2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 //#define BONDING_DEBUG 1
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <asm/system.h>
59 #include <asm/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV 0
88 #define BOND_LINK_ARP_INTERV 0
90 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
91 static int num_grat_arp
= 1;
92 static int miimon
= BOND_LINK_MON_INTERV
;
93 static int updelay
= 0;
94 static int downdelay
= 0;
95 static int use_carrier
= 1;
96 static char *mode
= NULL
;
97 static char *primary
= NULL
;
98 static char *lacp_rate
= 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(miimon
, int, 0);
111 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
112 module_param(updelay
, int, 0);
113 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
114 module_param(downdelay
, int, 0);
115 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
117 module_param(use_carrier
, int, 0);
118 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
119 "0 for off, 1 for on (default)");
120 module_param(mode
, charp
, 0);
121 MODULE_PARM_DESC(mode
, "Mode of operation : 0 for balance-rr, "
122 "1 for active-backup, 2 for balance-xor, "
123 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
124 "6 for balance-alb");
125 module_param(primary
, charp
, 0);
126 MODULE_PARM_DESC(primary
, "Primary network device to use");
127 module_param(lacp_rate
, charp
, 0);
128 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner "
130 module_param(xmit_hash_policy
, charp
, 0);
131 MODULE_PARM_DESC(xmit_hash_policy
, "XOR hashing method: 0 for layer 2 (default)"
132 ", 1 for layer 3+4");
133 module_param(arp_interval
, int, 0);
134 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
135 module_param_array(arp_ip_target
, charp
, NULL
, 0);
136 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
137 module_param(arp_validate
, charp
, 0);
138 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes: none (default), active, backup or all");
139 module_param(fail_over_mac
, charp
, 0);
140 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
142 /*----------------------------- Global variables ----------------------------*/
144 static const char * const version
=
145 DRV_DESCRIPTION
": v" DRV_VERSION
" (" DRV_RELDATE
")\n";
147 LIST_HEAD(bond_dev_list
);
149 #ifdef CONFIG_PROC_FS
150 static struct proc_dir_entry
*bond_proc_dir
= NULL
;
153 extern struct rw_semaphore bonding_rwsem
;
154 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
] = { 0, } ;
155 static int arp_ip_count
= 0;
156 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
157 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
158 static int lacp_fast
= 0;
161 struct bond_parm_tbl bond_lacp_tbl
[] = {
162 { "slow", AD_LACP_SLOW
},
163 { "fast", AD_LACP_FAST
},
167 struct bond_parm_tbl bond_mode_tbl
[] = {
168 { "balance-rr", BOND_MODE_ROUNDROBIN
},
169 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
170 { "balance-xor", BOND_MODE_XOR
},
171 { "broadcast", BOND_MODE_BROADCAST
},
172 { "802.3ad", BOND_MODE_8023AD
},
173 { "balance-tlb", BOND_MODE_TLB
},
174 { "balance-alb", BOND_MODE_ALB
},
178 struct bond_parm_tbl xmit_hashtype_tbl
[] = {
179 { "layer2", BOND_XMIT_POLICY_LAYER2
},
180 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
181 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
185 struct bond_parm_tbl arp_validate_tbl
[] = {
186 { "none", BOND_ARP_VALIDATE_NONE
},
187 { "active", BOND_ARP_VALIDATE_ACTIVE
},
188 { "backup", BOND_ARP_VALIDATE_BACKUP
},
189 { "all", BOND_ARP_VALIDATE_ALL
},
193 struct bond_parm_tbl fail_over_mac_tbl
[] = {
194 { "none", BOND_FOM_NONE
},
195 { "active", BOND_FOM_ACTIVE
},
196 { "follow", BOND_FOM_FOLLOW
},
200 /*-------------------------- Forward declarations ---------------------------*/
202 static void bond_send_gratuitous_arp(struct bonding
*bond
);
203 static void bond_deinit(struct net_device
*bond_dev
);
205 /*---------------------------- General routines -----------------------------*/
207 static const char *bond_mode_name(int mode
)
210 case BOND_MODE_ROUNDROBIN
:
211 return "load balancing (round-robin)";
212 case BOND_MODE_ACTIVEBACKUP
:
213 return "fault-tolerance (active-backup)";
215 return "load balancing (xor)";
216 case BOND_MODE_BROADCAST
:
217 return "fault-tolerance (broadcast)";
218 case BOND_MODE_8023AD
:
219 return "IEEE 802.3ad Dynamic link aggregation";
221 return "transmit load balancing";
223 return "adaptive load balancing";
229 /*---------------------------------- VLAN -----------------------------------*/
232 * bond_add_vlan - add a new vlan id on bond
233 * @bond: bond that got the notification
234 * @vlan_id: the vlan id to add
236 * Returns -ENOMEM if allocation failed.
238 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
240 struct vlan_entry
*vlan
;
242 dprintk("bond: %s, vlan id %d\n",
243 (bond
? bond
->dev
->name
: "None"), vlan_id
);
245 vlan
= kmalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
250 INIT_LIST_HEAD(&vlan
->vlan_list
);
251 vlan
->vlan_id
= vlan_id
;
254 write_lock_bh(&bond
->lock
);
256 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
258 write_unlock_bh(&bond
->lock
);
260 dprintk("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
266 * bond_del_vlan - delete a vlan id from bond
267 * @bond: bond that got the notification
268 * @vlan_id: the vlan id to delete
270 * returns -ENODEV if @vlan_id was not found in @bond.
272 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
274 struct vlan_entry
*vlan
;
277 dprintk("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
279 write_lock_bh(&bond
->lock
);
281 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
282 if (vlan
->vlan_id
== vlan_id
) {
283 list_del(&vlan
->vlan_list
);
285 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
286 (bond
->params
.mode
== BOND_MODE_ALB
)) {
287 bond_alb_clear_vlan(bond
, vlan_id
);
290 dprintk("removed VLAN ID %d from bond %s\n", vlan_id
,
295 if (list_empty(&bond
->vlan_list
) &&
296 (bond
->slave_cnt
== 0)) {
297 /* Last VLAN removed and no slaves, so
298 * restore block on adding VLANs. This will
299 * be removed once new slaves that are not
300 * VLAN challenged will be added.
302 bond
->dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
310 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id
,
314 write_unlock_bh(&bond
->lock
);
319 * bond_has_challenged_slaves
320 * @bond: the bond we're working on
322 * Searches the slave list. Returns 1 if a vlan challenged slave
323 * was found, 0 otherwise.
325 * Assumes bond->lock is held.
327 static int bond_has_challenged_slaves(struct bonding
*bond
)
332 bond_for_each_slave(bond
, slave
, i
) {
333 if (slave
->dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
334 dprintk("found VLAN challenged slave - %s\n",
340 dprintk("no VLAN challenged slaves found\n");
345 * bond_next_vlan - safely skip to the next item in the vlans list.
346 * @bond: the bond we're working on
347 * @curr: item we're advancing from
349 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
350 * or @curr->next otherwise (even if it is @curr itself again).
352 * Caller must hold bond->lock
354 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
356 struct vlan_entry
*next
, *last
;
358 if (list_empty(&bond
->vlan_list
)) {
363 next
= list_entry(bond
->vlan_list
.next
,
364 struct vlan_entry
, vlan_list
);
366 last
= list_entry(bond
->vlan_list
.prev
,
367 struct vlan_entry
, vlan_list
);
369 next
= list_entry(bond
->vlan_list
.next
,
370 struct vlan_entry
, vlan_list
);
372 next
= list_entry(curr
->vlan_list
.next
,
373 struct vlan_entry
, vlan_list
);
381 * bond_dev_queue_xmit - Prepare skb for xmit.
383 * @bond: bond device that got this skb for tx.
384 * @skb: hw accel VLAN tagged skb to transmit
385 * @slave_dev: slave that is supposed to xmit this skbuff
387 * When the bond gets an skb to transmit that is
388 * already hardware accelerated VLAN tagged, and it
389 * needs to relay this skb to a slave that is not
390 * hw accel capable, the skb needs to be "unaccelerated",
391 * i.e. strip the hwaccel tag and re-insert it as part
394 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
)
396 unsigned short uninitialized_var(vlan_id
);
398 if (!list_empty(&bond
->vlan_list
) &&
399 !(slave_dev
->features
& NETIF_F_HW_VLAN_TX
) &&
400 vlan_get_tag(skb
, &vlan_id
) == 0) {
401 skb
->dev
= slave_dev
;
402 skb
= vlan_put_tag(skb
, vlan_id
);
404 /* vlan_put_tag() frees the skb in case of error,
405 * so return success here so the calling functions
406 * won't attempt to free is again.
411 skb
->dev
= slave_dev
;
421 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
422 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
424 * a. This operation is performed in IOCTL context,
425 * b. The operation is protected by the RTNL semaphore in the 8021q code,
426 * c. Holding a lock with BH disabled while directly calling a base driver
427 * entry point is generally a BAD idea.
429 * The design of synchronization/protection for this operation in the 8021q
430 * module is good for one or more VLAN devices over a single physical device
431 * and cannot be extended for a teaming solution like bonding, so there is a
432 * potential race condition here where a net device from the vlan group might
433 * be referenced (either by a base driver or the 8021q code) while it is being
434 * removed from the system. However, it turns out we're not making matters
435 * worse, and if it works for regular VLAN usage it will work here too.
439 * bond_vlan_rx_register - Propagates registration to slaves
440 * @bond_dev: bonding net device that got called
441 * @grp: vlan group being registered
443 static void bond_vlan_rx_register(struct net_device
*bond_dev
, struct vlan_group
*grp
)
445 struct bonding
*bond
= bond_dev
->priv
;
451 bond_for_each_slave(bond
, slave
, i
) {
452 struct net_device
*slave_dev
= slave
->dev
;
454 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
455 slave_dev
->vlan_rx_register
) {
456 slave_dev
->vlan_rx_register(slave_dev
, grp
);
462 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
463 * @bond_dev: bonding net device that got called
464 * @vid: vlan id being added
466 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
468 struct bonding
*bond
= bond_dev
->priv
;
472 bond_for_each_slave(bond
, slave
, i
) {
473 struct net_device
*slave_dev
= slave
->dev
;
475 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
476 slave_dev
->vlan_rx_add_vid
) {
477 slave_dev
->vlan_rx_add_vid(slave_dev
, vid
);
481 res
= bond_add_vlan(bond
, vid
);
483 printk(KERN_ERR DRV_NAME
484 ": %s: Error: Failed to add vlan id %d\n",
485 bond_dev
->name
, vid
);
490 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
491 * @bond_dev: bonding net device that got called
492 * @vid: vlan id being removed
494 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
496 struct bonding
*bond
= bond_dev
->priv
;
498 struct net_device
*vlan_dev
;
501 bond_for_each_slave(bond
, slave
, i
) {
502 struct net_device
*slave_dev
= slave
->dev
;
504 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
505 slave_dev
->vlan_rx_kill_vid
) {
506 /* Save and then restore vlan_dev in the grp array,
507 * since the slave's driver might clear it.
509 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vid
);
510 slave_dev
->vlan_rx_kill_vid(slave_dev
, vid
);
511 vlan_group_set_device(bond
->vlgrp
, vid
, vlan_dev
);
515 res
= bond_del_vlan(bond
, vid
);
517 printk(KERN_ERR DRV_NAME
518 ": %s: Error: Failed to remove vlan id %d\n",
519 bond_dev
->name
, vid
);
523 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
525 struct vlan_entry
*vlan
;
527 write_lock_bh(&bond
->lock
);
529 if (list_empty(&bond
->vlan_list
)) {
533 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
534 slave_dev
->vlan_rx_register
) {
535 slave_dev
->vlan_rx_register(slave_dev
, bond
->vlgrp
);
538 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
539 !(slave_dev
->vlan_rx_add_vid
)) {
543 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
544 slave_dev
->vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
548 write_unlock_bh(&bond
->lock
);
551 static void bond_del_vlans_from_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
553 struct vlan_entry
*vlan
;
554 struct net_device
*vlan_dev
;
556 write_lock_bh(&bond
->lock
);
558 if (list_empty(&bond
->vlan_list
)) {
562 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
563 !(slave_dev
->vlan_rx_kill_vid
)) {
567 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
568 /* Save and then restore vlan_dev in the grp array,
569 * since the slave's driver might clear it.
571 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
572 slave_dev
->vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
573 vlan_group_set_device(bond
->vlgrp
, vlan
->vlan_id
, vlan_dev
);
577 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
578 slave_dev
->vlan_rx_register
) {
579 slave_dev
->vlan_rx_register(slave_dev
, NULL
);
583 write_unlock_bh(&bond
->lock
);
586 /*------------------------------- Link status -------------------------------*/
589 * Set the carrier state for the master according to the state of its
590 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
591 * do special 802.3ad magic.
593 * Returns zero if carrier state does not change, nonzero if it does.
595 static int bond_set_carrier(struct bonding
*bond
)
600 if (bond
->slave_cnt
== 0)
603 if (bond
->params
.mode
== BOND_MODE_8023AD
)
604 return bond_3ad_set_carrier(bond
);
606 bond_for_each_slave(bond
, slave
, i
) {
607 if (slave
->link
== BOND_LINK_UP
) {
608 if (!netif_carrier_ok(bond
->dev
)) {
609 netif_carrier_on(bond
->dev
);
617 if (netif_carrier_ok(bond
->dev
)) {
618 netif_carrier_off(bond
->dev
);
625 * Get link speed and duplex from the slave's base driver
626 * using ethtool. If for some reason the call fails or the
627 * values are invalid, fake speed and duplex to 100/Full
630 static int bond_update_speed_duplex(struct slave
*slave
)
632 struct net_device
*slave_dev
= slave
->dev
;
633 struct ethtool_cmd etool
;
636 /* Fake speed and duplex */
637 slave
->speed
= SPEED_100
;
638 slave
->duplex
= DUPLEX_FULL
;
640 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
643 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
647 switch (etool
.speed
) {
657 switch (etool
.duplex
) {
665 slave
->speed
= etool
.speed
;
666 slave
->duplex
= etool
.duplex
;
672 * if <dev> supports MII link status reporting, check its link status.
674 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
675 * depening upon the setting of the use_carrier parameter.
677 * Return either BMSR_LSTATUS, meaning that the link is up (or we
678 * can't tell and just pretend it is), or 0, meaning that the link is
681 * If reporting is non-zero, instead of faking link up, return -1 if
682 * both ETHTOOL and MII ioctls fail (meaning the device does not
683 * support them). If use_carrier is set, return whatever it says.
684 * It'd be nice if there was a good way to tell if a driver supports
685 * netif_carrier, but there really isn't.
687 static int bond_check_dev_link(struct bonding
*bond
, struct net_device
*slave_dev
, int reporting
)
689 static int (* ioctl
)(struct net_device
*, struct ifreq
*, int);
691 struct mii_ioctl_data
*mii
;
693 if (bond
->params
.use_carrier
) {
694 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
697 ioctl
= slave_dev
->do_ioctl
;
699 /* TODO: set pointer to correct ioctl on a per team member */
700 /* bases to make this more efficient. that is, once */
701 /* we determine the correct ioctl, we will always */
702 /* call it and not the others for that team */
706 * We cannot assume that SIOCGMIIPHY will also read a
707 * register; not all network drivers (e.g., e100)
711 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
712 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
714 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
715 mii
->reg_num
= MII_BMSR
;
716 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0) {
717 return (mii
->val_out
& BMSR_LSTATUS
);
723 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
724 * attempt to get link status from it if the above MII ioctls fail.
