810 bond_for_each_slave(bond
, slave
, i
) {
811 err
= dev_set_allmulti(slave
->dev
, inc
);
820 * Add a Multicast address to slaves
823 static void bond_mc_add(struct bonding
*bond
, void *addr
, int alen
)
825 if (USES_PRIMARY(bond
->params
.mode
)) {
826 /* write lock already acquired */
827 if (bond
->curr_active_slave
) {
828 dev_mc_add(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
833 bond_for_each_slave(bond
, slave
, i
) {
834 dev_mc_add(slave
->dev
, addr
, alen
, 0);
840 * Remove a multicast address from slave
843 static void bond_mc_delete(struct bonding
*bond
, void *addr
, int alen
)
845 if (USES_PRIMARY(bond
->params
.mode
)) {
846 /* write lock already acquired */
847 if (bond
->curr_active_slave
) {
848 dev_mc_delete(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
853 bond_for_each_slave(bond
, slave
, i
) {
854 dev_mc_delete(slave
->dev
, addr
, alen
, 0);
861 * Retrieve the list of registered multicast addresses for the bonding
862 * device and retransmit an IGMP JOIN request to the current active
865 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
867 struct in_device
*in_dev
;
868 struct ip_mc_list
*im
;
871 in_dev
= __in_dev_get_rcu(bond
->dev
);
873 for (im
= in_dev
->mc_list
; im
; im
= im
->next
) {
874 ip_mc_rejoin_group(im
);
882 * Totally destroys the mc_list in bond
884 static void bond_mc_list_destroy(struct bonding
*bond
)
886 struct dev_mc_list
*dmi
;
890 bond
->mc_list
= dmi
->next
;
894 bond
->mc_list
= NULL
;
898 * Copy all the Multicast addresses from src to the bonding device dst
900 static int bond_mc_list_copy(struct dev_mc_list
*mc_list
, struct bonding
*bond
,
903 struct dev_mc_list
*dmi
, *new_dmi
;
905 for (dmi
= mc_list
; dmi
; dmi
= dmi
->next
) {
906 new_dmi
= kmalloc(sizeof(struct dev_mc_list
), gfp_flag
);
909 /* FIXME: Potential memory leak !!! */
913 new_dmi
->next
= bond
->mc_list
;
914 bond
->mc_list
= new_dmi
;
915 new_dmi
->dmi_addrlen
= dmi
->dmi_addrlen
;
916 memcpy(new_dmi
->dmi_addr
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
917 new_dmi
->dmi_users
= dmi
->dmi_users
;
918 new_dmi
->dmi_gusers
= dmi
->dmi_gusers
;
925 * flush all members of flush->mc_list from device dev->mc_list
927 static void bond_mc_list_flush(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
929 struct bonding
*bond
= netdev_priv(bond_dev
);
930 struct dev_mc_list
*dmi
;
932 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
933 dev_mc_delete(slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
936 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
937 /* del lacpdu mc addr from mc list */
938 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
940 dev_mc_delete(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
944 /*--------------------------- Active slave change ---------------------------*/
947 * Update the mc list and multicast-related flags for the new and
948 * old active slaves (if any) according to the multicast mode, and
949 * promiscuous flags unconditionally.
951 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
, struct slave
*old_active
)
953 struct dev_mc_list
*dmi
;
955 if (!USES_PRIMARY(bond
->params
.mode
)) {
956 /* nothing to do - mc list is already up-to-date on
963 if (bond
->dev
->flags
& IFF_PROMISC
) {
964 dev_set_promiscuity(old_active
->dev
, -1);
967 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
968 dev_set_allmulti(old_active
->dev
, -1);
971 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
972 dev_mc_delete(old_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
977 /* FIXME: Signal errors upstream. */
978 if (bond
->dev
->flags
& IFF_PROMISC
) {
979 dev_set_promiscuity(new_active
->dev
, 1);
982 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
983 dev_set_allmulti(new_active
->dev
, 1);
986 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
987 dev_mc_add(new_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
989 bond_resend_igmp_join_requests(bond
);
994 * bond_do_fail_over_mac
996 * Perform special MAC address swapping for fail_over_mac settings
998 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1000 static void bond_do_fail_over_mac(struct bonding
*bond
,
1001 struct slave
*new_active
,
1002 struct slave
*old_active
)
1003 __releases(&bond
->curr_slave_lock
)
1004 __releases(&bond
->lock
)
1005 __acquires(&bond
->lock
)
1006 __acquires(&bond
->curr_slave_lock
)
1008 u8 tmp_mac
[ETH_ALEN
];
1009 struct sockaddr saddr
;
1012 switch (bond
->params
.fail_over_mac
) {
1013 case BOND_FOM_ACTIVE
:
1015 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
1016 new_active
->dev
->addr_len
);
1018 case BOND_FOM_FOLLOW
:
1020 * if new_active && old_active, swap them
1021 * if just old_active, do nothing (going to no active slave)
1022 * if just new_active, set new_active to bond's MAC
1027 write_unlock_bh(&bond
->curr_slave_lock
);
1028 read_unlock(&bond
->lock
);
1031 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
1032 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
1034 saddr
.sa_family
= new_active
->dev
->type
;
1036 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
1037 saddr
.sa_family
= bond
->dev
->type
;
1040 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
1042 printk(KERN_ERR DRV_NAME
1043 ": %s: Error %d setting MAC of slave %s\n",
1044 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1051 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1052 saddr
.sa_family
= old_active
->dev
->type
;
1054 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
1056 printk(KERN_ERR DRV_NAME
1057 ": %s: Error %d setting MAC of slave %s\n",
1058 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1060 read_lock(&bond
->lock
);
1061 write_lock_bh(&bond
->curr_slave_lock
);
1064 printk(KERN_ERR DRV_NAME
1065 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1066 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1074 * find_best_interface - select the best available slave to be the active one
1075 * @bond: our bonding struct
1077 * Warning: Caller must hold curr_slave_lock for writing.
1079 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1081 struct slave
*new_active
, *old_active
;
1082 struct slave
*bestslave
= NULL
;
1083 int mintime
= bond
->params
.updelay
;
1086 new_active
= old_active
= bond
->curr_active_slave
;
1088 if (!new_active
) { /* there were no active slaves left */
1089 if (bond
->slave_cnt
> 0) { /* found one slave */
1090 new_active
= bond
->first_slave
;
1092 return NULL
; /* still no slave, return NULL */
1096 /* first try the primary link; if arping, a link must tx/rx traffic
1097 * before it can be considered the curr_active_slave - also, we would skip
1098 * slaves between the curr_active_slave and primary_slave that may be up
1101 if ((bond
->primary_slave
) &&
1102 (!bond
->params
.arp_interval
) &&
1103 (IS_UP(bond
->primary_slave
->dev
))) {
1104 new_active
= bond
->primary_slave
;
1107 /* remember where to stop iterating over the slaves */
1108 old_active
= new_active
;
1110 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1111 if (IS_UP(new_active
->dev
)) {
1112 if (new_active
->link
== BOND_LINK_UP
) {
1114 } else if (new_active
->link
== BOND_LINK_BACK
) {
1115 /* link up, but waiting for stabilization */
1116 if (new_active
->delay
< mintime
) {
1117 mintime
= new_active
->delay
;
1118 bestslave
= new_active
;
1128 * change_active_interface - change the active slave into the specified one
1129 * @bond: our bonding struct
1130 * @new: the new slave to make the active one
1132 * Set the new slave to the bond's settings and unset them on the old
1133 * curr_active_slave.
1134 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1136 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1137 * because it is apparently the best available slave we have, even though its
1138 * updelay hasn't timed out yet.
1140 * If new_active is not NULL, caller must hold bond->lock for read and
1141 * curr_slave_lock for write_bh.
1143 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1145 struct slave
*old_active
= bond
->curr_active_slave
;
1147 if (old_active
== new_active
) {
1152 new_active
->jiffies
= jiffies
;
1154 if (new_active
->link
== BOND_LINK_BACK
) {
1155 if (USES_PRIMARY(bond
->params
.mode
)) {
1156 printk(KERN_INFO DRV_NAME
1157 ": %s: making interface %s the new "
1158 "active one %d ms earlier.\n",
1159 bond
->dev
->name
, new_active
->dev
->name
,
1160 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1163 new_active
->delay
= 0;
1164 new_active
->link
= BOND_LINK_UP
;
1166 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1167 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1170 if (bond_is_lb(bond
))
1171 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1173 if (USES_PRIMARY(bond
->params
.mode
)) {
1174 printk(KERN_INFO DRV_NAME
1175 ": %s: making interface %s the new "
1177 bond
->dev
->name
, new_active
->dev
->name
);
1182 if (USES_PRIMARY(bond
->params
.mode
)) {
1183 bond_mc_swap(bond
, new_active
, old_active
);
1186 if (bond_is_lb(bond
)) {
1187 bond_alb_handle_active_change(bond
, new_active
);
1189 bond_set_slave_inactive_flags(old_active
);
1191 bond_set_slave_active_flags(new_active
);
1193 bond
->curr_active_slave
= new_active
;
1196 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1198 bond_set_slave_inactive_flags(old_active
);
1202 bond_set_slave_active_flags(new_active
);
1204 if (bond
->params
.fail_over_mac
)
1205 bond_do_fail_over_mac(bond
, new_active
,
1208 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1209 bond_send_gratuitous_arp(bond
);
1211 bond
->send_unsol_na
= bond
->params
.num_unsol_na
;
1212 bond_send_unsolicited_na(bond
);
1214 write_unlock_bh(&bond
->curr_slave_lock
);
1215 read_unlock(&bond
->lock
);
1217 netdev_bonding_change(bond
->dev
);
1219 read_lock(&bond
->lock
);
1220 write_lock_bh(&bond
->curr_slave_lock
);
1226 * bond_select_active_slave - select a new active slave, if needed
1227 * @bond: our bonding struct
1229 * This functions shoud be called when one of the following occurs:
1230 * - The old curr_active_slave has been released or lost its link.
1231 * - The primary_slave has got its link back.
1232 * - A slave has got its link back and there's no old curr_active_slave.
1234 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1236 void bond_select_active_slave(struct bonding
*bond
)
1238 struct slave
*best_slave
;
1241 best_slave
= bond_find_best_slave(bond
);
1242 if (best_slave
!= bond
->curr_active_slave
) {
1243 bond_change_active_slave(bond
, best_slave
);
1244 rv
= bond_set_carrier(bond
);
1248 if (netif_carrier_ok(bond
->dev
)) {
1249 printk(KERN_INFO DRV_NAME
1250 ": %s: first active interface up!\n",
1253 printk(KERN_INFO DRV_NAME
": %s: "
1254 "now running without any active interface !\n",
1260 /*--------------------------- slave list handling ---------------------------*/
1263 * This function attaches the slave to the end of list.
1265 * bond->lock held for writing by caller.
1267 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1269 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1270 new_slave
->next
= new_slave
;
1271 new_slave
->prev
= new_slave
;
1272 bond
->first_slave
= new_slave
;
1274 new_slave
->next
= bond
->first_slave
;
1275 new_slave
->prev
= bond
->first_slave
->prev
;
1276 new_slave
->next
->prev
= new_slave
;
1277 new_slave
->prev
->next
= new_slave
;
1284 * This function detaches the slave from the list.
1285 * WARNING: no check is made to verify if the slave effectively
1286 * belongs to <bond>.
1287 * Nothing is freed on return, structures are just unchained.
1288 * If any slave pointer in bond was pointing to <slave>,
1289 * it should be changed by the calling function.
1291 * bond->lock held for writing by caller.
1293 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1296 slave
->next
->prev
= slave
->prev
;
1300 slave
->prev
->next
= slave
->next
;
1303 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1304 if (bond
->slave_cnt
> 1) { /* there are more slave */
1305 bond
->first_slave
= slave
->next
;
1307 bond
->first_slave
= NULL
; /* slave was the last one */
1316 /*---------------------------------- IOCTL ----------------------------------*/
1318 static int bond_sethwaddr(struct net_device
*bond_dev
,
1319 struct net_device
*slave_dev
)
1321 pr_debug("bond_dev=%p\n", bond_dev
);
1322 pr_debug("slave_dev=%p\n", slave_dev
);
1323 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1324 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1328 #define BOND_VLAN_FEATURES \
1329 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1330 NETIF_F_HW_VLAN_FILTER)
1333 * Compute the common dev->feature set available to all slaves. Some
1334 * feature bits are managed elsewhere, so preserve those feature bits
1335 * on the master device.
1337 static int bond_compute_features(struct bonding
*bond
)
1339 struct slave
*slave
;
1340 struct net_device
*bond_dev
= bond
->dev
;
1341 unsigned long features
= bond_dev
->features
;
1342 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1343 bond_dev
->hard_header_len
);
1346 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1347 features
|= NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1349 if (!bond
->first_slave
)
1352 features
&= ~NETIF_F_ONE_FOR_ALL
;
1354 bond_for_each_slave(bond
, slave
, i
) {
1355 features
= netdev_increment_features(features
,
1356 slave
->dev
->features
,
1357 NETIF_F_ONE_FOR_ALL
);
1358 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1359 max_hard_header_len
= slave
->dev
->hard_header_len
;
1363 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1364 bond_dev
->features
= netdev_fix_features(features
, NULL
);
1365 bond_dev
->hard_header_len
= max_hard_header_len
;
1370 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1371 struct net_device
*slave_dev
)
1373 struct bonding
*bond
= netdev_priv(bond_dev
);
1375 bond_dev
->header_ops
= slave_dev
->header_ops
;
1377 bond_dev
->type
= slave_dev
->type
;
1378 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1379 bond_dev
->addr_len
= slave_dev
->addr_len
;
1381 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1382 slave_dev
->addr_len
);
1383 bond
->setup_by_slave
= 1;
1386 /* enslave device <slave> to bond device <master> */
1387 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1389 struct bonding
*bond
= netdev_priv(bond_dev
);
1390 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1391 struct slave
*new_slave
= NULL
;
1392 struct dev_mc_list
*dmi
;
1393 struct sockaddr addr
;
1395 int old_features
= bond_dev
->features
;
1398 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1399 slave_ops
->ndo_do_ioctl
== NULL
) {
1400 printk(KERN_WARNING DRV_NAME
1401 ": %s: Warning: no link monitoring support for %s\n",
1402 bond_dev
->name
, slave_dev
->name
);
1405 /* bond must be initialized by bond_open() before enslaving */
1406 if (!(bond_dev
->flags
& IFF_UP
)) {
1407 printk(KERN_WARNING DRV_NAME
1408 " %s: master_dev is not up in bond_enslave\n",
1412 /* already enslaved */
1413 if (slave_dev
->flags
& IFF_SLAVE
) {
1414 pr_debug("Error, Device was already enslaved\n");
1418 /* vlan challenged mutual exclusion */
1419 /* no need to lock since we're protected by rtnl_lock */
1420 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1421 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1422 if (!list_empty(&bond
->vlan_list
)) {
1423 printk(KERN_ERR DRV_NAME
1424 ": %s: Error: cannot enslave VLAN "
1425 "challenged slave %s on VLAN enabled "
1426 "bond %s\n", bond_dev
->name
, slave_dev
->name
,
1430 printk(KERN_WARNING DRV_NAME
1431 ": %s: Warning: enslaved VLAN challenged "
1432 "slave %s. Adding VLANs will be blocked as "
1433 "long as %s is part of bond %s\n",
1434 bond_dev
->name
, slave_dev
->name
, slave_dev
->name
,
1436 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1439 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1440 if (bond
->slave_cnt
== 0) {
1441 /* First slave, and it is not VLAN challenged,
1442 * so remove the block of adding VLANs over the bond.
