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 int bond_set_promiscuity(struct bonding
*bond
, int inc
)
778 if (USES_PRIMARY(bond
->params
.mode
)) {
779 /* write lock already acquired */
780 if (bond
->curr_active_slave
) {
781 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
787 bond_for_each_slave(bond
, slave
, i
) {
788 err
= dev_set_promiscuity(slave
->dev
, inc
);
797 * Push the allmulti flag down to all slaves
799 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
802 if (USES_PRIMARY(bond
->params
.mode
)) {
803 /* write lock already acquired */
804 if (bond
->curr_active_slave
) {
805 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
811 bond_for_each_slave(bond
, slave
, i
) {
812 err
= dev_set_allmulti(slave
->dev
, inc
);
821 * Add a Multicast address to slaves
824 static void bond_mc_add(struct bonding
*bond
, void *addr
, int alen
)
826 if (USES_PRIMARY(bond
->params
.mode
)) {
827 /* write lock already acquired */
828 if (bond
->curr_active_slave
) {
829 dev_mc_add(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
834 bond_for_each_slave(bond
, slave
, i
) {
835 dev_mc_add(slave
->dev
, addr
, alen
, 0);
841 * Remove a multicast address from slave
844 static void bond_mc_delete(struct bonding
*bond
, void *addr
, int alen
)
846 if (USES_PRIMARY(bond
->params
.mode
)) {
847 /* write lock already acquired */
848 if (bond
->curr_active_slave
) {
849 dev_mc_delete(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
854 bond_for_each_slave(bond
, slave
, i
) {
855 dev_mc_delete(slave
->dev
, addr
, alen
, 0);
862 * Retrieve the list of registered multicast addresses for the bonding
863 * device and retransmit an IGMP JOIN request to the current active
866 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
868 struct in_device
*in_dev
;
869 struct ip_mc_list
*im
;
872 in_dev
= __in_dev_get_rcu(bond
->dev
);
874 for (im
= in_dev
->mc_list
; im
; im
= im
->next
) {
875 ip_mc_rejoin_group(im
);
883 * Totally destroys the mc_list in bond
885 static void bond_mc_list_destroy(struct bonding
*bond
)
887 struct dev_mc_list
*dmi
;
891 bond
->mc_list
= dmi
->next
;
895 bond
->mc_list
= NULL
;
899 * Copy all the Multicast addresses from src to the bonding device dst
901 static int bond_mc_list_copy(struct dev_mc_list
*mc_list
, struct bonding
*bond
,
904 struct dev_mc_list
*dmi
, *new_dmi
;
906 for (dmi
= mc_list
; dmi
; dmi
= dmi
->next
) {
907 new_dmi
= kmalloc(sizeof(struct dev_mc_list
), gfp_flag
);
910 /* FIXME: Potential memory leak !!! */
914 new_dmi
->next
= bond
->mc_list
;
915 bond
->mc_list
= new_dmi
;
916 new_dmi
->dmi_addrlen
= dmi
->dmi_addrlen
;
917 memcpy(new_dmi
->dmi_addr
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
918 new_dmi
->dmi_users
= dmi
->dmi_users
;
919 new_dmi
->dmi_gusers
= dmi
->dmi_gusers
;
926 * flush all members of flush->mc_list from device dev->mc_list
928 static void bond_mc_list_flush(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
930 struct bonding
*bond
= bond_dev
->priv
;
931 struct dev_mc_list
*dmi
;
933 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
934 dev_mc_delete(slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
937 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
938 /* del lacpdu mc addr from mc list */
939 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
941 dev_mc_delete(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
945 /*--------------------------- Active slave change ---------------------------*/
948 * Update the mc list and multicast-related flags for the new and
949 * old active slaves (if any) according to the multicast mode, and
950 * promiscuous flags unconditionally.
952 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
, struct slave
*old_active
)
954 struct dev_mc_list
*dmi
;
956 if (!USES_PRIMARY(bond
->params
.mode
)) {
957 /* nothing to do - mc list is already up-to-date on
964 if (bond
->dev
->flags
& IFF_PROMISC
) {
965 dev_set_promiscuity(old_active
->dev
, -1);
968 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
969 dev_set_allmulti(old_active
->dev
, -1);
972 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
973 dev_mc_delete(old_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
978 /* FIXME: Signal errors upstream. */
979 if (bond
->dev
->flags
& IFF_PROMISC
) {
980 dev_set_promiscuity(new_active
->dev
, 1);
983 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
984 dev_set_allmulti(new_active
->dev
, 1);
987 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
988 dev_mc_add(new_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
990 bond_resend_igmp_join_requests(bond
);
995 * bond_do_fail_over_mac
997 * Perform special MAC address swapping for fail_over_mac settings
999 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1001 static void bond_do_fail_over_mac(struct bonding
*bond
,
1002 struct slave
*new_active
,
1003 struct slave
*old_active
)
1005 u8 tmp_mac
[ETH_ALEN
];
1006 struct sockaddr saddr
;
1009 switch (bond
->params
.fail_over_mac
) {
1010 case BOND_FOM_ACTIVE
:
1012 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
1013 new_active
->dev
->addr_len
);
1015 case BOND_FOM_FOLLOW
:
1017 * if new_active && old_active, swap them
1018 * if just old_active, do nothing (going to no active slave)
1019 * if just new_active, set new_active to bond's MAC
1024 write_unlock_bh(&bond
->curr_slave_lock
);
1025 read_unlock(&bond
->lock
);
1028 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
1029 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
1031 saddr
.sa_family
= new_active
->dev
->type
;
1033 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
1034 saddr
.sa_family
= bond
->dev
->type
;
1037 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
1039 printk(KERN_ERR DRV_NAME
1040 ": %s: Error %d setting MAC of slave %s\n",
1041 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1048 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1049 saddr
.sa_family
= old_active
->dev
->type
;
1051 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
1053 printk(KERN_ERR DRV_NAME
1054 ": %s: Error %d setting MAC of slave %s\n",
1055 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1057 read_lock(&bond
->lock
);
1058 write_lock_bh(&bond
->curr_slave_lock
);
1061 printk(KERN_ERR DRV_NAME
1062 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1063 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1071 * find_best_interface - select the best available slave to be the active one
1072 * @bond: our bonding struct
1074 * Warning: Caller must hold curr_slave_lock for writing.
1076 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1078 struct slave
*new_active
, *old_active
;
1079 struct slave
*bestslave
= NULL
;
1080 int mintime
= bond
->params
.updelay
;
1083 new_active
= old_active
= bond
->curr_active_slave
;
1085 if (!new_active
) { /* there were no active slaves left */
1086 if (bond
->slave_cnt
> 0) { /* found one slave */
1087 new_active
= bond
->first_slave
;
1089 return NULL
; /* still no slave, return NULL */
1093 /* first try the primary link; if arping, a link must tx/rx traffic
1094 * before it can be considered the curr_active_slave - also, we would skip
1095 * slaves between the curr_active_slave and primary_slave that may be up
1098 if ((bond
->primary_slave
) &&
1099 (!bond
->params
.arp_interval
) &&
1100 (IS_UP(bond
->primary_slave
->dev
))) {
1101 new_active
= bond
->primary_slave
;
1104 /* remember where to stop iterating over the slaves */
1105 old_active
= new_active
;
1107 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1108 if (IS_UP(new_active
->dev
)) {
1109 if (new_active
->link
== BOND_LINK_UP
) {
1111 } else if (new_active
->link
== BOND_LINK_BACK
) {
1112 /* link up, but waiting for stabilization */
1113 if (new_active
->delay
< mintime
) {
1114 mintime
= new_active
->delay
;
1115 bestslave
= new_active
;
1125 * change_active_interface - change the active slave into the specified one
1126 * @bond: our bonding struct
1127 * @new: the new slave to make the active one
1129 * Set the new slave to the bond's settings and unset them on the old
1130 * curr_active_slave.
1131 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1133 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1134 * because it is apparently the best available slave we have, even though its
1135 * updelay hasn't timed out yet.
1137 * If new_active is not NULL, caller must hold bond->lock for read and
1138 * curr_slave_lock for write_bh.
1140 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1142 struct slave
*old_active
= bond
->curr_active_slave
;
1144 if (old_active
== new_active
) {
1149 new_active
->jiffies
= jiffies
;
1151 if (new_active
->link
== BOND_LINK_BACK
) {
1152 if (USES_PRIMARY(bond
->params
.mode
)) {
1153 printk(KERN_INFO DRV_NAME
1154 ": %s: making interface %s the new "
1155 "active one %d ms earlier.\n",
1156 bond
->dev
->name
, new_active
->dev
->name
,
1157 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1160 new_active
->delay
= 0;
1161 new_active
->link
= BOND_LINK_UP
;
1163 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1164 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1167 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1168 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1169 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1172 if (USES_PRIMARY(bond
->params
.mode
)) {
1173 printk(KERN_INFO DRV_NAME
1174 ": %s: making interface %s the new "
1176 bond
->dev
->name
, new_active
->dev
->name
);
1181 if (USES_PRIMARY(bond
->params
.mode
)) {
1182 bond_mc_swap(bond
, new_active
, old_active
);
1185 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1186 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1187 bond_alb_handle_active_change(bond
, new_active
);
1189 bond_set_slave_inactive_flags(old_active
);
1191 bond_set_slave_active_flags(new_active
);
1193 bond
->curr_active_slave
= new_active
;
1196 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1198 bond_set_slave_inactive_flags(old_active
);
1202 bond_set_slave_active_flags(new_active
);
1204 if (bond
->params
.fail_over_mac
)
1205 bond_do_fail_over_mac(bond
, new_active
,
1208 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1209 bond_send_gratuitous_arp(bond
);
1211 write_unlock_bh(&bond
->curr_slave_lock
);
1212 read_unlock(&bond
->lock
);
1214 netdev_bonding_change(bond
->dev
);
1216 read_lock(&bond
->lock
);
1217 write_lock_bh(&bond
->curr_slave_lock
);
1223 * bond_select_active_slave - select a new active slave, if needed
1224 * @bond: our bonding struct
1226 * This functions shoud be called when one of the following occurs:
1227 * - The old curr_active_slave has been released or lost its link.
1228 * - The primary_slave has got its link back.
1229 * - A slave has got its link back and there's no old curr_active_slave.
1231 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1233 void bond_select_active_slave(struct bonding
*bond
)
1235 struct slave
*best_slave
;
1238 best_slave
= bond_find_best_slave(bond
);
1239 if (best_slave
!= bond
->curr_active_slave
) {
1240 bond_change_active_slave(bond
, best_slave
);
1241 rv
= bond_set_carrier(bond
);
1245 if (netif_carrier_ok(bond
->dev
)) {
1246 printk(KERN_INFO DRV_NAME
1247 ": %s: first active interface up!\n",
1250 printk(KERN_INFO DRV_NAME
": %s: "
1251 "now running without any active interface !\n",
1257 /*--------------------------- slave list handling ---------------------------*/
1260 * This function attaches the slave to the end of list.
1262 * bond->lock held for writing by caller.
1264 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1266 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1267 new_slave
->next
= new_slave
;
1268 new_slave
->prev
= new_slave
;
1269 bond
->first_slave
= new_slave
;
1271 new_slave
->next
= bond
->first_slave
;
1272 new_slave
->prev
= bond
->first_slave
->prev
;
1273 new_slave
->next
->prev
= new_slave
;
1274 new_slave
->prev
->next
= new_slave
;
1281 * This function detaches the slave from the list.
1282 * WARNING: no check is made to verify if the slave effectively
1283 * belongs to <bond>.
1284 * Nothing is freed on return, structures are just unchained.
1285 * If any slave pointer in bond was pointing to <slave>,
1286 * it should be changed by the calling function.
1288 * bond->lock held for writing by caller.
1290 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1293 slave
->next
->prev
= slave
->prev
;
1297 slave
->prev
->next
= slave
->next
;
1300 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1301 if (bond
->slave_cnt
> 1) { /* there are more slave */
1302 bond
->first_slave
= slave
->next
;
1304 bond
->first_slave
= NULL
; /* slave was the last one */
1313 /*---------------------------------- IOCTL ----------------------------------*/
1315 static int bond_sethwaddr(struct net_device
*bond_dev
,
1316 struct net_device
*slave_dev
)
1318 dprintk("bond_dev=%p\n", bond_dev
);
1319 dprintk("slave_dev=%p\n", slave_dev
);
1320 dprintk("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1321 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1325 #define BOND_VLAN_FEATURES \
1326 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1327 NETIF_F_HW_VLAN_FILTER)
1330 * Compute the common dev->feature set available to all slaves. Some
1331 * feature bits are managed elsewhere, so preserve those feature bits
1332 * on the master device.
