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/sched.h>
39 #include <linux/types.h>
40 #include <linux/fcntl.h>
41 #include <linux/interrupt.h>
42 #include <linux/ptrace.h>
43 #include <linux/ioport.h>
47 #include <linux/tcp.h>
48 #include <linux/udp.h>
49 #include <linux/slab.h>
50 #include <linux/string.h>
51 #include <linux/init.h>
52 #include <linux/timer.h>
53 #include <linux/socket.h>
54 #include <linux/ctype.h>
55 #include <linux/inet.h>
56 #include <linux/bitops.h>
57 #include <asm/system.h>
60 #include <asm/uaccess.h>
61 #include <linux/errno.h>
62 #include <linux/netdevice.h>
63 #include <linux/inetdevice.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 <net/route.h>
82 /*---------------------------- Module parameters ----------------------------*/
84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
85 #define BOND_LINK_MON_INTERV 0
86 #define BOND_LINK_ARP_INTERV 0
88 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
89 static int miimon
= BOND_LINK_MON_INTERV
;
90 static int updelay
= 0;
91 static int downdelay
= 0;
92 static int use_carrier
= 1;
93 static char *mode
= NULL
;
94 static char *primary
= NULL
;
95 static char *lacp_rate
= NULL
;
96 static char *xmit_hash_policy
= NULL
;
97 static int arp_interval
= BOND_LINK_ARP_INTERV
;
98 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
] = { NULL
, };
99 struct bond_params bonding_defaults
;
101 module_param(max_bonds
, int, 0);
102 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
103 module_param(miimon
, int, 0);
104 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
105 module_param(updelay
, int, 0);
106 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
107 module_param(downdelay
, int, 0);
108 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
110 module_param(use_carrier
, int, 0);
111 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
112 "0 for off, 1 for on (default)");
113 module_param(mode
, charp
, 0);
114 MODULE_PARM_DESC(mode
, "Mode of operation : 0 for balance-rr, "
115 "1 for active-backup, 2 for balance-xor, "
116 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
117 "6 for balance-alb");
118 module_param(primary
, charp
, 0);
119 MODULE_PARM_DESC(primary
, "Primary network device to use");
120 module_param(lacp_rate
, charp
, 0);
121 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner "
123 module_param(xmit_hash_policy
, charp
, 0);
124 MODULE_PARM_DESC(xmit_hash_policy
, "XOR hashing method: 0 for layer 2 (default)"
125 ", 1 for layer 3+4");
126 module_param(arp_interval
, int, 0);
127 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
128 module_param_array(arp_ip_target
, charp
, NULL
, 0);
129 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
131 /*----------------------------- Global variables ----------------------------*/
133 static const char * const version
=
134 DRV_DESCRIPTION
": v" DRV_VERSION
" (" DRV_RELDATE
")\n";
136 LIST_HEAD(bond_dev_list
);
138 #ifdef CONFIG_PROC_FS
139 static struct proc_dir_entry
*bond_proc_dir
= NULL
;
142 extern struct rw_semaphore bonding_rwsem
;
143 static u32 arp_target
[BOND_MAX_ARP_TARGETS
] = { 0, } ;
144 static int arp_ip_count
= 0;
145 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
146 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
147 static int lacp_fast
= 0;
150 struct bond_parm_tbl bond_lacp_tbl
[] = {
151 { "slow", AD_LACP_SLOW
},
152 { "fast", AD_LACP_FAST
},
156 struct bond_parm_tbl bond_mode_tbl
[] = {
157 { "balance-rr", BOND_MODE_ROUNDROBIN
},
158 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
159 { "balance-xor", BOND_MODE_XOR
},
160 { "broadcast", BOND_MODE_BROADCAST
},
161 { "802.3ad", BOND_MODE_8023AD
},
162 { "balance-tlb", BOND_MODE_TLB
},
163 { "balance-alb", BOND_MODE_ALB
},
167 struct bond_parm_tbl xmit_hashtype_tbl
[] = {
168 { "layer2", BOND_XMIT_POLICY_LAYER2
},
169 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
173 /*-------------------------- Forward declarations ---------------------------*/
175 static void bond_send_gratuitous_arp(struct bonding
*bond
);
177 /*---------------------------- General routines -----------------------------*/
179 const char *bond_mode_name(int mode
)
182 case BOND_MODE_ROUNDROBIN
:
183 return "load balancing (round-robin)";
184 case BOND_MODE_ACTIVEBACKUP
:
185 return "fault-tolerance (active-backup)";
187 return "load balancing (xor)";
188 case BOND_MODE_BROADCAST
:
189 return "fault-tolerance (broadcast)";
190 case BOND_MODE_8023AD
:
191 return "IEEE 802.3ad Dynamic link aggregation";
193 return "transmit load balancing";
195 return "adaptive load balancing";
201 /*---------------------------------- VLAN -----------------------------------*/
204 * bond_add_vlan - add a new vlan id on bond
205 * @bond: bond that got the notification
206 * @vlan_id: the vlan id to add
208 * Returns -ENOMEM if allocation failed.
210 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
212 struct vlan_entry
*vlan
;
214 dprintk("bond: %s, vlan id %d\n",
215 (bond
? bond
->dev
->name
: "None"), vlan_id
);
217 vlan
= kmalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
222 INIT_LIST_HEAD(&vlan
->vlan_list
);
223 vlan
->vlan_id
= vlan_id
;
226 write_lock_bh(&bond
->lock
);
228 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
230 write_unlock_bh(&bond
->lock
);
232 dprintk("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
238 * bond_del_vlan - delete a vlan id from bond
239 * @bond: bond that got the notification
240 * @vlan_id: the vlan id to delete
242 * returns -ENODEV if @vlan_id was not found in @bond.
244 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
246 struct vlan_entry
*vlan
, *next
;
249 dprintk("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
251 write_lock_bh(&bond
->lock
);
253 list_for_each_entry_safe(vlan
, next
, &bond
->vlan_list
, vlan_list
) {
254 if (vlan
->vlan_id
== vlan_id
) {
255 list_del(&vlan
->vlan_list
);
257 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
258 (bond
->params
.mode
== BOND_MODE_ALB
)) {
259 bond_alb_clear_vlan(bond
, vlan_id
);
262 dprintk("removed VLAN ID %d from bond %s\n", vlan_id
,
267 if (list_empty(&bond
->vlan_list
) &&
268 (bond
->slave_cnt
== 0)) {
269 /* Last VLAN removed and no slaves, so
270 * restore block on adding VLANs. This will
271 * be removed once new slaves that are not
272 * VLAN challenged will be added.
274 bond
->dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
282 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id
,
286 write_unlock_bh(&bond
->lock
);
291 * bond_has_challenged_slaves
292 * @bond: the bond we're working on
294 * Searches the slave list. Returns 1 if a vlan challenged slave
295 * was found, 0 otherwise.
297 * Assumes bond->lock is held.
299 static int bond_has_challenged_slaves(struct bonding
*bond
)
304 bond_for_each_slave(bond
, slave
, i
) {
305 if (slave
->dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
306 dprintk("found VLAN challenged slave - %s\n",
312 dprintk("no VLAN challenged slaves found\n");
317 * bond_next_vlan - safely skip to the next item in the vlans list.
318 * @bond: the bond we're working on
319 * @curr: item we're advancing from
321 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
322 * or @curr->next otherwise (even if it is @curr itself again).
324 * Caller must hold bond->lock
326 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
328 struct vlan_entry
*next
, *last
;
330 if (list_empty(&bond
->vlan_list
)) {
335 next
= list_entry(bond
->vlan_list
.next
,
336 struct vlan_entry
, vlan_list
);
338 last
= list_entry(bond
->vlan_list
.prev
,
339 struct vlan_entry
, vlan_list
);
341 next
= list_entry(bond
->vlan_list
.next
,
342 struct vlan_entry
, vlan_list
);
344 next
= list_entry(curr
->vlan_list
.next
,
345 struct vlan_entry
, vlan_list
);
353 * bond_dev_queue_xmit - Prepare skb for xmit.
355 * @bond: bond device that got this skb for tx.
356 * @skb: hw accel VLAN tagged skb to transmit
357 * @slave_dev: slave that is supposed to xmit this skbuff
359 * When the bond gets an skb to transmit that is
360 * already hardware accelerated VLAN tagged, and it
361 * needs to relay this skb to a slave that is not
362 * hw accel capable, the skb needs to be "unaccelerated",
363 * i.e. strip the hwaccel tag and re-insert it as part
366 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
)
368 unsigned short vlan_id
;
370 if (!list_empty(&bond
->vlan_list
) &&
371 !(slave_dev
->features
& NETIF_F_HW_VLAN_TX
) &&
372 vlan_get_tag(skb
, &vlan_id
) == 0) {
373 skb
->dev
= slave_dev
;
374 skb
= vlan_put_tag(skb
, vlan_id
);
376 /* vlan_put_tag() frees the skb in case of error,
377 * so return success here so the calling functions
378 * won't attempt to free is again.
383 skb
->dev
= slave_dev
;
393 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
394 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
396 * a. This operation is performed in IOCTL context,
397 * b. The operation is protected by the RTNL semaphore in the 8021q code,
398 * c. Holding a lock with BH disabled while directly calling a base driver
399 * entry point is generally a BAD idea.
401 * The design of synchronization/protection for this operation in the 8021q
402 * module is good for one or more VLAN devices over a single physical device
403 * and cannot be extended for a teaming solution like bonding, so there is a
404 * potential race condition here where a net device from the vlan group might
405 * be referenced (either by a base driver or the 8021q code) while it is being
406 * removed from the system. However, it turns out we're not making matters
407 * worse, and if it works for regular VLAN usage it will work here too.
411 * bond_vlan_rx_register - Propagates registration to slaves
412 * @bond_dev: bonding net device that got called
413 * @grp: vlan group being registered
415 static void bond_vlan_rx_register(struct net_device
*bond_dev
, struct vlan_group
*grp
)
417 struct bonding
*bond
= bond_dev
->priv
;
423 bond_for_each_slave(bond
, slave
, i
) {
424 struct net_device
*slave_dev
= slave
->dev
;
426 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
427 slave_dev
->vlan_rx_register
) {
428 slave_dev
->vlan_rx_register(slave_dev
, grp
);
434 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
435 * @bond_dev: bonding net device that got called
436 * @vid: vlan id being added
438 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
440 struct bonding
*bond
= bond_dev
->priv
;
444 bond_for_each_slave(bond
, slave
, i
) {
445 struct net_device
*slave_dev
= slave
->dev
;
447 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
448 slave_dev
->vlan_rx_add_vid
) {
449 slave_dev
->vlan_rx_add_vid(slave_dev
, vid
);
453 res
= bond_add_vlan(bond
, vid
);
455 printk(KERN_ERR DRV_NAME
456 ": %s: Error: Failed to add vlan id %d\n",
457 bond_dev
->name
, vid
);
462 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
463 * @bond_dev: bonding net device that got called
464 * @vid: vlan id being removed
466 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
468 struct bonding
*bond
= bond_dev
->priv
;
470 struct net_device
*vlan_dev
;
473 bond_for_each_slave(bond
, slave
, i
) {
474 struct net_device
*slave_dev
= slave
->dev
;
476 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
477 slave_dev
->vlan_rx_kill_vid
) {
478 /* Save and then restore vlan_dev in the grp array,
479 * since the slave's driver might clear it.