1444 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1449 * Old ifenslave binaries are no longer supported. These can
1450 * be identified with moderate accurary by the state of the slave:
1451 * the current ifenslave will set the interface down prior to
1452 * enslaving it; the old ifenslave will not.
1454 if ((slave_dev
->flags
& IFF_UP
)) {
1455 printk(KERN_ERR DRV_NAME
": %s is up. "
1456 "This may be due to an out of date ifenslave.\n",
1459 goto err_undo_flags
;
1462 /* set bonding device ether type by slave - bonding netdevices are
1463 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1464 * there is a need to override some of the type dependent attribs/funcs.
1466 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1467 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1469 if (bond
->slave_cnt
== 0) {
1470 if (slave_dev
->type
!= ARPHRD_ETHER
)
1471 bond_setup_by_slave(bond_dev
, slave_dev
);
1472 } else if (bond_dev
->type
!= slave_dev
->type
) {
1473 printk(KERN_ERR DRV_NAME
": %s ether type (%d) is different "
1474 "from other slaves (%d), can not enslave it.\n",
1476 slave_dev
->type
, bond_dev
->type
);
1478 goto err_undo_flags
;
1481 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1482 if (bond
->slave_cnt
== 0) {
1483 printk(KERN_WARNING DRV_NAME
1484 ": %s: Warning: The first slave device "
1485 "specified does not support setting the MAC "
1486 "address. Setting fail_over_mac to active.",
1488 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1489 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1490 printk(KERN_ERR DRV_NAME
1491 ": %s: Error: The slave device specified "
1492 "does not support setting the MAC address, "
1493 "but fail_over_mac is not set to active.\n"
1496 goto err_undo_flags
;
1500 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1503 goto err_undo_flags
;
1506 /* save slave's original flags before calling
1507 * netdev_set_master and dev_open
1509 new_slave
->original_flags
= slave_dev
->flags
;
1512 * Save slave's original ("permanent") mac address for modes
1513 * that need it, and for restoring it upon release, and then
1514 * set it to the master's address
1516 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1518 if (!bond
->params
.fail_over_mac
) {
1520 * Set slave to master's mac address. The application already
1521 * set the master's mac address to that of the first slave
1523 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1524 addr
.sa_family
= slave_dev
->type
;
1525 res
= dev_set_mac_address(slave_dev
, &addr
);
1527 pr_debug("Error %d calling set_mac_address\n", res
);
1532 res
= netdev_set_master(slave_dev
, bond_dev
);
1534 pr_debug("Error %d calling netdev_set_master\n", res
);
1535 goto err_restore_mac
;
1537 /* open the slave since the application closed it */
1538 res
= dev_open(slave_dev
);
1540 pr_debug("Openning slave %s failed\n", slave_dev
->name
);
1541 goto err_unset_master
;
1544 new_slave
->dev
= slave_dev
;
1545 slave_dev
->priv_flags
|= IFF_BONDING
;
1547 if (bond_is_lb(bond
)) {
1548 /* bond_alb_init_slave() must be called before all other stages since
1549 * it might fail and we do not want to have to undo everything
1551 res
= bond_alb_init_slave(bond
, new_slave
);
1557 /* If the mode USES_PRIMARY, then the new slave gets the
1558 * master's promisc (and mc) settings only if it becomes the
1559 * curr_active_slave, and that is taken care of later when calling
1560 * bond_change_active()
1562 if (!USES_PRIMARY(bond
->params
.mode
)) {
1563 /* set promiscuity level to new slave */
1564 if (bond_dev
->flags
& IFF_PROMISC
) {
1565 res
= dev_set_promiscuity(slave_dev
, 1);
1570 /* set allmulti level to new slave */
1571 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1572 res
= dev_set_allmulti(slave_dev
, 1);
1577 netif_addr_lock_bh(bond_dev
);
1578 /* upload master's mc_list to new slave */
1579 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
1580 dev_mc_add (slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
1582 netif_addr_unlock_bh(bond_dev
);
1585 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1586 /* add lacpdu mc addr to mc list */
1587 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1589 dev_mc_add(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
1592 bond_add_vlans_on_slave(bond
, slave_dev
);
1594 write_lock_bh(&bond
->lock
);
1596 bond_attach_slave(bond
, new_slave
);
1598 new_slave
->delay
= 0;
1599 new_slave
->link_failure_count
= 0;
1601 bond_compute_features(bond
);
1603 write_unlock_bh(&bond
->lock
);
1605 read_lock(&bond
->lock
);
1607 new_slave
->last_arp_rx
= jiffies
;
1609 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1610 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1612 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1614 * miimon is set but a bonded network driver
1615 * does not support ETHTOOL/MII and
1616 * arp_interval is not set. Note: if
1617 * use_carrier is enabled, we will never go
1618 * here (because netif_carrier is always
1619 * supported); thus, we don't need to change
1620 * the messages for netif_carrier.
1622 printk(KERN_WARNING DRV_NAME
1623 ": %s: Warning: MII and ETHTOOL support not "
1624 "available for interface %s, and "
1625 "arp_interval/arp_ip_target module parameters "
1626 "not specified, thus bonding will not detect "
1627 "link failures! see bonding.txt for details.\n",
1628 bond_dev
->name
, slave_dev
->name
);
1629 } else if (link_reporting
== -1) {
1630 /* unable get link status using mii/ethtool */
1631 printk(KERN_WARNING DRV_NAME
1632 ": %s: Warning: can't get link status from "
1633 "interface %s; the network driver associated "
1634 "with this interface does not support MII or "
1635 "ETHTOOL link status reporting, thus miimon "
1636 "has no effect on this interface.\n",
1637 bond_dev
->name
, slave_dev
->name
);
1641 /* check for initial state */
1642 if (!bond
->params
.miimon
||
1643 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1644 if (bond
->params
.updelay
) {
1645 pr_debug("Initial state of slave_dev is "
1646 "BOND_LINK_BACK\n");
1647 new_slave
->link
= BOND_LINK_BACK
;
1648 new_slave
->delay
= bond
->params
.updelay
;
1650 pr_debug("Initial state of slave_dev is "
1652 new_slave
->link
= BOND_LINK_UP
;
1654 new_slave
->jiffies
= jiffies
;
1656 pr_debug("Initial state of slave_dev is "
1657 "BOND_LINK_DOWN\n");
1658 new_slave
->link
= BOND_LINK_DOWN
;
1661 if (bond_update_speed_duplex(new_slave
) &&
1662 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1663 printk(KERN_WARNING DRV_NAME
1664 ": %s: Warning: failed to get speed and duplex from %s, "
1665 "assumed to be 100Mb/sec and Full.\n",
1666 bond_dev
->name
, new_slave
->dev
->name
);
1668 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1669 printk(KERN_WARNING DRV_NAME
1670 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1671 "support in base driver for proper aggregator "
1672 "selection.\n", bond_dev
->name
);
1676 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1677 /* if there is a primary slave, remember it */
1678 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1679 bond
->primary_slave
= new_slave
;
1683 write_lock_bh(&bond
->curr_slave_lock
);
1685 switch (bond
->params
.mode
) {
1686 case BOND_MODE_ACTIVEBACKUP
:
1687 bond_set_slave_inactive_flags(new_slave
);
1688 bond_select_active_slave(bond
);
1690 case BOND_MODE_8023AD
:
1691 /* in 802.3ad mode, the internal mechanism
1692 * will activate the slaves in the selected
1695 bond_set_slave_inactive_flags(new_slave
);
1696 /* if this is the first slave */
1697 if (bond
->slave_cnt
== 1) {
1698 SLAVE_AD_INFO(new_slave
).id
= 1;
1699 /* Initialize AD with the number of times that the AD timer is called in 1 second
1700 * can be called only after the mac address of the bond is set
1702 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1703 bond
->params
.lacp_fast
);
1705 SLAVE_AD_INFO(new_slave
).id
=
1706 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1709 bond_3ad_bind_slave(new_slave
);
1713 new_slave
->state
= BOND_STATE_ACTIVE
;
1714 bond_set_slave_inactive_flags(new_slave
);
1715 bond_select_active_slave(bond
);
1718 pr_debug("This slave is always active in trunk mode\n");
1720 /* always active in trunk mode */
1721 new_slave
->state
= BOND_STATE_ACTIVE
;
1723 /* In trunking mode there is little meaning to curr_active_slave
1724 * anyway (it holds no special properties of the bond device),
1725 * so we can change it without calling change_active_interface()
1727 if (!bond
->curr_active_slave
) {
1728 bond
->curr_active_slave
= new_slave
;
1731 } /* switch(bond_mode) */
1733 write_unlock_bh(&bond
->curr_slave_lock
);
1735 bond_set_carrier(bond
);
1737 read_unlock(&bond
->lock
);
1739 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1743 printk(KERN_INFO DRV_NAME
1744 ": %s: enslaving %s as a%s interface with a%s link.\n",
1745 bond_dev
->name
, slave_dev
->name
,
1746 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1747 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1749 /* enslave is successful */
1752 /* Undo stages on error */
1754 dev_close(slave_dev
);
1757 netdev_set_master(slave_dev
, NULL
);
1760 if (!bond
->params
.fail_over_mac
) {
1761 /* XXX TODO - fom follow mode needs to change master's
1762 * MAC if this slave's MAC is in use by the bond, or at
1763 * least print a warning.
1765 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1766 addr
.sa_family
= slave_dev
->type
;
1767 dev_set_mac_address(slave_dev
, &addr
);
1774 bond_dev
->features
= old_features
;
1780 * Try to release the slave device <slave> from the bond device <master>
1781 * It is legal to access curr_active_slave without a lock because all the function
1784 * The rules for slave state should be:
1785 * for Active/Backup:
1786 * Active stays on all backups go down
1787 * for Bonded connections:
1788 * The first up interface should be left on and all others downed.
1790 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1792 struct bonding
*bond
= netdev_priv(bond_dev
);
1793 struct slave
*slave
, *oldcurrent
;
1794 struct sockaddr addr
;
1795 int mac_addr_differ
;
1797 /* slave is not a slave or master is not master of this slave */
1798 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1799 (slave_dev
->master
!= bond_dev
)) {
1800 printk(KERN_ERR DRV_NAME
1801 ": %s: Error: cannot release %s.\n",
1802 bond_dev
->name
, slave_dev
->name
);
1806 write_lock_bh(&bond
->lock
);
1808 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1810 /* not a slave of this bond */
1811 printk(KERN_INFO DRV_NAME
1812 ": %s: %s not enslaved\n",
1813 bond_dev
->name
, slave_dev
->name
);
1814 write_unlock_bh(&bond
->lock
);
1818 if (!bond
->params
.fail_over_mac
) {
1819 mac_addr_differ
= memcmp(bond_dev
->dev_addr
, slave
->perm_hwaddr
,
1821 if (!mac_addr_differ
&& (bond
->slave_cnt
> 1))
1822 printk(KERN_WARNING DRV_NAME
1823 ": %s: Warning: the permanent HWaddr of %s - "
1824 "%pM - is still in use by %s. "
1825 "Set the HWaddr of %s to a different address "
1826 "to avoid conflicts.\n",
1827 bond_dev
->name
, slave_dev
->name
,
1829 bond_dev
->name
, slave_dev
->name
);
1832 /* Inform AD package of unbinding of slave. */
1833 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1834 /* must be called before the slave is
1835 * detached from the list
1837 bond_3ad_unbind_slave(slave
);
1840 printk(KERN_INFO DRV_NAME
1841 ": %s: releasing %s interface %s\n",
1843 (slave
->state
== BOND_STATE_ACTIVE
)
1844 ? "active" : "backup",
1847 oldcurrent
= bond
->curr_active_slave
;
1849 bond
->current_arp_slave
= NULL
;
1851 /* release the slave from its bond */
1852 bond_detach_slave(bond
, slave
);
1854 bond_compute_features(bond
);
1856 if (bond
->primary_slave
== slave
) {
1857 bond
->primary_slave
= NULL
;
1860 if (oldcurrent
== slave
) {
1861 bond_change_active_slave(bond
, NULL
);
1864 if (bond_is_lb(bond
)) {
1865 /* Must be called only after the slave has been
1866 * detached from the list and the curr_active_slave
1867 * has been cleared (if our_slave == old_current),
1868 * but before a new active slave is selected.