1334 static int bond_compute_features(struct bonding
*bond
)
1336 struct slave
*slave
;
1337 struct net_device
*bond_dev
= bond
->dev
;
1338 unsigned long features
= bond_dev
->features
;
1339 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1340 bond_dev
->hard_header_len
);
1343 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1344 features
|= NETIF_F_SG
| NETIF_F_FRAGLIST
| NETIF_F_HIGHDMA
|
1345 NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1347 bond_for_each_slave(bond
, slave
, i
) {
1348 features
= netdev_compute_features(features
,
1349 slave
->dev
->features
);
1350 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1351 max_hard_header_len
= slave
->dev
->hard_header_len
;
1354 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1355 bond_dev
->features
= features
;
1356 bond_dev
->hard_header_len
= max_hard_header_len
;
1362 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1363 struct net_device
*slave_dev
)
1365 struct bonding
*bond
= bond_dev
->priv
;
1367 bond_dev
->neigh_setup
= slave_dev
->neigh_setup
;
1368 bond_dev
->header_ops
= slave_dev
->header_ops
;
1370 bond_dev
->type
= slave_dev
->type
;
1371 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1372 bond_dev
->addr_len
= slave_dev
->addr_len
;
1374 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1375 slave_dev
->addr_len
);
1376 bond
->setup_by_slave
= 1;
1379 /* enslave device <slave> to bond device <master> */
1380 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1382 struct bonding
*bond
= bond_dev
->priv
;
1383 struct slave
*new_slave
= NULL
;
1384 struct dev_mc_list
*dmi
;
1385 struct sockaddr addr
;
1387 int old_features
= bond_dev
->features
;
1390 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1391 slave_dev
->do_ioctl
== NULL
) {
1392 printk(KERN_WARNING DRV_NAME
1393 ": %s: Warning: no link monitoring support for %s\n",
1394 bond_dev
->name
, slave_dev
->name
);
1397 /* bond must be initialized by bond_open() before enslaving */
1398 if (!(bond_dev
->flags
& IFF_UP
)) {
1399 printk(KERN_WARNING DRV_NAME
1400 " %s: master_dev is not up in bond_enslave\n",
1404 /* already enslaved */
1405 if (slave_dev
->flags
& IFF_SLAVE
) {
1406 dprintk("Error, Device was already enslaved\n");
1410 /* vlan challenged mutual exclusion */
1411 /* no need to lock since we're protected by rtnl_lock */
1412 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1413 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1414 if (!list_empty(&bond
->vlan_list
)) {
1415 printk(KERN_ERR DRV_NAME
1416 ": %s: Error: cannot enslave VLAN "
1417 "challenged slave %s on VLAN enabled "
1418 "bond %s\n", bond_dev
->name
, slave_dev
->name
,
1422 printk(KERN_WARNING DRV_NAME
1423 ": %s: Warning: enslaved VLAN challenged "
1424 "slave %s. Adding VLANs will be blocked as "
1425 "long as %s is part of bond %s\n",
1426 bond_dev
->name
, slave_dev
->name
, slave_dev
->name
,
1428 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1431 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1432 if (bond
->slave_cnt
== 0) {
1433 /* First slave, and it is not VLAN challenged,
1434 * so remove the block of adding VLANs over the bond.
1436 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1441 * Old ifenslave binaries are no longer supported. These can
1442 * be identified with moderate accurary by the state of the slave:
1443 * the current ifenslave will set the interface down prior to
1444 * enslaving it; the old ifenslave will not.
1446 if ((slave_dev
->flags
& IFF_UP
)) {
1447 printk(KERN_ERR DRV_NAME
": %s is up. "
1448 "This may be due to an out of date ifenslave.\n",
1451 goto err_undo_flags
;
1454 /* set bonding device ether type by slave - bonding netdevices are
1455 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1456 * there is a need to override some of the type dependent attribs/funcs.
1458 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1459 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1461 if (bond
->slave_cnt
== 0) {
1462 if (slave_dev
->type
!= ARPHRD_ETHER
)
1463 bond_setup_by_slave(bond_dev
, slave_dev
);
1464 } else if (bond_dev
->type
!= slave_dev
->type
) {
1465 printk(KERN_ERR DRV_NAME
": %s ether type (%d) is different "
1466 "from other slaves (%d), can not enslave it.\n",
1468 slave_dev
->type
, bond_dev
->type
);
1470 goto err_undo_flags
;
1473 if (slave_dev
->set_mac_address
== NULL
) {
1474 if (bond
->slave_cnt
== 0) {
1475 printk(KERN_WARNING DRV_NAME
1476 ": %s: Warning: The first slave device "
1477 "specified does not support setting the MAC "
1478 "address. Setting fail_over_mac to active.",
1480 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1481 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1482 printk(KERN_ERR DRV_NAME
1483 ": %s: Error: The slave device specified "
1484 "does not support setting the MAC address, "
1485 "but fail_over_mac is not set to active.\n"
1488 goto err_undo_flags
;
1492 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1495 goto err_undo_flags
;
1498 /* save slave's original flags before calling
1499 * netdev_set_master and dev_open
1501 new_slave
->original_flags
= slave_dev
->flags
;
1504 * Save slave's original ("permanent") mac address for modes
1505 * that need it, and for restoring it upon release, and then
1506 * set it to the master's address
1508 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1510 if (!bond
->params
.fail_over_mac
) {
1512 * Set slave to master's mac address. The application already
1513 * set the master's mac address to that of the first slave
1515 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1516 addr
.sa_family
= slave_dev
->type
;
1517 res
= dev_set_mac_address(slave_dev
, &addr
);
1519 dprintk("Error %d calling set_mac_address\n", res
);
1524 res
= netdev_set_master(slave_dev
, bond_dev
);
1526 dprintk("Error %d calling netdev_set_master\n", res
);
1527 goto err_restore_mac
;
1529 /* open the slave since the application closed it */
1530 res
= dev_open(slave_dev
);
1532 dprintk("Openning slave %s failed\n", slave_dev
->name
);
1533 goto err_unset_master
;
1536 new_slave
->dev
= slave_dev
;
1537 slave_dev
->priv_flags
|= IFF_BONDING
;
1539 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1540 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1541 /* bond_alb_init_slave() must be called before all other stages since
1542 * it might fail and we do not want to have to undo everything
1544 res
= bond_alb_init_slave(bond
, new_slave
);
1550 /* If the mode USES_PRIMARY, then the new slave gets the
1551 * master's promisc (and mc) settings only if it becomes the
1552 * curr_active_slave, and that is taken care of later when calling
1553 * bond_change_active()
1555 if (!USES_PRIMARY(bond
->params
.mode
)) {
1556 /* set promiscuity level to new slave */
1557 if (bond_dev
->flags
& IFF_PROMISC
) {
1558 res
= dev_set_promiscuity(slave_dev
, 1);
1563 /* set allmulti level to new slave */
1564 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1565 res
= dev_set_allmulti(slave_dev
, 1);
1570 netif_addr_lock_bh(bond_dev
);
1571 /* upload master's mc_list to new slave */
1572 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
1573 dev_mc_add (slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
1575 netif_addr_unlock_bh(bond_dev
);
1578 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1579 /* add lacpdu mc addr to mc list */
1580 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1582 dev_mc_add(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
1585 bond_add_vlans_on_slave(bond
, slave_dev
);
1587 write_lock_bh(&bond
->lock
);
1589 bond_attach_slave(bond
, new_slave
);
1591 new_slave
->delay
= 0;
1592 new_slave
->link_failure_count
= 0;
1594 bond_compute_features(bond
);
1596 write_unlock_bh(&bond
->lock
);
1598 read_lock(&bond
->lock
);
1600 new_slave
->last_arp_rx
= jiffies
;
1602 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1603 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1605 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1607 * miimon is set but a bonded network driver
1608 * does not support ETHTOOL/MII and
1609 * arp_interval is not set. Note: if
1610 * use_carrier is enabled, we will never go
1611 * here (because netif_carrier is always
1612 * supported); thus, we don't need to change
1613 * the messages for netif_carrier.
1615 printk(KERN_WARNING DRV_NAME
1616 ": %s: Warning: MII and ETHTOOL support not "
1617 "available for interface %s, and "
1618 "arp_interval/arp_ip_target module parameters "
1619 "not specified, thus bonding will not detect "
1620 "link failures! see bonding.txt for details.\n",
1621 bond_dev
->name
, slave_dev
->name
);
1622 } else if (link_reporting
== -1) {
1623 /* unable get link status using mii/ethtool */
1624 printk(KERN_WARNING DRV_NAME
1625 ": %s: Warning: can't get link status from "
1626 "interface %s; the network driver associated "
1627 "with this interface does not support MII or "
1628 "ETHTOOL link status reporting, thus miimon "
1629 "has no effect on this interface.\n",
1630 bond_dev
->name
, slave_dev
->name
);
1634 /* check for initial state */
1635 if (!bond
->params
.miimon
||
1636 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1637 if (bond
->params
.updelay
) {
1638 dprintk("Initial state of slave_dev is "
1639 "BOND_LINK_BACK\n");
1640 new_slave
->link
= BOND_LINK_BACK
;
1641 new_slave
->delay
= bond
->params
.updelay
;
1643 dprintk("Initial state of slave_dev is "
1645 new_slave
->link
= BOND_LINK_UP
;
1647 new_slave
->jiffies
= jiffies
;
1649 dprintk("Initial state of slave_dev is "
1650 "BOND_LINK_DOWN\n");
1651 new_slave
->link
= BOND_LINK_DOWN
;
1654 if (bond_update_speed_duplex(new_slave
) &&
1655 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1656 printk(KERN_WARNING DRV_NAME
1657 ": %s: Warning: failed to get speed and duplex from %s, "
1658 "assumed to be 100Mb/sec and Full.\n",
1659 bond_dev
->name
, new_slave
->dev
->name
);
1661 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1662 printk(KERN_WARNING DRV_NAME
1663 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1664 "support in base driver for proper aggregator "
1665 "selection.\n", bond_dev
->name
);
1669 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1670 /* if there is a primary slave, remember it */
1671 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1672 bond
->primary_slave
= new_slave
;
1676 write_lock_bh(&bond
->curr_slave_lock
);
1678 switch (bond
->params
.mode
) {
1679 case BOND_MODE_ACTIVEBACKUP
:
1680 bond_set_slave_inactive_flags(new_slave
);
1681 bond_select_active_slave(bond
);
1683 case BOND_MODE_8023AD
:
1684 /* in 802.3ad mode, the internal mechanism
1685 * will activate the slaves in the selected
1688 bond_set_slave_inactive_flags(new_slave
);
1689 /* if this is the first slave */
1690 if (bond
->slave_cnt
== 1) {
1691 SLAVE_AD_INFO(new_slave
).id
= 1;
1692 /* Initialize AD with the number of times that the AD timer is called in 1 second
1693 * can be called only after the mac address of the bond is set
1695 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1696 bond
->params
.lacp_fast
);
1698 SLAVE_AD_INFO(new_slave
).id
=
1699 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1702 bond_3ad_bind_slave(new_slave
);
1706 new_slave
->state
= BOND_STATE_ACTIVE
;
1707 bond_set_slave_inactive_flags(new_slave
);
1710 dprintk("This slave is always active in trunk mode\n");
1712 /* always active in trunk mode */
1713 new_slave
->state
= BOND_STATE_ACTIVE
;
1715 /* In trunking mode there is little meaning to curr_active_slave
1716 * anyway (it holds no special properties of the bond device),
1717 * so we can change it without calling change_active_interface()
1719 if (!bond
->curr_active_slave
) {
1720 bond
->curr_active_slave
= new_slave
;
1723 } /* switch(bond_mode) */
1725 write_unlock_bh(&bond
->curr_slave_lock
);
1727 bond_set_carrier(bond
);
1729 read_unlock(&bond
->lock
);
1731 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1735 printk(KERN_INFO DRV_NAME
1736 ": %s: enslaving %s as a%s interface with a%s link.\n",
1737 bond_dev
->name
, slave_dev
->name
,
1738 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1739 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1741 /* enslave is successful */
1744 /* Undo stages on error */
1746 dev_close(slave_dev
);
1749 netdev_set_master(slave_dev
, NULL
);
1752 if (!bond
->params
.fail_over_mac
) {
1753 /* XXX TODO - fom follow mode needs to change master's
1754 * MAC if this slave's MAC is in use by the bond, or at
1755 * least print a warning.
1757 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1758 addr
.sa_family
= slave_dev
->type
;
1759 dev_set_mac_address(slave_dev
, &addr
);
1766 bond_dev
->features
= old_features
;
1772 * Try to release the slave device <slave> from the bond device <master>
1773 * It is legal to access curr_active_slave without a lock because all the function
1776 * The rules for slave state should be:
1777 * for Active/Backup:
1778 * Active stays on all backups go down
1779 * for Bonded connections:
1780 * The first up interface should be left on and all others downed.
1782 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1784 struct bonding
*bond
= bond_dev
->priv
;
1785 struct slave
*slave
, *oldcurrent
;
1786 struct sockaddr addr
;
1787 int mac_addr_differ
;
1788 DECLARE_MAC_BUF(mac
);
1790 /* slave is not a slave or master is not master of this slave */
1791 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1792 (slave_dev
->master
!= bond_dev
)) {
1793 printk(KERN_ERR DRV_NAME
1794 ": %s: Error: cannot release %s.\n",
1795 bond_dev
->name
, slave_dev
->name
);
1799 write_lock_bh(&bond
->lock
);
1801 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1803 /* not a slave of this bond */
1804 printk(KERN_INFO DRV_NAME
1805 ": %s: %s not enslaved\n",
1806 bond_dev
->name
, slave_dev
->name
);
1807 write_unlock_bh(&bond
->lock
);
1811 if (!bond
->params
.fail_over_mac
) {
1812 mac_addr_differ
= memcmp(bond_dev
->dev_addr
, slave
->perm_hwaddr
,
1814 if (!mac_addr_differ
&& (bond
->slave_cnt
> 1))
1815 printk(KERN_WARNING DRV_NAME
1816 ": %s: Warning: the permanent HWaddr of %s - "
1817 "%s - is still in use by %s. "
1818 "Set the HWaddr of %s to a different address "
1819 "to avoid conflicts.\n",
1820 bond_dev
->name
, slave_dev
->name
,
1821 print_mac(mac
, slave
->perm_hwaddr
),
1822 bond_dev
->name
, slave_dev
->name
);
1825 /* Inform AD package of unbinding of slave. */
1826 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1827 /* must be called before the slave is
1828 * detached from the list
1830 bond_3ad_unbind_slave(slave
);
1833 printk(KERN_INFO DRV_NAME
1834 ": %s: releasing %s interface %s\n",
1836 (slave
->state
== BOND_STATE_ACTIVE
)
1837 ? "active" : "backup",
1840 oldcurrent
= bond
->curr_active_slave
;
1842 bond
->current_arp_slave
= NULL
;
1844 /* release the slave from its bond */
1845 bond_detach_slave(bond
, slave
);
1847 bond_compute_features(bond
);
1849 if (bond
->primary_slave
== slave
) {
1850 bond
->primary_slave
= NULL
;
1853 if (oldcurrent
== slave
) {
1854 bond_change_active_slave(bond
, NULL
);
1857 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1858 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1859 /* Must be called only after the slave has been
1860 * detached from the list and the curr_active_slave
1861 * has been cleared (if our_slave == old_current),
1862 * but before a new active slave is selected.