481 vlan_dev
= bond
->vlgrp
->vlan_devices
[vid
];
482 slave_dev
->vlan_rx_kill_vid(slave_dev
, vid
);
483 bond
->vlgrp
->vlan_devices
[vid
] = vlan_dev
;
487 res
= bond_del_vlan(bond
, vid
);
489 printk(KERN_ERR DRV_NAME
490 ": %s: Error: Failed to remove vlan id %d\n",
491 bond_dev
->name
, vid
);
495 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
497 struct vlan_entry
*vlan
;
499 write_lock_bh(&bond
->lock
);
501 if (list_empty(&bond
->vlan_list
)) {
505 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
506 slave_dev
->vlan_rx_register
) {
507 slave_dev
->vlan_rx_register(slave_dev
, bond
->vlgrp
);
510 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
511 !(slave_dev
->vlan_rx_add_vid
)) {
515 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
516 slave_dev
->vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
520 write_unlock_bh(&bond
->lock
);
523 static void bond_del_vlans_from_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
525 struct vlan_entry
*vlan
;
526 struct net_device
*vlan_dev
;
528 write_lock_bh(&bond
->lock
);
530 if (list_empty(&bond
->vlan_list
)) {
534 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
535 !(slave_dev
->vlan_rx_kill_vid
)) {
539 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
540 /* Save and then restore vlan_dev in the grp array,
541 * since the slave's driver might clear it.
543 vlan_dev
= bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
];
544 slave_dev
->vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
545 bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
] = vlan_dev
;
549 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
550 slave_dev
->vlan_rx_register
) {
551 slave_dev
->vlan_rx_register(slave_dev
, NULL
);
555 write_unlock_bh(&bond
->lock
);
558 /*------------------------------- Link status -------------------------------*/
561 * Set the carrier state for the master according to the state of its
562 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
563 * do special 802.3ad magic.
565 * Returns zero if carrier state does not change, nonzero if it does.
567 static int bond_set_carrier(struct bonding
*bond
)
572 if (bond
->slave_cnt
== 0)
575 if (bond
->params
.mode
== BOND_MODE_8023AD
)
576 return bond_3ad_set_carrier(bond
);
578 bond_for_each_slave(bond
, slave
, i
) {
579 if (slave
->link
== BOND_LINK_UP
) {
580 if (!netif_carrier_ok(bond
->dev
)) {
581 netif_carrier_on(bond
->dev
);
589 if (netif_carrier_ok(bond
->dev
)) {
590 netif_carrier_off(bond
->dev
);
597 * Get link speed and duplex from the slave's base driver
598 * using ethtool. If for some reason the call fails or the
599 * values are invalid, fake speed and duplex to 100/Full
602 static int bond_update_speed_duplex(struct slave
*slave
)
604 struct net_device
*slave_dev
= slave
->dev
;
605 static int (* ioctl
)(struct net_device
*, struct ifreq
*, int);
607 struct ethtool_cmd etool
;
609 /* Fake speed and duplex */
610 slave
->speed
= SPEED_100
;
611 slave
->duplex
= DUPLEX_FULL
;
613 if (slave_dev
->ethtool_ops
) {
616 if (!slave_dev
->ethtool_ops
->get_settings
) {
620 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
628 ioctl
= slave_dev
->do_ioctl
;
629 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
630 etool
.cmd
= ETHTOOL_GSET
;
631 ifr
.ifr_data
= (char*)&etool
;
632 if (!ioctl
|| (IOCTL(slave_dev
, &ifr
, SIOCETHTOOL
) < 0)) {
637 switch (etool
.speed
) {
646 switch (etool
.duplex
) {
654 slave
->speed
= etool
.speed
;
655 slave
->duplex
= etool
.duplex
;
661 * if <dev> supports MII link status reporting, check its link status.
663 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
664 * depening upon the setting of the use_carrier parameter.
666 * Return either BMSR_LSTATUS, meaning that the link is up (or we
667 * can't tell and just pretend it is), or 0, meaning that the link is
670 * If reporting is non-zero, instead of faking link up, return -1 if
671 * both ETHTOOL and MII ioctls fail (meaning the device does not
672 * support them). If use_carrier is set, return whatever it says.
673 * It'd be nice if there was a good way to tell if a driver supports
674 * netif_carrier, but there really isn't.
676 static int bond_check_dev_link(struct bonding
*bond
, struct net_device
*slave_dev
, int reporting
)
678 static int (* ioctl
)(struct net_device
*, struct ifreq
*, int);
680 struct mii_ioctl_data
*mii
;
681 struct ethtool_value etool
;
683 if (bond
->params
.use_carrier
) {
684 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
687 ioctl
= slave_dev
->do_ioctl
;
689 /* TODO: set pointer to correct ioctl on a per team member */
690 /* bases to make this more efficient. that is, once */
691 /* we determine the correct ioctl, we will always */
692 /* call it and not the others for that team */
696 * We cannot assume that SIOCGMIIPHY will also read a
697 * register; not all network drivers (e.g., e100)
701 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
702 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
704 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
705 mii
->reg_num
= MII_BMSR
;
706 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0) {
707 return (mii
->val_out
& BMSR_LSTATUS
);
712 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
713 /* for a period of time so we attempt to get link status */
714 /* from it last if the above MII ioctls fail... */
715 if (slave_dev
->ethtool_ops
) {
716 if (slave_dev
->ethtool_ops
->get_link
) {
719 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
721 return link
? BMSR_LSTATUS
: 0;
726 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
727 etool
.cmd
= ETHTOOL_GLINK
;
728 ifr
.ifr_data
= (char*)&etool
;
729 if (IOCTL(slave_dev
, &ifr
, SIOCETHTOOL
) == 0) {
730 if (etool
.data
== 1) {
733 dprintk("SIOCETHTOOL shows link down\n");
740 * If reporting, report that either there's no dev->do_ioctl,
741 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
742 * cannot report link status). If not reporting, pretend
745 return (reporting
? -1 : BMSR_LSTATUS
);
748 /*----------------------------- Multicast list ------------------------------*/
751 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
753 static inline int bond_is_dmi_same(struct dev_mc_list
*dmi1
, struct dev_mc_list
*dmi2
)
755 return memcmp(dmi1
->dmi_addr
, dmi2
->dmi_addr
, dmi1
->dmi_addrlen
) == 0 &&
756 dmi1
->dmi_addrlen
== dmi2
->dmi_addrlen
;
760 * returns dmi entry if found, NULL otherwise
762 static struct dev_mc_list
*bond_mc_list_find_dmi(struct dev_mc_list
*dmi
, struct dev_mc_list
*mc_list
)
764 struct dev_mc_list
*idmi
;
766 for (idmi
= mc_list
; idmi
; idmi
= idmi
->next
) {
767 if (bond_is_dmi_same(dmi
, idmi
)) {
776 * Push the promiscuity flag down to appropriate slaves
778 static void bond_set_promiscuity(struct bonding
*bond
, int inc
)
780 if (USES_PRIMARY(bond
->params
.mode
)) {
781 /* write lock already acquired */
782 if (bond
->curr_active_slave
) {
783 dev_set_promiscuity(bond
->curr_active_slave
->dev
, inc
);
788 bond_for_each_slave(bond
, slave
, i
) {
789 dev_set_promiscuity(slave
->dev
, inc
);
795 * Push the allmulti flag down to all slaves
797 static void bond_set_allmulti(struct bonding
*bond
, int inc
)
799 if (USES_PRIMARY(bond
->params
.mode
)) {
800 /* write lock already acquired */
801 if (bond
->curr_active_slave
) {
802 dev_set_allmulti(bond
->curr_active_slave
->dev
, inc
);
807 bond_for_each_slave(bond
, slave
, i
) {
808 dev_set_allmulti(slave
->dev
, inc
);
814 * Add a Multicast address to slaves
817 static void bond_mc_add(struct bonding
*bond
, void *addr
, int alen
)
819 if (USES_PRIMARY(bond
->params
.mode
)) {
820 /* write lock already acquired */
821 if (bond
->curr_active_slave
) {
822 dev_mc_add(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
827 bond_for_each_slave(bond
, slave
, i
) {
828 dev_mc_add(slave
->dev
, addr
, alen
, 0);
834 * Remove a multicast address from slave
837 static void bond_mc_delete(struct bonding
*bond
, void *addr
, int alen
)
839 if (USES_PRIMARY(bond
->params
.mode
)) {
840 /* write lock already acquired */
841 if (bond
->curr_active_slave
) {
842 dev_mc_delete(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
847 bond_for_each_slave(bond
, slave
, i
) {
848 dev_mc_delete(slave
->dev
, addr
, alen
, 0);
854 * Totally destroys the mc_list in bond
856 static void bond_mc_list_destroy(struct bonding
*bond
)
858 struct dev_mc_list
*dmi
;
862 bond
->mc_list
= dmi
->next
;
869 * Copy all the Multicast addresses from src to the bonding device dst
871 static int bond_mc_list_copy(struct dev_mc_list
*mc_list
, struct bonding
*bond
,
874 struct dev_mc_list
*dmi
, *new_dmi
;
876 for (dmi
= mc_list
; dmi
; dmi
= dmi
->next
) {
877 new_dmi
= kmalloc(sizeof(struct dev_mc_list
), gfp_flag
);
880 /* FIXME: Potential memory leak !!! */
884 new_dmi
->next
= bond
->mc_list
;
885 bond
->mc_list
= new_dmi
;
886 new_dmi
->dmi_addrlen
= dmi
->dmi_addrlen
;
887 memcpy(new_dmi
->dmi_addr
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
888 new_dmi
->dmi_users
= dmi
->dmi_users
;
889 new_dmi
->dmi_gusers
= dmi
->dmi_gusers
;
896 * flush all members of flush->mc_list from device dev->mc_list
898 static void bond_mc_list_flush(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
900 struct bonding
*bond
= bond_dev
->priv
;
901 struct dev_mc_list
*dmi
;
903 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
904 dev_mc_delete(slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
907 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
908 /* del lacpdu mc addr from mc list */
909 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
911 dev_mc_delete(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
915 /*--------------------------- Active slave change ---------------------------*/
918 * Update the mc list and multicast-related flags for the new and
919 * old active slaves (if any) according to the multicast mode, and
920 * promiscuous flags unconditionally.
922 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
, struct slave
*old_active
)
924 struct dev_mc_list
*dmi
;
926 if (!USES_PRIMARY(bond
->params
.mode
)) {
927 /* nothing to do - mc list is already up-to-date on
934 if (bond
->dev
->flags
& IFF_PROMISC
) {
935 dev_set_promiscuity(old_active
->dev
, -1);
938 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
939 dev_set_allmulti(old_active
->dev
, -1);
942 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
943 dev_mc_delete(old_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
948 if (bond
->dev
->flags
& IFF_PROMISC
) {
949 dev_set_promiscuity(new_active
->dev
, 1);
952 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
953 dev_set_allmulti(new_active
->dev
, 1);
956 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
957 dev_mc_add(new_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
963 * find_best_interface - select the best available slave to be the active one
964 * @bond: our bonding struct
966 * Warning: Caller must hold curr_slave_lock for writing.
968 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
970 struct slave
*new_active
, *old_active
;
971 struct slave
*bestslave
= NULL
;
972 int mintime
= bond
->params
.updelay
;
975 new_active
= old_active
= bond
->curr_active_slave
;
977 if (!new_active
) { /* there were no active slaves left */
978 if (bond
->slave_cnt
> 0) { /* found one slave */
979 new_active
= bond
->first_slave
;
981 return NULL
; /* still no slave, return NULL */
985 /* first try the primary link; if arping, a link must tx/rx traffic
986 * before it can be considered the curr_active_slave - also, we would skip
987 * slaves between the curr_active_slave and primary_slave that may be up
990 if ((bond
->primary_slave
) &&
991 (!bond
->params
.arp_interval
) &&
992 (IS_UP(bond
->primary_slave
->dev
))) {
993 new_active
= bond
->primary_slave
;
996 /* remember where to stop iterating over the slaves */
997 old_active
= new_active
;
999 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1000 if (IS_UP(new_active
->dev
)) {
1001 if (new_active
->link
== BOND_LINK_UP
) {
1003 } else if (new_active
->link
== BOND_LINK_BACK
) {
1004 /* link up, but waiting for stabilization */
1005 if (new_active
->delay
< mintime
) {
1006 mintime
= new_active
->delay
;
1007 bestslave
= new_active
;
1017 * change_active_interface - change the active slave into the specified one
1018 * @bond: our bonding struct
1019 * @new: the new slave to make the active one
1021 * Set the new slave to the bond's settings and unset them on the old
1022 * curr_active_slave.