1870 write_unlock_bh(&bond
->lock
);
1871 bond_alb_deinit_slave(bond
, slave
);
1872 write_lock_bh(&bond
->lock
);
1875 if (oldcurrent
== slave
) {
1877 * Note that we hold RTNL over this sequence, so there
1878 * is no concern that another slave add/remove event
1881 write_unlock_bh(&bond
->lock
);
1882 read_lock(&bond
->lock
);
1883 write_lock_bh(&bond
->curr_slave_lock
);
1885 bond_select_active_slave(bond
);
1887 write_unlock_bh(&bond
->curr_slave_lock
);
1888 read_unlock(&bond
->lock
);
1889 write_lock_bh(&bond
->lock
);
1892 if (bond
->slave_cnt
== 0) {
1893 bond_set_carrier(bond
);
1895 /* if the last slave was removed, zero the mac address
1896 * of the master so it will be set by the application
1897 * to the mac address of the first slave
1899 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1901 if (list_empty(&bond
->vlan_list
)) {
1902 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1904 printk(KERN_WARNING DRV_NAME
1905 ": %s: Warning: clearing HW address of %s while it "
1906 "still has VLANs.\n",
1907 bond_dev
->name
, bond_dev
->name
);
1908 printk(KERN_WARNING DRV_NAME
1909 ": %s: When re-adding slaves, make sure the bond's "
1910 "HW address matches its VLANs'.\n",
1913 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1914 !bond_has_challenged_slaves(bond
)) {
1915 printk(KERN_INFO DRV_NAME
1916 ": %s: last VLAN challenged slave %s "
1917 "left bond %s. VLAN blocking is removed\n",
1918 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1919 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1922 write_unlock_bh(&bond
->lock
);
1924 /* must do this from outside any spinlocks */
1925 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1927 bond_del_vlans_from_slave(bond
, slave_dev
);
1929 /* If the mode USES_PRIMARY, then we should only remove its
1930 * promisc and mc settings if it was the curr_active_slave, but that was
1931 * already taken care of above when we detached the slave
1933 if (!USES_PRIMARY(bond
->params
.mode
)) {
1934 /* unset promiscuity level from slave */
1935 if (bond_dev
->flags
& IFF_PROMISC
) {
1936 dev_set_promiscuity(slave_dev
, -1);
1939 /* unset allmulti level from slave */
1940 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1941 dev_set_allmulti(slave_dev
, -1);
1944 /* flush master's mc_list from slave */
1945 netif_addr_lock_bh(bond_dev
);
1946 bond_mc_list_flush(bond_dev
, slave_dev
);
1947 netif_addr_unlock_bh(bond_dev
);
1950 netdev_set_master(slave_dev
, NULL
);
1952 /* close slave before restoring its mac address */
1953 dev_close(slave_dev
);
1955 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1956 /* restore original ("permanent") mac address */
1957 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1958 addr
.sa_family
= slave_dev
->type
;
1959 dev_set_mac_address(slave_dev
, &addr
);
1962 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
1963 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
1968 return 0; /* deletion OK */
1972 * Destroy a bonding device.
1973 * Must be under rtnl_lock when this function is called.
1975 void bond_destroy(struct bonding
*bond
)
1977 bond_deinit(bond
->dev
);
1978 bond_destroy_sysfs_entry(bond
);
1979 unregister_netdevice(bond
->dev
);
1982 static void bond_destructor(struct net_device
*bond_dev
)
1984 struct bonding
*bond
= netdev_priv(bond_dev
);
1987 destroy_workqueue(bond
->wq
);
1989 netif_addr_lock_bh(bond_dev
);
1990 bond_mc_list_destroy(bond
);
1991 netif_addr_unlock_bh(bond_dev
);
1993 free_netdev(bond_dev
);
1997 * First release a slave and than destroy the bond if no more slaves iare left.
1998 * Must be under rtnl_lock when this function is called.
2000 int bond_release_and_destroy(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2002 struct bonding
*bond
= netdev_priv(bond_dev
);
2005 ret
= bond_release(bond_dev
, slave_dev
);
2006 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2007 printk(KERN_INFO DRV_NAME
": %s: destroying bond %s.\n",
2008 bond_dev
->name
, bond_dev
->name
);
2015 * This function releases all slaves.
2017 static int bond_release_all(struct net_device
*bond_dev
)
2019 struct bonding
*bond
= netdev_priv(bond_dev
);
2020 struct slave
*slave
;
2021 struct net_device
*slave_dev
;
2022 struct sockaddr addr
;
2024 write_lock_bh(&bond
->lock
);
2026 netif_carrier_off(bond_dev
);
2028 if (bond
->slave_cnt
== 0) {
2032 bond
->current_arp_slave
= NULL
;
2033 bond
->primary_slave
= NULL
;
2034 bond_change_active_slave(bond
, NULL
);
2036 while ((slave
= bond
->first_slave
) != NULL
) {
2037 /* Inform AD package of unbinding of slave
2038 * before slave is detached from the list.
2040 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2041 bond_3ad_unbind_slave(slave
);
2044 slave_dev
= slave
->dev
;
2045 bond_detach_slave(bond
, slave
);
2047 /* now that the slave is detached, unlock and perform
2048 * all the undo steps that should not be called from
2051 write_unlock_bh(&bond
->lock
);
2053 if (bond_is_lb(bond
)) {
2054 /* must be called only after the slave
2055 * has been detached from the list
2057 bond_alb_deinit_slave(bond
, slave
);
2060 bond_compute_features(bond
);
2062 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2063 bond_del_vlans_from_slave(bond
, slave_dev
);
2065 /* If the mode USES_PRIMARY, then we should only remove its
2066 * promisc and mc settings if it was the curr_active_slave, but that was
2067 * already taken care of above when we detached the slave
2069 if (!USES_PRIMARY(bond
->params
.mode
)) {
2070 /* unset promiscuity level from slave */
2071 if (bond_dev
->flags
& IFF_PROMISC
) {
2072 dev_set_promiscuity(slave_dev
, -1);
2075 /* unset allmulti level from slave */
2076 if (bond_dev
->flags
& IFF_ALLMULTI
) {
2077 dev_set_allmulti(slave_dev
, -1);
2080 /* flush master's mc_list from slave */
2081 netif_addr_lock_bh(bond_dev
);
2082 bond_mc_list_flush(bond_dev
, slave_dev
);
2083 netif_addr_unlock_bh(bond_dev
);
2086 netdev_set_master(slave_dev
, NULL
);
2088 /* close slave before restoring its mac address */
2089 dev_close(slave_dev
);
2091 if (!bond
->params
.fail_over_mac
) {
2092 /* restore original ("permanent") mac address*/
2093 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2094 addr
.sa_family
= slave_dev
->type
;
2095 dev_set_mac_address(slave_dev
, &addr
);
2098 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2099 IFF_SLAVE_INACTIVE
);
2103 /* re-acquire the lock before getting the next slave */
2104 write_lock_bh(&bond
->lock
);
2107 /* zero the mac address of the master so it will be
2108 * set by the application to the mac address of the
2111 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2113 if (list_empty(&bond
->vlan_list
)) {
2114 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2116 printk(KERN_WARNING DRV_NAME
2117 ": %s: Warning: clearing HW address of %s while it "
2118 "still has VLANs.\n",
2119 bond_dev
->name
, bond_dev
->name
);
2120 printk(KERN_WARNING DRV_NAME
2121 ": %s: When re-adding slaves, make sure the bond's "
2122 "HW address matches its VLANs'.\n",
2126 printk(KERN_INFO DRV_NAME
2127 ": %s: released all slaves\n",
2131 write_unlock_bh(&bond
->lock
);
2137 * This function changes the active slave to slave <slave_dev>.
2138 * It returns -EINVAL in the following cases.
2139 * - <slave_dev> is not found in the list.
2140 * - There is not active slave now.
2141 * - <slave_dev> is already active.
2142 * - The link state of <slave_dev> is not BOND_LINK_UP.
2143 * - <slave_dev> is not running.
2144 * In these cases, this fuction does nothing.
2145 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2147 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2149 struct bonding
*bond
= netdev_priv(bond_dev
);
2150 struct slave
*old_active
= NULL
;
2151 struct slave
*new_active
= NULL
;
2154 if (!USES_PRIMARY(bond
->params
.mode
)) {
2158 /* Verify that master_dev is indeed the master of slave_dev */
2159 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
2160 (slave_dev
->master
!= bond_dev
)) {
2164 read_lock(&bond
->lock
);
2166 read_lock(&bond
->curr_slave_lock
);
2167 old_active
= bond
->curr_active_slave
;
2168 read_unlock(&bond
->curr_slave_lock
);
2170 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2173 * Changing to the current active: do nothing; return success.
2175 if (new_active
&& (new_active
== old_active
)) {
2176 read_unlock(&bond
->lock
);
2182 (new_active
->link
== BOND_LINK_UP
) &&
2183 IS_UP(new_active
->dev
)) {
2184 write_lock_bh(&bond
->curr_slave_lock
);
2185 bond_change_active_slave(bond
, new_active
);
2186 write_unlock_bh(&bond
->curr_slave_lock
);
2191 read_unlock(&bond
->lock
);
2196 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2198 struct bonding
*bond
= netdev_priv(bond_dev
);
2200 info
->bond_mode
= bond
->params
.mode
;
2201 info
->miimon
= bond
->params
.miimon
;
2203 read_lock(&bond
->lock
);
2204 info
->num_slaves
= bond
->slave_cnt
;
2205 read_unlock(&bond
->lock
);
2210 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2212 struct bonding
*bond
= netdev_priv(bond_dev
);
2213 struct slave
*slave
;
2214 int i
, res
= -ENODEV
;
2216 read_lock(&bond
->lock
);
2218 bond_for_each_slave(bond
, slave
, i
) {
2219 if (i
== (int)info
->slave_id
) {
2221 strcpy(info
->slave_name
, slave
->dev
->name
);
2222 info
->link
= slave
->link
;
2223 info
->state
= slave
->state
;
2224 info
->link_failure_count
= slave
->link_failure_count
;
2229 read_unlock(&bond
->lock
);
2234 /*-------------------------------- Monitoring -------------------------------*/
2237 static int bond_miimon_inspect(struct bonding
*bond
)
2239 struct slave
*slave
;
2240 int i
, link_state
, commit
= 0;
2241 bool ignore_updelay
;
2243 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2245 bond_for_each_slave(bond
, slave
, i
) {
2246 slave
->new_link
= BOND_LINK_NOCHANGE
;
2248 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2250 switch (slave
->link
) {
2255 slave
->link
= BOND_LINK_FAIL
;
2256 slave
->delay
= bond
->params
.downdelay
;
2258 printk(KERN_INFO DRV_NAME
2259 ": %s: link status down for %s"
2260 "interface %s, disabling it in %d ms.\n",
2262 (bond
->params
.mode
==
2263 BOND_MODE_ACTIVEBACKUP
) ?
2264 ((slave
->state
== BOND_STATE_ACTIVE
) ?
2265 "active " : "backup ") : "",
2267 bond
->params
.downdelay
* bond
->params
.miimon
);
2270 case BOND_LINK_FAIL
:
2273 * recovered before downdelay expired
2275 slave
->link
= BOND_LINK_UP
;
2276 slave
->jiffies
= jiffies
;
2277 printk(KERN_INFO DRV_NAME
2278 ": %s: link status up again after %d "
2279 "ms for interface %s.\n",
2281 (bond
->params
.downdelay
- slave
->delay
) *
2282 bond
->params
.miimon
,
2287 if (slave
->delay
<= 0) {
2288 slave
->new_link
= BOND_LINK_DOWN
;
2296 case BOND_LINK_DOWN
:
2300 slave
->link
= BOND_LINK_BACK
;
2301 slave
->delay
= bond
->params
.updelay
;
2304 printk(KERN_INFO DRV_NAME
2305 ": %s: link status up for "
2306 "interface %s, enabling it in %d ms.\n",
2307 bond
->dev
->name
, slave
->dev
->name
,
2308 ignore_updelay
? 0 :
2309 bond
->params
.updelay
*
2310 bond
->params
.miimon
);
2313 case BOND_LINK_BACK
:
2315 slave
->link
= BOND_LINK_DOWN
;
2316 printk(KERN_INFO DRV_NAME
2317 ": %s: link status down again after %d "
2318 "ms for interface %s.\n",
2320 (bond
->params
.updelay
- slave
->delay
) *
2321 bond
->params
.miimon
,
2330 if (slave
->delay
<= 0) {
2331 slave
->new_link
= BOND_LINK_UP
;
2333 ignore_updelay
= false;
2345 static void bond_miimon_commit(struct bonding
*bond
)
2347 struct slave
*slave
;
2350 bond_for_each_slave(bond
, slave
, i
) {
2351 switch (slave
->new_link
) {
2352 case BOND_LINK_NOCHANGE
:
2356 slave
->link
= BOND_LINK_UP
;
2357 slave
->jiffies
= jiffies
;
2359 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2360 /* prevent it from being the active one */
2361 slave
->state
= BOND_STATE_BACKUP
;
2362 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2363 /* make it immediately active */
2364 slave
->state
= BOND_STATE_ACTIVE
;
2365 } else if (slave
!= bond
->primary_slave
) {
2366 /* prevent it from being the active one */
2367 slave
->state
= BOND_STATE_BACKUP
;
2370 printk(KERN_INFO DRV_NAME
2371 ": %s: link status definitely "
2372 "up for interface %s.\n",
2373 bond
->dev
->name
, slave
->dev
->name
);
2375 /* notify ad that the link status has changed */
2376 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2377 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2379 if (bond_is_lb(bond
))
2380 bond_alb_handle_link_change(bond
, slave
,
2383 if (!bond
->curr_active_slave
||
2384 (slave
== bond
->primary_slave
))
2389 case BOND_LINK_DOWN
:
2390 if (slave
->link_failure_count
< UINT_MAX
)
2391 slave
->link_failure_count
++;
2393 slave
->link
= BOND_LINK_DOWN
;
2395 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2396 bond
->params
.mode
== BOND_MODE_8023AD
)
2397 bond_set_slave_inactive_flags(slave
);
2399 printk(KERN_INFO DRV_NAME
2400 ": %s: link status definitely down for "
2401 "interface %s, disabling it\n",
2402 bond
->dev
->name
, slave
->dev
->name
);
2404 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2405 bond_3ad_handle_link_change(slave
,
2408 if (bond
->params
.mode
== BOND_MODE_TLB
||
2409 bond
->params
.mode
== BOND_MODE_ALB
)
2410 bond_alb_handle_link_change(bond
, slave
,
2413 if (slave
== bond
->curr_active_slave
)
2419 printk(KERN_ERR DRV_NAME
2420 ": %s: invalid new link %d on slave %s\n",
2421 bond
->dev
->name
, slave
->new_link
,
2423 slave
->new_link
= BOND_LINK_NOCHANGE
;
2430 write_lock_bh(&bond
->curr_slave_lock
);
2431 bond_select_active_slave(bond
);
2432 write_unlock_bh(&bond
->curr_slave_lock
);
2435 bond_set_carrier(bond
);
2441 * Really a wrapper that splits the mii monitor into two phases: an
2442 * inspection, then (if inspection indicates something needs to be done)
2443 * an acquisition of appropriate locks followed by a commit phase to
2444 * implement whatever link state changes are indicated.