1864 write_unlock_bh(&bond
->lock
);
1865 bond_alb_deinit_slave(bond
, slave
);
1866 write_lock_bh(&bond
->lock
);
1869 if (oldcurrent
== slave
) {
1871 * Note that we hold RTNL over this sequence, so there
1872 * is no concern that another slave add/remove event
1875 write_unlock_bh(&bond
->lock
);
1876 read_lock(&bond
->lock
);
1877 write_lock_bh(&bond
->curr_slave_lock
);
1879 bond_select_active_slave(bond
);
1881 write_unlock_bh(&bond
->curr_slave_lock
);
1882 read_unlock(&bond
->lock
);
1883 write_lock_bh(&bond
->lock
);
1886 if (bond
->slave_cnt
== 0) {
1887 bond_set_carrier(bond
);
1889 /* if the last slave was removed, zero the mac address
1890 * of the master so it will be set by the application
1891 * to the mac address of the first slave
1893 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1895 if (list_empty(&bond
->vlan_list
)) {
1896 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1898 printk(KERN_WARNING DRV_NAME
1899 ": %s: Warning: clearing HW address of %s while it "
1900 "still has VLANs.\n",
1901 bond_dev
->name
, bond_dev
->name
);
1902 printk(KERN_WARNING DRV_NAME
1903 ": %s: When re-adding slaves, make sure the bond's "
1904 "HW address matches its VLANs'.\n",
1907 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1908 !bond_has_challenged_slaves(bond
)) {
1909 printk(KERN_INFO DRV_NAME
1910 ": %s: last VLAN challenged slave %s "
1911 "left bond %s. VLAN blocking is removed\n",
1912 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1913 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1916 write_unlock_bh(&bond
->lock
);
1918 /* must do this from outside any spinlocks */
1919 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1921 bond_del_vlans_from_slave(bond
, slave_dev
);
1923 /* If the mode USES_PRIMARY, then we should only remove its
1924 * promisc and mc settings if it was the curr_active_slave, but that was
1925 * already taken care of above when we detached the slave
1927 if (!USES_PRIMARY(bond
->params
.mode
)) {
1928 /* unset promiscuity level from slave */
1929 if (bond_dev
->flags
& IFF_PROMISC
) {
1930 dev_set_promiscuity(slave_dev
, -1);
1933 /* unset allmulti level from slave */
1934 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1935 dev_set_allmulti(slave_dev
, -1);
1938 /* flush master's mc_list from slave */
1939 netif_addr_lock_bh(bond_dev
);
1940 bond_mc_list_flush(bond_dev
, slave_dev
);
1941 netif_addr_unlock_bh(bond_dev
);
1944 netdev_set_master(slave_dev
, NULL
);
1946 /* close slave before restoring its mac address */
1947 dev_close(slave_dev
);
1949 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1950 /* restore original ("permanent") mac address */
1951 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1952 addr
.sa_family
= slave_dev
->type
;
1953 dev_set_mac_address(slave_dev
, &addr
);
1956 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
1957 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
1962 return 0; /* deletion OK */
1966 * Destroy a bonding device.
1967 * Must be under rtnl_lock when this function is called.
1969 void bond_destroy(struct bonding
*bond
)
1971 bond_deinit(bond
->dev
);
1972 bond_destroy_sysfs_entry(bond
);
1973 unregister_netdevice(bond
->dev
);
1977 * First release a slave and than destroy the bond if no more slaves iare left.
1978 * Must be under rtnl_lock when this function is called.
1980 int bond_release_and_destroy(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1982 struct bonding
*bond
= bond_dev
->priv
;
1985 ret
= bond_release(bond_dev
, slave_dev
);
1986 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
1987 printk(KERN_INFO DRV_NAME
": %s: destroying bond %s.\n",
1988 bond_dev
->name
, bond_dev
->name
);
1995 * This function releases all slaves.
1997 static int bond_release_all(struct net_device
*bond_dev
)
1999 struct bonding
*bond
= bond_dev
->priv
;
2000 struct slave
*slave
;
2001 struct net_device
*slave_dev
;
2002 struct sockaddr addr
;
2004 write_lock_bh(&bond
->lock
);
2006 netif_carrier_off(bond_dev
);
2008 if (bond
->slave_cnt
== 0) {
2012 bond
->current_arp_slave
= NULL
;
2013 bond
->primary_slave
= NULL
;
2014 bond_change_active_slave(bond
, NULL
);
2016 while ((slave
= bond
->first_slave
) != NULL
) {
2017 /* Inform AD package of unbinding of slave
2018 * before slave is detached from the list.
2020 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2021 bond_3ad_unbind_slave(slave
);
2024 slave_dev
= slave
->dev
;
2025 bond_detach_slave(bond
, slave
);
2027 /* now that the slave is detached, unlock and perform
2028 * all the undo steps that should not be called from
2031 write_unlock_bh(&bond
->lock
);
2033 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
2034 (bond
->params
.mode
== BOND_MODE_ALB
)) {
2035 /* must be called only after the slave
2036 * has been detached from the list
2038 bond_alb_deinit_slave(bond
, slave
);
2041 bond_compute_features(bond
);
2043 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2044 bond_del_vlans_from_slave(bond
, slave_dev
);
2046 /* If the mode USES_PRIMARY, then we should only remove its
2047 * promisc and mc settings if it was the curr_active_slave, but that was
2048 * already taken care of above when we detached the slave
2050 if (!USES_PRIMARY(bond
->params
.mode
)) {
2051 /* unset promiscuity level from slave */
2052 if (bond_dev
->flags
& IFF_PROMISC
) {
2053 dev_set_promiscuity(slave_dev
, -1);
2056 /* unset allmulti level from slave */
2057 if (bond_dev
->flags
& IFF_ALLMULTI
) {
2058 dev_set_allmulti(slave_dev
, -1);
2061 /* flush master's mc_list from slave */
2062 netif_addr_lock_bh(bond_dev
);
2063 bond_mc_list_flush(bond_dev
, slave_dev
);
2064 netif_addr_unlock_bh(bond_dev
);
2067 netdev_set_master(slave_dev
, NULL
);
2069 /* close slave before restoring its mac address */
2070 dev_close(slave_dev
);
2072 if (!bond
->params
.fail_over_mac
) {
2073 /* restore original ("permanent") mac address*/
2074 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2075 addr
.sa_family
= slave_dev
->type
;
2076 dev_set_mac_address(slave_dev
, &addr
);
2079 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2080 IFF_SLAVE_INACTIVE
);
2084 /* re-acquire the lock before getting the next slave */
2085 write_lock_bh(&bond
->lock
);
2088 /* zero the mac address of the master so it will be
2089 * set by the application to the mac address of the
2092 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2094 if (list_empty(&bond
->vlan_list
)) {
2095 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2097 printk(KERN_WARNING DRV_NAME
2098 ": %s: Warning: clearing HW address of %s while it "
2099 "still has VLANs.\n",
2100 bond_dev
->name
, bond_dev
->name
);
2101 printk(KERN_WARNING DRV_NAME
2102 ": %s: When re-adding slaves, make sure the bond's "
2103 "HW address matches its VLANs'.\n",
2107 printk(KERN_INFO DRV_NAME
2108 ": %s: released all slaves\n",
2112 write_unlock_bh(&bond
->lock
);
2118 * This function changes the active slave to slave <slave_dev>.
2119 * It returns -EINVAL in the following cases.
2120 * - <slave_dev> is not found in the list.
2121 * - There is not active slave now.
2122 * - <slave_dev> is already active.
2123 * - The link state of <slave_dev> is not BOND_LINK_UP.
2124 * - <slave_dev> is not running.
2125 * In these cases, this fuction does nothing.
2126 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2128 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2130 struct bonding
*bond
= bond_dev
->priv
;
2131 struct slave
*old_active
= NULL
;
2132 struct slave
*new_active
= NULL
;
2135 if (!USES_PRIMARY(bond
->params
.mode
)) {
2139 /* Verify that master_dev is indeed the master of slave_dev */
2140 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
2141 (slave_dev
->master
!= bond_dev
)) {
2145 read_lock(&bond
->lock
);
2147 read_lock(&bond
->curr_slave_lock
);
2148 old_active
= bond
->curr_active_slave
;
2149 read_unlock(&bond
->curr_slave_lock
);
2151 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2154 * Changing to the current active: do nothing; return success.
2156 if (new_active
&& (new_active
== old_active
)) {
2157 read_unlock(&bond
->lock
);
2163 (new_active
->link
== BOND_LINK_UP
) &&
2164 IS_UP(new_active
->dev
)) {
2165 write_lock_bh(&bond
->curr_slave_lock
);
2166 bond_change_active_slave(bond
, new_active
);
2167 write_unlock_bh(&bond
->curr_slave_lock
);
2172 read_unlock(&bond
->lock
);
2177 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2179 struct bonding
*bond
= bond_dev
->priv
;
2181 info
->bond_mode
= bond
->params
.mode
;
2182 info
->miimon
= bond
->params
.miimon
;
2184 read_lock(&bond
->lock
);
2185 info
->num_slaves
= bond
->slave_cnt
;
2186 read_unlock(&bond
->lock
);
2191 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2193 struct bonding
*bond
= bond_dev
->priv
;
2194 struct slave
*slave
;
2197 if (info
->slave_id
< 0) {
2201 read_lock(&bond
->lock
);
2203 bond_for_each_slave(bond
, slave
, i
) {
2204 if (i
== (int)info
->slave_id
) {
2210 read_unlock(&bond
->lock
);
2213 strcpy(info
->slave_name
, slave
->dev
->name
);
2214 info
->link
= slave
->link
;
2215 info
->state
= slave
->state
;
2216 info
->link_failure_count
= slave
->link_failure_count
;
2224 /*-------------------------------- Monitoring -------------------------------*/
2227 static int bond_miimon_inspect(struct bonding
*bond
)
2229 struct slave
*slave
;
2230 int i
, link_state
, commit
= 0;
2232 bond_for_each_slave(bond
, slave
, i
) {
2233 slave
->new_link
= BOND_LINK_NOCHANGE
;
2235 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2237 switch (slave
->link
) {
2242 slave
->link
= BOND_LINK_FAIL
;
2243 slave
->delay
= bond
->params
.downdelay
;
2245 printk(KERN_INFO DRV_NAME
2246 ": %s: link status down for %s"
2247 "interface %s, disabling it in %d ms.\n",
2249 (bond
->params
.mode
==
2250 BOND_MODE_ACTIVEBACKUP
) ?
2251 ((slave
->state
== BOND_STATE_ACTIVE
) ?
2252 "active " : "backup ") : "",
2254 bond
->params
.downdelay
* bond
->params
.miimon
);
2257 case BOND_LINK_FAIL
:
2260 * recovered before downdelay expired
2262 slave
->link
= BOND_LINK_UP
;
2263 slave
->jiffies
= jiffies
;
2264 printk(KERN_INFO DRV_NAME
2265 ": %s: link status up again after %d "
2266 "ms for interface %s.\n",
2268 (bond
->params
.downdelay
- slave
->delay
) *
2269 bond
->params
.miimon
,
2274 if (slave
->delay
<= 0) {
2275 slave
->new_link
= BOND_LINK_DOWN
;
2283 case BOND_LINK_DOWN
:
2287 slave
->link
= BOND_LINK_BACK
;
2288 slave
->delay
= bond
->params
.updelay
;
2291 printk(KERN_INFO DRV_NAME
2292 ": %s: link status up for "
2293 "interface %s, enabling it in %d ms.\n",
2294 bond
->dev
->name
, slave
->dev
->name
,
2295 bond
->params
.updelay
*
2296 bond
->params
.miimon
);
2299 case BOND_LINK_BACK
:
2301 slave
->link
= BOND_LINK_DOWN
;
2302 printk(KERN_INFO DRV_NAME
2303 ": %s: link status down again after %d "
2304 "ms for interface %s.\n",
2306 (bond
->params
.updelay
- slave
->delay
) *
2307 bond
->params
.miimon
,
2313 if (slave
->delay
<= 0) {
2314 slave
->new_link
= BOND_LINK_UP
;
2327 static void bond_miimon_commit(struct bonding
*bond
)
2329 struct slave
*slave
;
2332 bond_for_each_slave(bond
, slave
, i
) {
2333 switch (slave
->new_link
) {
2334 case BOND_LINK_NOCHANGE
:
2338 slave
->link
= BOND_LINK_UP
;
2339 slave
->jiffies
= jiffies
;
2341 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2342 /* prevent it from being the active one */
2343 slave
->state
= BOND_STATE_BACKUP
;
2344 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2345 /* make it immediately active */
2346 slave
->state
= BOND_STATE_ACTIVE
;
2347 } else if (slave
!= bond
->primary_slave
) {
2348 /* prevent it from being the active one */
2349 slave
->state
= BOND_STATE_BACKUP
;
2352 printk(KERN_INFO DRV_NAME
2353 ": %s: link status definitely "
2354 "up for interface %s.\n",
2355 bond
->dev
->name
, slave
->dev
->name
);
2357 /* notify ad that the link status has changed */
2358 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2359 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2361 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
2362 (bond
->params
.mode
== BOND_MODE_ALB
))
2363 bond_alb_handle_link_change(bond
, slave
,
2366 if (!bond
->curr_active_slave
||
2367 (slave
== bond
->primary_slave
))
2372 case BOND_LINK_DOWN
:
2373 slave
->link
= BOND_LINK_DOWN
;
2375 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2376 bond
->params
.mode
== BOND_MODE_8023AD
)
2377 bond_set_slave_inactive_flags(slave
);
2379 printk(KERN_INFO DRV_NAME
2380 ": %s: link status definitely down for "
2381 "interface %s, disabling it\n",
2382 bond
->dev
->name
, slave
->dev
->name
);
2384 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2385 bond_3ad_handle_link_change(slave
,
2388 if (bond
->params
.mode
== BOND_MODE_TLB
||
2389 bond
->params
.mode
== BOND_MODE_ALB
)
2390 bond_alb_handle_link_change(bond
, slave
,
2393 if (slave
== bond
->curr_active_slave
)
2399 printk(KERN_ERR DRV_NAME
2400 ": %s: invalid new link %d on slave %s\n",
2401 bond
->dev
->name
, slave
->new_link
,
2403 slave
->new_link
= BOND_LINK_NOCHANGE
;
2410 write_lock_bh(&bond
->curr_slave_lock
);
2411 bond_select_active_slave(bond
);
2412 write_unlock_bh(&bond
->curr_slave_lock
);
2415 bond_set_carrier(bond
);
2421 * Really a wrapper that splits the mii monitor into two phases: an
2422 * inspection, then (if inspection indicates something needs to be done)
2423 * an acquisition of appropriate locks followed by a commit phase to
2424 * implement whatever link state changes are indicated.