1023 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1025 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1026 * because it is apparently the best available slave we have, even though its
1027 * updelay hasn't timed out yet.
1029 * Warning: Caller must hold curr_slave_lock for writing.
1031 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1033 struct slave
*old_active
= bond
->curr_active_slave
;
1035 if (old_active
== new_active
) {
1040 if (new_active
->link
== BOND_LINK_BACK
) {
1041 if (USES_PRIMARY(bond
->params
.mode
)) {
1042 printk(KERN_INFO DRV_NAME
1043 ": %s: making interface %s the new "
1044 "active one %d ms earlier.\n",
1045 bond
->dev
->name
, new_active
->dev
->name
,
1046 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1049 new_active
->delay
= 0;
1050 new_active
->link
= BOND_LINK_UP
;
1051 new_active
->jiffies
= jiffies
;
1053 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1054 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1057 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1058 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1059 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1062 if (USES_PRIMARY(bond
->params
.mode
)) {
1063 printk(KERN_INFO DRV_NAME
1064 ": %s: making interface %s the new "
1066 bond
->dev
->name
, new_active
->dev
->name
);
1071 if (USES_PRIMARY(bond
->params
.mode
)) {
1072 bond_mc_swap(bond
, new_active
, old_active
);
1075 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1076 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1077 bond_alb_handle_active_change(bond
, new_active
);
1079 bond_set_slave_inactive_flags(old_active
);
1081 bond_set_slave_active_flags(new_active
);
1083 bond
->curr_active_slave
= new_active
;
1086 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1088 bond_set_slave_inactive_flags(old_active
);
1092 bond_set_slave_active_flags(new_active
);
1094 bond_send_gratuitous_arp(bond
);
1099 * bond_select_active_slave - select a new active slave, if needed
1100 * @bond: our bonding struct
1102 * This functions shoud be called when one of the following occurs:
1103 * - The old curr_active_slave has been released or lost its link.
1104 * - The primary_slave has got its link back.
1105 * - A slave has got its link back and there's no old curr_active_slave.
1107 * Warning: Caller must hold curr_slave_lock for writing.
1109 void bond_select_active_slave(struct bonding
*bond
)
1111 struct slave
*best_slave
;
1114 best_slave
= bond_find_best_slave(bond
);
1115 if (best_slave
!= bond
->curr_active_slave
) {
1116 bond_change_active_slave(bond
, best_slave
);
1117 rv
= bond_set_carrier(bond
);
1121 if (netif_carrier_ok(bond
->dev
)) {
1122 printk(KERN_INFO DRV_NAME
1123 ": %s: first active interface up!\n",
1126 printk(KERN_INFO DRV_NAME
": %s: "
1127 "now running without any active interface !\n",
1133 /*--------------------------- slave list handling ---------------------------*/
1136 * This function attaches the slave to the end of list.
1138 * bond->lock held for writing by caller.
1140 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1142 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1143 new_slave
->next
= new_slave
;
1144 new_slave
->prev
= new_slave
;
1145 bond
->first_slave
= new_slave
;
1147 new_slave
->next
= bond
->first_slave
;
1148 new_slave
->prev
= bond
->first_slave
->prev
;
1149 new_slave
->next
->prev
= new_slave
;
1150 new_slave
->prev
->next
= new_slave
;
1157 * This function detaches the slave from the list.
1158 * WARNING: no check is made to verify if the slave effectively
1159 * belongs to <bond>.
1160 * Nothing is freed on return, structures are just unchained.
1161 * If any slave pointer in bond was pointing to <slave>,
1162 * it should be changed by the calling function.
1164 * bond->lock held for writing by caller.
1166 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1169 slave
->next
->prev
= slave
->prev
;
1173 slave
->prev
->next
= slave
->next
;
1176 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1177 if (bond
->slave_cnt
> 1) { /* there are more slave */
1178 bond
->first_slave
= slave
->next
;
1180 bond
->first_slave
= NULL
; /* slave was the last one */
1189 /*---------------------------------- IOCTL ----------------------------------*/
1191 int bond_sethwaddr(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1193 dprintk("bond_dev=%p\n", bond_dev
);
1194 dprintk("slave_dev=%p\n", slave_dev
);
1195 dprintk("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1196 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1200 #define BOND_INTERSECT_FEATURES \
1201 (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_TSO | NETIF_F_UFO)
1204 * Compute the common dev->feature set available to all slaves. Some
1205 * feature bits are managed elsewhere, so preserve feature bits set on
1206 * master device that are not part of the examined set.
1208 static int bond_compute_features(struct bonding
*bond
)
1210 unsigned long features
= BOND_INTERSECT_FEATURES
;
1211 struct slave
*slave
;
1212 struct net_device
*bond_dev
= bond
->dev
;
1215 bond_for_each_slave(bond
, slave
, i
)
1216 features
&= (slave
->dev
->features
& BOND_INTERSECT_FEATURES
);
1218 if ((features
& NETIF_F_SG
) &&
1219 !(features
& NETIF_F_ALL_CSUM
))
1220 features
&= ~NETIF_F_SG
;
1223 * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1224 * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1225 * implies that all slaves also support scatter-gather
1226 * (NETIF_F_SG), which implies that features also includes
1227 * NETIF_F_SG. So no need to check whether we have an
1228 * illegal combination of NETIF_F_{TSO,UFO} and
1232 features
|= (bond_dev
->features
& ~BOND_INTERSECT_FEATURES
);
1233 bond_dev
->features
= features
;
1238 /* enslave device <slave> to bond device <master> */
1239 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1241 struct bonding
*bond
= bond_dev
->priv
;
1242 struct slave
*new_slave
= NULL
;
1243 struct dev_mc_list
*dmi
;
1244 struct sockaddr addr
;
1246 int old_features
= bond_dev
->features
;
1249 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1250 slave_dev
->do_ioctl
== NULL
) {
1251 printk(KERN_WARNING DRV_NAME
1252 ": %s: Warning: no link monitoring support for %s\n",
1253 bond_dev
->name
, slave_dev
->name
);
1256 /* bond must be initialized by bond_open() before enslaving */
1257 if (!(bond_dev
->flags
& IFF_UP
)) {
1258 dprintk("Error, master_dev is not up\n");
1262 /* already enslaved */
1263 if (slave_dev
->flags
& IFF_SLAVE
) {
1264 dprintk("Error, Device was already enslaved\n");
1268 /* vlan challenged mutual exclusion */
1269 /* no need to lock since we're protected by rtnl_lock */
1270 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1271 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1272 if (!list_empty(&bond
->vlan_list
)) {
1273 printk(KERN_ERR DRV_NAME
1274 ": %s: Error: cannot enslave VLAN "
1275 "challenged slave %s on VLAN enabled "
1276 "bond %s\n", bond_dev
->name
, slave_dev
->name
,
1280 printk(KERN_WARNING DRV_NAME
1281 ": %s: Warning: enslaved VLAN challenged "
1282 "slave %s. Adding VLANs will be blocked as "
1283 "long as %s is part of bond %s\n",
1284 bond_dev
->name
, slave_dev
->name
, slave_dev
->name
,
1286 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1289 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1290 if (bond
->slave_cnt
== 0) {
1291 /* First slave, and it is not VLAN challenged,
1292 * so remove the block of adding VLANs over the bond.
1294 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1299 * Old ifenslave binaries are no longer supported. These can
1300 * be identified with moderate accurary by the state of the slave:
1301 * the current ifenslave will set the interface down prior to
1302 * enslaving it; the old ifenslave will not.
1304 if ((slave_dev
->flags
& IFF_UP
)) {
1305 printk(KERN_ERR DRV_NAME
": %s is up. "
1306 "This may be due to an out of date ifenslave.\n",
1309 goto err_undo_flags
;
1312 if (slave_dev
->set_mac_address
== NULL
) {
1313 printk(KERN_ERR DRV_NAME
1314 ": %s: Error: The slave device you specified does "
1315 "not support setting the MAC address. "
1316 "Your kernel likely does not support slave "
1317 "devices.\n", bond_dev
->name
);
1319 goto err_undo_flags
;
1322 new_slave
= kmalloc(sizeof(struct slave
), GFP_KERNEL
);
1325 goto err_undo_flags
;
1328 memset(new_slave
, 0, sizeof(struct slave
));
1330 /* save slave's original flags before calling
1331 * netdev_set_master and dev_open
1333 new_slave
->original_flags
= slave_dev
->flags
;
1336 * Save slave's original ("permanent") mac address for modes
1337 * that need it, and for restoring it upon release, and then
1338 * set it to the master's address
1340 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1343 * Set slave to master's mac address. The application already
1344 * set the master's mac address to that of the first slave
1346 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1347 addr
.sa_family
= slave_dev
->type
;
1348 res
= dev_set_mac_address(slave_dev
, &addr
);
1350 dprintk("Error %d calling set_mac_address\n", res
);
1354 /* open the slave since the application closed it */
1355 res
= dev_open(slave_dev
);
1357 dprintk("Openning slave %s failed\n", slave_dev
->name
);
1358 goto err_restore_mac
;
1361 res
= netdev_set_master(slave_dev
, bond_dev
);
1363 dprintk("Error %d calling netdev_set_master\n", res
);
1367 new_slave
->dev
= slave_dev
;
1369 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1370 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1371 /* bond_alb_init_slave() must be called before all other stages since
1372 * it might fail and we do not want to have to undo everything
1374 res
= bond_alb_init_slave(bond
, new_slave
);
1376 goto err_unset_master
;
1380 /* If the mode USES_PRIMARY, then the new slave gets the
1381 * master's promisc (and mc) settings only if it becomes the
1382 * curr_active_slave, and that is taken care of later when calling
1383 * bond_change_active()
1385 if (!USES_PRIMARY(bond
->params
.mode
)) {
1386 /* set promiscuity level to new slave */
1387 if (bond_dev
->flags
& IFF_PROMISC
) {
1388 dev_set_promiscuity(slave_dev
, 1);
1391 /* set allmulti level to new slave */
1392 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1393 dev_set_allmulti(slave_dev
, 1);
1396 /* upload master's mc_list to new slave */
1397 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
1398 dev_mc_add (slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
1402 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1403 /* add lacpdu mc addr to mc list */
1404 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1406 dev_mc_add(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
1409 bond_add_vlans_on_slave(bond
, slave_dev
);
1411 write_lock_bh(&bond
->lock
);
1413 bond_attach_slave(bond
, new_slave
);
1415 new_slave
->delay
= 0;
1416 new_slave
->link_failure_count
= 0;
1418 bond_compute_features(bond
);
1420 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1421 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1423 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1425 * miimon is set but a bonded network driver
1426 * does not support ETHTOOL/MII and
1427 * arp_interval is not set. Note: if
1428 * use_carrier is enabled, we will never go
1429 * here (because netif_carrier is always
1430 * supported); thus, we don't need to change
1431 * the messages for netif_carrier.