2446 void bond_mii_monitor(struct work_struct
*work
)
2448 struct bonding
*bond
= container_of(work
, struct bonding
,
2451 read_lock(&bond
->lock
);
2452 if (bond
->kill_timers
)
2455 if (bond
->slave_cnt
== 0)
2458 if (bond
->send_grat_arp
) {
2459 read_lock(&bond
->curr_slave_lock
);
2460 bond_send_gratuitous_arp(bond
);
2461 read_unlock(&bond
->curr_slave_lock
);
2464 if (bond
->send_unsol_na
) {
2465 read_lock(&bond
->curr_slave_lock
);
2466 bond_send_unsolicited_na(bond
);
2467 read_unlock(&bond
->curr_slave_lock
);
2470 if (bond_miimon_inspect(bond
)) {
2471 read_unlock(&bond
->lock
);
2473 read_lock(&bond
->lock
);
2475 bond_miimon_commit(bond
);
2477 read_unlock(&bond
->lock
);
2478 rtnl_unlock(); /* might sleep, hold no other locks */
2479 read_lock(&bond
->lock
);
2483 if (bond
->params
.miimon
)
2484 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2485 msecs_to_jiffies(bond
->params
.miimon
));
2487 read_unlock(&bond
->lock
);
2490 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2492 struct in_device
*idev
;
2493 struct in_ifaddr
*ifa
;
2500 idev
= __in_dev_get_rcu(dev
);
2504 ifa
= idev
->ifa_list
;
2508 addr
= ifa
->ifa_local
;
2514 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2516 struct vlan_entry
*vlan
;
2518 if (ip
== bond
->master_ip
)
2521 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2522 if (ip
== vlan
->vlan_ip
)
2530 * We go to the (large) trouble of VLAN tagging ARP frames because
2531 * switches in VLAN mode (especially if ports are configured as
2532 * "native" to a VLAN) might not pass non-tagged frames.
2534 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2536 struct sk_buff
*skb
;
2538 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2539 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2541 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2542 NULL
, slave_dev
->dev_addr
, NULL
);
2545 printk(KERN_ERR DRV_NAME
": ARP packet allocation failed\n");
2549 skb
= vlan_put_tag(skb
, vlan_id
);
2551 printk(KERN_ERR DRV_NAME
": failed to insert VLAN tag\n");
2559 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2562 __be32
*targets
= bond
->params
.arp_targets
;
2563 struct vlan_entry
*vlan
;
2564 struct net_device
*vlan_dev
;
2568 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2571 pr_debug("basa: target %x\n", targets
[i
]);
2572 if (list_empty(&bond
->vlan_list
)) {
2573 pr_debug("basa: empty vlan: arp_send\n");
2574 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2575 bond
->master_ip
, 0);
2580 * If VLANs are configured, we do a route lookup to
2581 * determine which VLAN interface would be used, so we
2582 * can tag the ARP with the proper VLAN tag.
2584 memset(&fl
, 0, sizeof(fl
));
2585 fl
.fl4_dst
= targets
[i
];
2586 fl
.fl4_tos
= RTO_ONLINK
;
2588 rv
= ip_route_output_key(&init_net
, &rt
, &fl
);
2590 if (net_ratelimit()) {
2591 printk(KERN_WARNING DRV_NAME
2592 ": %s: no route to arp_ip_target %pI4\n",
2593 bond
->dev
->name
, &fl
.fl4_dst
);
2599 * This target is not on a VLAN
2601 if (rt
->u
.dst
.dev
== bond
->dev
) {
2603 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2604 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2605 bond
->master_ip
, 0);
2610 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2611 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2612 if (vlan_dev
== rt
->u
.dst
.dev
) {
2613 vlan_id
= vlan
->vlan_id
;
2614 pr_debug("basa: vlan match on %s %d\n",
2615 vlan_dev
->name
, vlan_id
);
2622 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2623 vlan
->vlan_ip
, vlan_id
);
2627 if (net_ratelimit()) {
2628 printk(KERN_WARNING DRV_NAME
2629 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2630 bond
->dev
->name
, &fl
.fl4_dst
,
2631 rt
->u
.dst
.dev
? rt
->u
.dst
.dev
->name
: "NULL");
2638 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2639 * for each VLAN above us.
2641 * Caller must hold curr_slave_lock for read or better
2643 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2645 struct slave
*slave
= bond
->curr_active_slave
;
2646 struct vlan_entry
*vlan
;
2647 struct net_device
*vlan_dev
;
2649 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond
->dev
->name
,
2650 slave
? slave
->dev
->name
: "NULL");
2652 if (!slave
|| !bond
->send_grat_arp
||
2653 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2656 bond
->send_grat_arp
--;
2658 if (bond
->master_ip
) {
2659 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2660 bond
->master_ip
, 0);
2663 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2664 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2665 if (vlan
->vlan_ip
) {
2666 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2667 vlan
->vlan_ip
, vlan
->vlan_id
);
2672 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2675 __be32
*targets
= bond
->params
.arp_targets
;
2677 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2678 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2679 &sip
, &tip
, i
, &targets
[i
], bond_has_this_ip(bond
, tip
));
2680 if (sip
== targets
[i
]) {
2681 if (bond_has_this_ip(bond
, tip
))
2682 slave
->last_arp_rx
= jiffies
;
2688 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2691 struct slave
*slave
;
2692 struct bonding
*bond
;
2693 unsigned char *arp_ptr
;
2696 if (dev_net(dev
) != &init_net
)
2699 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2702 bond
= netdev_priv(dev
);
2703 read_lock(&bond
->lock
);
2705 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2706 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2707 orig_dev
? orig_dev
->name
: "NULL");
2709 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2710 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2713 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2717 if (arp
->ar_hln
!= dev
->addr_len
||
2718 skb
->pkt_type
== PACKET_OTHERHOST
||
2719 skb
->pkt_type
== PACKET_LOOPBACK
||
2720 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2721 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2725 arp_ptr
= (unsigned char *)(arp
+ 1);
2726 arp_ptr
+= dev
->addr_len
;
2727 memcpy(&sip
, arp_ptr
, 4);
2728 arp_ptr
+= 4 + dev
->addr_len
;
2729 memcpy(&tip
, arp_ptr
, 4);
2731 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2732 bond
->dev
->name
, slave
->dev
->name
, slave
->state
,
2733 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2737 * Backup slaves won't see the ARP reply, but do come through
2738 * here for each ARP probe (so we swap the sip/tip to validate
2739 * the probe). In a "redundant switch, common router" type of
2740 * configuration, the ARP probe will (hopefully) travel from
2741 * the active, through one switch, the router, then the other
2742 * switch before reaching the backup.
2744 if (slave
->state
== BOND_STATE_ACTIVE
)
2745 bond_validate_arp(bond
, slave
, sip
, tip
);
2747 bond_validate_arp(bond
, slave
, tip
, sip
);
2750 read_unlock(&bond
->lock
);
2753 return NET_RX_SUCCESS
;
2757 * this function is called regularly to monitor each slave's link
2758 * ensuring that traffic is being sent and received when arp monitoring
2759 * is used in load-balancing mode. if the adapter has been dormant, then an
2760 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2761 * arp monitoring in active backup mode.
2763 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2765 struct bonding
*bond
= container_of(work
, struct bonding
,
2767 struct slave
*slave
, *oldcurrent
;
2768 int do_failover
= 0;
2772 read_lock(&bond
->lock
);
2774 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2776 if (bond
->kill_timers
) {
2780 if (bond
->slave_cnt
== 0) {
2784 read_lock(&bond
->curr_slave_lock
);
2785 oldcurrent
= bond
->curr_active_slave
;
2786 read_unlock(&bond
->curr_slave_lock
);
2788 /* see if any of the previous devices are up now (i.e. they have
2789 * xmt and rcv traffic). the curr_active_slave does not come into
2790 * the picture unless it is null. also, slave->jiffies is not needed
2791 * here because we send an arp on each slave and give a slave as
2792 * long as it needs to get the tx/rx within the delta.
2793 * TODO: what about up/down delay in arp mode? it wasn't here before
2796 bond_for_each_slave(bond
, slave
, i
) {
2797 if (slave
->link
!= BOND_LINK_UP
) {
2798 if (time_before_eq(jiffies
, dev_trans_start(slave
->dev
) + delta_in_ticks
) &&
2799 time_before_eq(jiffies
, slave
->dev
->last_rx
+ delta_in_ticks
)) {
2801 slave
->link
= BOND_LINK_UP
;
2802 slave
->state
= BOND_STATE_ACTIVE
;
2804 /* primary_slave has no meaning in round-robin
2805 * mode. the window of a slave being up and
2806 * curr_active_slave being null after enslaving
2810 printk(KERN_INFO DRV_NAME
2811 ": %s: link status definitely "
2812 "up for interface %s, ",
2817 printk(KERN_INFO DRV_NAME
2818 ": %s: interface %s is now up\n",
2824 /* slave->link == BOND_LINK_UP */
2826 /* not all switches will respond to an arp request
2827 * when the source ip is 0, so don't take the link down
2828 * if we don't know our ip yet
2830 if (time_after_eq(jiffies
, dev_trans_start(slave
->dev
) + 2*delta_in_ticks
) ||
2831 (time_after_eq(jiffies
, slave
->dev
->last_rx
+ 2*delta_in_ticks
))) {
2833 slave
->link
= BOND_LINK_DOWN
;
2834 slave
->state
= BOND_STATE_BACKUP
;
2836 if (slave
->link_failure_count
< UINT_MAX
) {
2837 slave
->link_failure_count
++;
2840 printk(KERN_INFO DRV_NAME
2841 ": %s: interface %s is now down.\n",
2845 if (slave
== oldcurrent
) {
2851 /* note: if switch is in round-robin mode, all links
2852 * must tx arp to ensure all links rx an arp - otherwise
2853 * links may oscillate or not come up at all; if switch is
2854 * in something like xor mode, there is nothing we can
2855 * do - all replies will be rx'ed on same link causing slaves
2856 * to be unstable during low/no traffic periods
2858 if (IS_UP(slave
->dev
)) {
2859 bond_arp_send_all(bond
, slave
);
2864 write_lock_bh(&bond
->curr_slave_lock
);
2866 bond_select_active_slave(bond
);
2868 write_unlock_bh(&bond
->curr_slave_lock
);
2872 if (bond
->params
.arp_interval
)
2873 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2875 read_unlock(&bond
->lock
);
2879 * Called to inspect slaves for active-backup mode ARP monitor link state
2880 * changes. Sets new_link in slaves to specify what action should take
2881 * place for the slave. Returns 0 if no changes are found, >0 if changes
2882 * to link states must be committed.
2884 * Called with bond->lock held for read.
2886 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2888 struct slave
*slave
;
2891 bond_for_each_slave(bond
, slave
, i
) {
2892 slave
->new_link
= BOND_LINK_NOCHANGE
;
2894 if (slave
->link
!= BOND_LINK_UP
) {
2895 if (time_before_eq(jiffies
, slave_last_rx(bond
, slave
) +
2897 slave
->new_link
= BOND_LINK_UP
;
2905 * Give slaves 2*delta after being enslaved or made
2906 * active. This avoids bouncing, as the last receive
2907 * times need a full ARP monitor cycle to be updated.
2909 if (!time_after_eq(jiffies
, slave
->jiffies
+
2910 2 * delta_in_ticks
))
2914 * Backup slave is down if:
2915 * - No current_arp_slave AND
2916 * - more than 3*delta since last receive AND
2917 * - the bond has an IP address
2919 * Note: a non-null current_arp_slave indicates
2920 * the curr_active_slave went down and we are
2921 * searching for a new one; under this condition
2922 * we only take the curr_active_slave down - this
2923 * gives each slave a chance to tx/rx traffic
2924 * before being taken out
2926 if (slave
->state
== BOND_STATE_BACKUP
&&
2927 !bond
->current_arp_slave
&&
2928 time_after(jiffies
, slave_last_rx(bond
, slave
) +
2929 3 * delta_in_ticks
)) {
2930 slave
->new_link
= BOND_LINK_DOWN
;
2935 * Active slave is down if:
2936 * - more than 2*delta since transmitting OR
2937 * - (more than 2*delta since receive AND
2938 * the bond has an IP address)
2940 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
2941 (time_after_eq(jiffies
, dev_trans_start(slave
->dev
) +
2942 2 * delta_in_ticks
) ||
2943 (time_after_eq(jiffies
, slave_last_rx(bond
, slave
)
2944 + 2 * delta_in_ticks
)))) {
2945 slave
->new_link
= BOND_LINK_DOWN
;
2950 read_lock(&bond
->curr_slave_lock
);
2953 * Trigger a commit if the primary option setting has changed.
2955 if (bond
->primary_slave
&&
2956 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
2957 (bond
->primary_slave
->link
== BOND_LINK_UP
))
2960 read_unlock(&bond
->curr_slave_lock
);
2966 * Called to commit link state changes noted by inspection step of
2967 * active-backup mode ARP monitor.
2969 * Called with RTNL and bond->lock for read.