726 if (slave_dev
->ethtool_ops
) {
727 if (slave_dev
->ethtool_ops
->get_link
) {
730 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
732 return link
? BMSR_LSTATUS
: 0;
737 * If reporting, report that either there's no dev->do_ioctl,
738 * or both SIOCGMIIREG and get_link failed (meaning that we
739 * cannot report link status). If not reporting, pretend
742 return (reporting
? -1 : BMSR_LSTATUS
);
745 /*----------------------------- Multicast list ------------------------------*/
748 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
750 static inline int bond_is_dmi_same(struct dev_mc_list
*dmi1
, struct dev_mc_list
*dmi2
)
752 return memcmp(dmi1
->dmi_addr
, dmi2
->dmi_addr
, dmi1
->dmi_addrlen
) == 0 &&
753 dmi1
->dmi_addrlen
== dmi2
->dmi_addrlen
;
757 * returns dmi entry if found, NULL otherwise
759 static struct dev_mc_list
*bond_mc_list_find_dmi(struct dev_mc_list
*dmi
, struct dev_mc_list
*mc_list
)
761 struct dev_mc_list
*idmi
;
763 for (idmi
= mc_list
; idmi
; idmi
= idmi
->next
) {
764 if (bond_is_dmi_same(dmi
, idmi
)) {
773 * Push the promiscuity flag down to appropriate slaves
775 static void 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 dev_set_promiscuity(bond
->curr_active_slave
->dev
, inc
);
785 bond_for_each_slave(bond
, slave
, i
) {
786 dev_set_promiscuity(slave
->dev
, inc
);
792 * Push the allmulti flag down to all slaves
794 static void bond_set_allmulti(struct bonding
*bond
, int inc
)
796 if (USES_PRIMARY(bond
->params
.mode
)) {
797 /* write lock already acquired */
798 if (bond
->curr_active_slave
) {
799 dev_set_allmulti(bond
->curr_active_slave
->dev
, inc
);
804 bond_for_each_slave(bond
, slave
, i
) {
805 dev_set_allmulti(slave
->dev
, inc
);
811 * Add a Multicast address to slaves
814 static void bond_mc_add(struct bonding
*bond
, void *addr
, int alen
)
816 if (USES_PRIMARY(bond
->params
.mode
)) {
817 /* write lock already acquired */
818 if (bond
->curr_active_slave
) {
819 dev_mc_add(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
824 bond_for_each_slave(bond
, slave
, i
) {
825 dev_mc_add(slave
->dev
, addr
, alen
, 0);
831 * Remove a multicast address from slave
834 static void bond_mc_delete(struct bonding
*bond
, void *addr
, int alen
)
836 if (USES_PRIMARY(bond
->params
.mode
)) {
837 /* write lock already acquired */
838 if (bond
->curr_active_slave
) {
839 dev_mc_delete(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
844 bond_for_each_slave(bond
, slave
, i
) {
845 dev_mc_delete(slave
->dev
, addr
, alen
, 0);
852 * Retrieve the list of registered multicast addresses for the bonding
853 * device and retransmit an IGMP JOIN request to the current active
856 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
858 struct in_device
*in_dev
;
859 struct ip_mc_list
*im
;
862 in_dev
= __in_dev_get_rcu(bond
->dev
);
864 for (im
= in_dev
->mc_list
; im
; im
= im
->next
) {
865 ip_mc_rejoin_group(im
);
873 * Totally destroys the mc_list in bond
875 static void bond_mc_list_destroy(struct bonding
*bond
)
877 struct dev_mc_list
*dmi
;
881 bond
->mc_list
= dmi
->next
;
885 bond
->mc_list
= NULL
;
889 * Copy all the Multicast addresses from src to the bonding device dst
891 static int bond_mc_list_copy(struct dev_mc_list
*mc_list
, struct bonding
*bond
,
894 struct dev_mc_list
*dmi
, *new_dmi
;
896 for (dmi
= mc_list
; dmi
; dmi
= dmi
->next
) {
897 new_dmi
= kmalloc(sizeof(struct dev_mc_list
), gfp_flag
);
900 /* FIXME: Potential memory leak !!! */
904 new_dmi
->next
= bond
->mc_list
;
905 bond
->mc_list
= new_dmi
;
906 new_dmi
->dmi_addrlen
= dmi
->dmi_addrlen
;
907 memcpy(new_dmi
->dmi_addr
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
908 new_dmi
->dmi_users
= dmi
->dmi_users
;
909 new_dmi
->dmi_gusers
= dmi
->dmi_gusers
;
916 * flush all members of flush->mc_list from device dev->mc_list
918 static void bond_mc_list_flush(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
920 struct bonding
*bond
= bond_dev
->priv
;
921 struct dev_mc_list
*dmi
;
923 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
924 dev_mc_delete(slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
927 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
928 /* del lacpdu mc addr from mc list */
929 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
931 dev_mc_delete(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
935 /*--------------------------- Active slave change ---------------------------*/
938 * Update the mc list and multicast-related flags for the new and
939 * old active slaves (if any) according to the multicast mode, and
940 * promiscuous flags unconditionally.
942 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
, struct slave
*old_active
)
944 struct dev_mc_list
*dmi
;
946 if (!USES_PRIMARY(bond
->params
.mode
)) {
947 /* nothing to do - mc list is already up-to-date on
954 if (bond
->dev
->flags
& IFF_PROMISC
) {
955 dev_set_promiscuity(old_active
->dev
, -1);
958 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
959 dev_set_allmulti(old_active
->dev
, -1);
962 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
963 dev_mc_delete(old_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
968 if (bond
->dev
->flags
& IFF_PROMISC
) {
969 dev_set_promiscuity(new_active
->dev
, 1);
972 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
973 dev_set_allmulti(new_active
->dev
, 1);
976 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
977 dev_mc_add(new_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
979 bond_resend_igmp_join_requests(bond
);
984 * bond_do_fail_over_mac
986 * Perform special MAC address swapping for fail_over_mac settings
988 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
990 static void bond_do_fail_over_mac(struct bonding
*bond
,
991 struct slave
*new_active
,
992 struct slave
*old_active
)
994 u8 tmp_mac
[ETH_ALEN
];
995 struct sockaddr saddr
;
998 switch (bond
->params
.fail_over_mac
) {
999 case BOND_FOM_ACTIVE
:
1001 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
1002 new_active
->dev
->addr_len
);
1004 case BOND_FOM_FOLLOW
:
1006 * if new_active && old_active, swap them
1007 * if just old_active, do nothing (going to no active slave)
1008 * if just new_active, set new_active to bond's MAC
1013 write_unlock_bh(&bond
->curr_slave_lock
);
1014 read_unlock(&bond
->lock
);
1017 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
1018 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
1020 saddr
.sa_family
= new_active
->dev
->type
;
1022 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
1023 saddr
.sa_family
= bond
->dev
->type
;
1026 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
1028 printk(KERN_ERR DRV_NAME
1029 ": %s: Error %d setting MAC of slave %s\n",
1030 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1037 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1038 saddr
.sa_family
= old_active
->dev
->type
;
1040 rv
= dev_set_mac_address(old_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
);
1046 read_lock(&bond
->lock
);
1047 write_lock_bh(&bond
->curr_slave_lock
);
1050 printk(KERN_ERR DRV_NAME
1051 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1052 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1060 * find_best_interface - select the best available slave to be the active one
1061 * @bond: our bonding struct
1063 * Warning: Caller must hold curr_slave_lock for writing.
1065 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1067 struct slave
*new_active
, *old_active
;
1068 struct slave
*bestslave
= NULL
;
1069 int mintime
= bond
->params
.updelay
;
1072 new_active
= old_active
= bond
->curr_active_slave
;
1074 if (!new_active
) { /* there were no active slaves left */
1075 if (bond
->slave_cnt
> 0) { /* found one slave */
1076 new_active
= bond
->first_slave
;
1078 return NULL
; /* still no slave, return NULL */
1082 /* first try the primary link; if arping, a link must tx/rx traffic
1083 * before it can be considered the curr_active_slave - also, we would skip
1084 * slaves between the curr_active_slave and primary_slave that may be up
1087 if ((bond
->primary_slave
) &&
1088 (!bond
->params
.arp_interval
) &&
1089 (IS_UP(bond
->primary_slave
->dev
))) {
1090 new_active
= bond
->primary_slave
;
1093 /* remember where to stop iterating over the slaves */
1094 old_active
= new_active
;
1096 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1097 if (IS_UP(new_active
->dev
)) {
1098 if (new_active
->link
== BOND_LINK_UP
) {
1100 } else if (new_active
->link
== BOND_LINK_BACK
) {
1101 /* link up, but waiting for stabilization */
1102 if (new_active
->delay
< mintime
) {
1103 mintime
= new_active
->delay
;
1104 bestslave
= new_active
;
1114 * change_active_interface - change the active slave into the specified one
1115 * @bond: our bonding struct
1116 * @new: the new slave to make the active one
1118 * Set the new slave to the bond's settings and unset them on the old
1119 * curr_active_slave.
1120 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1122 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1123 * because it is apparently the best available slave we have, even though its
1124 * updelay hasn't timed out yet.
1126 * If new_active is not NULL, caller must hold bond->lock for read and
1127 * curr_slave_lock for write_bh.
1129 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1131 struct slave
*old_active
= bond
->curr_active_slave
;
1133 if (old_active
== new_active
) {
1138 new_active
->jiffies
= jiffies
;
1140 if (new_active
->link
== BOND_LINK_BACK
) {
1141 if (USES_PRIMARY(bond
->params
.mode
)) {
1142 printk(KERN_INFO DRV_NAME
1143 ": %s: making interface %s the new "
1144 "active one %d ms earlier.\n",
1145 bond
->dev
->name
, new_active
->dev
->name
,
1146 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1149 new_active
->delay
= 0;
1150 new_active
->link
= BOND_LINK_UP
;
1152 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1153 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1156 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1157 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1158 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1161 if (USES_PRIMARY(bond
->params
.mode
)) {
1162 printk(KERN_INFO DRV_NAME
1163 ": %s: making interface %s the new "
1165 bond
->dev
->name
, new_active
->dev
->name
);
1170 if (USES_PRIMARY(bond
->params
.mode
)) {
1171 bond_mc_swap(bond
, new_active
, old_active
);
1174 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1175 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1176 bond_alb_handle_active_change(bond
, new_active
);
1178 bond_set_slave_inactive_flags(old_active
);
1180 bond_set_slave_active_flags(new_active
);
1182 bond
->curr_active_slave
= new_active
;
1185 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1187 bond_set_slave_inactive_flags(old_active
);
1191 bond_set_slave_active_flags(new_active
);
1193 if (bond
->params
.fail_over_mac
)
1194 bond_do_fail_over_mac(bond
, new_active
,
1197 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1198 bond_send_gratuitous_arp(bond
);
1200 write_unlock_bh(&bond
->curr_slave_lock
);
1201 read_unlock(&bond
->lock
);
1203 netdev_bonding_change(bond
->dev
);
1205 read_lock(&bond
->lock
);
1206 write_lock_bh(&bond
->curr_slave_lock
);
1212 * bond_select_active_slave - select a new active slave, if needed
1213 * @bond: our bonding struct
1215 * This functions shoud be called when one of the following occurs:
1216 * - The old curr_active_slave has been released or lost its link.
1217 * - The primary_slave has got its link back.
1218 * - A slave has got its link back and there's no old curr_active_slave.
1220 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1222 void bond_select_active_slave(struct bonding
*bond
)
1224 struct slave
*best_slave
;
1227 best_slave
= bond_find_best_slave(bond
);
1228 if (best_slave
!= bond
->curr_active_slave
) {
1229 bond_change_active_slave(bond
, best_slave
);
1230 rv
= bond_set_carrier(bond
);
1234 if (netif_carrier_ok(bond
->dev
)) {
1235 printk(KERN_INFO DRV_NAME
1236 ": %s: first active interface up!\n",
1239 printk(KERN_INFO DRV_NAME
": %s: "
1240 "now running without any active interface !\n",
1246 /*--------------------------- slave list handling ---------------------------*/
1249 * This function attaches the slave to the end of list.
1251 * bond->lock held for writing by caller.
1253 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1255 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1256 new_slave
->next
= new_slave
;
1257 new_slave
->prev
= new_slave
;
1258 bond
->first_slave
= new_slave
;
1260 new_slave
->next
= bond
->first_slave
;
1261 new_slave
->prev
= bond
->first_slave
->prev
;
1262 new_slave
->next
->prev
= new_slave
;
1263 new_slave
->prev
->next
= new_slave
;
1270 * This function detaches the slave from the list.
1271 * WARNING: no check is made to verify if the slave effectively
1272 * belongs to <bond>.
1273 * Nothing is freed on return, structures are just unchained.
1274 * If any slave pointer in bond was pointing to <slave>,
1275 * it should be changed by the calling function.
1277 * bond->lock held for writing by caller.
1279 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1282 slave
->next
->prev
= slave
->prev
;
1286 slave
->prev
->next
= slave
->next
;
1289 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1290 if (bond
->slave_cnt
> 1) { /* there are more slave */
1291 bond
->first_slave
= slave
->next
;
1293 bond
->first_slave
= NULL
; /* slave was the last one */
1302 /*---------------------------------- IOCTL ----------------------------------*/
1304 static int bond_sethwaddr(struct net_device
*bond_dev
,
1305 struct net_device
*slave_dev
)
1307 dprintk("bond_dev=%p\n", bond_dev
);
1308 dprintk("slave_dev=%p\n", slave_dev
);
1309 dprintk("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1310 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1314 #define BOND_VLAN_FEATURES \
1315 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1316 NETIF_F_HW_VLAN_FILTER)
1319 * Compute the common dev->feature set available to all slaves. Some
1320 * feature bits are managed elsewhere, so preserve those feature bits
1321 * on the master device.
1323 static int bond_compute_features(struct bonding
*bond
)
1325 struct slave
*slave
;
1326 struct net_device
*bond_dev
= bond
->dev
;
1327 unsigned long features
= bond_dev
->features
;
1328 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1329 bond_dev
->hard_header_len
);
1332 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1333 features
|= NETIF_F_SG
| NETIF_F_FRAGLIST
| NETIF_F_HIGHDMA
|
1334 NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1336 bond_for_each_slave(bond
, slave
, i
) {
1337 features
= netdev_compute_features(features
,
1338 slave
->dev
->features
);
1339 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1340 max_hard_header_len
= slave
->dev
->hard_header_len
;
1343 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1344 bond_dev
->features
= features
;
1345 bond_dev
->hard_header_len
= max_hard_header_len
;
1351 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1352 struct net_device
*slave_dev
)
1354 struct bonding
*bond
= bond_dev
->priv
;
1356 bond_dev
->neigh_setup
= slave_dev
->neigh_setup
;
1357 bond_dev
->header_ops
= slave_dev
->header_ops
;
1359 bond_dev
->type
= slave_dev
->type
;
1360 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1361 bond_dev
->addr_len
= slave_dev
->addr_len
;
1363 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1364 slave_dev
->addr_len
);
1365 bond
->setup_by_slave
= 1;
1368 /* enslave device <slave> to bond device <master> */
1369 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1371 struct bonding
*bond
= bond_dev
->priv
;
1372 struct slave
*new_slave
= NULL
;
1373 struct dev_mc_list
*dmi
;
1374 struct sockaddr addr
;
1376 int old_features
= bond_dev
->features
;
1379 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1380 slave_dev
->do_ioctl
== NULL
) {
1381 printk(KERN_WARNING DRV_NAME
1382 ": %s: Warning: no link monitoring support for %s\n",
1383 bond_dev
->name
, slave_dev
->name
);
1386 /* bond must be initialized by bond_open() before enslaving */
1387 if (!(bond_dev
->flags
& IFF_UP
)) {
1388 printk(KERN_WARNING DRV_NAME
1389 " %s: master_dev is not up in bond_enslave\n",
1393 /* already enslaved */
1394 if (slave_dev
->flags
& IFF_SLAVE
) {
1395 dprintk("Error, Device was already enslaved\n");
1399 /* vlan challenged mutual exclusion */
1400 /* no need to lock since we're protected by rtnl_lock */
1401 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1402 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1403 if (!list_empty(&bond
->vlan_list
)) {
1404 printk(KERN_ERR DRV_NAME
1405 ": %s: Error: cannot enslave VLAN "
1406 "challenged slave %s on VLAN enabled "
1407 "bond %s\n", bond_dev
->name
, slave_dev
->name
,
1411 printk(KERN_WARNING DRV_NAME
1412 ": %s: Warning: enslaved VLAN challenged "
1413 "slave %s. Adding VLANs will be blocked as "
1414 "long as %s is part of bond %s\n",
1415 bond_dev
->name
, slave_dev
->name
, slave_dev
->name
,
1417 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1420 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1421 if (bond
->slave_cnt
== 0) {
1422 /* First slave, and it is not VLAN challenged,
1423 * so remove the block of adding VLANs over the bond.