1433 printk(KERN_WARNING DRV_NAME
1434 ": %s: Warning: MII and ETHTOOL support not "
1435 "available for interface %s, and "
1436 "arp_interval/arp_ip_target module parameters "
1437 "not specified, thus bonding will not detect "
1438 "link failures! see bonding.txt for details.\n",
1439 bond_dev
->name
, slave_dev
->name
);
1440 } else if (link_reporting
== -1) {
1441 /* unable get link status using mii/ethtool */
1442 printk(KERN_WARNING DRV_NAME
1443 ": %s: Warning: can't get link status from "
1444 "interface %s; the network driver associated "
1445 "with this interface does not support MII or "
1446 "ETHTOOL link status reporting, thus miimon "
1447 "has no effect on this interface.\n",
1448 bond_dev
->name
, slave_dev
->name
);
1452 /* check for initial state */
1453 if (!bond
->params
.miimon
||
1454 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1455 if (bond
->params
.updelay
) {
1456 dprintk("Initial state of slave_dev is "
1457 "BOND_LINK_BACK\n");
1458 new_slave
->link
= BOND_LINK_BACK
;
1459 new_slave
->delay
= bond
->params
.updelay
;
1461 dprintk("Initial state of slave_dev is "
1463 new_slave
->link
= BOND_LINK_UP
;
1465 new_slave
->jiffies
= jiffies
;
1467 dprintk("Initial state of slave_dev is "
1468 "BOND_LINK_DOWN\n");
1469 new_slave
->link
= BOND_LINK_DOWN
;
1472 if (bond_update_speed_duplex(new_slave
) &&
1473 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1474 printk(KERN_WARNING DRV_NAME
1475 ": %s: Warning: failed to get speed and duplex from %s, "
1476 "assumed to be 100Mb/sec and Full.\n",
1477 bond_dev
->name
, new_slave
->dev
->name
);
1479 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1480 printk(KERN_WARNING DRV_NAME
1481 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1482 "support in base driver for proper aggregator "
1483 "selection.\n", bond_dev
->name
);
1487 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1488 /* if there is a primary slave, remember it */
1489 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1490 bond
->primary_slave
= new_slave
;
1494 switch (bond
->params
.mode
) {
1495 case BOND_MODE_ACTIVEBACKUP
:
1496 /* if we're in active-backup mode, we need one and
1497 * only one active interface. The backup interfaces
1498 * will have their SLAVE_INACTIVE flag set because we
1499 * need them to be drop all packets. Thus, since we
1500 * guarantee that curr_active_slave always point to
1501 * the last usable interface, we just have to verify
1502 * this interface's flag.
1504 if (((!bond
->curr_active_slave
) ||
1505 (bond
->curr_active_slave
->dev
->priv_flags
& IFF_SLAVE_INACTIVE
)) &&
1506 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1507 /* first slave or no active slave yet, and this link
1508 is OK, so make this interface the active one */
1509 bond_change_active_slave(bond
, new_slave
);
1510 printk(KERN_INFO DRV_NAME
1511 ": %s: first active interface up!\n",
1513 netif_carrier_on(bond
->dev
);
1516 dprintk("This is just a backup slave\n");
1517 bond_set_slave_inactive_flags(new_slave
);
1520 case BOND_MODE_8023AD
:
1521 /* in 802.3ad mode, the internal mechanism
1522 * will activate the slaves in the selected
1525 bond_set_slave_inactive_flags(new_slave
);
1526 /* if this is the first slave */
1527 if (bond
->slave_cnt
== 1) {
1528 SLAVE_AD_INFO(new_slave
).id
= 1;
1529 /* Initialize AD with the number of times that the AD timer is called in 1 second
1530 * can be called only after the mac address of the bond is set
1532 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1533 bond
->params
.lacp_fast
);
1535 SLAVE_AD_INFO(new_slave
).id
=
1536 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1539 bond_3ad_bind_slave(new_slave
);
1543 new_slave
->state
= BOND_STATE_ACTIVE
;
1544 if ((!bond
->curr_active_slave
) &&
1545 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1546 /* first slave or no active slave yet, and this link
1547 * is OK, so make this interface the active one
1549 bond_change_active_slave(bond
, new_slave
);
1551 bond_set_slave_inactive_flags(new_slave
);
1555 dprintk("This slave is always active in trunk mode\n");
1557 /* always active in trunk mode */
1558 new_slave
->state
= BOND_STATE_ACTIVE
;
1560 /* In trunking mode there is little meaning to curr_active_slave
1561 * anyway (it holds no special properties of the bond device),
1562 * so we can change it without calling change_active_interface()
1564 if (!bond
->curr_active_slave
) {
1565 bond
->curr_active_slave
= new_slave
;
1568 } /* switch(bond_mode) */
1570 bond_set_carrier(bond
);
1572 write_unlock_bh(&bond
->lock
);
1574 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1576 goto err_unset_master
;
1578 printk(KERN_INFO DRV_NAME
1579 ": %s: enslaving %s as a%s interface with a%s link.\n",
1580 bond_dev
->name
, slave_dev
->name
,
1581 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1582 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1584 /* enslave is successful */
1587 /* Undo stages on error */
1589 netdev_set_master(slave_dev
, NULL
);
1592 dev_close(slave_dev
);
1595 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1596 addr
.sa_family
= slave_dev
->type
;
1597 dev_set_mac_address(slave_dev
, &addr
);
1603 bond_dev
->features
= old_features
;
1609 * Try to release the slave device <slave> from the bond device <master>
1610 * It is legal to access curr_active_slave without a lock because all the function
1613 * The rules for slave state should be:
1614 * for Active/Backup:
1615 * Active stays on all backups go down
1616 * for Bonded connections:
1617 * The first up interface should be left on and all others downed.
1619 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1621 struct bonding
*bond
= bond_dev
->priv
;
1622 struct slave
*slave
, *oldcurrent
;
1623 struct sockaddr addr
;
1624 int mac_addr_differ
;
1626 /* slave is not a slave or master is not master of this slave */
1627 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1628 (slave_dev
->master
!= bond_dev
)) {
1629 printk(KERN_ERR DRV_NAME
1630 ": %s: Error: cannot release %s.\n",
1631 bond_dev
->name
, slave_dev
->name
);
1635 write_lock_bh(&bond
->lock
);
1637 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1639 /* not a slave of this bond */
1640 printk(KERN_INFO DRV_NAME
1641 ": %s: %s not enslaved\n",
1642 bond_dev
->name
, slave_dev
->name
);
1643 write_unlock_bh(&bond
->lock
);
1647 mac_addr_differ
= memcmp(bond_dev
->dev_addr
,
1650 if (!mac_addr_differ
&& (bond
->slave_cnt
> 1)) {
1651 printk(KERN_WARNING DRV_NAME
1652 ": %s: Warning: the permanent HWaddr of %s "
1653 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1654 "still in use by %s. Set the HWaddr of "
1655 "%s to a different address to avoid "
1659 slave
->perm_hwaddr
[0],
1660 slave
->perm_hwaddr
[1],
1661 slave
->perm_hwaddr
[2],
1662 slave
->perm_hwaddr
[3],
1663 slave
->perm_hwaddr
[4],
1664 slave
->perm_hwaddr
[5],
1669 /* Inform AD package of unbinding of slave. */
1670 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1671 /* must be called before the slave is
1672 * detached from the list
1674 bond_3ad_unbind_slave(slave
);
1677 printk(KERN_INFO DRV_NAME
1678 ": %s: releasing %s interface %s\n",
1680 (slave
->state
== BOND_STATE_ACTIVE
)
1681 ? "active" : "backup",
1684 oldcurrent
= bond
->curr_active_slave
;
1686 bond
->current_arp_slave
= NULL
;
1688 /* release the slave from its bond */
1689 bond_detach_slave(bond
, slave
);
1691 bond_compute_features(bond
);
1693 if (bond
->primary_slave
== slave
) {
1694 bond
->primary_slave
= NULL
;
1697 if (oldcurrent
== slave
) {
1698 bond_change_active_slave(bond
, NULL
);
1701 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1702 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1703 /* Must be called only after the slave has been
1704 * detached from the list and the curr_active_slave
1705 * has been cleared (if our_slave == old_current),
1706 * but before a new active slave is selected.
1708 bond_alb_deinit_slave(bond
, slave
);
1711 if (oldcurrent
== slave
)
1712 bond_select_active_slave(bond
);
1714 if (bond
->slave_cnt
== 0) {
1715 bond_set_carrier(bond
);
1717 /* if the last slave was removed, zero the mac address
1718 * of the master so it will be set by the application
1719 * to the mac address of the first slave
1721 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1723 if (list_empty(&bond
->vlan_list
)) {
1724 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1726 printk(KERN_WARNING DRV_NAME
1727 ": %s: Warning: clearing HW address of %s while it "
1728 "still has VLANs.\n",
1729 bond_dev
->name
, bond_dev
->name
);
1730 printk(KERN_WARNING DRV_NAME
1731 ": %s: When re-adding slaves, make sure the bond's "
1732 "HW address matches its VLANs'.\n",
1735 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1736 !bond_has_challenged_slaves(bond
)) {
1737 printk(KERN_INFO DRV_NAME
1738 ": %s: last VLAN challenged slave %s "
1739 "left bond %s. VLAN blocking is removed\n",
1740 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1741 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1744 write_unlock_bh(&bond
->lock
);
1746 /* must do this from outside any spinlocks */
1747 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1749 bond_del_vlans_from_slave(bond
, slave_dev
);
1751 /* If the mode USES_PRIMARY, then we should only remove its
1752 * promisc and mc settings if it was the curr_active_slave, but that was
1753 * already taken care of above when we detached the slave
1755 if (!USES_PRIMARY(bond
->params
.mode
)) {
1756 /* unset promiscuity level from slave */
1757 if (bond_dev
->flags
& IFF_PROMISC
) {
1758 dev_set_promiscuity(slave_dev
, -1);
1761 /* unset allmulti level from slave */
1762 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1763 dev_set_allmulti(slave_dev
, -1);
1766 /* flush master's mc_list from slave */
1767 bond_mc_list_flush(bond_dev
, slave_dev
);
1770 netdev_set_master(slave_dev
, NULL
);
1772 /* close slave before restoring its mac address */
1773 dev_close(slave_dev
);
1775 /* restore original ("permanent") mac address */
1776 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1777 addr
.sa_family
= slave_dev
->type
;
1778 dev_set_mac_address(slave_dev
, &addr
);
1780 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
1781 IFF_SLAVE_INACTIVE
);
1785 return 0; /* deletion OK */
1789 * This function releases all slaves.
1791 static int bond_release_all(struct net_device
*bond_dev
)
1793 struct bonding
*bond
= bond_dev
->priv
;
1794 struct slave
*slave
;
1795 struct net_device
*slave_dev
;
1796 struct sockaddr addr
;
1798 write_lock_bh(&bond
->lock
);
1800 netif_carrier_off(bond_dev
);
1802 if (bond
->slave_cnt
== 0) {
1806 bond
->current_arp_slave
= NULL
;
1807 bond
->primary_slave
= NULL
;
1808 bond_change_active_slave(bond
, NULL
);
1810 while ((slave
= bond
->first_slave
) != NULL
) {
1811 /* Inform AD package of unbinding of slave
1812 * before slave is detached from the list.