2971 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2973 struct slave
*slave
;
2976 bond_for_each_slave(bond
, slave
, i
) {
2977 switch (slave
->new_link
) {
2978 case BOND_LINK_NOCHANGE
:
2982 write_lock_bh(&bond
->curr_slave_lock
);
2984 if (!bond
->curr_active_slave
&&
2985 time_before_eq(jiffies
, dev_trans_start(slave
->dev
) +
2987 slave
->link
= BOND_LINK_UP
;
2988 bond_change_active_slave(bond
, slave
);
2989 bond
->current_arp_slave
= NULL
;
2991 printk(KERN_INFO DRV_NAME
2992 ": %s: %s is up and now the "
2993 "active interface\n",
2994 bond
->dev
->name
, slave
->dev
->name
);
2996 } else if (bond
->curr_active_slave
!= slave
) {
2997 /* this slave has just come up but we
2998 * already have a current slave; this can
2999 * also happen if bond_enslave adds a new
3000 * slave that is up while we are searching
3003 slave
->link
= BOND_LINK_UP
;
3004 bond_set_slave_inactive_flags(slave
);
3005 bond
->current_arp_slave
= NULL
;
3007 printk(KERN_INFO DRV_NAME
3008 ": %s: backup interface %s is now up\n",
3009 bond
->dev
->name
, slave
->dev
->name
);
3012 write_unlock_bh(&bond
->curr_slave_lock
);
3016 case BOND_LINK_DOWN
:
3017 if (slave
->link_failure_count
< UINT_MAX
)
3018 slave
->link_failure_count
++;
3020 slave
->link
= BOND_LINK_DOWN
;
3022 if (slave
== bond
->curr_active_slave
) {
3023 printk(KERN_INFO DRV_NAME
3024 ": %s: link status down for active "
3025 "interface %s, disabling it\n",
3026 bond
->dev
->name
, slave
->dev
->name
);
3028 bond_set_slave_inactive_flags(slave
);
3030 write_lock_bh(&bond
->curr_slave_lock
);
3032 bond_select_active_slave(bond
);
3033 if (bond
->curr_active_slave
)
3034 bond
->curr_active_slave
->jiffies
=
3037 write_unlock_bh(&bond
->curr_slave_lock
);
3039 bond
->current_arp_slave
= NULL
;
3041 } else if (slave
->state
== BOND_STATE_BACKUP
) {
3042 printk(KERN_INFO DRV_NAME
3043 ": %s: backup interface %s is now down\n",
3044 bond
->dev
->name
, slave
->dev
->name
);
3046 bond_set_slave_inactive_flags(slave
);
3051 printk(KERN_ERR DRV_NAME
3052 ": %s: impossible: new_link %d on slave %s\n",
3053 bond
->dev
->name
, slave
->new_link
,
3059 * No race with changes to primary via sysfs, as we hold rtnl.
3061 if (bond
->primary_slave
&&
3062 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
3063 (bond
->primary_slave
->link
== BOND_LINK_UP
)) {
3064 write_lock_bh(&bond
->curr_slave_lock
);
3065 bond_change_active_slave(bond
, bond
->primary_slave
);
3066 write_unlock_bh(&bond
->curr_slave_lock
);
3069 bond_set_carrier(bond
);
3073 * Send ARP probes for active-backup mode ARP monitor.
3075 * Called with bond->lock held for read.
3077 static void bond_ab_arp_probe(struct bonding
*bond
)
3079 struct slave
*slave
;
3082 read_lock(&bond
->curr_slave_lock
);
3084 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3085 printk("PROBE: c_arp %s && cas %s BAD\n",
3086 bond
->current_arp_slave
->dev
->name
,
3087 bond
->curr_active_slave
->dev
->name
);
3089 if (bond
->curr_active_slave
) {
3090 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3091 read_unlock(&bond
->curr_slave_lock
);
3095 read_unlock(&bond
->curr_slave_lock
);
3097 /* if we don't have a curr_active_slave, search for the next available
3098 * backup slave from the current_arp_slave and make it the candidate
3099 * for becoming the curr_active_slave
3102 if (!bond
->current_arp_slave
) {
3103 bond
->current_arp_slave
= bond
->first_slave
;
3104 if (!bond
->current_arp_slave
)
3108 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3110 /* search for next candidate */
3111 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3112 if (IS_UP(slave
->dev
)) {
3113 slave
->link
= BOND_LINK_BACK
;
3114 bond_set_slave_active_flags(slave
);
3115 bond_arp_send_all(bond
, slave
);
3116 slave
->jiffies
= jiffies
;
3117 bond
->current_arp_slave
= slave
;
3121 /* if the link state is up at this point, we
3122 * mark it down - this can happen if we have
3123 * simultaneous link failures and
3124 * reselect_active_interface doesn't make this
3125 * one the current slave so it is still marked
3126 * up when it is actually down
3128 if (slave
->link
== BOND_LINK_UP
) {
3129 slave
->link
= BOND_LINK_DOWN
;
3130 if (slave
->link_failure_count
< UINT_MAX
)
3131 slave
->link_failure_count
++;
3133 bond_set_slave_inactive_flags(slave
);
3135 printk(KERN_INFO DRV_NAME
3136 ": %s: backup interface %s is now down.\n",
3137 bond
->dev
->name
, slave
->dev
->name
);
3142 void bond_activebackup_arp_mon(struct work_struct
*work
)
3144 struct bonding
*bond
= container_of(work
, struct bonding
,
3148 read_lock(&bond
->lock
);
3150 if (bond
->kill_timers
)
3153 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3155 if (bond
->slave_cnt
== 0)
3158 if (bond
->send_grat_arp
) {
3159 read_lock(&bond
->curr_slave_lock
);
3160 bond_send_gratuitous_arp(bond
);
3161 read_unlock(&bond
->curr_slave_lock
);
3164 if (bond
->send_unsol_na
) {
3165 read_lock(&bond
->curr_slave_lock
);
3166 bond_send_unsolicited_na(bond
);
3167 read_unlock(&bond
->curr_slave_lock
);
3170 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3171 read_unlock(&bond
->lock
);
3173 read_lock(&bond
->lock
);
3175 bond_ab_arp_commit(bond
, delta_in_ticks
);
3177 read_unlock(&bond
->lock
);
3179 read_lock(&bond
->lock
);
3182 bond_ab_arp_probe(bond
);
3185 if (bond
->params
.arp_interval
) {
3186 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3189 read_unlock(&bond
->lock
);
3192 /*------------------------------ proc/seq_file-------------------------------*/
3194 #ifdef CONFIG_PROC_FS
3196 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3197 __acquires(&dev_base_lock
)
3198 __acquires(&bond
->lock
)
3200 struct bonding
*bond
= seq
->private;
3202 struct slave
*slave
;
3205 /* make sure the bond won't be taken away */
3206 read_lock(&dev_base_lock
);
3207 read_lock(&bond
->lock
);
3210 return SEQ_START_TOKEN
;
3213 bond_for_each_slave(bond
, slave
, i
) {
3214 if (++off
== *pos
) {
3222 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3224 struct bonding
*bond
= seq
->private;
3225 struct slave
*slave
= v
;
3228 if (v
== SEQ_START_TOKEN
) {
3229 return bond
->first_slave
;
3232 slave
= slave
->next
;
3234 return (slave
== bond
->first_slave
) ? NULL
: slave
;
3237 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
3238 __releases(&bond
->lock
)
3239 __releases(&dev_base_lock
)
3241 struct bonding
*bond
= seq
->private;
3243 read_unlock(&bond
->lock
);
3244 read_unlock(&dev_base_lock
);
3247 static void bond_info_show_master(struct seq_file
*seq
)
3249 struct bonding
*bond
= seq
->private;
3253 read_lock(&bond
->curr_slave_lock
);
3254 curr
= bond
->curr_active_slave
;
3255 read_unlock(&bond
->curr_slave_lock
);
3257 seq_printf(seq
, "Bonding Mode: %s",
3258 bond_mode_name(bond
->params
.mode
));
3260 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
&&
3261 bond
->params
.fail_over_mac
)
3262 seq_printf(seq
, " (fail_over_mac %s)",
3263 fail_over_mac_tbl
[bond
->params
.fail_over_mac
].modename
);
3265 seq_printf(seq
, "\n");
3267 if (bond
->params
.mode
== BOND_MODE_XOR
||
3268 bond
->params
.mode
== BOND_MODE_8023AD
) {
3269 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
3270 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
3271 bond
->params
.xmit_policy
);
3274 if (USES_PRIMARY(bond
->params
.mode
)) {
3275 seq_printf(seq
, "Primary Slave: %s\n",
3276 (bond
->primary_slave
) ?
3277 bond
->primary_slave
->dev
->name
: "None");
3279 seq_printf(seq
, "Currently Active Slave: %s\n",
3280 (curr
) ? curr
->dev
->name
: "None");
3283 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
3285 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
3286 seq_printf(seq
, "Up Delay (ms): %d\n",
3287 bond
->params
.updelay
* bond
->params
.miimon
);
3288 seq_printf(seq
, "Down Delay (ms): %d\n",
3289 bond
->params
.downdelay
* bond
->params
.miimon
);
3292 /* ARP information */
3293 if(bond
->params
.arp_interval
> 0) {
3295 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
3296 bond
->params
.arp_interval
);
3298 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
3300 for(i
= 0; (i
< BOND_MAX_ARP_TARGETS
) ;i
++) {
3301 if (!bond
->params
.arp_targets
[i
])
3304 seq_printf(seq
, ",");
3305 seq_printf(seq
, " %pI4", &bond
->params
.arp_targets
[i
]);
3308 seq_printf(seq
, "\n");
3311 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3312 struct ad_info ad_info
;
3314 seq_puts(seq
, "\n802.3ad info\n");
3315 seq_printf(seq
, "LACP rate: %s\n",
3316 (bond
->params
.lacp_fast
) ? "fast" : "slow");
3317 seq_printf(seq
, "Aggregator selection policy (ad_select): %s\n",
3318 ad_select_tbl
[bond
->params
.ad_select
].modename
);
3320 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3321 seq_printf(seq
, "bond %s has no active aggregator\n",
3324 seq_printf(seq
, "Active Aggregator Info:\n");
3326 seq_printf(seq
, "\tAggregator ID: %d\n",
3327 ad_info
.aggregator_id
);
3328 seq_printf(seq
, "\tNumber of ports: %d\n",
3330 seq_printf(seq
, "\tActor Key: %d\n",
3332 seq_printf(seq
, "\tPartner Key: %d\n",
3333 ad_info
.partner_key
);
3334 seq_printf(seq
, "\tPartner Mac Address: %pM\n",
3335 ad_info
.partner_system
);
3340 static void bond_info_show_slave(struct seq_file
*seq
, const struct slave
*slave
)
3342 struct bonding
*bond
= seq
->private;
3344 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
3345 seq_printf(seq
, "MII Status: %s\n",
3346 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
3347 seq_printf(seq
, "Link Failure Count: %u\n",
3348 slave
->link_failure_count
);
3350 seq_printf(seq
, "Permanent HW addr: %pM\n", slave
->perm_hwaddr
);
3352 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3353 const struct aggregator
*agg
3354 = SLAVE_AD_INFO(slave
).port
.aggregator
;
3357 seq_printf(seq
, "Aggregator ID: %d\n",
3358 agg
->aggregator_identifier
);
3360 seq_puts(seq
, "Aggregator ID: N/A\n");
3365 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
3367 if (v
== SEQ_START_TOKEN
) {
3368 seq_printf(seq
, "%s\n", version
);
3369 bond_info_show_master(seq
);
3371 bond_info_show_slave(seq
, v
);
3377 static const struct seq_operations bond_info_seq_ops
= {
3378 .start
= bond_info_seq_start
,
3379 .next
= bond_info_seq_next
,
3380 .stop
= bond_info_seq_stop
,
3381 .show
= bond_info_seq_show
,
3384 static int bond_info_open(struct inode
*inode
, struct file
*file
)
3386 struct seq_file
*seq
;
3387 struct proc_dir_entry
*proc
;
3390 res
= seq_open(file
, &bond_info_seq_ops
);
3392 /* recover the pointer buried in proc_dir_entry data */
3393 seq
= file
->private_data
;
3395 seq
->private = proc
->data
;
3401 static const struct file_operations bond_info_fops
= {
3402 .owner
= THIS_MODULE
,
3403 .open
= bond_info_open
,
3405 .llseek
= seq_lseek
,
3406 .release
= seq_release
,
3409 static int bond_create_proc_entry(struct bonding
*bond
)
3411 struct net_device
*bond_dev
= bond
->dev
;
3413 if (bond_proc_dir
) {
3414 bond
->proc_entry
= proc_create_data(bond_dev
->name
,
3415 S_IRUGO
, bond_proc_dir
,
3416 &bond_info_fops
, bond
);
3417 if (bond
->proc_entry
== NULL
) {
3418 printk(KERN_WARNING DRV_NAME
3419 ": Warning: Cannot create /proc/net/%s/%s\n",
3420 DRV_NAME
, bond_dev
->name
);
3422 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3429 static void bond_remove_proc_entry(struct bonding
*bond
)
3431 if (bond_proc_dir
&& bond
->proc_entry
) {
3432 remove_proc_entry(bond
->proc_file_name
, bond_proc_dir
);
3433 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3434 bond
->proc_entry
= NULL
;
3438 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3439 * Caller must hold rtnl_lock.
3441 static void bond_create_proc_dir(void)
3443 if (!bond_proc_dir
) {
3444 bond_proc_dir
= proc_mkdir(DRV_NAME
, init_net
.proc_net
);
3446 printk(KERN_WARNING DRV_NAME
3447 ": Warning: cannot create /proc/net/%s\n",
3452 /* Destroy the bonding directory under /proc/net, if empty.
3453 * Caller must hold rtnl_lock.