1425 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1430 * Old ifenslave binaries are no longer supported. These can
1431 * be identified with moderate accurary by the state of the slave:
1432 * the current ifenslave will set the interface down prior to
1433 * enslaving it; the old ifenslave will not.
1435 if ((slave_dev
->flags
& IFF_UP
)) {
1436 printk(KERN_ERR DRV_NAME
": %s is up. "
1437 "This may be due to an out of date ifenslave.\n",
1440 goto err_undo_flags
;
1443 /* set bonding device ether type by slave - bonding netdevices are
1444 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1445 * there is a need to override some of the type dependent attribs/funcs.
1447 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1448 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1450 if (bond
->slave_cnt
== 0) {
1451 if (slave_dev
->type
!= ARPHRD_ETHER
)
1452 bond_setup_by_slave(bond_dev
, slave_dev
);
1453 } else if (bond_dev
->type
!= slave_dev
->type
) {
1454 printk(KERN_ERR DRV_NAME
": %s ether type (%d) is different "
1455 "from other slaves (%d), can not enslave it.\n",
1457 slave_dev
->type
, bond_dev
->type
);
1459 goto err_undo_flags
;
1462 if (slave_dev
->set_mac_address
== NULL
) {
1463 if (bond
->slave_cnt
== 0) {
1464 printk(KERN_WARNING DRV_NAME
1465 ": %s: Warning: The first slave device "
1466 "specified does not support setting the MAC "
1467 "address. Setting fail_over_mac to active.",
1469 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1470 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1471 printk(KERN_ERR DRV_NAME
1472 ": %s: Error: The slave device specified "
1473 "does not support setting the MAC address, "
1474 "but fail_over_mac is not set to active.\n"
1477 goto err_undo_flags
;
1481 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1484 goto err_undo_flags
;
1487 /* save slave's original flags before calling
1488 * netdev_set_master and dev_open
1490 new_slave
->original_flags
= slave_dev
->flags
;
1493 * Save slave's original ("permanent") mac address for modes
1494 * that need it, and for restoring it upon release, and then
1495 * set it to the master's address
1497 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1499 if (!bond
->params
.fail_over_mac
) {
1501 * Set slave to master's mac address. The application already
1502 * set the master's mac address to that of the first slave
1504 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1505 addr
.sa_family
= slave_dev
->type
;
1506 res
= dev_set_mac_address(slave_dev
, &addr
);
1508 dprintk("Error %d calling set_mac_address\n", res
);
1513 res
= netdev_set_master(slave_dev
, bond_dev
);
1515 dprintk("Error %d calling netdev_set_master\n", res
);
1516 goto err_restore_mac
;
1518 /* open the slave since the application closed it */
1519 res
= dev_open(slave_dev
);
1521 dprintk("Openning slave %s failed\n", slave_dev
->name
);
1522 goto err_unset_master
;
1525 new_slave
->dev
= slave_dev
;
1526 slave_dev
->priv_flags
|= IFF_BONDING
;
1528 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1529 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1530 /* bond_alb_init_slave() must be called before all other stages since
1531 * it might fail and we do not want to have to undo everything
1533 res
= bond_alb_init_slave(bond
, new_slave
);
1539 /* If the mode USES_PRIMARY, then the new slave gets the
1540 * master's promisc (and mc) settings only if it becomes the
1541 * curr_active_slave, and that is taken care of later when calling
1542 * bond_change_active()
1544 if (!USES_PRIMARY(bond
->params
.mode
)) {
1545 /* set promiscuity level to new slave */
1546 if (bond_dev
->flags
& IFF_PROMISC
) {
1547 dev_set_promiscuity(slave_dev
, 1);
1550 /* set allmulti level to new slave */
1551 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1552 dev_set_allmulti(slave_dev
, 1);
1555 netif_tx_lock_bh(bond_dev
);
1556 /* upload master's mc_list to new slave */
1557 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
1558 dev_mc_add (slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
1560 netif_tx_unlock_bh(bond_dev
);
1563 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1564 /* add lacpdu mc addr to mc list */
1565 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1567 dev_mc_add(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
1570 bond_add_vlans_on_slave(bond
, slave_dev
);
1572 write_lock_bh(&bond
->lock
);
1574 bond_attach_slave(bond
, new_slave
);
1576 new_slave
->delay
= 0;
1577 new_slave
->link_failure_count
= 0;
1579 bond_compute_features(bond
);
1581 write_unlock_bh(&bond
->lock
);
1583 read_lock(&bond
->lock
);
1585 new_slave
->last_arp_rx
= jiffies
;
1587 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1588 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1590 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1592 * miimon is set but a bonded network driver
1593 * does not support ETHTOOL/MII and
1594 * arp_interval is not set. Note: if
1595 * use_carrier is enabled, we will never go
1596 * here (because netif_carrier is always
1597 * supported); thus, we don't need to change
1598 * the messages for netif_carrier.
1600 printk(KERN_WARNING DRV_NAME
1601 ": %s: Warning: MII and ETHTOOL support not "
1602 "available for interface %s, and "
1603 "arp_interval/arp_ip_target module parameters "
1604 "not specified, thus bonding will not detect "
1605 "link failures! see bonding.txt for details.\n",
1606 bond_dev
->name
, slave_dev
->name
);
1607 } else if (link_reporting
== -1) {
1608 /* unable get link status using mii/ethtool */
1609 printk(KERN_WARNING DRV_NAME
1610 ": %s: Warning: can't get link status from "
1611 "interface %s; the network driver associated "
1612 "with this interface does not support MII or "
1613 "ETHTOOL link status reporting, thus miimon "
1614 "has no effect on this interface.\n",
1615 bond_dev
->name
, slave_dev
->name
);
1619 /* check for initial state */
1620 if (!bond
->params
.miimon
||
1621 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1622 if (bond
->params
.updelay
) {
1623 dprintk("Initial state of slave_dev is "
1624 "BOND_LINK_BACK\n");
1625 new_slave
->link
= BOND_LINK_BACK
;
1626 new_slave
->delay
= bond
->params
.updelay
;
1628 dprintk("Initial state of slave_dev is "
1630 new_slave
->link
= BOND_LINK_UP
;
1632 new_slave
->jiffies
= jiffies
;
1634 dprintk("Initial state of slave_dev is "
1635 "BOND_LINK_DOWN\n");
1636 new_slave
->link
= BOND_LINK_DOWN
;
1639 if (bond_update_speed_duplex(new_slave
) &&
1640 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1641 printk(KERN_WARNING DRV_NAME
1642 ": %s: Warning: failed to get speed and duplex from %s, "
1643 "assumed to be 100Mb/sec and Full.\n",
1644 bond_dev
->name
, new_slave
->dev
->name
);
1646 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1647 printk(KERN_WARNING DRV_NAME
1648 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1649 "support in base driver for proper aggregator "
1650 "selection.\n", bond_dev
->name
);
1654 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1655 /* if there is a primary slave, remember it */
1656 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1657 bond
->primary_slave
= new_slave
;
1661 write_lock_bh(&bond
->curr_slave_lock
);
1663 switch (bond
->params
.mode
) {
1664 case BOND_MODE_ACTIVEBACKUP
:
1665 bond_set_slave_inactive_flags(new_slave
);
1666 bond_select_active_slave(bond
);
1668 case BOND_MODE_8023AD
:
1669 /* in 802.3ad mode, the internal mechanism
1670 * will activate the slaves in the selected
1673 bond_set_slave_inactive_flags(new_slave
);
1674 /* if this is the first slave */
1675 if (bond
->slave_cnt
== 1) {
1676 SLAVE_AD_INFO(new_slave
).id
= 1;
1677 /* Initialize AD with the number of times that the AD timer is called in 1 second
1678 * can be called only after the mac address of the bond is set
1680 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1681 bond
->params
.lacp_fast
);
1683 SLAVE_AD_INFO(new_slave
).id
=
1684 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1687 bond_3ad_bind_slave(new_slave
);
1691 new_slave
->state
= BOND_STATE_ACTIVE
;
1692 bond_set_slave_inactive_flags(new_slave
);
1695 dprintk("This slave is always active in trunk mode\n");
1697 /* always active in trunk mode */
1698 new_slave
->state
= BOND_STATE_ACTIVE
;
1700 /* In trunking mode there is little meaning to curr_active_slave
1701 * anyway (it holds no special properties of the bond device),
1702 * so we can change it without calling change_active_interface()
1704 if (!bond
->curr_active_slave
) {
1705 bond
->curr_active_slave
= new_slave
;
1708 } /* switch(bond_mode) */
1710 write_unlock_bh(&bond
->curr_slave_lock
);
1712 bond_set_carrier(bond
);
1714 read_unlock(&bond
->lock
);
1716 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1720 printk(KERN_INFO DRV_NAME
1721 ": %s: enslaving %s as a%s interface with a%s link.\n",
1722 bond_dev
->name
, slave_dev
->name
,
1723 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1724 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1726 /* enslave is successful */
1729 /* Undo stages on error */
1731 dev_close(slave_dev
);
1734 netdev_set_master(slave_dev
, NULL
);
1737 if (!bond
->params
.fail_over_mac
) {
1738 /* XXX TODO - fom follow mode needs to change master's
1739 * MAC if this slave's MAC is in use by the bond, or at
1740 * least print a warning.
1742 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1743 addr
.sa_family
= slave_dev
->type
;
1744 dev_set_mac_address(slave_dev
, &addr
);
1751 bond_dev
->features
= old_features
;
1757 * Try to release the slave device <slave> from the bond device <master>
1758 * It is legal to access curr_active_slave without a lock because all the function
1761 * The rules for slave state should be:
1762 * for Active/Backup:
1763 * Active stays on all backups go down
1764 * for Bonded connections:
1765 * The first up interface should be left on and all others downed.
1767 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1769 struct bonding
*bond
= bond_dev
->priv
;
1770 struct slave
*slave
, *oldcurrent
;
1771 struct sockaddr addr
;
1772 int mac_addr_differ
;
1773 DECLARE_MAC_BUF(mac
);
1775 /* slave is not a slave or master is not master of this slave */
1776 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1777 (slave_dev
->master
!= bond_dev
)) {
1778 printk(KERN_ERR DRV_NAME
1779 ": %s: Error: cannot release %s.\n",
1780 bond_dev
->name
, slave_dev
->name
);
1784 write_lock_bh(&bond
->lock
);
1786 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1788 /* not a slave of this bond */
1789 printk(KERN_INFO DRV_NAME
1790 ": %s: %s not enslaved\n",
1791 bond_dev
->name
, slave_dev
->name
);
1792 write_unlock_bh(&bond
->lock
);
1796 if (!bond
->params
.fail_over_mac
) {
1797 mac_addr_differ
= memcmp(bond_dev
->dev_addr
, slave
->perm_hwaddr
,
1799 if (!mac_addr_differ
&& (bond
->slave_cnt
> 1))
1800 printk(KERN_WARNING DRV_NAME
1801 ": %s: Warning: the permanent HWaddr of %s - "
1802 "%s - is still in use by %s. "
1803 "Set the HWaddr of %s to a different address "
1804 "to avoid conflicts.\n",
1805 bond_dev
->name
, slave_dev
->name
,
1806 print_mac(mac
, slave
->perm_hwaddr
),
1807 bond_dev
->name
, slave_dev
->name
);
1810 /* Inform AD package of unbinding of slave. */
1811 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1812 /* must be called before the slave is
1813 * detached from the list
1815 bond_3ad_unbind_slave(slave
);
1818 printk(KERN_INFO DRV_NAME
1819 ": %s: releasing %s interface %s\n",
1821 (slave
->state
== BOND_STATE_ACTIVE
)
1822 ? "active" : "backup",
1825 oldcurrent
= bond
->curr_active_slave
;
1827 bond
->current_arp_slave
= NULL
;
1829 /* release the slave from its bond */
1830 bond_detach_slave(bond
, slave
);
1832 bond_compute_features(bond
);
1834 if (bond
->primary_slave
== slave
) {
1835 bond
->primary_slave
= NULL
;
1838 if (oldcurrent
== slave
) {
1839 bond_change_active_slave(bond
, NULL
);
1842 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1843 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1844 /* Must be called only after the slave has been
1845 * detached from the list and the curr_active_slave
1846 * has been cleared (if our_slave == old_current),
1847 * but before a new active slave is selected.
1849 write_unlock_bh(&bond
->lock
);
1850 bond_alb_deinit_slave(bond
, slave
);
1851 write_lock_bh(&bond
->lock
);
1854 if (oldcurrent
== slave
) {
1856 * Note that we hold RTNL over this sequence, so there
1857 * is no concern that another slave add/remove event
1860 write_unlock_bh(&bond
->lock
);
1861 read_lock(&bond
->lock
);
1862 write_lock_bh(&bond
->curr_slave_lock
);
1864 bond_select_active_slave(bond
);
1866 write_unlock_bh(&bond
->curr_slave_lock
);
1867 read_unlock(&bond
->lock
);
1868 write_lock_bh(&bond
->lock
);
1871 if (bond
->slave_cnt
== 0) {
1872 bond_set_carrier(bond
);
1874 /* if the last slave was removed, zero the mac address
1875 * of the master so it will be set by the application
1876 * to the mac address of the first slave
1878 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1880 if (list_empty(&bond
->vlan_list
)) {
1881 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1883 printk(KERN_WARNING DRV_NAME
1884 ": %s: Warning: clearing HW address of %s while it "
1885 "still has VLANs.\n",
1886 bond_dev
->name
, bond_dev
->name
);
1887 printk(KERN_WARNING DRV_NAME
1888 ": %s: When re-adding slaves, make sure the bond's "
1889 "HW address matches its VLANs'.\n",
1892 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1893 !bond_has_challenged_slaves(bond
)) {
1894 printk(KERN_INFO DRV_NAME
1895 ": %s: last VLAN challenged slave %s "
1896 "left bond %s. VLAN blocking is removed\n",
1897 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1898 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1901 write_unlock_bh(&bond
->lock
);
1903 /* must do this from outside any spinlocks */
1904 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1906 bond_del_vlans_from_slave(bond
, slave_dev
);
1908 /* If the mode USES_PRIMARY, then we should only remove its
1909 * promisc and mc settings if it was the curr_active_slave, but that was
1910 * already taken care of above when we detached the slave
1912 if (!USES_PRIMARY(bond
->params
.mode
)) {
1913 /* unset promiscuity level from slave */
1914 if (bond_dev
->flags
& IFF_PROMISC
) {
1915 dev_set_promiscuity(slave_dev
, -1);
1918 /* unset allmulti level from slave */
1919 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1920 dev_set_allmulti(slave_dev
, -1);
1923 /* flush master's mc_list from slave */
1924 netif_tx_lock_bh(bond_dev
);
1925 bond_mc_list_flush(bond_dev
, slave_dev
);
1926 netif_tx_unlock_bh(bond_dev
);
1929 netdev_set_master(slave_dev
, NULL
);
1931 /* close slave before restoring its mac address */
1932 dev_close(slave_dev
);
1934 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1935 /* restore original ("permanent") mac address */
1936 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1937 addr
.sa_family
= slave_dev
->type
;
1938 dev_set_mac_address(slave_dev
, &addr
);
1941 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
1942 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
1947 return 0; /* deletion OK */
1951 * Destroy a bonding device.
1952 * Must be under rtnl_lock when this function is called.
1954 void bond_destroy(struct bonding
*bond
)
1956 bond_deinit(bond
->dev
);
1957 bond_destroy_sysfs_entry(bond
);
1958 unregister_netdevice(bond
->dev
);
1962 * First release a slave and than destroy the bond if no more slaves iare left.
1963 * Must be under rtnl_lock when this function is called.
1965 int bond_release_and_destroy(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1967 struct bonding
*bond
= bond_dev
->priv
;
1970 ret
= bond_release(bond_dev
, slave_dev
);
1971 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
1972 printk(KERN_INFO DRV_NAME
": %s: destroying bond %s.\n",
1973 bond_dev
->name
, bond_dev
->name
);
1980 * This function releases all slaves.