1814 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1815 bond_3ad_unbind_slave(slave
);
1818 slave_dev
= slave
->dev
;
1819 bond_detach_slave(bond
, slave
);
1821 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1822 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1823 /* must be called only after the slave
1824 * has been detached from the list
1826 bond_alb_deinit_slave(bond
, slave
);
1829 bond_compute_features(bond
);
1831 /* now that the slave is detached, unlock and perform
1832 * all the undo steps that should not be called from
1835 write_unlock_bh(&bond
->lock
);
1837 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1838 bond_del_vlans_from_slave(bond
, slave_dev
);
1840 /* If the mode USES_PRIMARY, then we should only remove its
1841 * promisc and mc settings if it was the curr_active_slave, but that was
1842 * already taken care of above when we detached the slave
1844 if (!USES_PRIMARY(bond
->params
.mode
)) {
1845 /* unset promiscuity level from slave */
1846 if (bond_dev
->flags
& IFF_PROMISC
) {
1847 dev_set_promiscuity(slave_dev
, -1);
1850 /* unset allmulti level from slave */
1851 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1852 dev_set_allmulti(slave_dev
, -1);
1855 /* flush master's mc_list from slave */
1856 bond_mc_list_flush(bond_dev
, slave_dev
);
1859 netdev_set_master(slave_dev
, NULL
);
1861 /* close slave before restoring its mac address */
1862 dev_close(slave_dev
);
1864 /* restore original ("permanent") mac address*/
1865 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1866 addr
.sa_family
= slave_dev
->type
;
1867 dev_set_mac_address(slave_dev
, &addr
);
1869 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
1870 IFF_SLAVE_INACTIVE
);
1874 /* re-acquire the lock before getting the next slave */
1875 write_lock_bh(&bond
->lock
);
1878 /* zero the mac address of the master so it will be
1879 * set by the application to the mac address of the
1882 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1884 if (list_empty(&bond
->vlan_list
)) {
1885 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1887 printk(KERN_WARNING DRV_NAME
1888 ": %s: Warning: clearing HW address of %s while it "
1889 "still has VLANs.\n",
1890 bond_dev
->name
, bond_dev
->name
);
1891 printk(KERN_WARNING DRV_NAME
1892 ": %s: When re-adding slaves, make sure the bond's "
1893 "HW address matches its VLANs'.\n",
1897 printk(KERN_INFO DRV_NAME
1898 ": %s: released all slaves\n",
1902 write_unlock_bh(&bond
->lock
);
1908 * This function changes the active slave to slave <slave_dev>.
1909 * It returns -EINVAL in the following cases.
1910 * - <slave_dev> is not found in the list.
1911 * - There is not active slave now.
1912 * - <slave_dev> is already active.
1913 * - The link state of <slave_dev> is not BOND_LINK_UP.
1914 * - <slave_dev> is not running.
1915 * In these cases, this fuction does nothing.
1916 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1918 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1920 struct bonding
*bond
= bond_dev
->priv
;
1921 struct slave
*old_active
= NULL
;
1922 struct slave
*new_active
= NULL
;
1925 if (!USES_PRIMARY(bond
->params
.mode
)) {
1929 /* Verify that master_dev is indeed the master of slave_dev */
1930 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1931 (slave_dev
->master
!= bond_dev
)) {
1935 write_lock_bh(&bond
->lock
);
1937 old_active
= bond
->curr_active_slave
;
1938 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
1941 * Changing to the current active: do nothing; return success.
1943 if (new_active
&& (new_active
== old_active
)) {
1944 write_unlock_bh(&bond
->lock
);
1950 (new_active
->link
== BOND_LINK_UP
) &&
1951 IS_UP(new_active
->dev
)) {
1952 bond_change_active_slave(bond
, new_active
);
1957 write_unlock_bh(&bond
->lock
);
1962 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1964 struct bonding
*bond
= bond_dev
->priv
;
1966 info
->bond_mode
= bond
->params
.mode
;
1967 info
->miimon
= bond
->params
.miimon
;
1969 read_lock_bh(&bond
->lock
);
1970 info
->num_slaves
= bond
->slave_cnt
;
1971 read_unlock_bh(&bond
->lock
);
1976 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
1978 struct bonding
*bond
= bond_dev
->priv
;
1979 struct slave
*slave
;
1982 if (info
->slave_id
< 0) {
1986 read_lock_bh(&bond
->lock
);
1988 bond_for_each_slave(bond
, slave
, i
) {
1989 if (i
== (int)info
->slave_id
) {
1995 read_unlock_bh(&bond
->lock
);
1998 strcpy(info
->slave_name
, slave
->dev
->name
);
1999 info
->link
= slave
->link
;
2000 info
->state
= slave
->state
;
2001 info
->link_failure_count
= slave
->link_failure_count
;
2009 /*-------------------------------- Monitoring -------------------------------*/
2011 /* this function is called regularly to monitor each slave's link. */
2012 void bond_mii_monitor(struct net_device
*bond_dev
)
2014 struct bonding
*bond
= bond_dev
->priv
;
2015 struct slave
*slave
, *oldcurrent
;
2016 int do_failover
= 0;
2020 read_lock(&bond
->lock
);
2022 delta_in_ticks
= (bond
->params
.miimon
* HZ
) / 1000;
2024 if (bond
->kill_timers
) {
2028 if (bond
->slave_cnt
== 0) {
2032 /* we will try to read the link status of each of our slaves, and
2033 * set their IFF_RUNNING flag appropriately. For each slave not
2034 * supporting MII status, we won't do anything so that a user-space
2035 * program could monitor the link itself if needed.
2038 read_lock(&bond
->curr_slave_lock
);
2039 oldcurrent
= bond
->curr_active_slave
;
2040 read_unlock(&bond
->curr_slave_lock
);
2042 bond_for_each_slave(bond
, slave
, i
) {
2043 struct net_device
*slave_dev
= slave
->dev
;
2045 u16 old_speed
= slave
->speed
;
2046 u8 old_duplex
= slave
->duplex
;
2048 link_state
= bond_check_dev_link(bond
, slave_dev
, 0);
2050 switch (slave
->link
) {
2051 case BOND_LINK_UP
: /* the link was up */
2052 if (link_state
== BMSR_LSTATUS
) {
2053 /* link stays up, nothing more to do */
2055 } else { /* link going down */
2056 slave
->link
= BOND_LINK_FAIL
;
2057 slave
->delay
= bond
->params
.downdelay
;
2059 if (slave
->link_failure_count
< UINT_MAX
) {
2060 slave
->link_failure_count
++;
2063 if (bond
->params
.downdelay
) {
2064 printk(KERN_INFO DRV_NAME
2065 ": %s: link status down for %s "
2066 "interface %s, disabling it in "
2070 ? ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
2071 ? ((slave
== oldcurrent
)
2072 ? "active " : "backup ")
2076 bond
->params
.downdelay
* bond
->params
.miimon
);
2079 /* no break ! fall through the BOND_LINK_FAIL test to
2080 ensure proper action to be taken
2082 case BOND_LINK_FAIL
: /* the link has just gone down */
2083 if (link_state
!= BMSR_LSTATUS
) {
2084 /* link stays down */
2085 if (slave
->delay
<= 0) {
2086 /* link down for too long time */
2087 slave
->link
= BOND_LINK_DOWN
;
2089 /* in active/backup mode, we must
2090 * completely disable this interface
2092 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) ||
2093 (bond
->params
.mode
== BOND_MODE_8023AD
)) {
2094 bond_set_slave_inactive_flags(slave
);
2097 printk(KERN_INFO DRV_NAME
2098 ": %s: link status definitely "
2099 "down for interface %s, "
2104 /* notify ad that the link status has changed */
2105 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2106 bond_3ad_handle_link_change(slave
, BOND_LINK_DOWN
);
2109 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
2110 (bond
->params
.mode
== BOND_MODE_ALB
)) {
2111 bond_alb_handle_link_change(bond
, slave
, BOND_LINK_DOWN
);
2114 if (slave
== oldcurrent
) {
2122 slave
->link
= BOND_LINK_UP
;
2123 slave
->jiffies
= jiffies
;
2124 printk(KERN_INFO DRV_NAME
2125 ": %s: link status up again after %d "
2126 "ms for interface %s.\n",
2128 (bond
->params
.downdelay
- slave
->delay
) * bond
->params
.miimon
,
2132 case BOND_LINK_DOWN
: /* the link was down */
2133 if (link_state
!= BMSR_LSTATUS
) {
2134 /* the link stays down, nothing more to do */
2136 } else { /* link going up */
2137 slave
->link
= BOND_LINK_BACK
;
2138 slave
->delay
= bond
->params
.updelay
;
2140 if (bond
->params
.updelay
) {
2141 /* if updelay == 0, no need to
2142 advertise about a 0 ms delay */
2143 printk(KERN_INFO DRV_NAME
2144 ": %s: link status up for "
2145 "interface %s, enabling it "
2149 bond
->params
.updelay
* bond
->params
.miimon
);
2152 /* no break ! fall through the BOND_LINK_BACK state in
2153 case there's something to do.
2155 case BOND_LINK_BACK
: /* the link has just come back */
2156 if (link_state
!= BMSR_LSTATUS
) {
2157 /* link down again */
2158 slave
->link
= BOND_LINK_DOWN
;
2160 printk(KERN_INFO DRV_NAME
2161 ": %s: link status down again after %d "
2162 "ms for interface %s.\n",
2164 (bond
->params
.updelay
- slave
->delay
) * bond
->params
.miimon
,
2168 if (slave
->delay
== 0) {
2169 /* now the link has been up for long time enough */
2170 slave
->link
= BOND_LINK_UP
;
2171 slave
->jiffies
= jiffies
;
2173 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2174 /* prevent it from being the active one */
2175 slave
->state
= BOND_STATE_BACKUP
;
2176 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2177 /* make it immediately active */
2178 slave
->state
= BOND_STATE_ACTIVE
;
2179 } else if (slave
!= bond
->primary_slave
) {
2180 /* prevent it from being the active one */
2181 slave
->state
= BOND_STATE_BACKUP
;
2184 printk(KERN_INFO DRV_NAME
2185 ": %s: link status definitely "
2186 "up for interface %s.\n",
2190 /* notify ad that the link status has changed */
2191 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2192 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2195 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
2196 (bond
->params
.mode
== BOND_MODE_ALB
)) {
2197 bond_alb_handle_link_change(bond
, slave
, BOND_LINK_UP
);
2200 if ((!oldcurrent
) ||
2201 (slave
== bond
->primary_slave
)) {
2210 /* Should not happen */
2211 printk(KERN_ERR DRV_NAME
2212 ": %s: Error: %s Illegal value (link=%d)\n",
2217 } /* end of switch (slave->link) */
2219 bond_update_speed_duplex(slave
);
2221 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2222 if (old_speed
!= slave
->speed
) {
2223 bond_3ad_adapter_speed_changed(slave
);
2226 if (old_duplex
!= slave
->duplex
) {
2227 bond_3ad_adapter_duplex_changed(slave
);
2234 write_lock(&bond
->curr_slave_lock
);
2236 bond_select_active_slave(bond
);
2238 write_unlock(&bond
->curr_slave_lock
);
2240 bond_set_carrier(bond
);
2243 if (bond
->params
.miimon
) {
2244 mod_timer(&bond
->mii_timer
, jiffies
+ delta_in_ticks
);
2247 read_unlock(&bond
->lock
);
2251 static u32
bond_glean_dev_ip(struct net_device
*dev
)
2253 struct in_device
*idev
;
2254 struct in_ifaddr
*ifa
;
2261 idev
= __in_dev_get_rcu(dev
);
2265 ifa
= idev
->ifa_list
;
2269 addr
= ifa
->ifa_local
;
2275 static int bond_has_ip(struct bonding
*bond
)
2277 struct vlan_entry
*vlan
, *vlan_next
;
2279 if (bond
->master_ip
)
2282 if (list_empty(&bond
->vlan_list
))
2285 list_for_each_entry_safe(vlan
, vlan_next
, &bond
->vlan_list
,
2295 * We go to the (large) trouble of VLAN tagging ARP frames because
2296 * switches in VLAN mode (especially if ports are configured as
2297 * "native" to a VLAN) might not pass non-tagged frames.