2426 void bond_mii_monitor(struct work_struct
*work
)
2428 struct bonding
*bond
= container_of(work
, struct bonding
,
2431 read_lock(&bond
->lock
);
2432 if (bond
->kill_timers
)
2435 if (bond
->slave_cnt
== 0)
2438 if (bond
->send_grat_arp
) {
2439 read_lock(&bond
->curr_slave_lock
);
2440 bond_send_gratuitous_arp(bond
);
2441 read_unlock(&bond
->curr_slave_lock
);
2444 if (bond_miimon_inspect(bond
)) {
2445 read_unlock(&bond
->lock
);
2447 read_lock(&bond
->lock
);
2449 bond_miimon_commit(bond
);
2451 read_unlock(&bond
->lock
);
2452 rtnl_unlock(); /* might sleep, hold no other locks */
2453 read_lock(&bond
->lock
);
2457 if (bond
->params
.miimon
)
2458 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2459 msecs_to_jiffies(bond
->params
.miimon
));
2461 read_unlock(&bond
->lock
);
2464 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2466 struct in_device
*idev
;
2467 struct in_ifaddr
*ifa
;
2474 idev
= __in_dev_get_rcu(dev
);
2478 ifa
= idev
->ifa_list
;
2482 addr
= ifa
->ifa_local
;
2488 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2490 struct vlan_entry
*vlan
;
2492 if (ip
== bond
->master_ip
)
2495 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2496 if (ip
== vlan
->vlan_ip
)
2504 * We go to the (large) trouble of VLAN tagging ARP frames because
2505 * switches in VLAN mode (especially if ports are configured as
2506 * "native" to a VLAN) might not pass non-tagged frames.
2508 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2510 struct sk_buff
*skb
;
2512 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2513 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2515 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2516 NULL
, slave_dev
->dev_addr
, NULL
);
2519 printk(KERN_ERR DRV_NAME
": ARP packet allocation failed\n");
2523 skb
= vlan_put_tag(skb
, vlan_id
);
2525 printk(KERN_ERR DRV_NAME
": failed to insert VLAN tag\n");
2533 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2536 __be32
*targets
= bond
->params
.arp_targets
;
2537 struct vlan_entry
*vlan
;
2538 struct net_device
*vlan_dev
;
2542 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2545 dprintk("basa: target %x\n", targets
[i
]);
2546 if (list_empty(&bond
->vlan_list
)) {
2547 dprintk("basa: empty vlan: arp_send\n");
2548 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2549 bond
->master_ip
, 0);
2554 * If VLANs are configured, we do a route lookup to
2555 * determine which VLAN interface would be used, so we
2556 * can tag the ARP with the proper VLAN tag.
2558 memset(&fl
, 0, sizeof(fl
));
2559 fl
.fl4_dst
= targets
[i
];
2560 fl
.fl4_tos
= RTO_ONLINK
;
2562 rv
= ip_route_output_key(&init_net
, &rt
, &fl
);
2564 if (net_ratelimit()) {
2565 printk(KERN_WARNING DRV_NAME
2566 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2567 bond
->dev
->name
, NIPQUAD(fl
.fl4_dst
));
2573 * This target is not on a VLAN
2575 if (rt
->u
.dst
.dev
== bond
->dev
) {
2577 dprintk("basa: rtdev == bond->dev: arp_send\n");
2578 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2579 bond
->master_ip
, 0);
2584 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2585 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2586 if (vlan_dev
== rt
->u
.dst
.dev
) {
2587 vlan_id
= vlan
->vlan_id
;
2588 dprintk("basa: vlan match on %s %d\n",
2589 vlan_dev
->name
, vlan_id
);
2596 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2597 vlan
->vlan_ip
, vlan_id
);
2601 if (net_ratelimit()) {
2602 printk(KERN_WARNING DRV_NAME
2603 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2604 bond
->dev
->name
, NIPQUAD(fl
.fl4_dst
),
2605 rt
->u
.dst
.dev
? rt
->u
.dst
.dev
->name
: "NULL");
2612 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2613 * for each VLAN above us.
2615 * Caller must hold curr_slave_lock for read or better
2617 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2619 struct slave
*slave
= bond
->curr_active_slave
;
2620 struct vlan_entry
*vlan
;
2621 struct net_device
*vlan_dev
;
2623 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond
->dev
->name
,
2624 slave
? slave
->dev
->name
: "NULL");
2626 if (!slave
|| !bond
->send_grat_arp
||
2627 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2630 bond
->send_grat_arp
--;
2632 if (bond
->master_ip
) {
2633 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2634 bond
->master_ip
, 0);
2637 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2638 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2639 if (vlan
->vlan_ip
) {
2640 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2641 vlan
->vlan_ip
, vlan
->vlan_id
);
2646 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2649 __be32
*targets
= bond
->params
.arp_targets
;
2651 targets
= bond
->params
.arp_targets
;
2652 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2653 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2654 "%u.%u.%u.%u bhti(tip) %d\n",
2655 NIPQUAD(sip
), NIPQUAD(tip
), i
, NIPQUAD(targets
[i
]),
2656 bond_has_this_ip(bond
, tip
));
2657 if (sip
== targets
[i
]) {
2658 if (bond_has_this_ip(bond
, tip
))
2659 slave
->last_arp_rx
= jiffies
;
2665 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2668 struct slave
*slave
;
2669 struct bonding
*bond
;
2670 unsigned char *arp_ptr
;
2673 if (dev_net(dev
) != &init_net
)
2676 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2680 read_lock(&bond
->lock
);
2682 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2683 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2684 orig_dev
? orig_dev
->name
: "NULL");
2686 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2687 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2690 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2694 if (arp
->ar_hln
!= dev
->addr_len
||
2695 skb
->pkt_type
== PACKET_OTHERHOST
||
2696 skb
->pkt_type
== PACKET_LOOPBACK
||
2697 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2698 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2702 arp_ptr
= (unsigned char *)(arp
+ 1);
2703 arp_ptr
+= dev
->addr_len
;
2704 memcpy(&sip
, arp_ptr
, 4);
2705 arp_ptr
+= 4 + dev
->addr_len
;
2706 memcpy(&tip
, arp_ptr
, 4);
2708 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2709 " tip %u.%u.%u.%u\n", bond
->dev
->name
, slave
->dev
->name
,
2710 slave
->state
, bond
->params
.arp_validate
,
2711 slave_do_arp_validate(bond
, slave
), NIPQUAD(sip
), NIPQUAD(tip
));
2714 * Backup slaves won't see the ARP reply, but do come through
2715 * here for each ARP probe (so we swap the sip/tip to validate
2716 * the probe). In a "redundant switch, common router" type of
2717 * configuration, the ARP probe will (hopefully) travel from
2718 * the active, through one switch, the router, then the other
2719 * switch before reaching the backup.
2721 if (slave
->state
== BOND_STATE_ACTIVE
)
2722 bond_validate_arp(bond
, slave
, sip
, tip
);
2724 bond_validate_arp(bond
, slave
, tip
, sip
);
2727 read_unlock(&bond
->lock
);
2730 return NET_RX_SUCCESS
;
2734 * this function is called regularly to monitor each slave's link
2735 * ensuring that traffic is being sent and received when arp monitoring
2736 * is used in load-balancing mode. if the adapter has been dormant, then an
2737 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2738 * arp monitoring in active backup mode.
2740 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2742 struct bonding
*bond
= container_of(work
, struct bonding
,
2744 struct slave
*slave
, *oldcurrent
;
2745 int do_failover
= 0;
2749 read_lock(&bond
->lock
);
2751 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2753 if (bond
->kill_timers
) {
2757 if (bond
->slave_cnt
== 0) {
2761 read_lock(&bond
->curr_slave_lock
);
2762 oldcurrent
= bond
->curr_active_slave
;
2763 read_unlock(&bond
->curr_slave_lock
);
2765 /* see if any of the previous devices are up now (i.e. they have
2766 * xmt and rcv traffic). the curr_active_slave does not come into
2767 * the picture unless it is null. also, slave->jiffies is not needed
2768 * here because we send an arp on each slave and give a slave as
2769 * long as it needs to get the tx/rx within the delta.
2770 * TODO: what about up/down delay in arp mode? it wasn't here before
2773 bond_for_each_slave(bond
, slave
, i
) {
2774 if (slave
->link
!= BOND_LINK_UP
) {
2775 if (time_before_eq(jiffies
, slave
->dev
->trans_start
+ delta_in_ticks
) &&
2776 time_before_eq(jiffies
, slave
->dev
->last_rx
+ delta_in_ticks
)) {
2778 slave
->link
= BOND_LINK_UP
;
2779 slave
->state
= BOND_STATE_ACTIVE
;
2781 /* primary_slave has no meaning in round-robin
2782 * mode. the window of a slave being up and
2783 * curr_active_slave being null after enslaving
2787 printk(KERN_INFO DRV_NAME
2788 ": %s: link status definitely "
2789 "up for interface %s, ",
2794 printk(KERN_INFO DRV_NAME
2795 ": %s: interface %s is now up\n",
2801 /* slave->link == BOND_LINK_UP */
2803 /* not all switches will respond to an arp request
2804 * when the source ip is 0, so don't take the link down
2805 * if we don't know our ip yet
2807 if (time_after_eq(jiffies
, slave
->dev
->trans_start
+ 2*delta_in_ticks
) ||
2808 (time_after_eq(jiffies
, slave
->dev
->last_rx
+ 2*delta_in_ticks
))) {
2810 slave
->link
= BOND_LINK_DOWN
;
2811 slave
->state
= BOND_STATE_BACKUP
;
2813 if (slave
->link_failure_count
< UINT_MAX
) {
2814 slave
->link_failure_count
++;
2817 printk(KERN_INFO DRV_NAME
2818 ": %s: interface %s is now down.\n",
2822 if (slave
== oldcurrent
) {
2828 /* note: if switch is in round-robin mode, all links
2829 * must tx arp to ensure all links rx an arp - otherwise
2830 * links may oscillate or not come up at all; if switch is
2831 * in something like xor mode, there is nothing we can
2832 * do - all replies will be rx'ed on same link causing slaves
2833 * to be unstable during low/no traffic periods
2835 if (IS_UP(slave
->dev
)) {
2836 bond_arp_send_all(bond
, slave
);
2841 write_lock_bh(&bond
->curr_slave_lock
);
2843 bond_select_active_slave(bond
);
2845 write_unlock_bh(&bond
->curr_slave_lock
);
2849 if (bond
->params
.arp_interval
)
2850 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2852 read_unlock(&bond
->lock
);
2856 * Called to inspect slaves for active-backup mode ARP monitor link state
2857 * changes. Sets new_link in slaves to specify what action should take
2858 * place for the slave. Returns 0 if no changes are found, >0 if changes
2859 * to link states must be committed.
2861 * Called with bond->lock held for read.
2863 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2865 struct slave
*slave
;
2868 bond_for_each_slave(bond
, slave
, i
) {
2869 slave
->new_link
= BOND_LINK_NOCHANGE
;
2871 if (slave
->link
!= BOND_LINK_UP
) {
2872 if (time_before_eq(jiffies
, slave_last_rx(bond
, slave
) +
2874 slave
->new_link
= BOND_LINK_UP
;
2882 * Give slaves 2*delta after being enslaved or made
2883 * active. This avoids bouncing, as the last receive
2884 * times need a full ARP monitor cycle to be updated.
2886 if (!time_after_eq(jiffies
, slave
->jiffies
+
2887 2 * delta_in_ticks
))
2891 * Backup slave is down if:
2892 * - No current_arp_slave AND
2893 * - more than 3*delta since last receive AND
2894 * - the bond has an IP address
2896 * Note: a non-null current_arp_slave indicates
2897 * the curr_active_slave went down and we are
2898 * searching for a new one; under this condition
2899 * we only take the curr_active_slave down - this
2900 * gives each slave a chance to tx/rx traffic
2901 * before being taken out
2903 if (slave
->state
== BOND_STATE_BACKUP
&&
2904 !bond
->current_arp_slave
&&
2905 time_after(jiffies
, slave_last_rx(bond
, slave
) +
2906 3 * delta_in_ticks
)) {
2907 slave
->new_link
= BOND_LINK_DOWN
;
2912 * Active slave is down if:
2913 * - more than 2*delta since transmitting OR
2914 * - (more than 2*delta since receive AND
2915 * the bond has an IP address)
2917 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
2918 (time_after_eq(jiffies
, slave
->dev
->trans_start
+
2919 2 * delta_in_ticks
) ||
2920 (time_after_eq(jiffies
, slave_last_rx(bond
, slave
)
2921 + 2 * delta_in_ticks
)))) {
2922 slave
->new_link
= BOND_LINK_DOWN
;
2927 read_lock(&bond
->curr_slave_lock
);
2930 * Trigger a commit if the primary option setting has changed.
2932 if (bond
->primary_slave
&&
2933 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
2934 (bond
->primary_slave
->link
== BOND_LINK_UP
))
2937 read_unlock(&bond
->curr_slave_lock
);
2943 * Called to commit link state changes noted by inspection step of
2944 * active-backup mode ARP monitor.