1982 static int bond_release_all(struct net_device
*bond_dev
)
1984 struct bonding
*bond
= bond_dev
->priv
;
1985 struct slave
*slave
;
1986 struct net_device
*slave_dev
;
1987 struct sockaddr addr
;
1989 write_lock_bh(&bond
->lock
);
1991 netif_carrier_off(bond_dev
);
1993 if (bond
->slave_cnt
== 0) {
1997 bond
->current_arp_slave
= NULL
;
1998 bond
->primary_slave
= NULL
;
1999 bond_change_active_slave(bond
, NULL
);
2001 while ((slave
= bond
->first_slave
) != NULL
) {
2002 /* Inform AD package of unbinding of slave
2003 * before slave is detached from the list.
2005 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2006 bond_3ad_unbind_slave(slave
);
2009 slave_dev
= slave
->dev
;
2010 bond_detach_slave(bond
, slave
);
2012 /* now that the slave is detached, unlock and perform
2013 * all the undo steps that should not be called from
2016 write_unlock_bh(&bond
->lock
);
2018 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
2019 (bond
->params
.mode
== BOND_MODE_ALB
)) {
2020 /* must be called only after the slave
2021 * has been detached from the list
2023 bond_alb_deinit_slave(bond
, slave
);
2026 bond_compute_features(bond
);
2028 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2029 bond_del_vlans_from_slave(bond
, slave_dev
);
2031 /* If the mode USES_PRIMARY, then we should only remove its
2032 * promisc and mc settings if it was the curr_active_slave, but that was
2033 * already taken care of above when we detached the slave
2035 if (!USES_PRIMARY(bond
->params
.mode
)) {
2036 /* unset promiscuity level from slave */
2037 if (bond_dev
->flags
& IFF_PROMISC
) {
2038 dev_set_promiscuity(slave_dev
, -1);
2041 /* unset allmulti level from slave */
2042 if (bond_dev
->flags
& IFF_ALLMULTI
) {
2043 dev_set_allmulti(slave_dev
, -1);
2046 /* flush master's mc_list from slave */
2047 netif_tx_lock_bh(bond_dev
);
2048 bond_mc_list_flush(bond_dev
, slave_dev
);
2049 netif_tx_unlock_bh(bond_dev
);
2052 netdev_set_master(slave_dev
, NULL
);
2054 /* close slave before restoring its mac address */
2055 dev_close(slave_dev
);
2057 if (!bond
->params
.fail_over_mac
) {
2058 /* restore original ("permanent") mac address*/
2059 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2060 addr
.sa_family
= slave_dev
->type
;
2061 dev_set_mac_address(slave_dev
, &addr
);
2064 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2065 IFF_SLAVE_INACTIVE
);
2069 /* re-acquire the lock before getting the next slave */
2070 write_lock_bh(&bond
->lock
);
2073 /* zero the mac address of the master so it will be
2074 * set by the application to the mac address of the
2077 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2079 if (list_empty(&bond
->vlan_list
)) {
2080 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2082 printk(KERN_WARNING DRV_NAME
2083 ": %s: Warning: clearing HW address of %s while it "
2084 "still has VLANs.\n",
2085 bond_dev
->name
, bond_dev
->name
);
2086 printk(KERN_WARNING DRV_NAME
2087 ": %s: When re-adding slaves, make sure the bond's "
2088 "HW address matches its VLANs'.\n",
2092 printk(KERN_INFO DRV_NAME
2093 ": %s: released all slaves\n",
2097 write_unlock_bh(&bond
->lock
);
2103 * This function changes the active slave to slave <slave_dev>.
2104 * It returns -EINVAL in the following cases.
2105 * - <slave_dev> is not found in the list.
2106 * - There is not active slave now.
2107 * - <slave_dev> is already active.
2108 * - The link state of <slave_dev> is not BOND_LINK_UP.
2109 * - <slave_dev> is not running.
2110 * In these cases, this fuction does nothing.
2111 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2113 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2115 struct bonding
*bond
= bond_dev
->priv
;
2116 struct slave
*old_active
= NULL
;
2117 struct slave
*new_active
= NULL
;
2120 if (!USES_PRIMARY(bond
->params
.mode
)) {
2124 /* Verify that master_dev is indeed the master of slave_dev */
2125 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
2126 (slave_dev
->master
!= bond_dev
)) {
2130 read_lock(&bond
->lock
);
2132 read_lock(&bond
->curr_slave_lock
);
2133 old_active
= bond
->curr_active_slave
;
2134 read_unlock(&bond
->curr_slave_lock
);
2136 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2139 * Changing to the current active: do nothing; return success.
2141 if (new_active
&& (new_active
== old_active
)) {
2142 read_unlock(&bond
->lock
);
2148 (new_active
->link
== BOND_LINK_UP
) &&
2149 IS_UP(new_active
->dev
)) {
2150 write_lock_bh(&bond
->curr_slave_lock
);
2151 bond_change_active_slave(bond
, new_active
);
2152 write_unlock_bh(&bond
->curr_slave_lock
);
2157 read_unlock(&bond
->lock
);
2162 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2164 struct bonding
*bond
= bond_dev
->priv
;
2166 info
->bond_mode
= bond
->params
.mode
;
2167 info
->miimon
= bond
->params
.miimon
;
2169 read_lock(&bond
->lock
);
2170 info
->num_slaves
= bond
->slave_cnt
;
2171 read_unlock(&bond
->lock
);
2176 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2178 struct bonding
*bond
= bond_dev
->priv
;
2179 struct slave
*slave
;
2182 if (info
->slave_id
< 0) {
2186 read_lock(&bond
->lock
);
2188 bond_for_each_slave(bond
, slave
, i
) {
2189 if (i
== (int)info
->slave_id
) {
2195 read_unlock(&bond
->lock
);
2198 strcpy(info
->slave_name
, slave
->dev
->name
);
2199 info
->link
= slave
->link
;
2200 info
->state
= slave
->state
;
2201 info
->link_failure_count
= slave
->link_failure_count
;
2209 /*-------------------------------- Monitoring -------------------------------*/
2212 * if !have_locks, return nonzero if a failover is necessary. if
2213 * have_locks, do whatever failover activities are needed.
2215 * This is to separate the inspection and failover steps for locking
2216 * purposes; failover requires rtnl, but acquiring it for every
2217 * inspection is undesirable, so a wrapper first does inspection, and
2218 * the acquires the necessary locks and calls again to perform
2219 * failover if needed. Since all locks are dropped, a complete
2220 * restart is needed between calls.
2222 static int __bond_mii_monitor(struct bonding
*bond
, int have_locks
)
2224 struct slave
*slave
, *oldcurrent
;
2225 int do_failover
= 0;
2228 if (bond
->slave_cnt
== 0)
2231 /* we will try to read the link status of each of our slaves, and
2232 * set their IFF_RUNNING flag appropriately. For each slave not
2233 * supporting MII status, we won't do anything so that a user-space
2234 * program could monitor the link itself if needed.
2237 read_lock(&bond
->curr_slave_lock
);
2238 oldcurrent
= bond
->curr_active_slave
;
2239 read_unlock(&bond
->curr_slave_lock
);
2241 bond_for_each_slave(bond
, slave
, i
) {
2242 struct net_device
*slave_dev
= slave
->dev
;
2244 u16 old_speed
= slave
->speed
;
2245 u8 old_duplex
= slave
->duplex
;
2247 link_state
= bond_check_dev_link(bond
, slave_dev
, 0);
2249 switch (slave
->link
) {
2250 case BOND_LINK_UP
: /* the link was up */
2251 if (link_state
== BMSR_LSTATUS
) {
2258 } else { /* link going down */
2259 slave
->link
= BOND_LINK_FAIL
;
2260 slave
->delay
= bond
->params
.downdelay
;
2262 if (slave
->link_failure_count
< UINT_MAX
) {
2263 slave
->link_failure_count
++;
2266 if (bond
->params
.downdelay
) {
2267 printk(KERN_INFO DRV_NAME
2268 ": %s: link status down for %s "
2269 "interface %s, disabling it in "
2273 ? ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
2274 ? ((slave
== oldcurrent
)
2275 ? "active " : "backup ")
2279 bond
->params
.downdelay
* bond
->params
.miimon
);
2282 /* no break ! fall through the BOND_LINK_FAIL test to
2283 ensure proper action to be taken
2285 case BOND_LINK_FAIL
: /* the link has just gone down */
2286 if (link_state
!= BMSR_LSTATUS
) {
2287 /* link stays down */
2288 if (slave
->delay
<= 0) {
2292 /* link down for too long time */
2293 slave
->link
= BOND_LINK_DOWN
;
2295 /* in active/backup mode, we must
2296 * completely disable this interface
2298 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) ||
2299 (bond
->params
.mode
== BOND_MODE_8023AD
)) {
2300 bond_set_slave_inactive_flags(slave
);
2303 printk(KERN_INFO DRV_NAME
2304 ": %s: link status definitely "
2305 "down for interface %s, "
2310 /* notify ad that the link status has changed */
2311 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2312 bond_3ad_handle_link_change(slave
, BOND_LINK_DOWN
);
2315 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
2316 (bond
->params
.mode
== BOND_MODE_ALB
)) {
2317 bond_alb_handle_link_change(bond
, slave
, BOND_LINK_DOWN
);
2320 if (slave
== oldcurrent
) {
2328 slave
->link
= BOND_LINK_UP
;
2329 slave
->jiffies
= jiffies
;
2330 printk(KERN_INFO DRV_NAME
2331 ": %s: link status up again after %d "
2332 "ms for interface %s.\n",
2334 (bond
->params
.downdelay
- slave
->delay
) * bond
->params
.miimon
,
2338 case BOND_LINK_DOWN
: /* the link was down */
2339 if (link_state
!= BMSR_LSTATUS
) {
2340 /* the link stays down, nothing more to do */
2342 } else { /* link going up */
2343 slave
->link
= BOND_LINK_BACK
;
2344 slave
->delay
= bond
->params
.updelay
;
2346 if (bond
->params
.updelay
) {
2347 /* if updelay == 0, no need to
2348 advertise about a 0 ms delay */
2349 printk(KERN_INFO DRV_NAME
2350 ": %s: link status up for "
2351 "interface %s, enabling it "
2355 bond
->params
.updelay
* bond
->params
.miimon
);
2358 /* no break ! fall through the BOND_LINK_BACK state in
2359 case there's something to do.
2361 case BOND_LINK_BACK
: /* the link has just come back */
2362 if (link_state
!= BMSR_LSTATUS
) {
2363 /* link down again */
2364 slave
->link
= BOND_LINK_DOWN
;
2366 printk(KERN_INFO DRV_NAME
2367 ": %s: link status down again after %d "
2368 "ms for interface %s.\n",
2370 (bond
->params
.updelay
- slave
->delay
) * bond
->params
.miimon
,
2374 if (slave
->delay
== 0) {
2378 /* now the link has been up for long time enough */
2379 slave
->link
= BOND_LINK_UP
;
2380 slave
->jiffies
= jiffies
;
2382 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2383 /* prevent it from being the active one */
2384 slave
->state
= BOND_STATE_BACKUP
;
2385 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2386 /* make it immediately active */
2387 slave
->state
= BOND_STATE_ACTIVE
;
2388 } else if (slave
!= bond
->primary_slave
) {
2389 /* prevent it from being the active one */
2390 slave
->state
= BOND_STATE_BACKUP
;
2393 printk(KERN_INFO DRV_NAME
2394 ": %s: link status definitely "
2395 "up for interface %s.\n",
2399 /* notify ad that the link status has changed */
2400 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2401 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2404 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
2405 (bond
->params
.mode
== BOND_MODE_ALB
)) {
2406 bond_alb_handle_link_change(bond
, slave
, BOND_LINK_UP
);
2409 if ((!oldcurrent
) ||
2410 (slave
== bond
->primary_slave
)) {
2419 /* Should not happen */
2420 printk(KERN_ERR DRV_NAME
2421 ": %s: Error: %s Illegal value (link=%d)\n",
2426 } /* end of switch (slave->link) */
2428 bond_update_speed_duplex(slave
);
2430 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2431 if (old_speed
!= slave
->speed
) {
2432 bond_3ad_adapter_speed_changed(slave
);
2435 if (old_duplex
!= slave
->duplex
) {
2436 bond_3ad_adapter_duplex_changed(slave
);
2445 write_lock_bh(&bond
->curr_slave_lock
);
2447 bond_select_active_slave(bond
);
2449 write_unlock_bh(&bond
->curr_slave_lock
);
2452 bond_set_carrier(bond
);
2461 * Really a wrapper that splits the mii monitor into two phases: an
2462 * inspection, then (if inspection indicates something needs to be
2463 * done) an acquisition of appropriate locks followed by another pass
2464 * to implement whatever link state changes are indicated.
2466 void bond_mii_monitor(struct work_struct
*work
)
2468 struct bonding
*bond
= container_of(work
, struct bonding
,
2470 unsigned long delay
;
2472 read_lock(&bond
->lock
);
2473 if (bond
->kill_timers
) {
2474 read_unlock(&bond
->lock
);
2478 if (bond
->send_grat_arp
) {
2479 read_lock(&bond
->curr_slave_lock
);
2480 bond_send_gratuitous_arp(bond
);
2481 read_unlock(&bond
->curr_slave_lock
);
2484 if (__bond_mii_monitor(bond
, 0)) {
2485 read_unlock(&bond
->lock
);
2487 read_lock(&bond
->lock
);
2488 __bond_mii_monitor(bond
, 1);
2489 read_unlock(&bond
->lock
);
2490 rtnl_unlock(); /* might sleep, hold no other locks */
2491 read_lock(&bond
->lock
);
2494 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2495 read_unlock(&bond
->lock
);
2496 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2499 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2501 struct in_device
*idev
;
2502 struct in_ifaddr
*ifa
;
2509 idev
= __in_dev_get_rcu(dev
);
2513 ifa
= idev
->ifa_list
;
2517 addr
= ifa
->ifa_local
;
2523 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2525 struct vlan_entry
*vlan
;
2527 if (ip
== bond
->master_ip
)
2530 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2531 if (ip
== vlan
->vlan_ip
)
2539 * We go to the (large) trouble of VLAN tagging ARP frames because
2540 * switches in VLAN mode (especially if ports are configured as
2541 * "native" to a VLAN) might not pass non-tagged frames.
2543 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2545 struct sk_buff
*skb
;
2547 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2548 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2550 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2551 NULL
, slave_dev
->dev_addr
, NULL
);
2554 printk(KERN_ERR DRV_NAME
": ARP packet allocation failed\n");
2558 skb
= vlan_put_tag(skb
, vlan_id
);
2560 printk(KERN_ERR DRV_NAME
": failed to insert VLAN tag\n");
2568 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2571 __be32
*targets
= bond
->params
.arp_targets
;
2572 struct vlan_entry
*vlan
;
2573 struct net_device
*vlan_dev
;
2577 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2580 dprintk("basa: target %x\n", targets
[i
]);
2581 if (list_empty(&bond
->vlan_list
)) {
2582 dprintk("basa: empty vlan: arp_send\n");
2583 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2584 bond
->master_ip
, 0);
2589 * If VLANs are configured, we do a route lookup to
2590 * determine which VLAN interface would be used, so we
2591 * can tag the ARP with the proper VLAN tag.
2593 memset(&fl
, 0, sizeof(fl
));
2594 fl
.fl4_dst
= targets
[i
];
2595 fl
.fl4_tos
= RTO_ONLINK
;
2597 rv
= ip_route_output_key(&init_net
, &rt
, &fl
);
2599 if (net_ratelimit()) {
2600 printk(KERN_WARNING DRV_NAME
2601 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2602 bond
->dev
->name
, NIPQUAD(fl
.fl4_dst
));
2608 * This target is not on a VLAN
2610 if (rt
->u
.dst
.dev
== bond
->dev
) {
2612 dprintk("basa: rtdev == bond->dev: arp_send\n");
2613 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2614 bond
->master_ip
, 0);
2619 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2620 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2621 if (vlan_dev
== rt
->u
.dst
.dev
) {
2622 vlan_id
= vlan
->vlan_id
;
2623 dprintk("basa: vlan match on %s %d\n",
2624 vlan_dev
->name
, vlan_id
);
2631 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2632 vlan
->vlan_ip
, vlan_id
);
2636 if (net_ratelimit()) {
2637 printk(KERN_WARNING DRV_NAME
2638 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2639 bond
->dev
->name
, NIPQUAD(fl
.fl4_dst
),
2640 rt
->u
.dst
.dev
? rt
->u
.dst
.dev
->name
: "NULL");
2647 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2648 * for each VLAN above us.