3455 static void bond_destroy_proc_dir(void)
3457 if (bond_proc_dir
) {
3458 remove_proc_entry(DRV_NAME
, init_net
.proc_net
);
3459 bond_proc_dir
= NULL
;
3463 #else /* !CONFIG_PROC_FS */
3465 static int bond_create_proc_entry(struct bonding
*bond
)
3469 static void bond_remove_proc_entry(struct bonding
*bond
)
3473 static void bond_create_proc_dir(void)
3477 static void bond_destroy_proc_dir(void)
3481 #endif /* CONFIG_PROC_FS */
3484 /*-------------------------- netdev event handling --------------------------*/
3487 * Change device name
3489 static int bond_event_changename(struct bonding
*bond
)
3491 bond_remove_proc_entry(bond
);
3492 bond_create_proc_entry(bond
);
3493 down_write(&(bonding_rwsem
));
3494 bond_destroy_sysfs_entry(bond
);
3495 bond_create_sysfs_entry(bond
);
3496 up_write(&(bonding_rwsem
));
3500 static int bond_master_netdev_event(unsigned long event
, struct net_device
*bond_dev
)
3502 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3505 case NETDEV_CHANGENAME
:
3506 return bond_event_changename(event_bond
);
3507 case NETDEV_UNREGISTER
:
3508 bond_release_all(event_bond
->dev
);
3517 static int bond_slave_netdev_event(unsigned long event
, struct net_device
*slave_dev
)
3519 struct net_device
*bond_dev
= slave_dev
->master
;
3520 struct bonding
*bond
= netdev_priv(bond_dev
);
3523 case NETDEV_UNREGISTER
:
3525 if (bond
->setup_by_slave
)
3526 bond_release_and_destroy(bond_dev
, slave_dev
);
3528 bond_release(bond_dev
, slave_dev
);
3532 if (bond
->params
.mode
== BOND_MODE_8023AD
|| bond_is_lb(bond
)) {
3533 struct slave
*slave
;
3535 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3537 u16 old_speed
= slave
->speed
;
3538 u16 old_duplex
= slave
->duplex
;
3540 bond_update_speed_duplex(slave
);
3542 if (bond_is_lb(bond
))
3545 if (old_speed
!= slave
->speed
)
3546 bond_3ad_adapter_speed_changed(slave
);
3547 if (old_duplex
!= slave
->duplex
)
3548 bond_3ad_adapter_duplex_changed(slave
);
3555 * ... Or is it this?
3558 case NETDEV_CHANGEMTU
:
3560 * TODO: Should slaves be allowed to
3561 * independently alter their MTU? For
3562 * an active-backup bond, slaves need
3563 * not be the same type of device, so
3564 * MTUs may vary. For other modes,
3565 * slaves arguably should have the
3566 * same MTUs. To do this, we'd need to
3567 * take over the slave's change_mtu
3568 * function for the duration of their
3572 case NETDEV_CHANGENAME
:
3574 * TODO: handle changing the primary's name
3577 case NETDEV_FEAT_CHANGE
:
3578 bond_compute_features(bond
);
3588 * bond_netdev_event: handle netdev notifier chain events.
3590 * This function receives events for the netdev chain. The caller (an
3591 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3592 * locks for us to safely manipulate the slave devices (RTNL lock,
3595 static int bond_netdev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3597 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3599 if (dev_net(event_dev
) != &init_net
)
3602 pr_debug("event_dev: %s, event: %lx\n",
3603 (event_dev
? event_dev
->name
: "None"),
3606 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3609 if (event_dev
->flags
& IFF_MASTER
) {
3610 pr_debug("IFF_MASTER\n");
3611 return bond_master_netdev_event(event
, event_dev
);
3614 if (event_dev
->flags
& IFF_SLAVE
) {
3615 pr_debug("IFF_SLAVE\n");
3616 return bond_slave_netdev_event(event
, event_dev
);
3623 * bond_inetaddr_event: handle inetaddr notifier chain events.
3625 * We keep track of device IPs primarily to use as source addresses in
3626 * ARP monitor probes (rather than spewing out broadcasts all the time).
3628 * We track one IP for the main device (if it has one), plus one per VLAN.
3630 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3632 struct in_ifaddr
*ifa
= ptr
;
3633 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3634 struct bonding
*bond
;
3635 struct vlan_entry
*vlan
;
3637 if (dev_net(ifa
->ifa_dev
->dev
) != &init_net
)
3640 list_for_each_entry(bond
, &bond_dev_list
, bond_list
) {
3641 if (bond
->dev
== event_dev
) {
3644 bond
->master_ip
= ifa
->ifa_local
;
3647 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3654 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3655 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3656 if (vlan_dev
== event_dev
) {
3659 vlan
->vlan_ip
= ifa
->ifa_local
;
3663 bond_glean_dev_ip(vlan_dev
);
3674 static struct notifier_block bond_netdev_notifier
= {
3675 .notifier_call
= bond_netdev_event
,
3678 static struct notifier_block bond_inetaddr_notifier
= {
3679 .notifier_call
= bond_inetaddr_event
,
3682 /*-------------------------- Packet type handling ---------------------------*/
3684 /* register to receive lacpdus on a bond */
3685 static void bond_register_lacpdu(struct bonding
*bond
)
3687 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3689 /* initialize packet type */
3690 pk_type
->type
= PKT_TYPE_LACPDU
;
3691 pk_type
->dev
= bond
->dev
;
3692 pk_type
->func
= bond_3ad_lacpdu_recv
;
3694 dev_add_pack(pk_type
);
3697 /* unregister to receive lacpdus on a bond */
3698 static void bond_unregister_lacpdu(struct bonding
*bond
)
3700 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3703 void bond_register_arp(struct bonding
*bond
)
3705 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3710 pt
->type
= htons(ETH_P_ARP
);
3711 pt
->dev
= bond
->dev
;
3712 pt
->func
= bond_arp_rcv
;
3716 void bond_unregister_arp(struct bonding
*bond
)
3718 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3720 dev_remove_pack(pt
);
3724 /*---------------------------- Hashing Policies -----------------------------*/
3727 * Hash for the output device based upon layer 2 and layer 3 data. If
3728 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3730 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
,
3731 struct net_device
*bond_dev
, int count
)
3733 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3734 struct iphdr
*iph
= ip_hdr(skb
);
3736 if (skb
->protocol
== htons(ETH_P_IP
)) {
3737 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3738 (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5])) % count
;
3741 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3745 * Hash for the output device based upon layer 3 and layer 4 data. If
3746 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3747 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3749 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
,
3750 struct net_device
*bond_dev
, int count
)
3752 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3753 struct iphdr
*iph
= ip_hdr(skb
);
3754 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3757 if (skb
->protocol
== htons(ETH_P_IP
)) {
3758 if (!(iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) &&
3759 (iph
->protocol
== IPPROTO_TCP
||
3760 iph
->protocol
== IPPROTO_UDP
)) {
3761 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3763 return (layer4_xor
^
3764 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3768 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3772 * Hash for the output device based upon layer 2 data
3774 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
,
3775 struct net_device
*bond_dev
, int count
)
3777 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3779 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3782 /*-------------------------- Device entry points ----------------------------*/
3784 static int bond_open(struct net_device
*bond_dev
)
3786 struct bonding
*bond
= netdev_priv(bond_dev
);
3788 bond
->kill_timers
= 0;
3790 if (bond_is_lb(bond
)) {
3791 /* bond_alb_initialize must be called before the timer
3794 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3795 /* something went wrong - fail the open operation */
3799 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3800 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3803 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3804 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3805 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3808 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3809 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3810 INIT_DELAYED_WORK(&bond
->arp_work
,
3811 bond_activebackup_arp_mon
);
3813 INIT_DELAYED_WORK(&bond
->arp_work
,
3814 bond_loadbalance_arp_mon
);
3816 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3817 if (bond
->params
.arp_validate
)
3818 bond_register_arp(bond
);
3821 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3822 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3823 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3824 /* register to receive LACPDUs */
3825 bond_register_lacpdu(bond
);
3826 bond_3ad_initiate_agg_selection(bond
, 1);
3832 static int bond_close(struct net_device
*bond_dev
)
3834 struct bonding
*bond
= netdev_priv(bond_dev
);
3836 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3837 /* Unregister the receive of LACPDUs */
3838 bond_unregister_lacpdu(bond
);
3841 if (bond
->params
.arp_validate
)
3842 bond_unregister_arp(bond
);
3844 write_lock_bh(&bond
->lock
);
3846 bond
->send_grat_arp
= 0;
3847 bond
->send_unsol_na
= 0;
3849 /* signal timers not to re-arm */
3850 bond
->kill_timers
= 1;
3852 write_unlock_bh(&bond
->lock
);
3854 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3855 cancel_delayed_work(&bond
->mii_work
);
3858 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3859 cancel_delayed_work(&bond
->arp_work
);
3862 switch (bond
->params
.mode
) {
3863 case BOND_MODE_8023AD
:
3864 cancel_delayed_work(&bond
->ad_work
);
3868 cancel_delayed_work(&bond
->alb_work
);
3875 if (bond_is_lb(bond
)) {
3876 /* Must be called only after all
3877 * slaves have been released
3879 bond_alb_deinitialize(bond
);
3885 static struct net_device_stats
*bond_get_stats(struct net_device
*bond_dev
)
3887 struct bonding
*bond
= netdev_priv(bond_dev
);
3888 struct net_device_stats
*stats
= &bond
->stats
;
3889 struct net_device_stats local_stats
;
3890 struct slave
*slave
;
3893 memset(&local_stats
, 0, sizeof(struct net_device_stats
));
3895 read_lock_bh(&bond
->lock
);
3897 bond_for_each_slave(bond
, slave
, i
) {
3898 const struct net_device_stats
*sstats
= dev_get_stats(slave
->dev
);
3900 local_stats
.rx_packets
+= sstats
->rx_packets
;
3901 local_stats
.rx_bytes
+= sstats
->rx_bytes
;
3902 local_stats
.rx_errors
+= sstats
->rx_errors
;
3903 local_stats
.rx_dropped
+= sstats
->rx_dropped
;
3905 local_stats
.tx_packets
+= sstats
->tx_packets
;
3906 local_stats
.tx_bytes
+= sstats
->tx_bytes
;
3907 local_stats
.tx_errors
+= sstats
->tx_errors
;
3908 local_stats
.tx_dropped
+= sstats
->tx_dropped
;
3910 local_stats
.multicast
+= sstats
->multicast
;
3911 local_stats
.collisions
+= sstats
->collisions
;
3913 local_stats
.rx_length_errors
+= sstats
->rx_length_errors
;
3914 local_stats
.rx_over_errors
+= sstats
->rx_over_errors
;
3915 local_stats
.rx_crc_errors
+= sstats
->rx_crc_errors
;
3916 local_stats
.rx_frame_errors
+= sstats
->rx_frame_errors
;
3917 local_stats
.rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3918 local_stats
.rx_missed_errors
+= sstats
->rx_missed_errors
;
3920 local_stats
.tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3921 local_stats
.tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3922 local_stats
.tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3923 local_stats
.tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3924 local_stats
.tx_window_errors
+= sstats
->tx_window_errors
;
3927 memcpy(stats
, &local_stats
, sizeof(struct net_device_stats
));
3929 read_unlock_bh(&bond
->lock
);
3934 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3936 struct net_device
*slave_dev
= NULL
;
3937 struct ifbond k_binfo
;
3938 struct ifbond __user
*u_binfo
= NULL
;
3939 struct ifslave k_sinfo
;
3940 struct ifslave __user
*u_sinfo
= NULL
;
3941 struct mii_ioctl_data
*mii
= NULL
;
3944 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3945 bond_dev
->name
, cmd
);
3957 * We do this again just in case we were called by SIOCGMIIREG
3958 * instead of SIOCGMIIPHY.
3965 if (mii
->reg_num
== 1) {
3966 struct bonding
*bond
= netdev_priv(bond_dev
);
3968 read_lock(&bond
->lock
);
3969 read_lock(&bond
->curr_slave_lock
);
3970 if (netif_carrier_ok(bond
->dev
)) {
3971 mii
->val_out
= BMSR_LSTATUS
;
3973 read_unlock(&bond
->curr_slave_lock
);
3974 read_unlock(&bond
->lock
);
3978 case BOND_INFO_QUERY_OLD
:
3979 case SIOCBONDINFOQUERY
:
3980 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3982 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
))) {
3986 res
= bond_info_query(bond_dev
, &k_binfo
);
3988 if (copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
))) {
3994 case BOND_SLAVE_INFO_QUERY_OLD
:
3995 case SIOCBONDSLAVEINFOQUERY
:
3996 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3998 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
))) {
4002 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
4004 if (copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
))) {
4015 if (!capable(CAP_NET_ADMIN
)) {
4019 down_write(&(bonding_rwsem
));
4020 slave_dev
= dev_get_by_name(&init_net
, ifr
->ifr_slave
);
4022 pr_debug("slave_dev=%p: \n", slave_dev
);
4027 pr_debug("slave_dev->name=%s: \n", slave_dev
->name
);
4029 case BOND_ENSLAVE_OLD
:
4030 case SIOCBONDENSLAVE
:
4031 res
= bond_enslave(bond_dev
, slave_dev
);
4033 case BOND_RELEASE_OLD
:
4034 case SIOCBONDRELEASE
:
4035 res
= bond_release(bond_dev
, slave_dev
);
4037 case BOND_SETHWADDR_OLD
:
4038 case SIOCBONDSETHWADDR
:
4039 res
= bond_sethwaddr(bond_dev
, slave_dev
);
4041 case BOND_CHANGE_ACTIVE_OLD
:
4042 case SIOCBONDCHANGEACTIVE
:
4043 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
4052 up_write(&(bonding_rwsem
));
4056 static void bond_set_multicast_list(struct net_device
*bond_dev
)
4058 struct bonding
*bond
= netdev_priv(bond_dev
);
4059 struct dev_mc_list
*dmi
;
4062 * Do promisc before checking multicast_mode
4064 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
)) {
4066 * FIXME: Need to handle the error when one of the multi-slaves
4069 bond_set_promiscuity(bond
, 1);
4072 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
)) {
4073 bond_set_promiscuity(bond
, -1);
4076 /* set allmulti flag to slaves */
4077 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
)) {
4079 * FIXME: Need to handle the error when one of the multi-slaves
4082 bond_set_allmulti(bond
, 1);
4085 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
)) {
4086 bond_set_allmulti(bond
, -1);
4089 read_lock(&bond
->lock
);
4091 bond
->flags
= bond_dev
->flags
;
4093 /* looking for addresses to add to slaves' mc list */
4094 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
4095 if (!bond_mc_list_find_dmi(dmi
, bond
->mc_list
)) {
4096 bond_mc_add(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4100 /* looking for addresses to delete from slaves' list */
4101 for (dmi
= bond
->mc_list
; dmi
; dmi
= dmi
->next
) {
4102 if (!bond_mc_list_find_dmi(dmi
, bond_dev
->mc_list
)) {
4103 bond_mc_delete(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4107 /* save master's multicast list */
4108 bond_mc_list_destroy(bond
);
4109 bond_mc_list_copy(bond_dev
->mc_list
, bond
, GFP_ATOMIC
);
4111 read_unlock(&bond
->lock
);
4114 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
4116 struct bonding
*bond
= netdev_priv(dev
);
4117 struct slave
*slave
= bond
->first_slave
;
4120 const struct net_device_ops
*slave_ops
4121 = slave
->dev
->netdev_ops
;
4122 if (slave_ops
->ndo_neigh_setup
)
4123 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
4129 * Change the MTU of all of a master's slaves to match the master
4131 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
4133 struct bonding
*bond
= netdev_priv(bond_dev
);
4134 struct slave
*slave
, *stop_at
;
4138 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
4139 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
4141 /* Can't hold bond->lock with bh disabled here since
4142 * some base drivers panic. On the other hand we can't
4143 * hold bond->lock without bh disabled because we'll
4144 * deadlock. The only solution is to rely on the fact
4145 * that we're under rtnl_lock here, and the slaves
4146 * list won't change. This doesn't solve the problem
4147 * of setting the slave's MTU while it is
4148 * transmitting, but the assumption is that the base
4149 * driver can handle that.