2946 * Called with RTNL and bond->lock for read.
2948 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2950 struct slave
*slave
;
2953 bond_for_each_slave(bond
, slave
, i
) {
2954 switch (slave
->new_link
) {
2955 case BOND_LINK_NOCHANGE
:
2959 write_lock_bh(&bond
->curr_slave_lock
);
2961 if (!bond
->curr_active_slave
&&
2962 time_before_eq(jiffies
, slave
->dev
->trans_start
+
2964 slave
->link
= BOND_LINK_UP
;
2965 bond_change_active_slave(bond
, slave
);
2966 bond
->current_arp_slave
= NULL
;
2968 printk(KERN_INFO DRV_NAME
2969 ": %s: %s is up and now the "
2970 "active interface\n",
2971 bond
->dev
->name
, slave
->dev
->name
);
2973 } else if (bond
->curr_active_slave
!= slave
) {
2974 /* this slave has just come up but we
2975 * already have a current slave; this can
2976 * also happen if bond_enslave adds a new
2977 * slave that is up while we are searching
2980 slave
->link
= BOND_LINK_UP
;
2981 bond_set_slave_inactive_flags(slave
);
2982 bond
->current_arp_slave
= NULL
;
2984 printk(KERN_INFO DRV_NAME
2985 ": %s: backup interface %s is now up\n",
2986 bond
->dev
->name
, slave
->dev
->name
);
2989 write_unlock_bh(&bond
->curr_slave_lock
);
2993 case BOND_LINK_DOWN
:
2994 if (slave
->link_failure_count
< UINT_MAX
)
2995 slave
->link_failure_count
++;
2997 slave
->link
= BOND_LINK_DOWN
;
2999 if (slave
== bond
->curr_active_slave
) {
3000 printk(KERN_INFO DRV_NAME
3001 ": %s: link status down for active "
3002 "interface %s, disabling it\n",
3003 bond
->dev
->name
, slave
->dev
->name
);
3005 bond_set_slave_inactive_flags(slave
);
3007 write_lock_bh(&bond
->curr_slave_lock
);
3009 bond_select_active_slave(bond
);
3010 if (bond
->curr_active_slave
)
3011 bond
->curr_active_slave
->jiffies
=
3014 write_unlock_bh(&bond
->curr_slave_lock
);
3016 bond
->current_arp_slave
= NULL
;
3018 } else if (slave
->state
== BOND_STATE_BACKUP
) {
3019 printk(KERN_INFO DRV_NAME
3020 ": %s: backup interface %s is now down\n",
3021 bond
->dev
->name
, slave
->dev
->name
);
3023 bond_set_slave_inactive_flags(slave
);
3028 printk(KERN_ERR DRV_NAME
3029 ": %s: impossible: new_link %d on slave %s\n",
3030 bond
->dev
->name
, slave
->new_link
,
3036 * No race with changes to primary via sysfs, as we hold rtnl.
3038 if (bond
->primary_slave
&&
3039 (bond
->primary_slave
!= bond
->curr_active_slave
) &&
3040 (bond
->primary_slave
->link
== BOND_LINK_UP
)) {
3041 write_lock_bh(&bond
->curr_slave_lock
);
3042 bond_change_active_slave(bond
, bond
->primary_slave
);
3043 write_unlock_bh(&bond
->curr_slave_lock
);
3046 bond_set_carrier(bond
);
3050 * Send ARP probes for active-backup mode ARP monitor.
3052 * Called with bond->lock held for read.
3054 static void bond_ab_arp_probe(struct bonding
*bond
)
3056 struct slave
*slave
;
3059 read_lock(&bond
->curr_slave_lock
);
3061 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3062 printk("PROBE: c_arp %s && cas %s BAD\n",
3063 bond
->current_arp_slave
->dev
->name
,
3064 bond
->curr_active_slave
->dev
->name
);
3066 if (bond
->curr_active_slave
) {
3067 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3068 read_unlock(&bond
->curr_slave_lock
);
3072 read_unlock(&bond
->curr_slave_lock
);
3074 /* if we don't have a curr_active_slave, search for the next available
3075 * backup slave from the current_arp_slave and make it the candidate
3076 * for becoming the curr_active_slave
3079 if (!bond
->current_arp_slave
) {
3080 bond
->current_arp_slave
= bond
->first_slave
;
3081 if (!bond
->current_arp_slave
)
3085 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3087 /* search for next candidate */
3088 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3089 if (IS_UP(slave
->dev
)) {
3090 slave
->link
= BOND_LINK_BACK
;
3091 bond_set_slave_active_flags(slave
);
3092 bond_arp_send_all(bond
, slave
);
3093 slave
->jiffies
= jiffies
;
3094 bond
->current_arp_slave
= slave
;
3098 /* if the link state is up at this point, we
3099 * mark it down - this can happen if we have
3100 * simultaneous link failures and
3101 * reselect_active_interface doesn't make this
3102 * one the current slave so it is still marked
3103 * up when it is actually down
3105 if (slave
->link
== BOND_LINK_UP
) {
3106 slave
->link
= BOND_LINK_DOWN
;
3107 if (slave
->link_failure_count
< UINT_MAX
)
3108 slave
->link_failure_count
++;
3110 bond_set_slave_inactive_flags(slave
);
3112 printk(KERN_INFO DRV_NAME
3113 ": %s: backup interface %s is now down.\n",
3114 bond
->dev
->name
, slave
->dev
->name
);
3119 void bond_activebackup_arp_mon(struct work_struct
*work
)
3121 struct bonding
*bond
= container_of(work
, struct bonding
,
3125 read_lock(&bond
->lock
);
3127 if (bond
->kill_timers
)
3130 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3132 if (bond
->slave_cnt
== 0)
3135 if (bond
->send_grat_arp
) {
3136 read_lock(&bond
->curr_slave_lock
);
3137 bond_send_gratuitous_arp(bond
);
3138 read_unlock(&bond
->curr_slave_lock
);
3141 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3142 read_unlock(&bond
->lock
);
3144 read_lock(&bond
->lock
);
3146 bond_ab_arp_commit(bond
, delta_in_ticks
);
3148 read_unlock(&bond
->lock
);
3150 read_lock(&bond
->lock
);
3153 bond_ab_arp_probe(bond
);
3156 if (bond
->params
.arp_interval
) {
3157 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3160 read_unlock(&bond
->lock
);
3163 /*------------------------------ proc/seq_file-------------------------------*/
3165 #ifdef CONFIG_PROC_FS
3167 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3169 struct bonding
*bond
= seq
->private;
3171 struct slave
*slave
;
3174 /* make sure the bond won't be taken away */
3175 read_lock(&dev_base_lock
);
3176 read_lock(&bond
->lock
);
3179 return SEQ_START_TOKEN
;
3182 bond_for_each_slave(bond
, slave
, i
) {
3183 if (++off
== *pos
) {
3191 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3193 struct bonding
*bond
= seq
->private;
3194 struct slave
*slave
= v
;
3197 if (v
== SEQ_START_TOKEN
) {
3198 return bond
->first_slave
;
3201 slave
= slave
->next
;
3203 return (slave
== bond
->first_slave
) ? NULL
: slave
;
3206 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
3208 struct bonding
*bond
= seq
->private;
3210 read_unlock(&bond
->lock
);
3211 read_unlock(&dev_base_lock
);
3214 static void bond_info_show_master(struct seq_file
*seq
)
3216 struct bonding
*bond
= seq
->private;
3221 read_lock(&bond
->curr_slave_lock
);
3222 curr
= bond
->curr_active_slave
;
3223 read_unlock(&bond
->curr_slave_lock
);
3225 seq_printf(seq
, "Bonding Mode: %s",
3226 bond_mode_name(bond
->params
.mode
));
3228 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
&&
3229 bond
->params
.fail_over_mac
)
3230 seq_printf(seq
, " (fail_over_mac %s)",
3231 fail_over_mac_tbl
[bond
->params
.fail_over_mac
].modename
);
3233 seq_printf(seq
, "\n");
3235 if (bond
->params
.mode
== BOND_MODE_XOR
||
3236 bond
->params
.mode
== BOND_MODE_8023AD
) {
3237 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
3238 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
3239 bond
->params
.xmit_policy
);
3242 if (USES_PRIMARY(bond
->params
.mode
)) {
3243 seq_printf(seq
, "Primary Slave: %s\n",
3244 (bond
->primary_slave
) ?
3245 bond
->primary_slave
->dev
->name
: "None");
3247 seq_printf(seq
, "Currently Active Slave: %s\n",
3248 (curr
) ? curr
->dev
->name
: "None");
3251 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
3253 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
3254 seq_printf(seq
, "Up Delay (ms): %d\n",
3255 bond
->params
.updelay
* bond
->params
.miimon
);
3256 seq_printf(seq
, "Down Delay (ms): %d\n",
3257 bond
->params
.downdelay
* bond
->params
.miimon
);
3260 /* ARP information */
3261 if(bond
->params
.arp_interval
> 0) {
3263 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
3264 bond
->params
.arp_interval
);
3266 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
3268 for(i
= 0; (i
< BOND_MAX_ARP_TARGETS
) ;i
++) {
3269 if (!bond
->params
.arp_targets
[i
])
3272 seq_printf(seq
, ",");
3273 target
= ntohl(bond
->params
.arp_targets
[i
]);
3274 seq_printf(seq
, " %d.%d.%d.%d", HIPQUAD(target
));
3277 seq_printf(seq
, "\n");
3280 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3281 struct ad_info ad_info
;
3282 DECLARE_MAC_BUF(mac
);
3284 seq_puts(seq
, "\n802.3ad info\n");
3285 seq_printf(seq
, "LACP rate: %s\n",
3286 (bond
->params
.lacp_fast
) ? "fast" : "slow");
3288 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3289 seq_printf(seq
, "bond %s has no active aggregator\n",
3292 seq_printf(seq
, "Active Aggregator Info:\n");
3294 seq_printf(seq
, "\tAggregator ID: %d\n",
3295 ad_info
.aggregator_id
);
3296 seq_printf(seq
, "\tNumber of ports: %d\n",
3298 seq_printf(seq
, "\tActor Key: %d\n",
3300 seq_printf(seq
, "\tPartner Key: %d\n",
3301 ad_info
.partner_key
);
3302 seq_printf(seq
, "\tPartner Mac Address: %s\n",
3303 print_mac(mac
, ad_info
.partner_system
));
3308 static void bond_info_show_slave(struct seq_file
*seq
, const struct slave
*slave
)
3310 struct bonding
*bond
= seq
->private;
3311 DECLARE_MAC_BUF(mac
);
3313 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
3314 seq_printf(seq
, "MII Status: %s\n",
3315 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
3316 seq_printf(seq
, "Link Failure Count: %u\n",
3317 slave
->link_failure_count
);
3320 "Permanent HW addr: %s\n",
3321 print_mac(mac
, slave
->perm_hwaddr
));
3323 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3324 const struct aggregator
*agg
3325 = SLAVE_AD_INFO(slave
).port
.aggregator
;
3328 seq_printf(seq
, "Aggregator ID: %d\n",
3329 agg
->aggregator_identifier
);
3331 seq_puts(seq
, "Aggregator ID: N/A\n");
3336 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
3338 if (v
== SEQ_START_TOKEN
) {
3339 seq_printf(seq
, "%s\n", version
);
3340 bond_info_show_master(seq
);
3342 bond_info_show_slave(seq
, v
);
3348 static struct seq_operations bond_info_seq_ops
= {
3349 .start
= bond_info_seq_start
,
3350 .next
= bond_info_seq_next
,
3351 .stop
= bond_info_seq_stop
,
3352 .show
= bond_info_seq_show
,
3355 static int bond_info_open(struct inode
*inode
, struct file
*file
)
3357 struct seq_file
*seq
;
3358 struct proc_dir_entry
*proc
;
3361 res
= seq_open(file
, &bond_info_seq_ops
);
3363 /* recover the pointer buried in proc_dir_entry data */
3364 seq
= file
->private_data
;
3366 seq
->private = proc
->data
;
3372 static const struct file_operations bond_info_fops
= {
3373 .owner
= THIS_MODULE
,
3374 .open
= bond_info_open
,
3376 .llseek
= seq_lseek
,
3377 .release
= seq_release
,
3380 static int bond_create_proc_entry(struct bonding
*bond
)
3382 struct net_device
*bond_dev
= bond
->dev
;
3384 if (bond_proc_dir
) {
3385 bond
->proc_entry
= proc_create_data(bond_dev
->name
,
3386 S_IRUGO
, bond_proc_dir
,
3387 &bond_info_fops
, bond
);
3388 if (bond
->proc_entry
== NULL
) {
3389 printk(KERN_WARNING DRV_NAME
3390 ": Warning: Cannot create /proc/net/%s/%s\n",
3391 DRV_NAME
, bond_dev
->name
);
3393 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3400 static void bond_remove_proc_entry(struct bonding
*bond
)
3402 if (bond_proc_dir
&& bond
->proc_entry
) {
3403 remove_proc_entry(bond
->proc_file_name
, bond_proc_dir
);
3404 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3405 bond
->proc_entry
= NULL
;
3409 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3410 * Caller must hold rtnl_lock.
3412 static void bond_create_proc_dir(void)
3414 int len
= strlen(DRV_NAME
);
3416 for (bond_proc_dir
= init_net
.proc_net
->subdir
; bond_proc_dir
;
3417 bond_proc_dir
= bond_proc_dir
->next
) {
3418 if ((bond_proc_dir
->namelen
== len
) &&
3419 !memcmp(bond_proc_dir
->name
, DRV_NAME
, len
)) {
3424 if (!bond_proc_dir
) {
3425 bond_proc_dir
= proc_mkdir(DRV_NAME
, init_net
.proc_net
);
3426 if (bond_proc_dir
) {
3427 bond_proc_dir
->owner
= THIS_MODULE
;
3429 printk(KERN_WARNING DRV_NAME
3430 ": Warning: cannot create /proc/net/%s\n",
3436 /* Destroy the bonding directory under /proc/net, if empty.