2650 * Caller must hold curr_slave_lock for read or better
2652 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2654 struct slave
*slave
= bond
->curr_active_slave
;
2655 struct vlan_entry
*vlan
;
2656 struct net_device
*vlan_dev
;
2658 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond
->dev
->name
,
2659 slave
? slave
->dev
->name
: "NULL");
2661 if (!slave
|| !bond
->send_grat_arp
||
2662 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2665 bond
->send_grat_arp
--;
2667 if (bond
->master_ip
) {
2668 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2669 bond
->master_ip
, 0);
2672 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2673 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2674 if (vlan
->vlan_ip
) {
2675 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2676 vlan
->vlan_ip
, vlan
->vlan_id
);
2681 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2684 __be32
*targets
= bond
->params
.arp_targets
;
2686 targets
= bond
->params
.arp_targets
;
2687 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2688 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2689 "%u.%u.%u.%u bhti(tip) %d\n",
2690 NIPQUAD(sip
), NIPQUAD(tip
), i
, NIPQUAD(targets
[i
]),
2691 bond_has_this_ip(bond
, tip
));
2692 if (sip
== targets
[i
]) {
2693 if (bond_has_this_ip(bond
, tip
))
2694 slave
->last_arp_rx
= jiffies
;
2700 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2703 struct slave
*slave
;
2704 struct bonding
*bond
;
2705 unsigned char *arp_ptr
;
2708 if (dev_net(dev
) != &init_net
)
2711 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2715 read_lock(&bond
->lock
);
2717 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2718 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2719 orig_dev
? orig_dev
->name
: "NULL");
2721 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2722 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2725 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2729 if (arp
->ar_hln
!= dev
->addr_len
||
2730 skb
->pkt_type
== PACKET_OTHERHOST
||
2731 skb
->pkt_type
== PACKET_LOOPBACK
||
2732 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2733 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2737 arp_ptr
= (unsigned char *)(arp
+ 1);
2738 arp_ptr
+= dev
->addr_len
;
2739 memcpy(&sip
, arp_ptr
, 4);
2740 arp_ptr
+= 4 + dev
->addr_len
;
2741 memcpy(&tip
, arp_ptr
, 4);
2743 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2744 " tip %u.%u.%u.%u\n", bond
->dev
->name
, slave
->dev
->name
,
2745 slave
->state
, bond
->params
.arp_validate
,
2746 slave_do_arp_validate(bond
, slave
), NIPQUAD(sip
), NIPQUAD(tip
));
2749 * Backup slaves won't see the ARP reply, but do come through
2750 * here for each ARP probe (so we swap the sip/tip to validate
2751 * the probe). In a "redundant switch, common router" type of
2752 * configuration, the ARP probe will (hopefully) travel from
2753 * the active, through one switch, the router, then the other
2754 * switch before reaching the backup.
2756 if (slave
->state
== BOND_STATE_ACTIVE
)
2757 bond_validate_arp(bond
, slave
, sip
, tip
);
2759 bond_validate_arp(bond
, slave
, tip
, sip
);
2762 read_unlock(&bond
->lock
);
2765 return NET_RX_SUCCESS
;
2769 * this function is called regularly to monitor each slave's link
2770 * ensuring that traffic is being sent and received when arp monitoring
2771 * is used in load-balancing mode. if the adapter has been dormant, then an
2772 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2773 * arp monitoring in active backup mode.
2775 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2777 struct bonding
*bond
= container_of(work
, struct bonding
,
2779 struct slave
*slave
, *oldcurrent
;
2780 int do_failover
= 0;
2784 read_lock(&bond
->lock
);
2786 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2788 if (bond
->kill_timers
) {
2792 if (bond
->slave_cnt
== 0) {
2796 read_lock(&bond
->curr_slave_lock
);
2797 oldcurrent
= bond
->curr_active_slave
;
2798 read_unlock(&bond
->curr_slave_lock
);
2800 /* see if any of the previous devices are up now (i.e. they have
2801 * xmt and rcv traffic). the curr_active_slave does not come into
2802 * the picture unless it is null. also, slave->jiffies is not needed
2803 * here because we send an arp on each slave and give a slave as
2804 * long as it needs to get the tx/rx within the delta.
2805 * TODO: what about up/down delay in arp mode? it wasn't here before
2808 bond_for_each_slave(bond
, slave
, i
) {
2809 if (slave
->link
!= BOND_LINK_UP
) {
2810 if (time_before_eq(jiffies
, slave
->dev
->trans_start
+ delta_in_ticks
) &&
2811 time_before_eq(jiffies
, slave
->dev
->last_rx
+ delta_in_ticks
)) {
2813 slave
->link
= BOND_LINK_UP
;
2814 slave
->state
= BOND_STATE_ACTIVE
;
2816 /* primary_slave has no meaning in round-robin
2817 * mode. the window of a slave being up and
2818 * curr_active_slave being null after enslaving
2822 printk(KERN_INFO DRV_NAME
2823 ": %s: link status definitely "
2824 "up for interface %s, ",
2829 printk(KERN_INFO DRV_NAME
2830 ": %s: interface %s is now up\n",
2836 /* slave->link == BOND_LINK_UP */
2838 /* not all switches will respond to an arp request
2839 * when the source ip is 0, so don't take the link down
2840 * if we don't know our ip yet
2842 if (time_after_eq(jiffies
, slave
->dev
->trans_start
+ 2*delta_in_ticks
) ||
2843 (time_after_eq(jiffies
, slave
->dev
->last_rx
+ 2*delta_in_ticks
))) {
2845 slave
->link
= BOND_LINK_DOWN
;
2846 slave
->state
= BOND_STATE_BACKUP
;
2848 if (slave
->link_failure_count
< UINT_MAX
) {
2849 slave
->link_failure_count
++;
2852 printk(KERN_INFO DRV_NAME
2853 ": %s: interface %s is now down.\n",
2857 if (slave
== oldcurrent
) {
2863 /* note: if switch is in round-robin mode, all links
2864 * must tx arp to ensure all links rx an arp - otherwise
2865 * links may oscillate or not come up at all; if switch is
2866 * in something like xor mode, there is nothing we can
2867 * do - all replies will be rx'ed on same link causing slaves
2868 * to be unstable during low/no traffic periods
2870 if (IS_UP(slave
->dev
)) {
2871 bond_arp_send_all(bond
, slave
);
2876 write_lock_bh(&bond
->curr_slave_lock
);
2878 bond_select_active_slave(bond
);
2880 write_unlock_bh(&bond
->curr_slave_lock
);
2884 if (bond
->params
.arp_interval
)
2885 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2887 read_unlock(&bond
->lock
);
2891 * Called to inspect slaves for active-backup mode ARP monitor link state
2892 * changes. Sets new_link in slaves to specify what action should take
2893 * place for the slave. Returns 0 if no changes are found, >0 if changes
2894 * to link states must be committed.
2896 * Called with bond->lock held for read.
2898 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2900 struct slave
*slave
;
2903 bond_for_each_slave(bond
, slave
, i
) {
2904 slave
->new_link
= BOND_LINK_NOCHANGE
;
2906 if (slave
->link
!= BOND_LINK_UP
) {
2907 if (time_before_eq(jiffies
, slave_last_rx(bond
, slave
) +
2909 slave
->new_link
= BOND_LINK_UP
;
2917 * Give slaves 2*delta after being enslaved or made
2918 * active. This avoids bouncing, as the last receive
2919 * times need a full ARP monitor cycle to be updated.
2921 if (!time_after_eq(jiffies
, slave
->jiffies
+
2922 2 * delta_in_ticks
))
2926 * Backup slave is down if:
2927 * - No current_arp_slave AND
2928 * - more than 3*delta since last receive AND
2929 * - the bond has an IP address
2931 * Note: a non-null current_arp_slave indicates
2932 * the curr_active_slave went down and we are
2933 * searching for a new one; under this condition
2934 * we only take the curr_active_slave down - this
2935 * gives each slave a chance to tx/rx traffic
2936 * before being taken out
2938 if (slave
->state
== BOND_STATE_BACKUP
&&
2939 !bond
->current_arp_slave
&&
2940 time_after(jiffies
, slave_last_rx(bond
, slave
) +
2941 3 * delta_in_ticks
)) {
2942 slave
->new_link
= BOND_LINK_DOWN
;
2947 * Active slave is down if:
2948 * - more than 2*delta since transmitting OR
2949 * - (more than 2*delta since receive AND
2950 * the bond has an IP address)
2952 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
2953 (time_after_eq(jiffies
, slave
->dev
->trans_start
+
2954 2 * delta_in_ticks
) ||
2955 (time_after_eq(jiffies
, slave_last_rx(bond
, slave
)
2956 + 2 * delta_in_ticks
)))) {
2957 slave
->new_link
= BOND_LINK_DOWN
;
2962 read_lock(&bond
->curr_slave_lock
);
2965 * Trigger a commit if the primary option setting has changed.
2967 if (bond
->primary_slave
&&
2968 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
2969 (bond
->primary_slave
->link
== BOND_LINK_UP
))
2972 read_unlock(&bond
->curr_slave_lock
);
2978 * Called to commit link state changes noted by inspection step of
2979 * active-backup mode ARP monitor.
2981 * Called with RTNL and bond->lock for read.
2983 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2985 struct slave
*slave
;
2988 bond_for_each_slave(bond
, slave
, i
) {
2989 switch (slave
->new_link
) {
2990 case BOND_LINK_NOCHANGE
:
2994 write_lock_bh(&bond
->curr_slave_lock
);
2996 if (!bond
->curr_active_slave
&&
2997 time_before_eq(jiffies
, slave
->dev
->trans_start
+
2999 slave
->link
= BOND_LINK_UP
;
3000 bond_change_active_slave(bond
, slave
);
3001 bond
->current_arp_slave
= NULL
;
3003 printk(KERN_INFO DRV_NAME
3004 ": %s: %s is up and now the "
3005 "active interface\n",
3006 bond
->dev
->name
, slave
->dev
->name
);
3008 } else if (bond
->curr_active_slave
!= slave
) {
3009 /* this slave has just come up but we
3010 * already have a current slave; this can
3011 * also happen if bond_enslave adds a new
3012 * slave that is up while we are searching
3015 slave
->link
= BOND_LINK_UP
;
3016 bond_set_slave_inactive_flags(slave
);
3017 bond
->current_arp_slave
= NULL
;
3019 printk(KERN_INFO DRV_NAME
3020 ": %s: backup interface %s is now up\n",
3021 bond
->dev
->name
, slave
->dev
->name
);
3024 write_unlock_bh(&bond
->curr_slave_lock
);
3028 case BOND_LINK_DOWN
:
3029 if (slave
->link_failure_count
< UINT_MAX
)
3030 slave
->link_failure_count
++;
3032 slave
->link
= BOND_LINK_DOWN
;
3034 if (slave
== bond
->curr_active_slave
) {
3035 printk(KERN_INFO DRV_NAME
3036 ": %s: link status down for active "
3037 "interface %s, disabling it\n",
3038 bond
->dev
->name
, slave
->dev
->name
);
3040 bond_set_slave_inactive_flags(slave
);
3042 write_lock_bh(&bond
->curr_slave_lock
);
3044 bond_select_active_slave(bond
);
3045 if (bond
->curr_active_slave
)
3046 bond
->curr_active_slave
->jiffies
=
3049 write_unlock_bh(&bond
->curr_slave_lock
);
3051 bond
->current_arp_slave
= NULL
;
3053 } else if (slave
->state
== BOND_STATE_BACKUP
) {
3054 printk(KERN_INFO DRV_NAME
3055 ": %s: backup interface %s is now down\n",
3056 bond
->dev
->name
, slave
->dev
->name
);
3058 bond_set_slave_inactive_flags(slave
);
3063 printk(KERN_ERR DRV_NAME
3064 ": %s: impossible: new_link %d on slave %s\n",
3065 bond
->dev
->name
, slave
->new_link
,
3071 * No race with changes to primary via sysfs, as we hold rtnl.
3073 if (bond
->primary_slave
&&
3074 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
3075 (bond
->primary_slave
->link
== BOND_LINK_UP
)) {
3076 write_lock_bh(&bond
->curr_slave_lock
);
3077 bond_change_active_slave(bond
, bond
->primary_slave
);
3078 write_unlock_bh(&bond
->curr_slave_lock
);
3081 bond_set_carrier(bond
);
3085 * Send ARP probes for active-backup mode ARP monitor.
3087 * Called with bond->lock held for read.
3089 static void bond_ab_arp_probe(struct bonding
*bond
)
3091 struct slave
*slave
;
3094 read_lock(&bond
->curr_slave_lock
);
3096 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3097 printk("PROBE: c_arp %s && cas %s BAD\n",
3098 bond
->current_arp_slave
->dev
->name
,
3099 bond
->curr_active_slave
->dev
->name
);
3101 if (bond
->curr_active_slave
) {
3102 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3103 read_unlock(&bond
->curr_slave_lock
);
3107 read_unlock(&bond
->curr_slave_lock
);
3109 /* if we don't have a curr_active_slave, search for the next available
3110 * backup slave from the current_arp_slave and make it the candidate
3111 * for becoming the curr_active_slave
3114 if (!bond
->current_arp_slave
) {
3115 bond
->current_arp_slave
= bond
->first_slave
;
3116 if (!bond
->current_arp_slave
)
3120 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3122 /* search for next candidate */
3123 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3124 if (IS_UP(slave
->dev
)) {
3125 slave
->link
= BOND_LINK_BACK
;
3126 bond_set_slave_active_flags(slave
);
3127 bond_arp_send_all(bond
, slave
);
3128 slave
->jiffies
= jiffies
;
3129 bond
->current_arp_slave
= slave
;
3133 /* if the link state is up at this point, we
3134 * mark it down - this can happen if we have
3135 * simultaneous link failures and
3136 * reselect_active_interface doesn't make this
3137 * one the current slave so it is still marked
3138 * up when it is actually down
3140 if (slave
->link
== BOND_LINK_UP
) {
3141 slave
->link
= BOND_LINK_DOWN
;
3142 if (slave
->link_failure_count
< UINT_MAX
)
3143 slave
->link_failure_count
++;
3145 bond_set_slave_inactive_flags(slave
);
3147 printk(KERN_INFO DRV_NAME
3148 ": %s: backup interface %s is now down.\n",
3149 bond
->dev
->name
, slave
->dev
->name
);
3154 void bond_activebackup_arp_mon(struct work_struct
*work
)
3156 struct bonding
*bond
= container_of(work
, struct bonding
,
3160 read_lock(&bond
->lock
);
3162 if (bond
->kill_timers
)
3165 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3167 if (bond
->slave_cnt
== 0)
3170 if (bond
->send_grat_arp
) {
3171 read_lock(&bond
->curr_slave_lock
);
3172 bond_send_gratuitous_arp(bond
);
3173 read_unlock(&bond
->curr_slave_lock
);
3176 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3177 read_unlock(&bond
->lock
);
3179 read_lock(&bond
->lock
);
3181 bond_ab_arp_commit(bond
, delta_in_ticks
);
3183 read_unlock(&bond
->lock
);
3185 read_lock(&bond
->lock
);
3188 bond_ab_arp_probe(bond
);
3191 if (bond
->params
.arp_interval
) {
3192 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3195 read_unlock(&bond
->lock
);
3198 /*------------------------------ proc/seq_file-------------------------------*/
3200 #ifdef CONFIG_PROC_FS
3202 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3204 struct bonding
*bond
= seq
->private;
3206 struct slave
*slave
;
3209 /* make sure the bond won't be taken away */
3210 read_lock(&dev_base_lock
);
3211 read_lock(&bond
->lock
);
3214 return SEQ_START_TOKEN
;
3217 bond_for_each_slave(bond
, slave
, i
) {
3218 if (++off
== *pos
) {
3226 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3228 struct bonding
*bond
= seq
->private;
3229 struct slave
*slave
= v
;
3232 if (v
== SEQ_START_TOKEN
) {
3233 return bond
->first_slave
;
3236 slave
= slave
->next
;
3238 return (slave
== bond
->first_slave
) ? NULL
: slave
;
3241 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
3243 struct bonding
*bond
= seq
->private;
3245 read_unlock(&bond
->lock
);
3246 read_unlock(&dev_base_lock
);
3249 static void bond_info_show_master(struct seq_file
*seq
)
3251 struct bonding
*bond
= seq
->private;
3256 read_lock(&bond
->curr_slave_lock
);
3257 curr
= bond
->curr_active_slave
;
3258 read_unlock(&bond
->curr_slave_lock
);
3260 seq_printf(seq
, "Bonding Mode: %s",
3261 bond_mode_name(bond
->params
.mode
));
3263 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
&&
3264 bond
->params
.fail_over_mac
)
3265 seq_printf(seq
, " (fail_over_mac %s)",
3266 fail_over_mac_tbl
[bond
->params
.fail_over_mac
].modename
);
3268 seq_printf(seq
, "\n");
3270 if (bond
->params
.mode
== BOND_MODE_XOR
||
3271 bond
->params
.mode
== BOND_MODE_8023AD
) {
3272 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
3273 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
3274 bond
->params
.xmit_policy
);
3277 if (USES_PRIMARY(bond
->params
.mode
)) {
3278 seq_printf(seq
, "Primary Slave: %s\n",
3279 (bond
->primary_slave
) ?