2299 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, u32 dest_ip
, u32 src_ip
, unsigned short vlan_id
)
2301 struct sk_buff
*skb
;
2303 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2304 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2306 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2307 NULL
, slave_dev
->dev_addr
, NULL
);
2310 printk(KERN_ERR DRV_NAME
": ARP packet allocation failed\n");
2314 skb
= vlan_put_tag(skb
, vlan_id
);
2316 printk(KERN_ERR DRV_NAME
": failed to insert VLAN tag\n");
2324 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2327 u32
*targets
= bond
->params
.arp_targets
;
2328 struct vlan_entry
*vlan
, *vlan_next
;
2329 struct net_device
*vlan_dev
;
2333 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2336 dprintk("basa: target %x\n", targets
[i
]);
2337 if (list_empty(&bond
->vlan_list
)) {
2338 dprintk("basa: empty vlan: arp_send\n");
2339 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2340 bond
->master_ip
, 0);
2345 * If VLANs are configured, we do a route lookup to
2346 * determine which VLAN interface would be used, so we
2347 * can tag the ARP with the proper VLAN tag.
2349 memset(&fl
, 0, sizeof(fl
));
2350 fl
.fl4_dst
= targets
[i
];
2351 fl
.fl4_tos
= RTO_ONLINK
;
2353 rv
= ip_route_output_key(&rt
, &fl
);
2355 if (net_ratelimit()) {
2356 printk(KERN_WARNING DRV_NAME
2357 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2358 bond
->dev
->name
, NIPQUAD(fl
.fl4_dst
));
2364 * This target is not on a VLAN
2366 if (rt
->u
.dst
.dev
== bond
->dev
) {
2368 dprintk("basa: rtdev == bond->dev: arp_send\n");
2369 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2370 bond
->master_ip
, 0);
2375 list_for_each_entry_safe(vlan
, vlan_next
, &bond
->vlan_list
,
2377 vlan_dev
= bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
];
2378 if (vlan_dev
== rt
->u
.dst
.dev
) {
2379 vlan_id
= vlan
->vlan_id
;
2380 dprintk("basa: vlan match on %s %d\n",
2381 vlan_dev
->name
, vlan_id
);
2388 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2389 vlan
->vlan_ip
, vlan_id
);
2393 if (net_ratelimit()) {
2394 printk(KERN_WARNING DRV_NAME
2395 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2396 bond
->dev
->name
, NIPQUAD(fl
.fl4_dst
),
2397 rt
->u
.dst
.dev
? rt
->u
.dst
.dev
->name
: "NULL");
2404 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2405 * for each VLAN above us.
2407 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2409 struct slave
*slave
= bond
->curr_active_slave
;
2410 struct vlan_entry
*vlan
;
2411 struct net_device
*vlan_dev
;
2413 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond
->dev
->name
,
2414 slave
? slave
->dev
->name
: "NULL");
2418 if (bond
->master_ip
) {
2419 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2420 bond
->master_ip
, 0);
2423 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2424 vlan_dev
= bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
];
2425 if (vlan
->vlan_ip
) {
2426 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2427 vlan
->vlan_ip
, vlan
->vlan_id
);
2433 * this function is called regularly to monitor each slave's link
2434 * ensuring that traffic is being sent and received when arp monitoring
2435 * is used in load-balancing mode. if the adapter has been dormant, then an
2436 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2437 * arp monitoring in active backup mode.
2439 void bond_loadbalance_arp_mon(struct net_device
*bond_dev
)
2441 struct bonding
*bond
= bond_dev
->priv
;
2442 struct slave
*slave
, *oldcurrent
;
2443 int do_failover
= 0;
2447 read_lock(&bond
->lock
);
2449 delta_in_ticks
= (bond
->params
.arp_interval
* HZ
) / 1000;
2451 if (bond
->kill_timers
) {
2455 if (bond
->slave_cnt
== 0) {
2459 read_lock(&bond
->curr_slave_lock
);
2460 oldcurrent
= bond
->curr_active_slave
;
2461 read_unlock(&bond
->curr_slave_lock
);
2463 /* see if any of the previous devices are up now (i.e. they have
2464 * xmt and rcv traffic). the curr_active_slave does not come into
2465 * the picture unless it is null. also, slave->jiffies is not needed
2466 * here because we send an arp on each slave and give a slave as
2467 * long as it needs to get the tx/rx within the delta.
2468 * TODO: what about up/down delay in arp mode? it wasn't here before
2471 bond_for_each_slave(bond
, slave
, i
) {
2472 if (slave
->link
!= BOND_LINK_UP
) {
2473 if (((jiffies
- slave
->dev
->trans_start
) <= delta_in_ticks
) &&
2474 ((jiffies
- slave
->dev
->last_rx
) <= delta_in_ticks
)) {
2476 slave
->link
= BOND_LINK_UP
;
2477 slave
->state
= BOND_STATE_ACTIVE
;
2479 /* primary_slave has no meaning in round-robin
2480 * mode. the window of a slave being up and
2481 * curr_active_slave being null after enslaving
2485 printk(KERN_INFO DRV_NAME
2486 ": %s: link status definitely "
2487 "up for interface %s, ",
2492 printk(KERN_INFO DRV_NAME
2493 ": %s: interface %s is now up\n",
2499 /* slave->link == BOND_LINK_UP */
2501 /* not all switches will respond to an arp request
2502 * when the source ip is 0, so don't take the link down
2503 * if we don't know our ip yet
2505 if (((jiffies
- slave
->dev
->trans_start
) >= (2*delta_in_ticks
)) ||
2506 (((jiffies
- slave
->dev
->last_rx
) >= (2*delta_in_ticks
)) &&
2507 bond_has_ip(bond
))) {
2509 slave
->link
= BOND_LINK_DOWN
;
2510 slave
->state
= BOND_STATE_BACKUP
;
2512 if (slave
->link_failure_count
< UINT_MAX
) {
2513 slave
->link_failure_count
++;
2516 printk(KERN_INFO DRV_NAME
2517 ": %s: interface %s is now down.\n",
2521 if (slave
== oldcurrent
) {
2527 /* note: if switch is in round-robin mode, all links
2528 * must tx arp to ensure all links rx an arp - otherwise
2529 * links may oscillate or not come up at all; if switch is
2530 * in something like xor mode, there is nothing we can
2531 * do - all replies will be rx'ed on same link causing slaves
2532 * to be unstable during low/no traffic periods
2534 if (IS_UP(slave
->dev
)) {
2535 bond_arp_send_all(bond
, slave
);
2540 write_lock(&bond
->curr_slave_lock
);
2542 bond_select_active_slave(bond
);
2544 write_unlock(&bond
->curr_slave_lock
);
2548 if (bond
->params
.arp_interval
) {
2549 mod_timer(&bond
->arp_timer
, jiffies
+ delta_in_ticks
);
2552 read_unlock(&bond
->lock
);
2556 * When using arp monitoring in active-backup mode, this function is
2557 * called to determine if any backup slaves have went down or a new
2558 * current slave needs to be found.
2559 * The backup slaves never generate traffic, they are considered up by merely
2560 * receiving traffic. If the current slave goes down, each backup slave will
2561 * be given the opportunity to tx/rx an arp before being taken down - this
2562 * prevents all slaves from being taken down due to the current slave not
2563 * sending any traffic for the backups to receive. The arps are not necessarily
2564 * necessary, any tx and rx traffic will keep the current slave up. While any
2565 * rx traffic will keep the backup slaves up, the current slave is responsible
2566 * for generating traffic to keep them up regardless of any other traffic they
2567 * may have received.
2568 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2570 void bond_activebackup_arp_mon(struct net_device
*bond_dev
)
2572 struct bonding
*bond
= bond_dev
->priv
;
2573 struct slave
*slave
;
2577 read_lock(&bond
->lock
);
2579 delta_in_ticks
= (bond
->params
.arp_interval
* HZ
) / 1000;
2581 if (bond
->kill_timers
) {
2585 if (bond
->slave_cnt
== 0) {
2589 /* determine if any slave has come up or any backup slave has
2591 * TODO: what about up/down delay in arp mode? it wasn't here before
2594 bond_for_each_slave(bond
, slave
, i
) {
2595 if (slave
->link
!= BOND_LINK_UP
) {
2596 if ((jiffies
- slave
->dev
->last_rx
) <= delta_in_ticks
) {
2598 slave
->link
= BOND_LINK_UP
;
2600 write_lock(&bond
->curr_slave_lock
);
2602 if ((!bond
->curr_active_slave
) &&
2603 ((jiffies
- slave
->dev
->trans_start
) <= delta_in_ticks
)) {
2604 bond_change_active_slave(bond
, slave
);
2605 bond
->current_arp_slave
= NULL
;
2606 } else if (bond
->curr_active_slave
!= slave
) {
2607 /* this slave has just come up but we
2608 * already have a current slave; this
2609 * can also happen if bond_enslave adds
2610 * a new slave that is up while we are
2611 * searching for a new slave
2613 bond_set_slave_inactive_flags(slave
);
2614 bond
->current_arp_slave
= NULL
;
2617 bond_set_carrier(bond
);
2619 if (slave
== bond
->curr_active_slave
) {
2620 printk(KERN_INFO DRV_NAME
2621 ": %s: %s is up and now the "
2622 "active interface\n",
2625 netif_carrier_on(bond
->dev
);
2627 printk(KERN_INFO DRV_NAME
2628 ": %s: backup interface %s is "
2634 write_unlock(&bond
->curr_slave_lock
);
2637 read_lock(&bond
->curr_slave_lock
);
2639 if ((slave
!= bond
->curr_active_slave
) &&
2640 (!bond
->current_arp_slave
) &&
2641 (((jiffies
- slave
->dev
->last_rx
) >= 3*delta_in_ticks
) &&
2642 bond_has_ip(bond
))) {
2643 /* a backup slave has gone down; three times
2644 * the delta allows the current slave to be
2645 * taken out before the backup slave.