3437 * Caller must hold rtnl_lock.
3439 static void bond_destroy_proc_dir(void)
3441 struct proc_dir_entry
*de
;
3443 if (!bond_proc_dir
) {
3447 /* verify that the /proc dir is empty */
3448 for (de
= bond_proc_dir
->subdir
; de
; de
= de
->next
) {
3449 /* ignore . and .. */
3450 if (*(de
->name
) != '.') {
3456 if (bond_proc_dir
->owner
== THIS_MODULE
) {
3457 bond_proc_dir
->owner
= NULL
;
3460 remove_proc_entry(DRV_NAME
, init_net
.proc_net
);
3461 bond_proc_dir
= NULL
;
3464 #endif /* CONFIG_PROC_FS */
3466 /*-------------------------- netdev event handling --------------------------*/
3469 * Change device name
3471 static int bond_event_changename(struct bonding
*bond
)
3473 #ifdef CONFIG_PROC_FS
3474 bond_remove_proc_entry(bond
);
3475 bond_create_proc_entry(bond
);
3477 down_write(&(bonding_rwsem
));
3478 bond_destroy_sysfs_entry(bond
);
3479 bond_create_sysfs_entry(bond
);
3480 up_write(&(bonding_rwsem
));
3484 static int bond_master_netdev_event(unsigned long event
, struct net_device
*bond_dev
)
3486 struct bonding
*event_bond
= bond_dev
->priv
;
3489 case NETDEV_CHANGENAME
:
3490 return bond_event_changename(event_bond
);
3491 case NETDEV_UNREGISTER
:
3492 bond_release_all(event_bond
->dev
);
3501 static int bond_slave_netdev_event(unsigned long event
, struct net_device
*slave_dev
)
3503 struct net_device
*bond_dev
= slave_dev
->master
;
3504 struct bonding
*bond
= bond_dev
->priv
;
3507 case NETDEV_UNREGISTER
:
3509 if (bond
->setup_by_slave
)
3510 bond_release_and_destroy(bond_dev
, slave_dev
);
3512 bond_release(bond_dev
, slave_dev
);
3517 * TODO: is this what we get if somebody
3518 * sets up a hierarchical bond, then rmmod's
3519 * one of the slave bonding devices?
3524 * ... Or is it this?
3527 case NETDEV_CHANGEMTU
:
3529 * TODO: Should slaves be allowed to
3530 * independently alter their MTU? For
3531 * an active-backup bond, slaves need
3532 * not be the same type of device, so
3533 * MTUs may vary. For other modes,
3534 * slaves arguably should have the
3535 * same MTUs. To do this, we'd need to
3536 * take over the slave's change_mtu
3537 * function for the duration of their
3541 case NETDEV_CHANGENAME
:
3543 * TODO: handle changing the primary's name
3546 case NETDEV_FEAT_CHANGE
:
3547 bond_compute_features(bond
);
3557 * bond_netdev_event: handle netdev notifier chain events.
3559 * This function receives events for the netdev chain. The caller (an
3560 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3561 * locks for us to safely manipulate the slave devices (RTNL lock,
3564 static int bond_netdev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3566 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3568 if (dev_net(event_dev
) != &init_net
)
3571 dprintk("event_dev: %s, event: %lx\n",
3572 (event_dev
? event_dev
->name
: "None"),
3575 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3578 if (event_dev
->flags
& IFF_MASTER
) {
3579 dprintk("IFF_MASTER\n");
3580 return bond_master_netdev_event(event
, event_dev
);
3583 if (event_dev
->flags
& IFF_SLAVE
) {
3584 dprintk("IFF_SLAVE\n");
3585 return bond_slave_netdev_event(event
, event_dev
);
3592 * bond_inetaddr_event: handle inetaddr notifier chain events.
3594 * We keep track of device IPs primarily to use as source addresses in
3595 * ARP monitor probes (rather than spewing out broadcasts all the time).
3597 * We track one IP for the main device (if it has one), plus one per VLAN.
3599 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3601 struct in_ifaddr
*ifa
= ptr
;
3602 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3603 struct bonding
*bond
;
3604 struct vlan_entry
*vlan
;
3606 if (dev_net(ifa
->ifa_dev
->dev
) != &init_net
)
3609 list_for_each_entry(bond
, &bond_dev_list
, bond_list
) {
3610 if (bond
->dev
== event_dev
) {
3613 bond
->master_ip
= ifa
->ifa_local
;
3616 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3623 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3624 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3625 if (vlan_dev
== event_dev
) {
3628 vlan
->vlan_ip
= ifa
->ifa_local
;
3632 bond_glean_dev_ip(vlan_dev
);
3643 static struct notifier_block bond_netdev_notifier
= {
3644 .notifier_call
= bond_netdev_event
,
3647 static struct notifier_block bond_inetaddr_notifier
= {
3648 .notifier_call
= bond_inetaddr_event
,
3651 /*-------------------------- Packet type handling ---------------------------*/
3653 /* register to receive lacpdus on a bond */
3654 static void bond_register_lacpdu(struct bonding
*bond
)
3656 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3658 /* initialize packet type */
3659 pk_type
->type
= PKT_TYPE_LACPDU
;
3660 pk_type
->dev
= bond
->dev
;
3661 pk_type
->func
= bond_3ad_lacpdu_recv
;
3663 dev_add_pack(pk_type
);
3666 /* unregister to receive lacpdus on a bond */
3667 static void bond_unregister_lacpdu(struct bonding
*bond
)
3669 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3672 void bond_register_arp(struct bonding
*bond
)
3674 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3679 pt
->type
= htons(ETH_P_ARP
);
3680 pt
->dev
= bond
->dev
;
3681 pt
->func
= bond_arp_rcv
;
3685 void bond_unregister_arp(struct bonding
*bond
)
3687 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3689 dev_remove_pack(pt
);
3693 /*---------------------------- Hashing Policies -----------------------------*/
3696 * Hash for the output device based upon layer 2 and layer 3 data. If
3697 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3699 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
,
3700 struct net_device
*bond_dev
, int count
)
3702 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3703 struct iphdr
*iph
= ip_hdr(skb
);
3705 if (skb
->protocol
== __constant_htons(ETH_P_IP
)) {
3706 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3707 (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5])) % count
;
3710 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3714 * Hash for the output device based upon layer 3 and layer 4 data. If
3715 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3716 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3718 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
,
3719 struct net_device
*bond_dev
, int count
)
3721 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3722 struct iphdr
*iph
= ip_hdr(skb
);
3723 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3726 if (skb
->protocol
== __constant_htons(ETH_P_IP
)) {
3727 if (!(iph
->frag_off
& __constant_htons(IP_MF
|IP_OFFSET
)) &&
3728 (iph
->protocol
== IPPROTO_TCP
||
3729 iph
->protocol
== IPPROTO_UDP
)) {
3730 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3732 return (layer4_xor
^
3733 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3737 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3741 * Hash for the output device based upon layer 2 data
3743 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
,
3744 struct net_device
*bond_dev
, int count
)
3746 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3748 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3751 /*-------------------------- Device entry points ----------------------------*/
3753 static int bond_open(struct net_device
*bond_dev
)
3755 struct bonding
*bond
= bond_dev
->priv
;
3757 bond
->kill_timers
= 0;
3759 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
3760 (bond
->params
.mode
== BOND_MODE_ALB
)) {
3761 /* bond_alb_initialize must be called before the timer
3764 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3765 /* something went wrong - fail the open operation */
3769 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3770 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3773 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3774 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3775 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3778 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3779 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3780 INIT_DELAYED_WORK(&bond
->arp_work
,
3781 bond_activebackup_arp_mon
);
3783 INIT_DELAYED_WORK(&bond
->arp_work
,
3784 bond_loadbalance_arp_mon
);
3786 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3787 if (bond
->params
.arp_validate
)
3788 bond_register_arp(bond
);
3791 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3792 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3793 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3794 /* register to receive LACPDUs */
3795 bond_register_lacpdu(bond
);
3801 static int bond_close(struct net_device
*bond_dev
)
3803 struct bonding
*bond
= bond_dev
->priv
;
3805 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3806 /* Unregister the receive of LACPDUs */
3807 bond_unregister_lacpdu(bond
);
3810 if (bond
->params
.arp_validate
)
3811 bond_unregister_arp(bond
);
3813 write_lock_bh(&bond
->lock
);
3815 bond
->send_grat_arp
= 0;
3817 /* signal timers not to re-arm */
3818 bond
->kill_timers
= 1;
3820 write_unlock_bh(&bond
->lock
);
3822 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3823 cancel_delayed_work(&bond
->mii_work
);
3826 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3827 cancel_delayed_work(&bond
->arp_work
);
3830 switch (bond
->params
.mode
) {
3831 case BOND_MODE_8023AD
:
3832 cancel_delayed_work(&bond
->ad_work
);
3836 cancel_delayed_work(&bond
->alb_work
);
3843 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
3844 (bond
->params
.mode
== BOND_MODE_ALB
)) {
3845 /* Must be called only after all
3846 * slaves have been released
3848 bond_alb_deinitialize(bond
);
3854 static struct net_device_stats
*bond_get_stats(struct net_device
*bond_dev
)
3856 struct bonding
*bond
= bond_dev
->priv
;
3857 struct net_device_stats
*stats
= &(bond
->stats
), *sstats
;
3858 struct net_device_stats local_stats
;
3859 struct slave
*slave
;
3862 memset(&local_stats
, 0, sizeof(struct net_device_stats
));
3864 read_lock_bh(&bond
->lock
);
3866 bond_for_each_slave(bond
, slave
, i
) {
3867 sstats
= slave
->dev
->get_stats(slave
->dev
);
3868 local_stats
.rx_packets
+= sstats
->rx_packets
;
3869 local_stats
.rx_bytes
+= sstats
->rx_bytes
;
3870 local_stats
.rx_errors
+= sstats
->rx_errors
;
3871 local_stats
.rx_dropped
+= sstats
->rx_dropped
;
3873 local_stats
.tx_packets
+= sstats
->tx_packets
;
3874 local_stats
.tx_bytes
+= sstats
->tx_bytes
;
3875 local_stats
.tx_errors
+= sstats
->tx_errors
;
3876 local_stats
.tx_dropped
+= sstats
->tx_dropped
;
3878 local_stats
.multicast
+= sstats
->multicast
;
3879 local_stats
.collisions
+= sstats
->collisions
;
3881 local_stats
.rx_length_errors
+= sstats
->rx_length_errors
;
3882 local_stats
.rx_over_errors
+= sstats
->rx_over_errors
;
3883 local_stats
.rx_crc_errors
+= sstats
->rx_crc_errors
;
3884 local_stats
.rx_frame_errors
+= sstats
->rx_frame_errors
;
3885 local_stats
.rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3886 local_stats
.rx_missed_errors
+= sstats
->rx_missed_errors
;
3888 local_stats
.tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3889 local_stats
.tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3890 local_stats
.tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3891 local_stats
.tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3892 local_stats
.tx_window_errors
+= sstats
->tx_window_errors
;
3895 memcpy(stats
, &local_stats
, sizeof(struct net_device_stats
));
3897 read_unlock_bh(&bond
->lock
);
3902 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3904 struct net_device
*slave_dev
= NULL
;
3905 struct ifbond k_binfo
;
3906 struct ifbond __user
*u_binfo
= NULL
;
3907 struct ifslave k_sinfo
;
3908 struct ifslave __user
*u_sinfo
= NULL
;
3909 struct mii_ioctl_data
*mii
= NULL
;
3912 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3913 bond_dev
->name
, cmd
);
3925 * We do this again just in case we were called by SIOCGMIIREG
3926 * instead of SIOCGMIIPHY.
3933 if (mii
->reg_num
== 1) {
3934 struct bonding
*bond
= bond_dev
->priv
;
3936 read_lock(&bond
->lock
);
3937 read_lock(&bond
->curr_slave_lock
);
3938 if (netif_carrier_ok(bond
->dev
)) {
3939 mii
->val_out
= BMSR_LSTATUS
;
3941 read_unlock(&bond
->curr_slave_lock
);
3942 read_unlock(&bond
->lock
);
3946 case BOND_INFO_QUERY_OLD
:
3947 case SIOCBONDINFOQUERY
:
3948 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3950 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
))) {
3954 res
= bond_info_query(bond_dev
, &k_binfo
);
3956 if (copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
))) {
3962 case BOND_SLAVE_INFO_QUERY_OLD
:
3963 case SIOCBONDSLAVEINFOQUERY
:
3964 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3966 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
))) {
3970 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3972 if (copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
))) {
3983 if (!capable(CAP_NET_ADMIN
)) {
3987 down_write(&(bonding_rwsem
));
3988 slave_dev
= dev_get_by_name(&init_net
, ifr
->ifr_slave
);
3990 dprintk("slave_dev=%p: \n", slave_dev
);
3995 dprintk("slave_dev->name=%s: \n", slave_dev
->name
);
3997 case BOND_ENSLAVE_OLD
:
3998 case SIOCBONDENSLAVE
:
3999 res
= bond_enslave(bond_dev
, slave_dev
);
4001 case BOND_RELEASE_OLD
:
4002 case SIOCBONDRELEASE
:
4003 res
= bond_release(bond_dev
, slave_dev
);
4005 case BOND_SETHWADDR_OLD
:
4006 case SIOCBONDSETHWADDR
:
4007 res
= bond_sethwaddr(bond_dev
, slave_dev
);
4009 case BOND_CHANGE_ACTIVE_OLD
:
4010 case SIOCBONDCHANGEACTIVE
:
4011 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
4020 up_write(&(bonding_rwsem
));
4024 static void bond_set_multicast_list(struct net_device
*bond_dev
)
4026 struct bonding
*bond
= bond_dev
->priv
;
4027 struct dev_mc_list
*dmi
;
4030 * Do promisc before checking multicast_mode
4032 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
)) {
4034 * FIXME: Need to handle the error when one of the multi-slaves
4037 bond_set_promiscuity(bond
, 1);
4040 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
)) {
4041 bond_set_promiscuity(bond
, -1);
4044 /* set allmulti flag to slaves */
4045 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
)) {
4047 * FIXME: Need to handle the error when one of the multi-slaves
4050 bond_set_allmulti(bond
, 1);
4053 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
)) {
4054 bond_set_allmulti(bond
, -1);
4057 read_lock(&bond
->lock
);
4059 bond
->flags
= bond_dev
->flags
;
4061 /* looking for addresses to add to slaves' mc list */
4062 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
4063 if (!bond_mc_list_find_dmi(dmi
, bond
->mc_list
)) {
4064 bond_mc_add(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4068 /* looking for addresses to delete from slaves' list */
4069 for (dmi
= bond
->mc_list
; dmi
; dmi
= dmi
->next
) {
4070 if (!bond_mc_list_find_dmi(dmi
, bond_dev
->mc_list
)) {
4071 bond_mc_delete(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
4075 /* save master's multicast list */
4076 bond_mc_list_destroy(bond
);
4077 bond_mc_list_copy(bond_dev
->mc_list
, bond
, GFP_ATOMIC
);
4079 read_unlock(&bond
->lock
);
4083 * Change the MTU of all of a master's slaves to match the master
4085 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
4087 struct bonding
*bond
= bond_dev
->priv
;
4088 struct slave
*slave
, *stop_at
;
4092 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond
,
4093 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
4095 /* Can't hold bond->lock with bh disabled here since
4096 * some base drivers panic. On the other hand we can't
4097 * hold bond->lock without bh disabled because we'll
4098 * deadlock. The only solution is to rely on the fact
4099 * that we're under rtnl_lock here, and the slaves
4100 * list won't change. This doesn't solve the problem
4101 * of setting the slave's MTU while it is
4102 * transmitting, but the assumption is that the base
4103 * driver can handle that.