3280 bond
->primary_slave
->dev
->name
: "None");
3282 seq_printf(seq
, "Currently Active Slave: %s\n",
3283 (curr
) ? curr
->dev
->name
: "None");
3286 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
3288 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
3289 seq_printf(seq
, "Up Delay (ms): %d\n",
3290 bond
->params
.updelay
* bond
->params
.miimon
);
3291 seq_printf(seq
, "Down Delay (ms): %d\n",
3292 bond
->params
.downdelay
* bond
->params
.miimon
);
3295 /* ARP information */
3296 if(bond
->params
.arp_interval
> 0) {
3298 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
3299 bond
->params
.arp_interval
);
3301 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
3303 for(i
= 0; (i
< BOND_MAX_ARP_TARGETS
) ;i
++) {
3304 if (!bond
->params
.arp_targets
[i
])
3307 seq_printf(seq
, ",");
3308 target
= ntohl(bond
->params
.arp_targets
[i
]);
3309 seq_printf(seq
, " %d.%d.%d.%d", HIPQUAD(target
));
3312 seq_printf(seq
, "\n");
3315 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3316 struct ad_info ad_info
;
3317 DECLARE_MAC_BUF(mac
);
3319 seq_puts(seq
, "\n802.3ad info\n");
3320 seq_printf(seq
, "LACP rate: %s\n",
3321 (bond
->params
.lacp_fast
) ? "fast" : "slow");
3323 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3324 seq_printf(seq
, "bond %s has no active aggregator\n",
3327 seq_printf(seq
, "Active Aggregator Info:\n");
3329 seq_printf(seq
, "\tAggregator ID: %d\n",
3330 ad_info
.aggregator_id
);
3331 seq_printf(seq
, "\tNumber of ports: %d\n",
3333 seq_printf(seq
, "\tActor Key: %d\n",
3335 seq_printf(seq
, "\tPartner Key: %d\n",
3336 ad_info
.partner_key
);
3337 seq_printf(seq
, "\tPartner Mac Address: %s\n",
3338 print_mac(mac
, ad_info
.partner_system
));
3343 static void bond_info_show_slave(struct seq_file
*seq
, const struct slave
*slave
)
3345 struct bonding
*bond
= seq
->private;
3346 DECLARE_MAC_BUF(mac
);
3348 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
3349 seq_printf(seq
, "MII Status: %s\n",
3350 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
3351 seq_printf(seq
, "Link Failure Count: %u\n",
3352 slave
->link_failure_count
);
3355 "Permanent HW addr: %s\n",
3356 print_mac(mac
, slave
->perm_hwaddr
));
3358 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3359 const struct aggregator
*agg
3360 = SLAVE_AD_INFO(slave
).port
.aggregator
;
3363 seq_printf(seq
, "Aggregator ID: %d\n",
3364 agg
->aggregator_identifier
);
3366 seq_puts(seq
, "Aggregator ID: N/A\n");
3371 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
3373 if (v
== SEQ_START_TOKEN
) {
3374 seq_printf(seq
, "%s\n", version
);
3375 bond_info_show_master(seq
);
3377 bond_info_show_slave(seq
, v
);
3383 static struct seq_operations bond_info_seq_ops
= {
3384 .start
= bond_info_seq_start
,
3385 .next
= bond_info_seq_next
,
3386 .stop
= bond_info_seq_stop
,
3387 .show
= bond_info_seq_show
,
3390 static int bond_info_open(struct inode
*inode
, struct file
*file
)
3392 struct seq_file
*seq
;
3393 struct proc_dir_entry
*proc
;
3396 res
= seq_open(file
, &bond_info_seq_ops
);
3398 /* recover the pointer buried in proc_dir_entry data */
3399 seq
= file
->private_data
;
3401 seq
->private = proc
->data
;
3407 static const struct file_operations bond_info_fops
= {
3408 .owner
= THIS_MODULE
,
3409 .open
= bond_info_open
,
3411 .llseek
= seq_lseek
,
3412 .release
= seq_release
,
3415 static int bond_create_proc_entry(struct bonding
*bond
)
3417 struct net_device
*bond_dev
= bond
->dev
;
3419 if (bond_proc_dir
) {
3420 bond
->proc_entry
= proc_create_data(bond_dev
->name
,
3421 S_IRUGO
, bond_proc_dir
,
3422 &bond_info_fops
, bond
);
3423 if (bond
->proc_entry
== NULL
) {
3424 printk(KERN_WARNING DRV_NAME
3425 ": Warning: Cannot create /proc/net/%s/%s\n",
3426 DRV_NAME
, bond_dev
->name
);
3428 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3435 static void bond_remove_proc_entry(struct bonding
*bond
)
3437 if (bond_proc_dir
&& bond
->proc_entry
) {
3438 remove_proc_entry(bond
->proc_file_name
, bond_proc_dir
);
3439 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3440 bond
->proc_entry
= NULL
;
3444 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3445 * Caller must hold rtnl_lock.
3447 static void bond_create_proc_dir(void)
3449 int len
= strlen(DRV_NAME
);
3451 for (bond_proc_dir
= init_net
.proc_net
->subdir
; bond_proc_dir
;
3452 bond_proc_dir
= bond_proc_dir
->next
) {
3453 if ((bond_proc_dir
->namelen
== len
) &&
3454 !memcmp(bond_proc_dir
->name
, DRV_NAME
, len
)) {
3459 if (!bond_proc_dir
) {
3460 bond_proc_dir
= proc_mkdir(DRV_NAME
, init_net
.proc_net
);
3461 if (bond_proc_dir
) {
3462 bond_proc_dir
->owner
= THIS_MODULE
;
3464 printk(KERN_WARNING DRV_NAME
3465 ": Warning: cannot create /proc/net/%s\n",
3471 /* Destroy the bonding directory under /proc/net, if empty.
3472 * Caller must hold rtnl_lock.
3474 static void bond_destroy_proc_dir(void)
3476 struct proc_dir_entry
*de
;
3478 if (!bond_proc_dir
) {
3482 /* verify that the /proc dir is empty */
3483 for (de
= bond_proc_dir
->subdir
; de
; de
= de
->next
) {
3484 /* ignore . and .. */
3485 if (*(de
->name
) != '.') {
3491 if (bond_proc_dir
->owner
== THIS_MODULE
) {
3492 bond_proc_dir
->owner
= NULL
;
3495 remove_proc_entry(DRV_NAME
, init_net
.proc_net
);
3496 bond_proc_dir
= NULL
;
3499 #endif /* CONFIG_PROC_FS */
3501 /*-------------------------- netdev event handling --------------------------*/
3504 * Change device name
3506 static int bond_event_changename(struct bonding
*bond
)
3508 #ifdef CONFIG_PROC_FS
3509 bond_remove_proc_entry(bond
);
3510 bond_create_proc_entry(bond
);
3512 down_write(&(bonding_rwsem
));
3513 bond_destroy_sysfs_entry(bond
);
3514 bond_create_sysfs_entry(bond
);
3515 up_write(&(bonding_rwsem
));
3519 static int bond_master_netdev_event(unsigned long event
, struct net_device
*bond_dev
)
3521 struct bonding
*event_bond
= bond_dev
->priv
;
3524 case NETDEV_CHANGENAME
:
3525 return bond_event_changename(event_bond
);
3526 case NETDEV_UNREGISTER
:
3527 bond_release_all(event_bond
->dev
);
3536 static int bond_slave_netdev_event(unsigned long event
, struct net_device
*slave_dev
)
3538 struct net_device
*bond_dev
= slave_dev
->master
;
3539 struct bonding
*bond
= bond_dev
->priv
;
3542 case NETDEV_UNREGISTER
:
3544 if (bond
->setup_by_slave
)
3545 bond_release_and_destroy(bond_dev
, slave_dev
);
3547 bond_release(bond_dev
, slave_dev
);
3552 * TODO: is this what we get if somebody
3553 * sets up a hierarchical bond, then rmmod's
3554 * one of the slave bonding devices?
3559 * ... Or is it this?
3562 case NETDEV_CHANGEMTU
:
3564 * TODO: Should slaves be allowed to
3565 * independently alter their MTU? For
3566 * an active-backup bond, slaves need
3567 * not be the same type of device, so
3568 * MTUs may vary. For other modes,
3569 * slaves arguably should have the
3570 * same MTUs. To do this, we'd need to
3571 * take over the slave's change_mtu
3572 * function for the duration of their
3576 case NETDEV_CHANGENAME
:
3578 * TODO: handle changing the primary's name
3581 case NETDEV_FEAT_CHANGE
:
3582 bond_compute_features(bond
);
3592 * bond_netdev_event: handle netdev notifier chain events.
3594 * This function receives events for the netdev chain. The caller (an
3595 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3596 * locks for us to safely manipulate the slave devices (RTNL lock,
3599 static int bond_netdev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3601 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3603 if (dev_net(event_dev
) != &init_net
)
3606 dprintk("event_dev: %s, event: %lx\n",
3607 (event_dev
? event_dev
->name
: "None"),
3610 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3613 if (event_dev
->flags
& IFF_MASTER
) {
3614 dprintk("IFF_MASTER\n");
3615 return bond_master_netdev_event(event
, event_dev
);
3618 if (event_dev
->flags
& IFF_SLAVE
) {
3619 dprintk("IFF_SLAVE\n");
3620 return bond_slave_netdev_event(event
, event_dev
);
3627 * bond_inetaddr_event: handle inetaddr notifier chain events.
3629 * We keep track of device IPs primarily to use as source addresses in
3630 * ARP monitor probes (rather than spewing out broadcasts all the time).
3632 * We track one IP for the main device (if it has one), plus one per VLAN.
3634 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3636 struct in_ifaddr
*ifa
= ptr
;
3637 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3638 struct bonding
*bond
;
3639 struct vlan_entry
*vlan
;
3641 if (dev_net(ifa
->ifa_dev
->dev
) != &init_net
)
3644 list_for_each_entry(bond
, &bond_dev_list
, bond_list
) {
3645 if (bond
->dev
== event_dev
) {
3648 bond
->master_ip
= ifa
->ifa_local
;
3651 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3658 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3659 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3660 if (vlan_dev
== event_dev
) {
3663 vlan
->vlan_ip
= ifa
->ifa_local
;
3667 bond_glean_dev_ip(vlan_dev
);
3678 static struct notifier_block bond_netdev_notifier
= {
3679 .notifier_call
= bond_netdev_event
,
3682 static struct notifier_block bond_inetaddr_notifier
= {
3683 .notifier_call
= bond_inetaddr_event
,
3686 /*-------------------------- Packet type handling ---------------------------*/
3688 /* register to receive lacpdus on a bond */
3689 static void bond_register_lacpdu(struct bonding
*bond
)
3691 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3693 /* initialize packet type */
3694 pk_type
->type
= PKT_TYPE_LACPDU
;
3695 pk_type
->dev
= bond
->dev
;
3696 pk_type
->func
= bond_3ad_lacpdu_recv
;
3698 dev_add_pack(pk_type
);
3701 /* unregister to receive lacpdus on a bond */
3702 static void bond_unregister_lacpdu(struct bonding
*bond
)
3704 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3707 void bond_register_arp(struct bonding
*bond
)
3709 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3714 pt
->type
= htons(ETH_P_ARP
);
3715 pt
->dev
= bond
->dev
;
3716 pt
->func
= bond_arp_rcv
;
3720 void bond_unregister_arp(struct bonding
*bond
)
3722 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3724 dev_remove_pack(pt
);
3728 /*---------------------------- Hashing Policies -----------------------------*/
3731 * Hash for the output device based upon layer 2 and layer 3 data. If
3732 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3734 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
,
3735 struct net_device
*bond_dev
, int count
)
3737 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3738 struct iphdr
*iph
= ip_hdr(skb
);
3740 if (skb
->protocol
== __constant_htons(ETH_P_IP
)) {
3741 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3742 (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5])) % count
;
3745 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3749 * Hash for the output device based upon layer 3 and layer 4 data. If
3750 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3751 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3753 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
,
3754 struct net_device
*bond_dev
, int count
)
3756 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3757 struct iphdr
*iph
= ip_hdr(skb
);
3758 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3761 if (skb
->protocol
== __constant_htons(ETH_P_IP
)) {
3762 if (!(iph
->frag_off
& __constant_htons(IP_MF
|IP_OFFSET
)) &&
3763 (iph
->protocol
== IPPROTO_TCP
||
3764 iph
->protocol
== IPPROTO_UDP
)) {
3765 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3767 return (layer4_xor
^
3768 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3772 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3776 * Hash for the output device based upon layer 2 data
3778 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
,
3779 struct net_device
*bond_dev
, int count
)
3781 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3783 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3786 /*-------------------------- Device entry points ----------------------------*/
3788 static int bond_open(struct net_device
*bond_dev
)
3790 struct bonding
*bond
= bond_dev
->priv
;
3792 bond
->kill_timers
= 0;
3794 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
3795 (bond
->params
.mode
== BOND_MODE_ALB
)) {
3796 /* bond_alb_initialize must be called before the timer
3799 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3800 /* something went wrong - fail the open operation */
3804 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3805 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3808 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3809 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3810 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3813 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3814 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3815 INIT_DELAYED_WORK(&bond
->arp_work
,
3816 bond_activebackup_arp_mon
);
3818 INIT_DELAYED_WORK(&bond
->arp_work
,
3819 bond_loadbalance_arp_mon
);
3821 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3822 if (bond
->params
.arp_validate
)
3823 bond_register_arp(bond
);
3826 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3827 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3828 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3829 /* register to receive LACPDUs */
3830 bond_register_lacpdu(bond
);
3836 static int bond_close(struct net_device
*bond_dev
)
3838 struct bonding
*bond
= bond_dev
->priv
;
3840 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3841 /* Unregister the receive of LACPDUs */
3842 bond_unregister_lacpdu(bond
);
3845 if (bond
->params
.arp_validate
)
3846 bond_unregister_arp(bond
);
3848 write_lock_bh(&bond
->lock
);
3850 bond
->send_grat_arp
= 0;
3852 /* signal timers not to re-arm */
3853 bond
->kill_timers
= 1;
3855 write_unlock_bh(&bond
->lock
);
3857 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3858 cancel_delayed_work(&bond
->mii_work
);
3861 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3862 cancel_delayed_work(&bond
->arp_work
);
3865 switch (bond
->params
.mode
) {
3866 case BOND_MODE_8023AD
:
3867 cancel_delayed_work(&bond
->ad_work
);
3871 cancel_delayed_work(&bond
->alb_work
);
3878 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
3879 (bond
->params
.mode
== BOND_MODE_ALB
)) {
3880 /* Must be called only after all
3881 * slaves have been released
3883 bond_alb_deinitialize(bond
);
3889 static struct net_device_stats
*bond_get_stats(struct net_device
*bond_dev
)
3891 struct bonding
*bond
= bond_dev
->priv
;
3892 struct net_device_stats
*stats
= &(bond
->stats
), *sstats
;
3893 struct net_device_stats local_stats
;
3894 struct slave
*slave
;
3897 memset(&local_stats
, 0, sizeof(struct net_device_stats
));
3899 read_lock_bh(&bond
->lock
);
3901 bond_for_each_slave(bond
, slave
, i
) {
3902 sstats
= slave
->dev
->get_stats(slave
->dev
);
3903 local_stats
.rx_packets
+= sstats
->rx_packets
;
3904 local_stats
.rx_bytes
+= sstats
->rx_bytes
;
3905 local_stats
.rx_errors
+= sstats
->rx_errors
;
3906 local_stats
.rx_dropped
+= sstats
->rx_dropped
;
3908 local_stats
.tx_packets
+= sstats
->tx_packets
;
3909 local_stats
.tx_bytes
+= sstats
->tx_bytes
;
3910 local_stats
.tx_errors
+= sstats
->tx_errors
;
3911 local_stats
.tx_dropped
+= sstats
->tx_dropped
;
3913 local_stats
.multicast
+= sstats
->multicast
;
3914 local_stats
.collisions
+= sstats
->collisions
;
3916 local_stats
.rx_length_errors
+= sstats
->rx_length_errors
;
3917 local_stats
.rx_over_errors
+= sstats
->rx_over_errors
;
3918 local_stats
.rx_crc_errors
+= sstats
->rx_crc_errors
;
3919 local_stats
.rx_frame_errors
+= sstats
->rx_frame_errors
;
3920 local_stats
.rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3921 local_stats
.rx_missed_errors
+= sstats
->rx_missed_errors
;
3923 local_stats
.tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3924 local_stats
.tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3925 local_stats
.tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3926 local_stats
.tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3927 local_stats
.tx_window_errors
+= sstats
->tx_window_errors
;
3930 memcpy(stats
, &local_stats
, sizeof(struct net_device_stats
));
3932 read_unlock_bh(&bond
->lock
);
3937 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3939 struct net_device
*slave_dev
= NULL
;
3940 struct ifbond k_binfo
;
3941 struct ifbond __user
*u_binfo
= NULL
;
3942 struct ifslave k_sinfo
;
3943 struct ifslave __user
*u_sinfo
= NULL
;
3944 struct mii_ioctl_data
*mii
= NULL
;
3947 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3948 bond_dev
->name
, cmd
);
3960 * We do this again just in case we were called by SIOCGMIIREG
3961 * instead of SIOCGMIIPHY.