4151 * TODO: figure out a way to safely iterate the slaves
4152 * list, but without holding a lock around the actual
4153 * call to the base driver.
4156 bond_for_each_slave(bond
, slave
, i
) {
4157 pr_debug("s %p s->p %p c_m %p\n", slave
,
4158 slave
->prev
, slave
->dev
->netdev_ops
->ndo_change_mtu
);
4160 res
= dev_set_mtu(slave
->dev
, new_mtu
);
4163 /* If we failed to set the slave's mtu to the new value
4164 * we must abort the operation even in ACTIVE_BACKUP
4165 * mode, because if we allow the backup slaves to have
4166 * different mtu values than the active slave we'll
4167 * need to change their mtu when doing a failover. That
4168 * means changing their mtu from timer context, which
4169 * is probably not a good idea.
4171 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4176 bond_dev
->mtu
= new_mtu
;
4181 /* unwind from head to the slave that failed */
4183 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4186 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
4188 pr_debug("unwind err %d dev %s\n", tmp_res
,
4199 * Note that many devices must be down to change the HW address, and
4200 * downing the master releases all slaves. We can make bonds full of
4201 * bonding devices to test this, however.
4203 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4205 struct bonding
*bond
= netdev_priv(bond_dev
);
4206 struct sockaddr
*sa
= addr
, tmp_sa
;
4207 struct slave
*slave
, *stop_at
;
4211 if (bond
->params
.mode
== BOND_MODE_ALB
)
4212 return bond_alb_set_mac_address(bond_dev
, addr
);
4215 pr_debug("bond=%p, name=%s\n", bond
, (bond_dev
? bond_dev
->name
: "None"));
4218 * If fail_over_mac is set to active, do nothing and return
4219 * success. Returning an error causes ifenslave to fail.
4221 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4224 if (!is_valid_ether_addr(sa
->sa_data
)) {
4225 return -EADDRNOTAVAIL
;
4228 /* Can't hold bond->lock with bh disabled here since
4229 * some base drivers panic. On the other hand we can't
4230 * hold bond->lock without bh disabled because we'll
4231 * deadlock. The only solution is to rely on the fact
4232 * that we're under rtnl_lock here, and the slaves
4233 * list won't change. This doesn't solve the problem
4234 * of setting the slave's hw address while it is
4235 * transmitting, but the assumption is that the base
4236 * driver can handle that.
4238 * TODO: figure out a way to safely iterate the slaves
4239 * list, but without holding a lock around the actual
4240 * call to the base driver.
4243 bond_for_each_slave(bond
, slave
, i
) {
4244 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
4245 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
4247 if (slave_ops
->ndo_set_mac_address
== NULL
) {
4249 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
4253 res
= dev_set_mac_address(slave
->dev
, addr
);
4255 /* TODO: consider downing the slave
4257 * User should expect communications
4258 * breakage anyway until ARP finish
4261 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4267 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4271 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4272 tmp_sa
.sa_family
= bond_dev
->type
;
4274 /* unwind from head to the slave that failed */
4276 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4279 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4281 pr_debug("unwind err %d dev %s\n", tmp_res
,
4289 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4291 struct bonding
*bond
= netdev_priv(bond_dev
);
4292 struct slave
*slave
, *start_at
;
4293 int i
, slave_no
, res
= 1;
4295 read_lock(&bond
->lock
);
4297 if (!BOND_IS_OK(bond
)) {
4302 * Concurrent TX may collide on rr_tx_counter; we accept that
4303 * as being rare enough not to justify using an atomic op here
4305 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4307 bond_for_each_slave(bond
, slave
, i
) {
4315 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4316 if (IS_UP(slave
->dev
) &&
4317 (slave
->link
== BOND_LINK_UP
) &&
4318 (slave
->state
== BOND_STATE_ACTIVE
)) {
4319 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4326 /* no suitable interface, frame not sent */
4329 read_unlock(&bond
->lock
);
4335 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4336 * the bond has a usable interface.
4338 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4340 struct bonding
*bond
= netdev_priv(bond_dev
);
4343 read_lock(&bond
->lock
);
4344 read_lock(&bond
->curr_slave_lock
);
4346 if (!BOND_IS_OK(bond
)) {
4350 if (!bond
->curr_active_slave
)
4353 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4357 /* no suitable interface, frame not sent */
4360 read_unlock(&bond
->curr_slave_lock
);
4361 read_unlock(&bond
->lock
);
4366 * In bond_xmit_xor() , we determine the output device by using a pre-
4367 * determined xmit_hash_policy(), If the selected device is not enabled,
4368 * find the next active slave.
4370 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4372 struct bonding
*bond
= netdev_priv(bond_dev
);
4373 struct slave
*slave
, *start_at
;
4378 read_lock(&bond
->lock
);
4380 if (!BOND_IS_OK(bond
)) {
4384 slave_no
= bond
->xmit_hash_policy(skb
, bond_dev
, bond
->slave_cnt
);
4386 bond_for_each_slave(bond
, slave
, i
) {
4395 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4396 if (IS_UP(slave
->dev
) &&
4397 (slave
->link
== BOND_LINK_UP
) &&
4398 (slave
->state
== BOND_STATE_ACTIVE
)) {
4399 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4406 /* no suitable interface, frame not sent */
4409 read_unlock(&bond
->lock
);
4414 * in broadcast mode, we send everything to all usable interfaces.
4416 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4418 struct bonding
*bond
= netdev_priv(bond_dev
);
4419 struct slave
*slave
, *start_at
;
4420 struct net_device
*tx_dev
= NULL
;
4424 read_lock(&bond
->lock
);
4426 if (!BOND_IS_OK(bond
)) {
4430 read_lock(&bond
->curr_slave_lock
);
4431 start_at
= bond
->curr_active_slave
;
4432 read_unlock(&bond
->curr_slave_lock
);
4438 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4439 if (IS_UP(slave
->dev
) &&
4440 (slave
->link
== BOND_LINK_UP
) &&
4441 (slave
->state
== BOND_STATE_ACTIVE
)) {
4443 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4445 printk(KERN_ERR DRV_NAME
4446 ": %s: Error: bond_xmit_broadcast(): "
4447 "skb_clone() failed\n",
4452 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4454 dev_kfree_skb(skb2
);
4458 tx_dev
= slave
->dev
;
4463 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4468 /* no suitable interface, frame not sent */
4471 /* frame sent to all suitable interfaces */
4472 read_unlock(&bond
->lock
);
4476 /*------------------------- Device initialization ---------------------------*/
4478 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4480 switch (bond
->params
.xmit_policy
) {
4481 case BOND_XMIT_POLICY_LAYER23
:
4482 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4484 case BOND_XMIT_POLICY_LAYER34
:
4485 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4487 case BOND_XMIT_POLICY_LAYER2
:
4489 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4494 static int bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4496 const struct bonding
*bond
= netdev_priv(dev
);
4498 switch (bond
->params
.mode
) {
4499 case BOND_MODE_ROUNDROBIN
:
4500 return bond_xmit_roundrobin(skb
, dev
);
4501 case BOND_MODE_ACTIVEBACKUP
:
4502 return bond_xmit_activebackup(skb
, dev
);
4504 return bond_xmit_xor(skb
, dev
);
4505 case BOND_MODE_BROADCAST
:
4506 return bond_xmit_broadcast(skb
, dev
);
4507 case BOND_MODE_8023AD
:
4508 return bond_3ad_xmit_xor(skb
, dev
);
4511 return bond_alb_xmit(skb
, dev
);
4513 /* Should never happen, mode already checked */
4514 printk(KERN_ERR DRV_NAME
": %s: Error: Unknown bonding mode %d\n",
4515 dev
->name
, bond
->params
.mode
);
4518 return NETDEV_TX_OK
;
4524 * set bond mode specific net device operations
4526 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4528 struct net_device
*bond_dev
= bond
->dev
;
4531 case BOND_MODE_ROUNDROBIN
:
4533 case BOND_MODE_ACTIVEBACKUP
:
4536 bond_set_xmit_hash_policy(bond
);
4538 case BOND_MODE_BROADCAST
:
4540 case BOND_MODE_8023AD
:
4541 bond_set_master_3ad_flags(bond
);
4542 bond_set_xmit_hash_policy(bond
);
4545 bond_set_master_alb_flags(bond
);
4550 /* Should never happen, mode already checked */
4551 printk(KERN_ERR DRV_NAME
4552 ": %s: Error: Unknown bonding mode %d\n",
4559 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4560 struct ethtool_drvinfo
*drvinfo
)
4562 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4563 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4564 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4567 static const struct ethtool_ops bond_ethtool_ops
= {
4568 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4569 .get_link
= ethtool_op_get_link
,
4570 .get_tx_csum
= ethtool_op_get_tx_csum
,
4571 .get_sg
= ethtool_op_get_sg
,
4572 .get_tso
= ethtool_op_get_tso
,
4573 .get_ufo
= ethtool_op_get_ufo
,
4574 .get_flags
= ethtool_op_get_flags
,
4577 static const struct net_device_ops bond_netdev_ops
= {
4578 .ndo_open
= bond_open
,
4579 .ndo_stop
= bond_close
,
4580 .ndo_start_xmit
= bond_start_xmit
,
4581 .ndo_get_stats
= bond_get_stats
,
4582 .ndo_do_ioctl
= bond_do_ioctl
,
4583 .ndo_set_multicast_list
= bond_set_multicast_list
,
4584 .ndo_change_mtu
= bond_change_mtu
,
4585 .ndo_set_mac_address
= bond_set_mac_address
,
4586 .ndo_neigh_setup
= bond_neigh_setup
,
4587 .ndo_vlan_rx_register
= bond_vlan_rx_register
,
4588 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4589 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4593 * Does not allocate but creates a /proc entry.
4596 static int bond_init(struct net_device
*bond_dev
, struct bond_params
*params
)
4598 struct bonding
*bond
= netdev_priv(bond_dev
);
4600 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4602 /* initialize rwlocks */
4603 rwlock_init(&bond
->lock
);
4604 rwlock_init(&bond
->curr_slave_lock
);
4606 bond
->params
= *params
; /* copy params struct */
4608 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4612 /* Initialize pointers */
4613 bond
->first_slave
= NULL
;
4614 bond
->curr_active_slave
= NULL
;
4615 bond
->current_arp_slave
= NULL
;
4616 bond
->primary_slave
= NULL
;
4617 bond
->dev
= bond_dev
;
4618 bond
->send_grat_arp
= 0;
4619 bond
->send_unsol_na
= 0;
4620 bond
->setup_by_slave
= 0;
4621 INIT_LIST_HEAD(&bond
->vlan_list
);
4623 /* Initialize the device entry points */
4624 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4625 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4626 bond_set_mode_ops(bond
, bond
->params
.mode
);
4628 bond_dev
->destructor
= bond_destructor
;
4630 /* Initialize the device options */
4631 bond_dev
->tx_queue_len
= 0;
4632 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4633 bond_dev
->priv_flags
|= IFF_BONDING
;
4634 if (bond
->params
.arp_interval
)
4635 bond_dev
->priv_flags
|= IFF_MASTER_ARPMON
;
4637 /* At first, we block adding VLANs. That's the only way to
4638 * prevent problems that occur when adding VLANs over an
4639 * empty bond. The block will be removed once non-challenged
4640 * slaves are enslaved.
4642 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4644 /* don't acquire bond device's netif_tx_lock when
4646 bond_dev
->features
|= NETIF_F_LLTX
;
4648 /* By default, we declare the bond to be fully
4649 * VLAN hardware accelerated capable. Special
4650 * care is taken in the various xmit functions
4651 * when there are slaves that are not hw accel
4654 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4655 NETIF_F_HW_VLAN_RX
|
4656 NETIF_F_HW_VLAN_FILTER
);
4658 bond_create_proc_entry(bond
);
4659 list_add_tail(&bond
->bond_list
, &bond_dev_list
);
4664 static void bond_work_cancel_all(struct bonding
*bond
)
4666 write_lock_bh(&bond
->lock
);
4667 bond
->kill_timers
= 1;
4668 write_unlock_bh(&bond
->lock
);
4670 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4671 cancel_delayed_work(&bond
->mii_work
);
4673 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4674 cancel_delayed_work(&bond
->arp_work
);
4676 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4677 delayed_work_pending(&bond
->alb_work
))
4678 cancel_delayed_work(&bond
->alb_work
);
4680 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4681 delayed_work_pending(&bond
->ad_work
))
4682 cancel_delayed_work(&bond
->ad_work
);
4685 /* De-initialize device specific data.
4686 * Caller must hold rtnl_lock.
4688 static void bond_deinit(struct net_device
*bond_dev
)
4690 struct bonding
*bond
= netdev_priv(bond_dev
);
4692 list_del(&bond
->bond_list
);
4694 bond_work_cancel_all(bond
);
4696 bond_remove_proc_entry(bond
);
4699 /* Unregister and free all bond devices.
4700 * Caller must hold rtnl_lock.