2646 * note: a non-null current_arp_slave indicates
2647 * the curr_active_slave went down and we are
2648 * searching for a new one; under this
2649 * condition we only take the curr_active_slave
2650 * down - this gives each slave a chance to
2651 * tx/rx traffic before being taken out
2654 read_unlock(&bond
->curr_slave_lock
);
2656 slave
->link
= BOND_LINK_DOWN
;
2658 if (slave
->link_failure_count
< UINT_MAX
) {
2659 slave
->link_failure_count
++;
2662 bond_set_slave_inactive_flags(slave
);
2664 printk(KERN_INFO DRV_NAME
2665 ": %s: backup interface %s is now down\n",
2669 read_unlock(&bond
->curr_slave_lock
);
2674 read_lock(&bond
->curr_slave_lock
);
2675 slave
= bond
->curr_active_slave
;
2676 read_unlock(&bond
->curr_slave_lock
);
2679 /* if we have sent traffic in the past 2*arp_intervals but
2680 * haven't xmit and rx traffic in that time interval, select
2681 * a different slave. slave->jiffies is only updated when
2682 * a slave first becomes the curr_active_slave - not necessarily
2683 * after every arp; this ensures the slave has a full 2*delta
2684 * before being taken out. if a primary is being used, check
2685 * if it is up and needs to take over as the curr_active_slave
2687 if ((((jiffies
- slave
->dev
->trans_start
) >= (2*delta_in_ticks
)) ||
2688 (((jiffies
- slave
->dev
->last_rx
) >= (2*delta_in_ticks
)) &&
2689 bond_has_ip(bond
))) &&
2690 ((jiffies
- slave
->jiffies
) >= 2*delta_in_ticks
)) {
2692 slave
->link
= BOND_LINK_DOWN
;
2694 if (slave
->link_failure_count
< UINT_MAX
) {
2695 slave
->link_failure_count
++;
2698 printk(KERN_INFO DRV_NAME
2699 ": %s: link status down for active interface "
2700 "%s, disabling it\n",
2704 write_lock(&bond
->curr_slave_lock
);
2706 bond_select_active_slave(bond
);
2707 slave
= bond
->curr_active_slave
;
2709 write_unlock(&bond
->curr_slave_lock
);
2711 bond
->current_arp_slave
= slave
;
2714 slave
->jiffies
= jiffies
;
2716 } else if ((bond
->primary_slave
) &&
2717 (bond
->primary_slave
!= slave
) &&
2718 (bond
->primary_slave
->link
== BOND_LINK_UP
)) {
2719 /* at this point, slave is the curr_active_slave */
2720 printk(KERN_INFO DRV_NAME
2721 ": %s: changing from interface %s to primary "
2725 bond
->primary_slave
->dev
->name
);
2727 /* primary is up so switch to it */
2728 write_lock(&bond
->curr_slave_lock
);
2729 bond_change_active_slave(bond
, bond
->primary_slave
);
2730 write_unlock(&bond
->curr_slave_lock
);
2732 slave
= bond
->primary_slave
;
2733 slave
->jiffies
= jiffies
;
2735 bond
->current_arp_slave
= NULL
;
2738 /* the current slave must tx an arp to ensure backup slaves
2741 if (slave
&& bond_has_ip(bond
)) {
2742 bond_arp_send_all(bond
, slave
);
2746 /* if we don't have a curr_active_slave, search for the next available
2747 * backup slave from the current_arp_slave and make it the candidate
2748 * for becoming the curr_active_slave
2751 if (!bond
->current_arp_slave
) {
2752 bond
->current_arp_slave
= bond
->first_slave
;
2755 if (bond
->current_arp_slave
) {
2756 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
2758 /* search for next candidate */
2759 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
2760 if (IS_UP(slave
->dev
)) {
2761 slave
->link
= BOND_LINK_BACK
;
2762 bond_set_slave_active_flags(slave
);
2763 bond_arp_send_all(bond
, slave
);
2764 slave
->jiffies
= jiffies
;
2765 bond
->current_arp_slave
= slave
;
2769 /* if the link state is up at this point, we
2770 * mark it down - this can happen if we have
2771 * simultaneous link failures and
2772 * reselect_active_interface doesn't make this
2773 * one the current slave so it is still marked
2774 * up when it is actually down
2776 if (slave
->link
== BOND_LINK_UP
) {
2777 slave
->link
= BOND_LINK_DOWN
;
2778 if (slave
->link_failure_count
< UINT_MAX
) {
2779 slave
->link_failure_count
++;
2782 bond_set_slave_inactive_flags(slave
);
2784 printk(KERN_INFO DRV_NAME
2785 ": %s: backup interface %s is "
2795 if (bond
->params
.arp_interval
) {
2796 mod_timer(&bond
->arp_timer
, jiffies
+ delta_in_ticks
);
2799 read_unlock(&bond
->lock
);
2802 /*------------------------------ proc/seq_file-------------------------------*/
2804 #ifdef CONFIG_PROC_FS
2806 #define SEQ_START_TOKEN ((void *)1)
2808 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2810 struct bonding
*bond
= seq
->private;
2812 struct slave
*slave
;
2815 /* make sure the bond won't be taken away */
2816 read_lock(&dev_base_lock
);
2817 read_lock_bh(&bond
->lock
);
2820 return SEQ_START_TOKEN
;
2823 bond_for_each_slave(bond
, slave
, i
) {
2824 if (++off
== *pos
) {
2832 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2834 struct bonding
*bond
= seq
->private;
2835 struct slave
*slave
= v
;
2838 if (v
== SEQ_START_TOKEN
) {
2839 return bond
->first_slave
;
2842 slave
= slave
->next
;
2844 return (slave
== bond
->first_slave
) ? NULL
: slave
;
2847 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
2849 struct bonding
*bond
= seq
->private;
2851 read_unlock_bh(&bond
->lock
);
2852 read_unlock(&dev_base_lock
);
2855 static void bond_info_show_master(struct seq_file
*seq
)
2857 struct bonding
*bond
= seq
->private;
2862 read_lock(&bond
->curr_slave_lock
);
2863 curr
= bond
->curr_active_slave
;
2864 read_unlock(&bond
->curr_slave_lock
);
2866 seq_printf(seq
, "Bonding Mode: %s\n",
2867 bond_mode_name(bond
->params
.mode
));
2869 if (bond
->params
.mode
== BOND_MODE_XOR
||
2870 bond
->params
.mode
== BOND_MODE_8023AD
) {
2871 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
2872 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
2873 bond
->params
.xmit_policy
);
2876 if (USES_PRIMARY(bond
->params
.mode
)) {
2877 seq_printf(seq
, "Primary Slave: %s\n",
2878 (bond
->primary_slave
) ?
2879 bond
->primary_slave
->dev
->name
: "None");
2881 seq_printf(seq
, "Currently Active Slave: %s\n",
2882 (curr
) ? curr
->dev
->name
: "None");
2885 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
2887 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
2888 seq_printf(seq
, "Up Delay (ms): %d\n",
2889 bond
->params
.updelay
* bond
->params
.miimon
);
2890 seq_printf(seq
, "Down Delay (ms): %d\n",
2891 bond
->params
.downdelay
* bond
->params
.miimon
);
2894 /* ARP information */
2895 if(bond
->params
.arp_interval
> 0) {
2897 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
2898 bond
->params
.arp_interval
);
2900 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
2902 for(i
= 0; (i
< BOND_MAX_ARP_TARGETS
) ;i
++) {
2903 if (!bond
->params
.arp_targets
[i
])
2906 seq_printf(seq
, ",");
2907 target
= ntohl(bond
->params
.arp_targets
[i
]);
2908 seq_printf(seq
, " %d.%d.%d.%d", HIPQUAD(target
));
2911 seq_printf(seq
, "\n");
2914 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2915 struct ad_info ad_info
;
2917 seq_puts(seq
, "\n802.3ad info\n");
2918 seq_printf(seq
, "LACP rate: %s\n",
2919 (bond
->params
.lacp_fast
) ? "fast" : "slow");
2921 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
2922 seq_printf(seq
, "bond %s has no active aggregator\n",
2925 seq_printf(seq
, "Active Aggregator Info:\n");
2927 seq_printf(seq
, "\tAggregator ID: %d\n",
2928 ad_info
.aggregator_id
);
2929 seq_printf(seq
, "\tNumber of ports: %d\n",
2931 seq_printf(seq
, "\tActor Key: %d\n",
2933 seq_printf(seq
, "\tPartner Key: %d\n",
2934 ad_info
.partner_key
);
2935 seq_printf(seq
, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2936 ad_info
.partner_system
[0],
2937 ad_info
.partner_system
[1],
2938 ad_info
.partner_system
[2],
2939 ad_info
.partner_system
[3],
2940 ad_info
.partner_system
[4],
2941 ad_info
.partner_system
[5]);
2946 static void bond_info_show_slave(struct seq_file
*seq
, const struct slave
*slave
)
2948 struct bonding
*bond
= seq
->private;
2950 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
2951 seq_printf(seq
, "MII Status: %s\n",
2952 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
2953 seq_printf(seq
, "Link Failure Count: %d\n",
2954 slave
->link_failure_count
);
2957 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
2958 slave
->perm_hwaddr
[0], slave
->perm_hwaddr
[1],
2959 slave
->perm_hwaddr
[2], slave
->perm_hwaddr
[3],
2960 slave
->perm_hwaddr
[4], slave
->perm_hwaddr
[5]);
2962 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2963 const struct aggregator
*agg
2964 = SLAVE_AD_INFO(slave
).port
.aggregator
;
2967 seq_printf(seq
, "Aggregator ID: %d\n",
2968 agg
->aggregator_identifier
);
2970 seq_puts(seq
, "Aggregator ID: N/A\n");
2975 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
2977 if (v
== SEQ_START_TOKEN
) {
2978 seq_printf(seq
, "%s\n", version
);
2979 bond_info_show_master(seq
);
2981 bond_info_show_slave(seq
, v
);
2987 static struct seq_operations bond_info_seq_ops
= {
2988 .start
= bond_info_seq_start
,
2989 .next
= bond_info_seq_next
,
2990 .stop
= bond_info_seq_stop
,
2991 .show
= bond_info_seq_show
,
2994 static int bond_info_open(struct inode
*inode
, struct file
*file
)
2996 struct seq_file
*seq
;
2997 struct proc_dir_entry
*proc
;
3000 res
= seq_open(file
, &bond_info_seq_ops
);
3002 /* recover the pointer buried in proc_dir_entry data */
3003 seq
= file
->private_data
;
3005 seq
->private = proc
->data
;
3011 static struct file_operations bond_info_fops
= {
3012 .owner
= THIS_MODULE
,
3013 .open
= bond_info_open
,
3015 .llseek
= seq_lseek
,
3016 .release
= seq_release
,
3019 static int bond_create_proc_entry(struct bonding
*bond
)
3021 struct net_device
*bond_dev
= bond
->dev
;
3023 if (bond_proc_dir
) {
3024 bond
->proc_entry
= create_proc_entry(bond_dev
->name
,
3027 if (bond
->proc_entry
== NULL
) {
3028 printk(KERN_WARNING DRV_NAME
3029 ": Warning: Cannot create /proc/net/%s/%s\n",
3030 DRV_NAME
, bond_dev
->name
);
3032 bond
->proc_entry
->data
= bond
;
3033 bond
->proc_entry
->proc_fops
= &bond_info_fops
;
3034 bond
->proc_entry
->owner
= THIS_MODULE
;
3035 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3042 static void bond_remove_proc_entry(struct bonding
*bond
)
3044 if (bond_proc_dir
&& bond
->proc_entry
) {
3045 remove_proc_entry(bond
->proc_file_name
, bond_proc_dir
);
3046 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3047 bond
->proc_entry
= NULL
;
3051 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3052 * Caller must hold rtnl_lock.
3054 static void bond_create_proc_dir(void)
3056 int len
= strlen(DRV_NAME
);
3058 for (bond_proc_dir
= proc_net
->subdir
; bond_proc_dir
;
3059 bond_proc_dir
= bond_proc_dir
->next
) {
3060 if ((bond_proc_dir
->namelen
== len
) &&
3061 !memcmp(bond_proc_dir
->name
, DRV_NAME
, len
)) {
3066 if (!bond_proc_dir
) {
3067 bond_proc_dir
= proc_mkdir(DRV_NAME
, proc_net
);
3068 if (bond_proc_dir
) {
3069 bond_proc_dir
->owner
= THIS_MODULE
;
3071 printk(KERN_WARNING DRV_NAME
3072 ": Warning: cannot create /proc/net/%s\n",
3078 /* Destroy the bonding directory under /proc/net, if empty.
3079 * Caller must hold rtnl_lock.
3081 static void bond_destroy_proc_dir(void)
3083 struct proc_dir_entry
*de
;
3085 if (!bond_proc_dir
) {
3089 /* verify that the /proc dir is empty */
3090 for (de
= bond_proc_dir
->subdir
; de
; de
= de
->next
) {
3091 /* ignore . and .. */
3092 if (*(de
->name
) != '.') {
3098 if (bond_proc_dir
->owner
== THIS_MODULE
) {
3099 bond_proc_dir
->owner
= NULL
;
3102 remove_proc_entry(DRV_NAME
, proc_net
);
3103 bond_proc_dir
= NULL
;
3106 #endif /* CONFIG_PROC_FS */
3108 /*-------------------------- netdev event handling --------------------------*/
3111 * Change device name
3113 static int bond_event_changename(struct bonding
*bond
)
3115 #ifdef CONFIG_PROC_FS
3116 bond_remove_proc_entry(bond
);
3117 bond_create_proc_entry(bond
);
3119 down_write(&(bonding_rwsem
));
3120 bond_destroy_sysfs_entry(bond
);
3121 bond_create_sysfs_entry(bond
);
3122 up_write(&(bonding_rwsem
));
3126 static int bond_master_netdev_event(unsigned long event
, struct net_device
*bond_dev
)
3128 struct bonding
*event_bond
= bond_dev
->priv
;
3131 case NETDEV_CHANGENAME
:
3132 return bond_event_changename(event_bond
);
3133 case NETDEV_UNREGISTER
:
3135 * TODO: remove a bond from the list?