3968 if (mii
->reg_num
== 1) {
3969 struct bonding
*bond
= bond_dev
->priv
;
3971 read_lock(&bond
->lock
);
3972 read_lock(&bond
->curr_slave_lock
);
3973 if (netif_carrier_ok(bond
->dev
)) {
3974 mii
->val_out
= BMSR_LSTATUS
;
3976 read_unlock(&bond
->curr_slave_lock
);
3977 read_unlock(&bond
->lock
);
3981 case BOND_INFO_QUERY_OLD
:
3982 case SIOCBONDINFOQUERY
:
3983 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3985 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
))) {
3989 res
= bond_info_query(bond_dev
, &k_binfo
);
3991 if (copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
))) {
3997 case BOND_SLAVE_INFO_QUERY_OLD
:
3998 case SIOCBONDSLAVEINFOQUERY
:
3999 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
4001 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
))) {
4005 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
4007 if (copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
))) {
4018 if (!capable(CAP_NET_ADMIN
)) {
4022 down_write(&(bonding_rwsem
));
4023 slave_dev
= dev_get_by_name(&init_net
, ifr
->ifr_slave
);
4025 dprintk("slave_dev=%p: \n", slave_dev
);
4030 dprintk("slave_dev->name=%s: \n", slave_dev
->name
);
4032 case BOND_ENSLAVE_OLD
:
4033 case SIOCBONDENSLAVE
:
4034 res
= bond_enslave(bond_dev
, slave_dev
);
4036 case BOND_RELEASE_OLD
:
4037 case SIOCBONDRELEASE
:
4038 res
= bond_release(bond_dev
, slave_dev
);
4040 case BOND_SETHWADDR_OLD
:
4041 case SIOCBONDSETHWADDR
:
4042 res
= bond_sethwaddr(bond_dev
, slave_dev
);
4044 case BOND_CHANGE_ACTIVE_OLD
:
4045 case SIOCBONDCHANGEACTIVE
:
4046 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
4055 up_write(&(bonding_rwsem
));
4059 static void bond_set_multicast_list(struct net_device
*bond_dev
)
4061 struct bonding
*bond
= bond_dev
->priv
;
4062 struct dev_mc_list
*dmi
;
4065 * Do promisc before checking multicast_mode
4067 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
)) {
4068 bond_set_promiscuity(bond
, 1);
4071 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
)) {
4072 bond_set_promiscuity(bond
, -1);
4075 /* set allmulti flag to slaves */
4076 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
)) {
4077 bond_set_allmulti(bond
, 1);
4080 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
)) {
4081 bond_set_allmulti(bond
, -1);
4084 read_lock(&bond
->lock
);
4086 bond
->flags
= bond_dev
->flags
;
4088 /* looking for addresses to add to slaves' mc list */
4089 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
4090 if (!bond_mc_list_find_dmi(dmi
, bond
->mc_list
)) {
4091 bond_mc_add(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4095 /* looking for addresses to delete from slaves' list */
4096 for (dmi
= bond
->mc_list
; dmi
; dmi
= dmi
->next
) {
4097 if (!bond_mc_list_find_dmi(dmi
, bond_dev
->mc_list
)) {
4098 bond_mc_delete(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4102 /* save master's multicast list */
4103 bond_mc_list_destroy(bond
);
4104 bond_mc_list_copy(bond_dev
->mc_list
, bond
, GFP_ATOMIC
);
4106 read_unlock(&bond
->lock
);
4110 * Change the MTU of all of a master's slaves to match the master
4112 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
4114 struct bonding
*bond
= bond_dev
->priv
;
4115 struct slave
*slave
, *stop_at
;
4119 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond
,
4120 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
4122 /* Can't hold bond->lock with bh disabled here since
4123 * some base drivers panic. On the other hand we can't
4124 * hold bond->lock without bh disabled because we'll
4125 * deadlock. The only solution is to rely on the fact
4126 * that we're under rtnl_lock here, and the slaves
4127 * list won't change. This doesn't solve the problem
4128 * of setting the slave's MTU while it is
4129 * transmitting, but the assumption is that the base
4130 * driver can handle that.
4132 * TODO: figure out a way to safely iterate the slaves
4133 * list, but without holding a lock around the actual
4134 * call to the base driver.
4137 bond_for_each_slave(bond
, slave
, i
) {
4138 dprintk("s %p s->p %p c_m %p\n", slave
,
4139 slave
->prev
, slave
->dev
->change_mtu
);
4141 res
= dev_set_mtu(slave
->dev
, new_mtu
);
4144 /* If we failed to set the slave's mtu to the new value
4145 * we must abort the operation even in ACTIVE_BACKUP
4146 * mode, because if we allow the backup slaves to have
4147 * different mtu values than the active slave we'll
4148 * need to change their mtu when doing a failover. That
4149 * means changing their mtu from timer context, which
4150 * is probably not a good idea.
4152 dprintk("err %d %s\n", res
, slave
->dev
->name
);
4157 bond_dev
->mtu
= new_mtu
;
4162 /* unwind from head to the slave that failed */
4164 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4167 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
4169 dprintk("unwind err %d dev %s\n", tmp_res
,
4180 * Note that many devices must be down to change the HW address, and
4181 * downing the master releases all slaves. We can make bonds full of
4182 * bonding devices to test this, however.
4184 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4186 struct bonding
*bond
= bond_dev
->priv
;
4187 struct sockaddr
*sa
= addr
, tmp_sa
;
4188 struct slave
*slave
, *stop_at
;
4192 dprintk("bond=%p, name=%s\n", bond
, (bond_dev
? bond_dev
->name
: "None"));
4195 * If fail_over_mac is set to active, do nothing and return
4196 * success. Returning an error causes ifenslave to fail.
4198 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4201 if (!is_valid_ether_addr(sa
->sa_data
)) {
4202 return -EADDRNOTAVAIL
;
4205 /* Can't hold bond->lock with bh disabled here since
4206 * some base drivers panic. On the other hand we can't
4207 * hold bond->lock without bh disabled because we'll
4208 * deadlock. The only solution is to rely on the fact
4209 * that we're under rtnl_lock here, and the slaves
4210 * list won't change. This doesn't solve the problem
4211 * of setting the slave's hw address while it is
4212 * transmitting, but the assumption is that the base
4213 * driver can handle that.
4215 * TODO: figure out a way to safely iterate the slaves
4216 * list, but without holding a lock around the actual
4217 * call to the base driver.
4220 bond_for_each_slave(bond
, slave
, i
) {
4221 dprintk("slave %p %s\n", slave
, slave
->dev
->name
);
4223 if (slave
->dev
->set_mac_address
== NULL
) {
4225 dprintk("EOPNOTSUPP %s\n", slave
->dev
->name
);
4229 res
= dev_set_mac_address(slave
->dev
, addr
);
4231 /* TODO: consider downing the slave
4233 * User should expect communications
4234 * breakage anyway until ARP finish
4237 dprintk("err %d %s\n", res
, slave
->dev
->name
);
4243 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4247 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4248 tmp_sa
.sa_family
= bond_dev
->type
;
4250 /* unwind from head to the slave that failed */
4252 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4255 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4257 dprintk("unwind err %d dev %s\n", tmp_res
,
4265 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4267 struct bonding
*bond
= bond_dev
->priv
;
4268 struct slave
*slave
, *start_at
;
4269 int i
, slave_no
, res
= 1;
4271 read_lock(&bond
->lock
);
4273 if (!BOND_IS_OK(bond
)) {
4278 * Concurrent TX may collide on rr_tx_counter; we accept that
4279 * as being rare enough not to justify using an atomic op here
4281 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4283 bond_for_each_slave(bond
, slave
, i
) {
4291 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4292 if (IS_UP(slave
->dev
) &&
4293 (slave
->link
== BOND_LINK_UP
) &&
4294 (slave
->state
== BOND_STATE_ACTIVE
)) {
4295 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4302 /* no suitable interface, frame not sent */
4305 read_unlock(&bond
->lock
);
4311 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4312 * the bond has a usable interface.
4314 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4316 struct bonding
*bond
= bond_dev
->priv
;
4319 read_lock(&bond
->lock
);
4320 read_lock(&bond
->curr_slave_lock
);
4322 if (!BOND_IS_OK(bond
)) {
4326 if (!bond
->curr_active_slave
)
4329 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4333 /* no suitable interface, frame not sent */
4336 read_unlock(&bond
->curr_slave_lock
);
4337 read_unlock(&bond
->lock
);
4342 * In bond_xmit_xor() , we determine the output device by using a pre-
4343 * determined xmit_hash_policy(), If the selected device is not enabled,
4344 * find the next active slave.
4346 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4348 struct bonding
*bond
= bond_dev
->priv
;
4349 struct slave
*slave
, *start_at
;
4354 read_lock(&bond
->lock
);
4356 if (!BOND_IS_OK(bond
)) {
4360 slave_no
= bond
->xmit_hash_policy(skb
, bond_dev
, bond
->slave_cnt
);
4362 bond_for_each_slave(bond
, slave
, i
) {
4371 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4372 if (IS_UP(slave
->dev
) &&
4373 (slave
->link
== BOND_LINK_UP
) &&
4374 (slave
->state
== BOND_STATE_ACTIVE
)) {
4375 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4382 /* no suitable interface, frame not sent */
4385 read_unlock(&bond
->lock
);
4390 * in broadcast mode, we send everything to all usable interfaces.
4392 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4394 struct bonding
*bond
= bond_dev
->priv
;
4395 struct slave
*slave
, *start_at
;
4396 struct net_device
*tx_dev
= NULL
;
4400 read_lock(&bond
->lock
);
4402 if (!BOND_IS_OK(bond
)) {
4406 read_lock(&bond
->curr_slave_lock
);
4407 start_at
= bond
->curr_active_slave
;
4408 read_unlock(&bond
->curr_slave_lock
);
4414 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4415 if (IS_UP(slave
->dev
) &&
4416 (slave
->link
== BOND_LINK_UP
) &&
4417 (slave
->state
== BOND_STATE_ACTIVE
)) {
4419 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4421 printk(KERN_ERR DRV_NAME
4422 ": %s: Error: bond_xmit_broadcast(): "
4423 "skb_clone() failed\n",
4428 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4430 dev_kfree_skb(skb2
);
4434 tx_dev
= slave
->dev
;
4439 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4444 /* no suitable interface, frame not sent */
4447 /* frame sent to all suitable interfaces */
4448 read_unlock(&bond
->lock
);
4452 /*------------------------- Device initialization ---------------------------*/
4454 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4456 switch (bond
->params
.xmit_policy
) {
4457 case BOND_XMIT_POLICY_LAYER23
:
4458 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4460 case BOND_XMIT_POLICY_LAYER34
:
4461 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4463 case BOND_XMIT_POLICY_LAYER2
:
4465 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4471 * set bond mode specific net device operations
4473 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4475 struct net_device
*bond_dev
= bond
->dev
;
4478 case BOND_MODE_ROUNDROBIN
:
4479 bond_dev
->hard_start_xmit
= bond_xmit_roundrobin
;
4481 case BOND_MODE_ACTIVEBACKUP
:
4482 bond_dev
->hard_start_xmit
= bond_xmit_activebackup
;
4485 bond_dev
->hard_start_xmit
= bond_xmit_xor
;
4486 bond_set_xmit_hash_policy(bond
);
4488 case BOND_MODE_BROADCAST
:
4489 bond_dev
->hard_start_xmit
= bond_xmit_broadcast
;
4491 case BOND_MODE_8023AD
:
4492 bond_set_master_3ad_flags(bond
);
4493 bond_dev
->hard_start_xmit
= bond_3ad_xmit_xor
;
4494 bond_set_xmit_hash_policy(bond
);
4497 bond_set_master_alb_flags(bond
);
4500 bond_dev
->hard_start_xmit
= bond_alb_xmit
;
4501 bond_dev
->set_mac_address
= bond_alb_set_mac_address
;
4504 /* Should never happen, mode already checked */
4505 printk(KERN_ERR DRV_NAME
4506 ": %s: Error: Unknown bonding mode %d\n",
4513 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4514 struct ethtool_drvinfo
*drvinfo
)
4516 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4517 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4518 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4521 static const struct ethtool_ops bond_ethtool_ops
= {
4522 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4526 * Does not allocate but creates a /proc entry.
4529 static int bond_init(struct net_device
*bond_dev
, struct bond_params
*params
)
4531 struct bonding
*bond
= bond_dev
->priv
;
4533 dprintk("Begin bond_init for %s\n", bond_dev
->name
);
4535 /* initialize rwlocks */
4536 rwlock_init(&bond
->lock
);
4537 rwlock_init(&bond
->curr_slave_lock
);
4539 bond
->params
= *params
; /* copy params struct */
4541 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4545 /* Initialize pointers */
4546 bond
->first_slave
= NULL
;
4547 bond
->curr_active_slave
= NULL
;
4548 bond
->current_arp_slave
= NULL
;
4549 bond
->primary_slave
= NULL
;
4550 bond
->dev
= bond_dev
;
4551 bond
->send_grat_arp
= 0;
4552 bond
->setup_by_slave
= 0;
4553 INIT_LIST_HEAD(&bond
->vlan_list
);
4555 /* Initialize the device entry points */
4556 bond_dev
->open
= bond_open
;
4557 bond_dev
->stop
= bond_close
;
4558 bond_dev
->get_stats
= bond_get_stats
;
4559 bond_dev
->do_ioctl
= bond_do_ioctl
;
4560 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4561 bond_dev
->set_multicast_list
= bond_set_multicast_list
;
4562 bond_dev
->change_mtu
= bond_change_mtu
;
4563 bond_dev
->set_mac_address
= bond_set_mac_address
;
4564 bond_dev
->validate_addr
= NULL
;
4566 bond_set_mode_ops(bond
, bond
->params
.mode
);
4568 bond_dev
->destructor
= free_netdev
;
4570 /* Initialize the device options */
4571 bond_dev
->tx_queue_len
= 0;
4572 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4573 bond_dev
->priv_flags
|= IFF_BONDING
;
4575 /* At first, we block adding VLANs. That's the only way to
4576 * prevent problems that occur when adding VLANs over an
4577 * empty bond. The block will be removed once non-challenged
4578 * slaves are enslaved.