4105 * TODO: figure out a way to safely iterate the slaves
4106 * list, but without holding a lock around the actual
4107 * call to the base driver.
4110 bond_for_each_slave(bond
, slave
, i
) {
4111 dprintk("s %p s->p %p c_m %p\n", slave
,
4112 slave
->prev
, slave
->dev
->change_mtu
);
4114 res
= dev_set_mtu(slave
->dev
, new_mtu
);
4117 /* If we failed to set the slave's mtu to the new value
4118 * we must abort the operation even in ACTIVE_BACKUP
4119 * mode, because if we allow the backup slaves to have
4120 * different mtu values than the active slave we'll
4121 * need to change their mtu when doing a failover. That
4122 * means changing their mtu from timer context, which
4123 * is probably not a good idea.
4125 dprintk("err %d %s\n", res
, slave
->dev
->name
);
4130 bond_dev
->mtu
= new_mtu
;
4135 /* unwind from head to the slave that failed */
4137 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4140 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
4142 dprintk("unwind err %d dev %s\n", tmp_res
,
4153 * Note that many devices must be down to change the HW address, and
4154 * downing the master releases all slaves. We can make bonds full of
4155 * bonding devices to test this, however.
4157 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4159 struct bonding
*bond
= bond_dev
->priv
;
4160 struct sockaddr
*sa
= addr
, tmp_sa
;
4161 struct slave
*slave
, *stop_at
;
4165 dprintk("bond=%p, name=%s\n", bond
, (bond_dev
? bond_dev
->name
: "None"));
4168 * If fail_over_mac is set to active, do nothing and return
4169 * success. Returning an error causes ifenslave to fail.
4171 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4174 if (!is_valid_ether_addr(sa
->sa_data
)) {
4175 return -EADDRNOTAVAIL
;
4178 /* Can't hold bond->lock with bh disabled here since
4179 * some base drivers panic. On the other hand we can't
4180 * hold bond->lock without bh disabled because we'll
4181 * deadlock. The only solution is to rely on the fact
4182 * that we're under rtnl_lock here, and the slaves
4183 * list won't change. This doesn't solve the problem
4184 * of setting the slave's hw address while it is
4185 * transmitting, but the assumption is that the base
4186 * driver can handle that.
4188 * TODO: figure out a way to safely iterate the slaves
4189 * list, but without holding a lock around the actual
4190 * call to the base driver.
4193 bond_for_each_slave(bond
, slave
, i
) {
4194 dprintk("slave %p %s\n", slave
, slave
->dev
->name
);
4196 if (slave
->dev
->set_mac_address
== NULL
) {
4198 dprintk("EOPNOTSUPP %s\n", slave
->dev
->name
);
4202 res
= dev_set_mac_address(slave
->dev
, addr
);
4204 /* TODO: consider downing the slave
4206 * User should expect communications
4207 * breakage anyway until ARP finish
4210 dprintk("err %d %s\n", res
, slave
->dev
->name
);
4216 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4220 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4221 tmp_sa
.sa_family
= bond_dev
->type
;
4223 /* unwind from head to the slave that failed */
4225 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4228 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4230 dprintk("unwind err %d dev %s\n", tmp_res
,
4238 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4240 struct bonding
*bond
= bond_dev
->priv
;
4241 struct slave
*slave
, *start_at
;
4242 int i
, slave_no
, res
= 1;
4244 read_lock(&bond
->lock
);
4246 if (!BOND_IS_OK(bond
)) {
4251 * Concurrent TX may collide on rr_tx_counter; we accept that
4252 * as being rare enough not to justify using an atomic op here
4254 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4256 bond_for_each_slave(bond
, slave
, i
) {
4264 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4265 if (IS_UP(slave
->dev
) &&
4266 (slave
->link
== BOND_LINK_UP
) &&
4267 (slave
->state
== BOND_STATE_ACTIVE
)) {
4268 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4275 /* no suitable interface, frame not sent */
4278 read_unlock(&bond
->lock
);
4284 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4285 * the bond has a usable interface.
4287 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4289 struct bonding
*bond
= bond_dev
->priv
;
4292 read_lock(&bond
->lock
);
4293 read_lock(&bond
->curr_slave_lock
);
4295 if (!BOND_IS_OK(bond
)) {
4299 if (!bond
->curr_active_slave
)
4302 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4306 /* no suitable interface, frame not sent */
4309 read_unlock(&bond
->curr_slave_lock
);
4310 read_unlock(&bond
->lock
);
4315 * In bond_xmit_xor() , we determine the output device by using a pre-
4316 * determined xmit_hash_policy(), If the selected device is not enabled,
4317 * find the next active slave.
4319 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4321 struct bonding
*bond
= bond_dev
->priv
;
4322 struct slave
*slave
, *start_at
;
4327 read_lock(&bond
->lock
);
4329 if (!BOND_IS_OK(bond
)) {
4333 slave_no
= bond
->xmit_hash_policy(skb
, bond_dev
, bond
->slave_cnt
);
4335 bond_for_each_slave(bond
, slave
, i
) {
4344 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4345 if (IS_UP(slave
->dev
) &&
4346 (slave
->link
== BOND_LINK_UP
) &&
4347 (slave
->state
== BOND_STATE_ACTIVE
)) {
4348 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4355 /* no suitable interface, frame not sent */
4358 read_unlock(&bond
->lock
);
4363 * in broadcast mode, we send everything to all usable interfaces.
4365 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4367 struct bonding
*bond
= bond_dev
->priv
;
4368 struct slave
*slave
, *start_at
;
4369 struct net_device
*tx_dev
= NULL
;
4373 read_lock(&bond
->lock
);
4375 if (!BOND_IS_OK(bond
)) {
4379 read_lock(&bond
->curr_slave_lock
);
4380 start_at
= bond
->curr_active_slave
;
4381 read_unlock(&bond
->curr_slave_lock
);
4387 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4388 if (IS_UP(slave
->dev
) &&
4389 (slave
->link
== BOND_LINK_UP
) &&
4390 (slave
->state
== BOND_STATE_ACTIVE
)) {
4392 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4394 printk(KERN_ERR DRV_NAME
4395 ": %s: Error: bond_xmit_broadcast(): "
4396 "skb_clone() failed\n",
4401 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4403 dev_kfree_skb(skb2
);
4407 tx_dev
= slave
->dev
;
4412 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4417 /* no suitable interface, frame not sent */
4420 /* frame sent to all suitable interfaces */
4421 read_unlock(&bond
->lock
);
4425 /*------------------------- Device initialization ---------------------------*/
4427 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4429 switch (bond
->params
.xmit_policy
) {
4430 case BOND_XMIT_POLICY_LAYER23
:
4431 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4433 case BOND_XMIT_POLICY_LAYER34
:
4434 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4436 case BOND_XMIT_POLICY_LAYER2
:
4438 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4444 * set bond mode specific net device operations
4446 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4448 struct net_device
*bond_dev
= bond
->dev
;
4451 case BOND_MODE_ROUNDROBIN
:
4452 bond_dev
->hard_start_xmit
= bond_xmit_roundrobin
;
4454 case BOND_MODE_ACTIVEBACKUP
:
4455 bond_dev
->hard_start_xmit
= bond_xmit_activebackup
;
4458 bond_dev
->hard_start_xmit
= bond_xmit_xor
;
4459 bond_set_xmit_hash_policy(bond
);
4461 case BOND_MODE_BROADCAST
:
4462 bond_dev
->hard_start_xmit
= bond_xmit_broadcast
;
4464 case BOND_MODE_8023AD
:
4465 bond_set_master_3ad_flags(bond
);
4466 bond_dev
->hard_start_xmit
= bond_3ad_xmit_xor
;
4467 bond_set_xmit_hash_policy(bond
);
4470 bond_set_master_alb_flags(bond
);
4473 bond_dev
->hard_start_xmit
= bond_alb_xmit
;
4474 bond_dev
->set_mac_address
= bond_alb_set_mac_address
;
4477 /* Should never happen, mode already checked */
4478 printk(KERN_ERR DRV_NAME
4479 ": %s: Error: Unknown bonding mode %d\n",
4486 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4487 struct ethtool_drvinfo
*drvinfo
)
4489 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4490 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4491 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4494 static const struct ethtool_ops bond_ethtool_ops
= {
4495 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4499 * Does not allocate but creates a /proc entry.
4502 static int bond_init(struct net_device
*bond_dev
, struct bond_params
*params
)
4504 struct bonding
*bond
= bond_dev
->priv
;
4506 dprintk("Begin bond_init for %s\n", bond_dev
->name
);
4508 /* initialize rwlocks */
4509 rwlock_init(&bond
->lock
);
4510 rwlock_init(&bond
->curr_slave_lock
);
4512 bond
->params
= *params
; /* copy params struct */
4514 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4518 /* Initialize pointers */
4519 bond
->first_slave
= NULL
;
4520 bond
->curr_active_slave
= NULL
;
4521 bond
->current_arp_slave
= NULL
;
4522 bond
->primary_slave
= NULL
;
4523 bond
->dev
= bond_dev
;
4524 bond
->send_grat_arp
= 0;
4525 bond
->setup_by_slave
= 0;
4526 INIT_LIST_HEAD(&bond
->vlan_list
);
4528 /* Initialize the device entry points */
4529 bond_dev
->open
= bond_open
;
4530 bond_dev
->stop
= bond_close
;
4531 bond_dev
->get_stats
= bond_get_stats
;
4532 bond_dev
->do_ioctl
= bond_do_ioctl
;
4533 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4534 bond_dev
->set_multicast_list
= bond_set_multicast_list
;
4535 bond_dev
->change_mtu
= bond_change_mtu
;
4536 bond_dev
->set_mac_address
= bond_set_mac_address
;
4537 bond_dev
->validate_addr
= NULL
;
4539 bond_set_mode_ops(bond
, bond
->params
.mode
);
4541 bond_dev
->destructor
= free_netdev
;
4543 /* Initialize the device options */
4544 bond_dev
->tx_queue_len
= 0;
4545 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4546 bond_dev
->priv_flags
|= IFF_BONDING
;
4548 /* At first, we block adding VLANs. That's the only way to
4549 * prevent problems that occur when adding VLANs over an
4550 * empty bond. The block will be removed once non-challenged
4551 * slaves are enslaved.
4553 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4555 /* don't acquire bond device's netif_tx_lock when
4557 bond_dev
->features
|= NETIF_F_LLTX
;
4559 /* By default, we declare the bond to be fully
4560 * VLAN hardware accelerated capable. Special
4561 * care is taken in the various xmit functions
4562 * when there are slaves that are not hw accel
4565 bond_dev
->vlan_rx_register
= bond_vlan_rx_register
;
4566 bond_dev
->vlan_rx_add_vid
= bond_vlan_rx_add_vid
;
4567 bond_dev
->vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
;
4568 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4569 NETIF_F_HW_VLAN_RX
|
4570 NETIF_F_HW_VLAN_FILTER
);
4572 #ifdef CONFIG_PROC_FS
4573 bond_create_proc_entry(bond
);
4575 list_add_tail(&bond
->bond_list
, &bond_dev_list
);
4580 /* De-initialize device specific data.
4581 * Caller must hold rtnl_lock.
4583 static void bond_deinit(struct net_device
*bond_dev
)
4585 struct bonding
*bond
= bond_dev
->priv
;
4587 list_del(&bond
->bond_list
);
4589 #ifdef CONFIG_PROC_FS
4590 bond_remove_proc_entry(bond
);
4594 static void bond_work_cancel_all(struct bonding
*bond
)
4596 write_lock_bh(&bond
->lock
);
4597 bond
->kill_timers
= 1;
4598 write_unlock_bh(&bond
->lock
);
4600 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4601 cancel_delayed_work(&bond
->mii_work
);
4603 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4604 cancel_delayed_work(&bond
->arp_work
);
4606 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4607 delayed_work_pending(&bond
->alb_work
))
4608 cancel_delayed_work(&bond
->alb_work
);
4610 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4611 delayed_work_pending(&bond
->ad_work
))
4612 cancel_delayed_work(&bond
->ad_work
);
4615 /* Unregister and free all bond devices.