4702 static void bond_free_all(void)
4704 struct bonding
*bond
, *nxt
;
4706 list_for_each_entry_safe(bond
, nxt
, &bond_dev_list
, bond_list
) {
4707 struct net_device
*bond_dev
= bond
->dev
;
4709 bond_work_cancel_all(bond
);
4710 /* Release the bonded slaves */
4711 bond_release_all(bond_dev
);
4715 bond_destroy_proc_dir();
4718 /*------------------------- Module initialization ---------------------------*/
4721 * Convert string input module parms. Accept either the
4722 * number of the mode or its string name. A bit complicated because
4723 * some mode names are substrings of other names, and calls from sysfs
4724 * may have whitespace in the name (trailing newlines, for example).
4726 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4728 int modeint
= -1, i
, rv
;
4729 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4731 for (p
= (char *)buf
; *p
; p
++)
4732 if (!(isdigit(*p
) || isspace(*p
)))
4736 rv
= sscanf(buf
, "%20s", modestr
);
4738 rv
= sscanf(buf
, "%d", &modeint
);
4743 for (i
= 0; tbl
[i
].modename
; i
++) {
4744 if (modeint
== tbl
[i
].mode
)
4746 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4753 static int bond_check_params(struct bond_params
*params
)
4755 int arp_validate_value
, fail_over_mac_value
;
4758 * Convert string parameters.
4761 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4762 if (bond_mode
== -1) {
4763 printk(KERN_ERR DRV_NAME
4764 ": Error: Invalid bonding mode \"%s\"\n",
4765 mode
== NULL
? "NULL" : mode
);
4770 if (xmit_hash_policy
) {
4771 if ((bond_mode
!= BOND_MODE_XOR
) &&
4772 (bond_mode
!= BOND_MODE_8023AD
)) {
4773 printk(KERN_INFO DRV_NAME
4774 ": xor_mode param is irrelevant in mode %s\n",
4775 bond_mode_name(bond_mode
));
4777 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4779 if (xmit_hashtype
== -1) {
4780 printk(KERN_ERR DRV_NAME
4781 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4782 xmit_hash_policy
== NULL
? "NULL" :
4790 if (bond_mode
!= BOND_MODE_8023AD
) {
4791 printk(KERN_INFO DRV_NAME
4792 ": lacp_rate param is irrelevant in mode %s\n",
4793 bond_mode_name(bond_mode
));
4795 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4796 if (lacp_fast
== -1) {
4797 printk(KERN_ERR DRV_NAME
4798 ": Error: Invalid lacp rate \"%s\"\n",
4799 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4806 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4807 if (params
->ad_select
== -1) {
4808 printk(KERN_ERR DRV_NAME
4809 ": Error: Invalid ad_select \"%s\"\n",
4810 ad_select
== NULL
? "NULL" : ad_select
);
4814 if (bond_mode
!= BOND_MODE_8023AD
) {
4815 printk(KERN_WARNING DRV_NAME
4816 ": ad_select param only affects 802.3ad mode\n");
4819 params
->ad_select
= BOND_AD_STABLE
;
4822 if (max_bonds
< 0 || max_bonds
> INT_MAX
) {
4823 printk(KERN_WARNING DRV_NAME
4824 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4825 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4826 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4827 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4831 printk(KERN_WARNING DRV_NAME
4832 ": Warning: miimon module parameter (%d), "
4833 "not in range 0-%d, so it was reset to %d\n",
4834 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4835 miimon
= BOND_LINK_MON_INTERV
;
4839 printk(KERN_WARNING DRV_NAME
4840 ": Warning: updelay module parameter (%d), "
4841 "not in range 0-%d, so it was reset to 0\n",
4846 if (downdelay
< 0) {
4847 printk(KERN_WARNING DRV_NAME
4848 ": Warning: downdelay module parameter (%d), "
4849 "not in range 0-%d, so it was reset to 0\n",
4850 downdelay
, INT_MAX
);
4854 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4855 printk(KERN_WARNING DRV_NAME
4856 ": Warning: use_carrier module parameter (%d), "
4857 "not of valid value (0/1), so it was set to 1\n",
4862 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
4863 printk(KERN_WARNING DRV_NAME
4864 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4865 "was reset to 1 \n", num_grat_arp
);
4869 if (num_unsol_na
< 0 || num_unsol_na
> 255) {
4870 printk(KERN_WARNING DRV_NAME
4871 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4872 "was reset to 1 \n", num_unsol_na
);
4876 /* reset values for 802.3ad */
4877 if (bond_mode
== BOND_MODE_8023AD
) {
4879 printk(KERN_WARNING DRV_NAME
4880 ": Warning: miimon must be specified, "
4881 "otherwise bonding will not detect link "
4882 "failure, speed and duplex which are "
4883 "essential for 802.3ad operation\n");
4884 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4889 /* reset values for TLB/ALB */
4890 if ((bond_mode
== BOND_MODE_TLB
) ||
4891 (bond_mode
== BOND_MODE_ALB
)) {
4893 printk(KERN_WARNING DRV_NAME
4894 ": Warning: miimon must be specified, "
4895 "otherwise bonding will not detect link "
4896 "failure and link speed which are essential "
4897 "for TLB/ALB load balancing\n");
4898 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4903 if (bond_mode
== BOND_MODE_ALB
) {
4904 printk(KERN_NOTICE DRV_NAME
4905 ": In ALB mode you might experience client "
4906 "disconnections upon reconnection of a link if the "
4907 "bonding module updelay parameter (%d msec) is "
4908 "incompatible with the forwarding delay time of the "
4914 if (updelay
|| downdelay
) {
4915 /* just warn the user the up/down delay will have
4916 * no effect since miimon is zero...
4918 printk(KERN_WARNING DRV_NAME
4919 ": Warning: miimon module parameter not set "
4920 "and updelay (%d) or downdelay (%d) module "
4921 "parameter is set; updelay and downdelay have "
4922 "no effect unless miimon is set\n",
4923 updelay
, downdelay
);
4926 /* don't allow arp monitoring */
4928 printk(KERN_WARNING DRV_NAME
4929 ": Warning: miimon (%d) and arp_interval (%d) "
4930 "can't be used simultaneously, disabling ARP "
4932 miimon
, arp_interval
);
4936 if ((updelay
% miimon
) != 0) {
4937 printk(KERN_WARNING DRV_NAME
4938 ": Warning: updelay (%d) is not a multiple "
4939 "of miimon (%d), updelay rounded to %d ms\n",
4940 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4945 if ((downdelay
% miimon
) != 0) {
4946 printk(KERN_WARNING DRV_NAME
4947 ": Warning: downdelay (%d) is not a multiple "
4948 "of miimon (%d), downdelay rounded to %d ms\n",
4950 (downdelay
/ miimon
) * miimon
);
4953 downdelay
/= miimon
;
4956 if (arp_interval
< 0) {
4957 printk(KERN_WARNING DRV_NAME
4958 ": Warning: arp_interval module parameter (%d) "
4959 ", not in range 0-%d, so it was reset to %d\n",
4960 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4961 arp_interval
= BOND_LINK_ARP_INTERV
;
4964 for (arp_ip_count
= 0;
4965 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4967 /* not complete check, but should be good enough to
4969 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4970 printk(KERN_WARNING DRV_NAME
4971 ": Warning: bad arp_ip_target module parameter "
4972 "(%s), ARP monitoring will not be performed\n",
4973 arp_ip_target
[arp_ip_count
]);
4976 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4977 arp_target
[arp_ip_count
] = ip
;
4981 if (arp_interval
&& !arp_ip_count
) {
4982 /* don't allow arping if no arp_ip_target given... */
4983 printk(KERN_WARNING DRV_NAME
4984 ": Warning: arp_interval module parameter (%d) "
4985 "specified without providing an arp_ip_target "
4986 "parameter, arp_interval was reset to 0\n",
4992 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4993 printk(KERN_ERR DRV_NAME
4994 ": arp_validate only supported in active-backup mode\n");
4997 if (!arp_interval
) {
4998 printk(KERN_ERR DRV_NAME
4999 ": arp_validate requires arp_interval\n");
5003 arp_validate_value
= bond_parse_parm(arp_validate
,
5005 if (arp_validate_value
== -1) {
5006 printk(KERN_ERR DRV_NAME
5007 ": Error: invalid arp_validate \"%s\"\n",
5008 arp_validate
== NULL
? "NULL" : arp_validate
);
5012 arp_validate_value
= 0;
5015 printk(KERN_INFO DRV_NAME
5016 ": MII link monitoring set to %d ms\n",
5018 } else if (arp_interval
) {
5021 printk(KERN_INFO DRV_NAME
5022 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5024 arp_validate_tbl
[arp_validate_value
].modename
,
5027 for (i
= 0; i
< arp_ip_count
; i
++)
5028 printk (" %s", arp_ip_target
[i
]);
5032 } else if (max_bonds
) {
5033 /* miimon and arp_interval not set, we need one so things
5034 * work as expected, see bonding.txt for details
5036 printk(KERN_WARNING DRV_NAME
5037 ": Warning: either miimon or arp_interval and "
5038 "arp_ip_target module parameters must be specified, "
5039 "otherwise bonding will not detect link failures! see "
5040 "bonding.txt for details.\n");
5043 if (primary
&& !USES_PRIMARY(bond_mode
)) {
5044 /* currently, using a primary only makes sense
5045 * in active backup, TLB or ALB modes
5047 printk(KERN_WARNING DRV_NAME
5048 ": Warning: %s primary device specified but has no "
5049 "effect in %s mode\n",
5050 primary
, bond_mode_name(bond_mode
));
5054 if (fail_over_mac
) {
5055 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
5057 if (fail_over_mac_value
== -1) {
5058 printk(KERN_ERR DRV_NAME
5059 ": Error: invalid fail_over_mac \"%s\"\n",
5060 arp_validate
== NULL
? "NULL" : arp_validate
);
5064 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
5065 printk(KERN_WARNING DRV_NAME
5066 ": Warning: fail_over_mac only affects "
5067 "active-backup mode.\n");
5069 fail_over_mac_value
= BOND_FOM_NONE
;
5072 /* fill params struct with the proper values */
5073 params
->mode
= bond_mode
;
5074 params
->xmit_policy
= xmit_hashtype
;
5075 params
->miimon
= miimon
;
5076 params
->num_grat_arp
= num_grat_arp
;
5077 params
->num_unsol_na
= num_unsol_na
;
5078 params
->arp_interval
= arp_interval
;
5079 params
->arp_validate
= arp_validate_value
;
5080 params
->updelay
= updelay
;
5081 params
->downdelay
= downdelay
;
5082 params
->use_carrier
= use_carrier
;
5083 params
->lacp_fast
= lacp_fast
;
5084 params
->primary
[0] = 0;
5085 params
->fail_over_mac
= fail_over_mac_value
;
5088 strncpy(params
->primary
, primary
, IFNAMSIZ
);
5089 params
->primary
[IFNAMSIZ
- 1] = 0;
5092 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
5097 static struct lock_class_key bonding_netdev_xmit_lock_key
;
5098 static struct lock_class_key bonding_netdev_addr_lock_key
;
5100 static void bond_set_lockdep_class_one(struct net_device
*dev
,
5101 struct netdev_queue
*txq
,
5104 lockdep_set_class(&txq
->_xmit_lock
,
5105 &bonding_netdev_xmit_lock_key
);
5108 static void bond_set_lockdep_class(struct net_device
*dev
)
5110 lockdep_set_class(&dev
->addr_list_lock
,
5111 &bonding_netdev_addr_lock_key
);
5112 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
5115 /* Create a new bond based on the specified name and bonding parameters.
5116 * If name is NULL, obtain a suitable "bond%d" name for us.
5117 * Caller must NOT hold rtnl_lock; we need to release it here before we
5118 * set up our sysfs entries.
5120 int bond_create(char *name
, struct bond_params
*params
)
5122 struct net_device
*bond_dev
;
5123 struct bonding
*bond
;
5127 down_write(&bonding_rwsem
);
5129 /* Check to see if the bond already exists. */
5131 list_for_each_entry(bond
, &bond_dev_list
, bond_list
)
5132 if (strnicmp(bond
->dev
->name
, name
, IFNAMSIZ
) == 0) {
5133 printk(KERN_ERR DRV_NAME
5134 ": cannot add bond %s; it already exists\n",
5141 bond_dev
= alloc_netdev(sizeof(struct bonding
), name
? name
: "",
5144 printk(KERN_ERR DRV_NAME
5145 ": %s: eek! can't alloc netdev!\n",
5152 res
= dev_alloc_name(bond_dev
, "bond%d");
5157 /* bond_init() must be called after dev_alloc_name() (for the
5158 * /proc files), but before register_netdevice(), because we
5159 * need to set function pointers.
5162 res
= bond_init(bond_dev
, params
);
5167 res
= register_netdevice(bond_dev
);
5172 bond_set_lockdep_class(bond_dev
);
5174 netif_carrier_off(bond_dev
);
5176 up_write(&bonding_rwsem
);
5177 rtnl_unlock(); /* allows sysfs registration of net device */
5178 res
= bond_create_sysfs_entry(netdev_priv(bond_dev
));
5186 down_write(&bonding_rwsem
);
5187 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)
5203 printk(KERN_INFO
"%s", version
);
5205 res
= bond_check_params(&bonding_defaults
);
5210 bond_create_proc_dir();
5212 init_rwsem(&bonding_rwsem
);
5214 for (i
= 0; i
< max_bonds
; i
++) {
5215 res
= bond_create(NULL
, &bonding_defaults
);
5220 res
= bond_create_sysfs();
5224 register_netdevice_notifier(&bond_netdev_notifier
);
5225 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5226 bond_register_ipv6_notifier();
5238 static void __exit
bonding_exit(void)
5240 unregister_netdevice_notifier(&bond_netdev_notifier
);
5241 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5242 bond_unregister_ipv6_notifier();
5244 bond_destroy_sysfs();
5251 module_init(bonding_init
);
5252 module_exit(bonding_exit
);
5253 MODULE_LICENSE("GPL");
5254 MODULE_VERSION(DRV_VERSION
);
5255 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5256 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5257 MODULE_SUPPORTED_DEVICE("most ethernet devices");