4580 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4582 /* don't acquire bond device's netif_tx_lock when
4584 bond_dev
->features
|= NETIF_F_LLTX
;
4586 /* By default, we declare the bond to be fully
4587 * VLAN hardware accelerated capable. Special
4588 * care is taken in the various xmit functions
4589 * when there are slaves that are not hw accel
4592 bond_dev
->vlan_rx_register
= bond_vlan_rx_register
;
4593 bond_dev
->vlan_rx_add_vid
= bond_vlan_rx_add_vid
;
4594 bond_dev
->vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
;
4595 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4596 NETIF_F_HW_VLAN_RX
|
4597 NETIF_F_HW_VLAN_FILTER
);
4599 #ifdef CONFIG_PROC_FS
4600 bond_create_proc_entry(bond
);
4602 list_add_tail(&bond
->bond_list
, &bond_dev_list
);
4607 /* De-initialize device specific data.
4608 * Caller must hold rtnl_lock.
4610 static void bond_deinit(struct net_device
*bond_dev
)
4612 struct bonding
*bond
= bond_dev
->priv
;
4614 list_del(&bond
->bond_list
);
4616 #ifdef CONFIG_PROC_FS
4617 bond_remove_proc_entry(bond
);
4621 static void bond_work_cancel_all(struct bonding
*bond
)
4623 write_lock_bh(&bond
->lock
);
4624 bond
->kill_timers
= 1;
4625 write_unlock_bh(&bond
->lock
);
4627 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4628 cancel_delayed_work(&bond
->mii_work
);
4630 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4631 cancel_delayed_work(&bond
->arp_work
);
4633 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4634 delayed_work_pending(&bond
->alb_work
))
4635 cancel_delayed_work(&bond
->alb_work
);
4637 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4638 delayed_work_pending(&bond
->ad_work
))
4639 cancel_delayed_work(&bond
->ad_work
);
4642 /* Unregister and free all bond devices.
4643 * Caller must hold rtnl_lock.
4645 static void bond_free_all(void)
4647 struct bonding
*bond
, *nxt
;
4649 list_for_each_entry_safe(bond
, nxt
, &bond_dev_list
, bond_list
) {
4650 struct net_device
*bond_dev
= bond
->dev
;
4652 bond_work_cancel_all(bond
);
4653 netif_tx_lock_bh(bond_dev
);
4654 bond_mc_list_destroy(bond
);
4655 netif_tx_unlock_bh(bond_dev
);
4656 /* Release the bonded slaves */
4657 bond_release_all(bond_dev
);
4661 #ifdef CONFIG_PROC_FS
4662 bond_destroy_proc_dir();
4666 /*------------------------- Module initialization ---------------------------*/
4669 * Convert string input module parms. Accept either the
4670 * number of the mode or its string name. A bit complicated because
4671 * some mode names are substrings of other names, and calls from sysfs
4672 * may have whitespace in the name (trailing newlines, for example).
4674 int bond_parse_parm(const char *buf
, struct bond_parm_tbl
*tbl
)
4676 int mode
= -1, i
, rv
;
4677 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4679 for (p
= (char *)buf
; *p
; p
++)
4680 if (!(isdigit(*p
) || isspace(*p
)))
4684 rv
= sscanf(buf
, "%20s", modestr
);
4686 rv
= sscanf(buf
, "%d", &mode
);
4691 for (i
= 0; tbl
[i
].modename
; i
++) {
4692 if (mode
== tbl
[i
].mode
)
4694 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4701 static int bond_check_params(struct bond_params
*params
)
4703 int arp_validate_value
, fail_over_mac_value
;
4706 * Convert string parameters.
4709 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4710 if (bond_mode
== -1) {
4711 printk(KERN_ERR DRV_NAME
4712 ": Error: Invalid bonding mode \"%s\"\n",
4713 mode
== NULL
? "NULL" : mode
);
4718 if (xmit_hash_policy
) {
4719 if ((bond_mode
!= BOND_MODE_XOR
) &&
4720 (bond_mode
!= BOND_MODE_8023AD
)) {
4721 printk(KERN_INFO DRV_NAME
4722 ": xor_mode param is irrelevant in mode %s\n",
4723 bond_mode_name(bond_mode
));
4725 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4727 if (xmit_hashtype
== -1) {
4728 printk(KERN_ERR DRV_NAME
4729 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4730 xmit_hash_policy
== NULL
? "NULL" :
4738 if (bond_mode
!= BOND_MODE_8023AD
) {
4739 printk(KERN_INFO DRV_NAME
4740 ": lacp_rate param is irrelevant in mode %s\n",
4741 bond_mode_name(bond_mode
));
4743 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4744 if (lacp_fast
== -1) {
4745 printk(KERN_ERR DRV_NAME
4746 ": Error: Invalid lacp rate \"%s\"\n",
4747 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4753 if (max_bonds
< 0 || max_bonds
> INT_MAX
) {
4754 printk(KERN_WARNING DRV_NAME
4755 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4756 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4757 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4758 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4762 printk(KERN_WARNING DRV_NAME
4763 ": Warning: miimon module parameter (%d), "
4764 "not in range 0-%d, so it was reset to %d\n",
4765 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4766 miimon
= BOND_LINK_MON_INTERV
;
4770 printk(KERN_WARNING DRV_NAME
4771 ": Warning: updelay module parameter (%d), "
4772 "not in range 0-%d, so it was reset to 0\n",
4777 if (downdelay
< 0) {
4778 printk(KERN_WARNING DRV_NAME
4779 ": Warning: downdelay module parameter (%d), "
4780 "not in range 0-%d, so it was reset to 0\n",
4781 downdelay
, INT_MAX
);
4785 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4786 printk(KERN_WARNING DRV_NAME
4787 ": Warning: use_carrier module parameter (%d), "
4788 "not of valid value (0/1), so it was set to 1\n",
4793 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
4794 printk(KERN_WARNING DRV_NAME
4795 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4796 "was reset to 1 \n", num_grat_arp
);
4800 /* reset values for 802.3ad */
4801 if (bond_mode
== BOND_MODE_8023AD
) {
4803 printk(KERN_WARNING DRV_NAME
4804 ": Warning: miimon must be specified, "
4805 "otherwise bonding will not detect link "
4806 "failure, speed and duplex which are "
4807 "essential for 802.3ad operation\n");
4808 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4813 /* reset values for TLB/ALB */
4814 if ((bond_mode
== BOND_MODE_TLB
) ||
4815 (bond_mode
== BOND_MODE_ALB
)) {
4817 printk(KERN_WARNING DRV_NAME
4818 ": Warning: miimon must be specified, "
4819 "otherwise bonding will not detect link "
4820 "failure and link speed which are essential "
4821 "for TLB/ALB load balancing\n");
4822 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4827 if (bond_mode
== BOND_MODE_ALB
) {
4828 printk(KERN_NOTICE DRV_NAME
4829 ": In ALB mode you might experience client "
4830 "disconnections upon reconnection of a link if the "
4831 "bonding module updelay parameter (%d msec) is "
4832 "incompatible with the forwarding delay time of the "
4838 if (updelay
|| downdelay
) {
4839 /* just warn the user the up/down delay will have
4840 * no effect since miimon is zero...
4842 printk(KERN_WARNING DRV_NAME
4843 ": Warning: miimon module parameter not set "
4844 "and updelay (%d) or downdelay (%d) module "
4845 "parameter is set; updelay and downdelay have "
4846 "no effect unless miimon is set\n",
4847 updelay
, downdelay
);
4850 /* don't allow arp monitoring */
4852 printk(KERN_WARNING DRV_NAME
4853 ": Warning: miimon (%d) and arp_interval (%d) "
4854 "can't be used simultaneously, disabling ARP "
4856 miimon
, arp_interval
);
4860 if ((updelay
% miimon
) != 0) {
4861 printk(KERN_WARNING DRV_NAME
4862 ": Warning: updelay (%d) is not a multiple "
4863 "of miimon (%d), updelay rounded to %d ms\n",
4864 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4869 if ((downdelay
% miimon
) != 0) {
4870 printk(KERN_WARNING DRV_NAME
4871 ": Warning: downdelay (%d) is not a multiple "
4872 "of miimon (%d), downdelay rounded to %d ms\n",
4874 (downdelay
/ miimon
) * miimon
);
4877 downdelay
/= miimon
;
4880 if (arp_interval
< 0) {
4881 printk(KERN_WARNING DRV_NAME
4882 ": Warning: arp_interval module parameter (%d) "
4883 ", not in range 0-%d, so it was reset to %d\n",
4884 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4885 arp_interval
= BOND_LINK_ARP_INTERV
;
4888 for (arp_ip_count
= 0;
4889 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4891 /* not complete check, but should be good enough to
4893 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4894 printk(KERN_WARNING DRV_NAME
4895 ": Warning: bad arp_ip_target module parameter "
4896 "(%s), ARP monitoring will not be performed\n",
4897 arp_ip_target
[arp_ip_count
]);
4900 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4901 arp_target
[arp_ip_count
] = ip
;
4905 if (arp_interval
&& !arp_ip_count
) {
4906 /* don't allow arping if no arp_ip_target given... */
4907 printk(KERN_WARNING DRV_NAME
4908 ": Warning: arp_interval module parameter (%d) "
4909 "specified without providing an arp_ip_target "
4910 "parameter, arp_interval was reset to 0\n",
4916 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4917 printk(KERN_ERR DRV_NAME
4918 ": arp_validate only supported in active-backup mode\n");
4921 if (!arp_interval
) {
4922 printk(KERN_ERR DRV_NAME
4923 ": arp_validate requires arp_interval\n");
4927 arp_validate_value
= bond_parse_parm(arp_validate
,
4929 if (arp_validate_value
== -1) {
4930 printk(KERN_ERR DRV_NAME
4931 ": Error: invalid arp_validate \"%s\"\n",
4932 arp_validate
== NULL
? "NULL" : arp_validate
);
4936 arp_validate_value
= 0;
4939 printk(KERN_INFO DRV_NAME
4940 ": MII link monitoring set to %d ms\n",
4942 } else if (arp_interval
) {
4945 printk(KERN_INFO DRV_NAME
4946 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4948 arp_validate_tbl
[arp_validate_value
].modename
,
4951 for (i
= 0; i
< arp_ip_count
; i
++)
4952 printk (" %s", arp_ip_target
[i
]);
4956 } else if (max_bonds
) {
4957 /* miimon and arp_interval not set, we need one so things
4958 * work as expected, see bonding.txt for details
4960 printk(KERN_WARNING DRV_NAME
4961 ": Warning: either miimon or arp_interval and "
4962 "arp_ip_target module parameters must be specified, "
4963 "otherwise bonding will not detect link failures! see "
4964 "bonding.txt for details.\n");
4967 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4968 /* currently, using a primary only makes sense
4969 * in active backup, TLB or ALB modes
4971 printk(KERN_WARNING DRV_NAME
4972 ": Warning: %s primary device specified but has no "
4973 "effect in %s mode\n",
4974 primary
, bond_mode_name(bond_mode
));
4978 if (fail_over_mac
) {
4979 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4981 if (fail_over_mac_value
== -1) {
4982 printk(KERN_ERR DRV_NAME
4983 ": Error: invalid fail_over_mac \"%s\"\n",
4984 arp_validate
== NULL
? "NULL" : arp_validate
);
4988 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4989 printk(KERN_WARNING DRV_NAME
4990 ": Warning: fail_over_mac only affects "
4991 "active-backup mode.\n");
4993 fail_over_mac_value
= BOND_FOM_NONE
;
4996 /* fill params struct with the proper values */
4997 params
->mode
= bond_mode
;
4998 params
->xmit_policy
= xmit_hashtype
;
4999 params
->miimon
= miimon
;
5000 params
->num_grat_arp
= num_grat_arp
;
5001 params
->arp_interval
= arp_interval
;
5002 params
->arp_validate
= arp_validate_value
;
5003 params
->updelay
= updelay
;
5004 params
->downdelay
= downdelay
;
5005 params
->use_carrier
= use_carrier
;
5006 params
->lacp_fast
= lacp_fast
;
5007 params
->primary
[0] = 0;
5008 params
->fail_over_mac
= fail_over_mac_value
;
5011 strncpy(params
->primary
, primary
, IFNAMSIZ
);
5012 params
->primary
[IFNAMSIZ
- 1] = 0;
5015 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
5020 static struct lock_class_key bonding_netdev_xmit_lock_key
;
5022 static void bond_set_lockdep_class_one(struct netdev_queue
*txq
)
5024 lockdep_set_class(&txq
->_xmit_lock
,
5025 &bonding_netdev_xmit_lock_key
);
5028 static void bond_set_lockdep_class(struct net_device
*dev
)
5030 bond_set_lockdep_class_one(&dev
->tx_queue
);
5033 /* Create a new bond based on the specified name and bonding parameters.
5034 * If name is NULL, obtain a suitable "bond%d" name for us.
5035 * Caller must NOT hold rtnl_lock; we need to release it here before we
5036 * set up our sysfs entries.
5038 int bond_create(char *name
, struct bond_params
*params
)
5040 struct net_device
*bond_dev
;
5041 struct bonding
*bond
;
5045 down_write(&bonding_rwsem
);
5047 /* Check to see if the bond already exists. */
5049 list_for_each_entry(bond
, &bond_dev_list
, bond_list
)
5050 if (strnicmp(bond
->dev
->name
, name
, IFNAMSIZ
) == 0) {
5051 printk(KERN_ERR DRV_NAME
5052 ": cannot add bond %s; it already exists\n",
5059 bond_dev
= alloc_netdev(sizeof(struct bonding
), name
? name
: "",
5062 printk(KERN_ERR DRV_NAME
5063 ": %s: eek! can't alloc netdev!\n",
5070 res
= dev_alloc_name(bond_dev
, "bond%d");
5075 /* bond_init() must be called after dev_alloc_name() (for the
5076 * /proc files), but before register_netdevice(), because we
5077 * need to set function pointers.
5080 res
= bond_init(bond_dev
, params
);
5085 res
= register_netdevice(bond_dev
);
5090 bond_set_lockdep_class(bond_dev
);
5092 netif_carrier_off(bond_dev
);
5094 up_write(&bonding_rwsem
);
5095 rtnl_unlock(); /* allows sysfs registration of net device */
5096 res
= bond_create_sysfs_entry(bond_dev
->priv
);
5099 down_write(&bonding_rwsem
);
5100 bond_deinit(bond_dev
);
5101 unregister_netdevice(bond_dev
);
5108 bond_deinit(bond_dev
);
5110 free_netdev(bond_dev
);
5112 up_write(&bonding_rwsem
);
5117 static int __init
bonding_init(void)
5121 struct bonding
*bond
;
5123 printk(KERN_INFO
"%s", version
);
5125 res
= bond_check_params(&bonding_defaults
);
5130 #ifdef CONFIG_PROC_FS
5131 bond_create_proc_dir();
5134 init_rwsem(&bonding_rwsem
);
5136 for (i
= 0; i
< max_bonds
; i
++) {
5137 res
= bond_create(NULL
, &bonding_defaults
);
5142 res
= bond_create_sysfs();
5146 register_netdevice_notifier(&bond_netdev_notifier
);
5147 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5151 list_for_each_entry(bond
, &bond_dev_list
, bond_list
) {
5152 bond_work_cancel_all(bond
);
5153 destroy_workqueue(bond
->wq
);
5156 bond_destroy_sysfs();
5166 static void __exit
bonding_exit(void)
5168 unregister_netdevice_notifier(&bond_netdev_notifier
);
5169 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5171 bond_destroy_sysfs();
5178 module_init(bonding_init
);
5179 module_exit(bonding_exit
);
5180 MODULE_LICENSE("GPL");
5181 MODULE_VERSION(DRV_VERSION
);
5182 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5183 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5184 MODULE_SUPPORTED_DEVICE("most ethernet devices");