4616 * Caller must hold rtnl_lock.
4618 static void bond_free_all(void)
4620 struct bonding
*bond
, *nxt
;
4622 list_for_each_entry_safe(bond
, nxt
, &bond_dev_list
, bond_list
) {
4623 struct net_device
*bond_dev
= bond
->dev
;
4625 bond_work_cancel_all(bond
);
4626 netif_addr_lock_bh(bond_dev
);
4627 bond_mc_list_destroy(bond
);
4628 netif_addr_unlock_bh(bond_dev
);
4629 /* Release the bonded slaves */
4630 bond_release_all(bond_dev
);
4634 #ifdef CONFIG_PROC_FS
4635 bond_destroy_proc_dir();
4639 /*------------------------- Module initialization ---------------------------*/
4642 * Convert string input module parms. Accept either the
4643 * number of the mode or its string name. A bit complicated because
4644 * some mode names are substrings of other names, and calls from sysfs
4645 * may have whitespace in the name (trailing newlines, for example).
4647 int bond_parse_parm(const char *buf
, struct bond_parm_tbl
*tbl
)
4649 int mode
= -1, i
, rv
;
4650 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4652 for (p
= (char *)buf
; *p
; p
++)
4653 if (!(isdigit(*p
) || isspace(*p
)))
4657 rv
= sscanf(buf
, "%20s", modestr
);
4659 rv
= sscanf(buf
, "%d", &mode
);
4664 for (i
= 0; tbl
[i
].modename
; i
++) {
4665 if (mode
== tbl
[i
].mode
)
4667 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4674 static int bond_check_params(struct bond_params
*params
)
4676 int arp_validate_value
, fail_over_mac_value
;
4679 * Convert string parameters.
4682 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4683 if (bond_mode
== -1) {
4684 printk(KERN_ERR DRV_NAME
4685 ": Error: Invalid bonding mode \"%s\"\n",
4686 mode
== NULL
? "NULL" : mode
);
4691 if (xmit_hash_policy
) {
4692 if ((bond_mode
!= BOND_MODE_XOR
) &&
4693 (bond_mode
!= BOND_MODE_8023AD
)) {
4694 printk(KERN_INFO DRV_NAME
4695 ": xor_mode param is irrelevant in mode %s\n",
4696 bond_mode_name(bond_mode
));
4698 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4700 if (xmit_hashtype
== -1) {
4701 printk(KERN_ERR DRV_NAME
4702 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4703 xmit_hash_policy
== NULL
? "NULL" :
4711 if (bond_mode
!= BOND_MODE_8023AD
) {
4712 printk(KERN_INFO DRV_NAME
4713 ": lacp_rate param is irrelevant in mode %s\n",
4714 bond_mode_name(bond_mode
));
4716 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4717 if (lacp_fast
== -1) {
4718 printk(KERN_ERR DRV_NAME
4719 ": Error: Invalid lacp rate \"%s\"\n",
4720 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4726 if (max_bonds
< 0 || max_bonds
> INT_MAX
) {
4727 printk(KERN_WARNING DRV_NAME
4728 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4729 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4730 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4731 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4735 printk(KERN_WARNING DRV_NAME
4736 ": Warning: miimon module parameter (%d), "
4737 "not in range 0-%d, so it was reset to %d\n",
4738 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4739 miimon
= BOND_LINK_MON_INTERV
;
4743 printk(KERN_WARNING DRV_NAME
4744 ": Warning: updelay module parameter (%d), "
4745 "not in range 0-%d, so it was reset to 0\n",
4750 if (downdelay
< 0) {
4751 printk(KERN_WARNING DRV_NAME
4752 ": Warning: downdelay module parameter (%d), "
4753 "not in range 0-%d, so it was reset to 0\n",
4754 downdelay
, INT_MAX
);
4758 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4759 printk(KERN_WARNING DRV_NAME
4760 ": Warning: use_carrier module parameter (%d), "
4761 "not of valid value (0/1), so it was set to 1\n",
4766 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
4767 printk(KERN_WARNING DRV_NAME
4768 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4769 "was reset to 1 \n", num_grat_arp
);
4773 /* reset values for 802.3ad */
4774 if (bond_mode
== BOND_MODE_8023AD
) {
4776 printk(KERN_WARNING DRV_NAME
4777 ": Warning: miimon must be specified, "
4778 "otherwise bonding will not detect link "
4779 "failure, speed and duplex which are "
4780 "essential for 802.3ad operation\n");
4781 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4786 /* reset values for TLB/ALB */
4787 if ((bond_mode
== BOND_MODE_TLB
) ||
4788 (bond_mode
== BOND_MODE_ALB
)) {
4790 printk(KERN_WARNING DRV_NAME
4791 ": Warning: miimon must be specified, "
4792 "otherwise bonding will not detect link "
4793 "failure and link speed which are essential "
4794 "for TLB/ALB load balancing\n");
4795 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4800 if (bond_mode
== BOND_MODE_ALB
) {
4801 printk(KERN_NOTICE DRV_NAME
4802 ": In ALB mode you might experience client "
4803 "disconnections upon reconnection of a link if the "
4804 "bonding module updelay parameter (%d msec) is "
4805 "incompatible with the forwarding delay time of the "
4811 if (updelay
|| downdelay
) {
4812 /* just warn the user the up/down delay will have
4813 * no effect since miimon is zero...
4815 printk(KERN_WARNING DRV_NAME
4816 ": Warning: miimon module parameter not set "
4817 "and updelay (%d) or downdelay (%d) module "
4818 "parameter is set; updelay and downdelay have "
4819 "no effect unless miimon is set\n",
4820 updelay
, downdelay
);
4823 /* don't allow arp monitoring */
4825 printk(KERN_WARNING DRV_NAME
4826 ": Warning: miimon (%d) and arp_interval (%d) "
4827 "can't be used simultaneously, disabling ARP "
4829 miimon
, arp_interval
);
4833 if ((updelay
% miimon
) != 0) {
4834 printk(KERN_WARNING DRV_NAME
4835 ": Warning: updelay (%d) is not a multiple "
4836 "of miimon (%d), updelay rounded to %d ms\n",
4837 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4842 if ((downdelay
% miimon
) != 0) {
4843 printk(KERN_WARNING DRV_NAME
4844 ": Warning: downdelay (%d) is not a multiple "
4845 "of miimon (%d), downdelay rounded to %d ms\n",
4847 (downdelay
/ miimon
) * miimon
);
4850 downdelay
/= miimon
;
4853 if (arp_interval
< 0) {
4854 printk(KERN_WARNING DRV_NAME
4855 ": Warning: arp_interval module parameter (%d) "
4856 ", not in range 0-%d, so it was reset to %d\n",
4857 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4858 arp_interval
= BOND_LINK_ARP_INTERV
;
4861 for (arp_ip_count
= 0;
4862 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4864 /* not complete check, but should be good enough to
4866 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4867 printk(KERN_WARNING DRV_NAME
4868 ": Warning: bad arp_ip_target module parameter "
4869 "(%s), ARP monitoring will not be performed\n",
4870 arp_ip_target
[arp_ip_count
]);
4873 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4874 arp_target
[arp_ip_count
] = ip
;
4878 if (arp_interval
&& !arp_ip_count
) {
4879 /* don't allow arping if no arp_ip_target given... */
4880 printk(KERN_WARNING DRV_NAME
4881 ": Warning: arp_interval module parameter (%d) "
4882 "specified without providing an arp_ip_target "
4883 "parameter, arp_interval was reset to 0\n",
4889 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4890 printk(KERN_ERR DRV_NAME
4891 ": arp_validate only supported in active-backup mode\n");
4894 if (!arp_interval
) {
4895 printk(KERN_ERR DRV_NAME
4896 ": arp_validate requires arp_interval\n");
4900 arp_validate_value
= bond_parse_parm(arp_validate
,
4902 if (arp_validate_value
== -1) {
4903 printk(KERN_ERR DRV_NAME
4904 ": Error: invalid arp_validate \"%s\"\n",
4905 arp_validate
== NULL
? "NULL" : arp_validate
);
4909 arp_validate_value
= 0;
4912 printk(KERN_INFO DRV_NAME
4913 ": MII link monitoring set to %d ms\n",
4915 } else if (arp_interval
) {
4918 printk(KERN_INFO DRV_NAME
4919 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4921 arp_validate_tbl
[arp_validate_value
].modename
,
4924 for (i
= 0; i
< arp_ip_count
; i
++)
4925 printk (" %s", arp_ip_target
[i
]);
4929 } else if (max_bonds
) {
4930 /* miimon and arp_interval not set, we need one so things
4931 * work as expected, see bonding.txt for details
4933 printk(KERN_WARNING DRV_NAME
4934 ": Warning: either miimon or arp_interval and "
4935 "arp_ip_target module parameters must be specified, "
4936 "otherwise bonding will not detect link failures! see "
4937 "bonding.txt for details.\n");
4940 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4941 /* currently, using a primary only makes sense
4942 * in active backup, TLB or ALB modes
4944 printk(KERN_WARNING DRV_NAME
4945 ": Warning: %s primary device specified but has no "
4946 "effect in %s mode\n",
4947 primary
, bond_mode_name(bond_mode
));
4951 if (fail_over_mac
) {
4952 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4954 if (fail_over_mac_value
== -1) {
4955 printk(KERN_ERR DRV_NAME
4956 ": Error: invalid fail_over_mac \"%s\"\n",
4957 arp_validate
== NULL
? "NULL" : arp_validate
);
4961 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4962 printk(KERN_WARNING DRV_NAME
4963 ": Warning: fail_over_mac only affects "
4964 "active-backup mode.\n");
4966 fail_over_mac_value
= BOND_FOM_NONE
;
4969 /* fill params struct with the proper values */
4970 params
->mode
= bond_mode
;
4971 params
->xmit_policy
= xmit_hashtype
;
4972 params
->miimon
= miimon
;
4973 params
->num_grat_arp
= num_grat_arp
;
4974 params
->arp_interval
= arp_interval
;
4975 params
->arp_validate
= arp_validate_value
;
4976 params
->updelay
= updelay
;
4977 params
->downdelay
= downdelay
;
4978 params
->use_carrier
= use_carrier
;
4979 params
->lacp_fast
= lacp_fast
;
4980 params
->primary
[0] = 0;
4981 params
->fail_over_mac
= fail_over_mac_value
;
4984 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4985 params
->primary
[IFNAMSIZ
- 1] = 0;
4988 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4993 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4994 static struct lock_class_key bonding_netdev_addr_lock_key
;
4996 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4997 struct netdev_queue
*txq
,
5000 lockdep_set_class(&txq
->_xmit_lock
,
5001 &bonding_netdev_xmit_lock_key
);
5004 static void bond_set_lockdep_class(struct net_device
*dev
)
5006 lockdep_set_class(&dev
->addr_list_lock
,
5007 &bonding_netdev_addr_lock_key
);
5008 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
5011 /* Create a new bond based on the specified name and bonding parameters.
5012 * If name is NULL, obtain a suitable "bond%d" name for us.
5013 * Caller must NOT hold rtnl_lock; we need to release it here before we
5014 * set up our sysfs entries.
5016 int bond_create(char *name
, struct bond_params
*params
)
5018 struct net_device
*bond_dev
;
5019 struct bonding
*bond
;
5023 down_write(&bonding_rwsem
);
5025 /* Check to see if the bond already exists. */
5027 list_for_each_entry(bond
, &bond_dev_list
, bond_list
)
5028 if (strnicmp(bond
->dev
->name
, name
, IFNAMSIZ
) == 0) {
5029 printk(KERN_ERR DRV_NAME
5030 ": cannot add bond %s; it already exists\n",
5037 bond_dev
= alloc_netdev(sizeof(struct bonding
), name
? name
: "",
5040 printk(KERN_ERR DRV_NAME
5041 ": %s: eek! can't alloc netdev!\n",
5048 res
= dev_alloc_name(bond_dev
, "bond%d");
5053 /* bond_init() must be called after dev_alloc_name() (for the
5054 * /proc files), but before register_netdevice(), because we
5055 * need to set function pointers.
5058 res
= bond_init(bond_dev
, params
);
5063 res
= register_netdevice(bond_dev
);
5068 bond_set_lockdep_class(bond_dev
);
5070 netif_carrier_off(bond_dev
);
5072 up_write(&bonding_rwsem
);
5073 rtnl_unlock(); /* allows sysfs registration of net device */
5074 res
= bond_create_sysfs_entry(bond_dev
->priv
);
5077 down_write(&bonding_rwsem
);
5078 bond_deinit(bond_dev
);
5079 unregister_netdevice(bond_dev
);
5086 bond_deinit(bond_dev
);
5088 free_netdev(bond_dev
);
5090 up_write(&bonding_rwsem
);
5095 static int __init
bonding_init(void)
5099 struct bonding
*bond
;
5101 printk(KERN_INFO
"%s", version
);
5103 res
= bond_check_params(&bonding_defaults
);
5108 #ifdef CONFIG_PROC_FS
5109 bond_create_proc_dir();
5112 init_rwsem(&bonding_rwsem
);
5114 for (i
= 0; i
< max_bonds
; i
++) {
5115 res
= bond_create(NULL
, &bonding_defaults
);
5120 res
= bond_create_sysfs();
5124 register_netdevice_notifier(&bond_netdev_notifier
);
5125 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5129 list_for_each_entry(bond
, &bond_dev_list
, bond_list
) {
5130 bond_work_cancel_all(bond
);
5131 destroy_workqueue(bond
->wq
);
5134 bond_destroy_sysfs();
5144 static void __exit
bonding_exit(void)
5146 unregister_netdevice_notifier(&bond_netdev_notifier
);
5147 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5149 bond_destroy_sysfs();
5156 module_init(bonding_init
);
5157 module_exit(bonding_exit
);
5158 MODULE_LICENSE("GPL");
5159 MODULE_VERSION(DRV_VERSION
);
5160 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5161 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5162 MODULE_SUPPORTED_DEVICE("most ethernet devices");