3145 static int bond_slave_netdev_event(unsigned long event
, struct net_device
*slave_dev
)
3147 struct net_device
*bond_dev
= slave_dev
->master
;
3148 struct bonding
*bond
= bond_dev
->priv
;
3151 case NETDEV_UNREGISTER
:
3153 bond_release(bond_dev
, slave_dev
);
3158 * TODO: is this what we get if somebody
3159 * sets up a hierarchical bond, then rmmod's
3160 * one of the slave bonding devices?
3165 * ... Or is it this?
3168 case NETDEV_CHANGEMTU
:
3170 * TODO: Should slaves be allowed to
3171 * independently alter their MTU? For
3172 * an active-backup bond, slaves need
3173 * not be the same type of device, so
3174 * MTUs may vary. For other modes,
3175 * slaves arguably should have the
3176 * same MTUs. To do this, we'd need to
3177 * take over the slave's change_mtu
3178 * function for the duration of their
3182 case NETDEV_CHANGENAME
:
3184 * TODO: handle changing the primary's name
3187 case NETDEV_FEAT_CHANGE
:
3188 bond_compute_features(bond
);
3198 * bond_netdev_event: handle netdev notifier chain events.
3200 * This function receives events for the netdev chain. The caller (an
3201 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3202 * locks for us to safely manipulate the slave devices (RTNL lock,
3205 static int bond_netdev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3207 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3209 dprintk("event_dev: %s, event: %lx\n",
3210 (event_dev
? event_dev
->name
: "None"),
3213 if (event_dev
->flags
& IFF_MASTER
) {
3214 dprintk("IFF_MASTER\n");
3215 return bond_master_netdev_event(event
, event_dev
);
3218 if (event_dev
->flags
& IFF_SLAVE
) {
3219 dprintk("IFF_SLAVE\n");
3220 return bond_slave_netdev_event(event
, event_dev
);
3227 * bond_inetaddr_event: handle inetaddr notifier chain events.
3229 * We keep track of device IPs primarily to use as source addresses in
3230 * ARP monitor probes (rather than spewing out broadcasts all the time).
3232 * We track one IP for the main device (if it has one), plus one per VLAN.
3234 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3236 struct in_ifaddr
*ifa
= ptr
;
3237 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3238 struct bonding
*bond
, *bond_next
;
3239 struct vlan_entry
*vlan
, *vlan_next
;
3241 list_for_each_entry_safe(bond
, bond_next
, &bond_dev_list
, bond_list
) {
3242 if (bond
->dev
== event_dev
) {
3245 bond
->master_ip
= ifa
->ifa_local
;
3248 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3255 if (list_empty(&bond
->vlan_list
))
3258 list_for_each_entry_safe(vlan
, vlan_next
, &bond
->vlan_list
,
3260 vlan_dev
= bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
];
3261 if (vlan_dev
== event_dev
) {
3264 vlan
->vlan_ip
= ifa
->ifa_local
;
3268 bond_glean_dev_ip(vlan_dev
);
3279 static struct notifier_block bond_netdev_notifier
= {
3280 .notifier_call
= bond_netdev_event
,
3283 static struct notifier_block bond_inetaddr_notifier
= {
3284 .notifier_call
= bond_inetaddr_event
,
3287 /*-------------------------- Packet type handling ---------------------------*/
3289 /* register to receive lacpdus on a bond */
3290 static void bond_register_lacpdu(struct bonding
*bond
)
3292 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3294 /* initialize packet type */
3295 pk_type
->type
= PKT_TYPE_LACPDU
;
3296 pk_type
->dev
= bond
->dev
;
3297 pk_type
->func
= bond_3ad_lacpdu_recv
;
3299 dev_add_pack(pk_type
);
3302 /* unregister to receive lacpdus on a bond */
3303 static void bond_unregister_lacpdu(struct bonding
*bond
)
3305 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3308 /*---------------------------- Hashing Policies -----------------------------*/
3311 * Hash for the the output device based upon layer 3 and layer 4 data. If
3312 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3313 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3315 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
,
3316 struct net_device
*bond_dev
, int count
)
3318 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3319 struct iphdr
*iph
= skb
->nh
.iph
;
3320 u16
*layer4hdr
= (u16
*)((u32
*)iph
+ iph
->ihl
);
3323 if (skb
->protocol
== __constant_htons(ETH_P_IP
)) {
3324 if (!(iph
->frag_off
& __constant_htons(IP_MF
|IP_OFFSET
)) &&
3325 (iph
->protocol
== IPPROTO_TCP
||
3326 iph
->protocol
== IPPROTO_UDP
)) {
3327 layer4_xor
= htons((*layer4hdr
^ *(layer4hdr
+ 1)));
3329 return (layer4_xor
^
3330 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3334 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3338 * Hash for the output device based upon layer 2 data
3340 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
,
3341 struct net_device
*bond_dev
, int count
)
3343 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3345 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3348 /*-------------------------- Device entry points ----------------------------*/
3350 static int bond_open(struct net_device
*bond_dev
)
3352 struct bonding
*bond
= bond_dev
->priv
;
3353 struct timer_list
*mii_timer
= &bond
->mii_timer
;
3354 struct timer_list
*arp_timer
= &bond
->arp_timer
;
3356 bond
->kill_timers
= 0;
3358 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
3359 (bond
->params
.mode
== BOND_MODE_ALB
)) {
3360 struct timer_list
*alb_timer
= &(BOND_ALB_INFO(bond
).alb_timer
);
3362 /* bond_alb_initialize must be called before the timer
3365 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3366 /* something went wrong - fail the open operation */
3370 init_timer(alb_timer
);
3371 alb_timer
->expires
= jiffies
+ 1;
3372 alb_timer
->data
= (unsigned long)bond
;
3373 alb_timer
->function
= (void *)&bond_alb_monitor
;
3374 add_timer(alb_timer
);
3377 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3378 init_timer(mii_timer
);
3379 mii_timer
->expires
= jiffies
+ 1;
3380 mii_timer
->data
= (unsigned long)bond_dev
;
3381 mii_timer
->function
= (void *)&bond_mii_monitor
;
3382 add_timer(mii_timer
);
3385 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3386 init_timer(arp_timer
);
3387 arp_timer
->expires
= jiffies
+ 1;
3388 arp_timer
->data
= (unsigned long)bond_dev
;
3389 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
3390 arp_timer
->function
= (void *)&bond_activebackup_arp_mon
;
3392 arp_timer
->function
= (void *)&bond_loadbalance_arp_mon
;
3394 add_timer(arp_timer
);
3397 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3398 struct timer_list
*ad_timer
= &(BOND_AD_INFO(bond
).ad_timer
);
3399 init_timer(ad_timer
);
3400 ad_timer
->expires
= jiffies
+ 1;
3401 ad_timer
->data
= (unsigned long)bond
;
3402 ad_timer
->function
= (void *)&bond_3ad_state_machine_handler
;
3403 add_timer(ad_timer
);
3405 /* register to receive LACPDUs */
3406 bond_register_lacpdu(bond
);
3412 static int bond_close(struct net_device
*bond_dev
)
3414 struct bonding
*bond
= bond_dev
->priv
;
3416 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3417 /* Unregister the receive of LACPDUs */
3418 bond_unregister_lacpdu(bond
);
3421 write_lock_bh(&bond
->lock
);
3423 bond_mc_list_destroy(bond
);
3425 /* signal timers not to re-arm */
3426 bond
->kill_timers
= 1;
3428 write_unlock_bh(&bond
->lock
);
3430 /* del_timer_sync must run without holding the bond->lock
3431 * because a running timer might be trying to hold it too
3434 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3435 del_timer_sync(&bond
->mii_timer
);
3438 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3439 del_timer_sync(&bond
->arp_timer
);
3442 switch (bond
->params
.mode
) {
3443 case BOND_MODE_8023AD
:
3444 del_timer_sync(&(BOND_AD_INFO(bond
).ad_timer
));
3448 del_timer_sync(&(BOND_ALB_INFO(bond
).alb_timer
));
3454 /* Release the bonded slaves */
3455 bond_release_all(bond_dev
);
3457 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
3458 (bond
->params
.mode
== BOND_MODE_ALB
)) {
3459 /* Must be called only after all
3460 * slaves have been released
3462 bond_alb_deinitialize(bond
);
3468 static struct net_device_stats
*bond_get_stats(struct net_device
*bond_dev
)
3470 struct bonding
*bond
= bond_dev
->priv
;
3471 struct net_device_stats
*stats
= &(bond
->stats
), *sstats
;
3472 struct slave
*slave
;
3475 memset(stats
, 0, sizeof(struct net_device_stats
));
3477 read_lock_bh(&bond
->lock
);
3479 bond_for_each_slave(bond
, slave
, i
) {
3480 sstats
= slave
->dev
->get_stats(slave
->dev
);
3482 stats
->rx_packets
+= sstats
->rx_packets
;
3483 stats
->rx_bytes
+= sstats
->rx_bytes
;
3484 stats
->rx_errors
+= sstats
->rx_errors
;
3485 stats
->rx_dropped
+= sstats
->rx_dropped
;
3487 stats
->tx_packets
+= sstats
->tx_packets
;
3488 stats
->tx_bytes
+= sstats
->tx_bytes
;
3489 stats
->tx_errors
+= sstats
->tx_errors
;
3490 stats
->tx_dropped
+= sstats
->tx_dropped
;
3492 stats
->multicast
+= sstats
->multicast
;
3493 stats
->collisions
+= sstats
->collisions
;
3495 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3496 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3497 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3498 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3499 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3500 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3502 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3503 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3504 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3505 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3506 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3509 read_unlock_bh(&bond
->lock
);
3514 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3516 struct net_device
*slave_dev
= NULL
;
3517 struct ifbond k_binfo
;
3518 struct ifbond __user
*u_binfo
= NULL
;
3519 struct ifslave k_sinfo
;
3520 struct ifslave __user
*u_sinfo
= NULL
;
3521 struct mii_ioctl_data
*mii
= NULL
;
3524 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3525 bond_dev
->name
, cmd
);
3537 * We do this again just in case we were called by SIOCGMIIREG
3538 * instead of SIOCGMIIPHY.
3545 if (mii
->reg_num
== 1) {
3546 struct bonding
*bond
= bond_dev
->priv
;
3548 read_lock_bh(&bond
->lock
);
3549 read_lock(&bond
->curr_slave_lock
);
3550 if (netif_carrier_ok(bond
->dev
)) {
3551 mii
->val_out
= BMSR_LSTATUS
;
3553 read_unlock(&bond
->curr_slave_lock
);
3554 read_unlock_bh(&bond
->lock
);
3558 case BOND_INFO_QUERY_OLD
:
3559 case SIOCBONDINFOQUERY
:
3560 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3562 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
))) {
3566 res
= bond_info_query(bond_dev
, &k_binfo
);
3568 if (copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
))) {
3574 case BOND_SLAVE_INFO_QUERY_OLD
:
3575 case SIOCBONDSLAVEINFOQUERY
:
3576 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3578 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
))) {
3582 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3584 if (copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
))) {
3595 if (!capable(CAP_NET_ADMIN
)) {
3599 down_write(&(bonding_rwsem
));
3600 slave_dev
= dev_get_by_name(ifr
->ifr_slave
);
3602 dprintk("slave_dev=%p: \n", slave_dev
);
3607 dprintk("slave_dev->name=%s: \n", slave_dev
->name
);
3609 case BOND_ENSLAVE_OLD
:
3610 case SIOCBONDENSLAVE
:
3611 res
= bond_enslave(bond_dev
, slave_dev
);
3613 case BOND_RELEASE_OLD
:
3614 case SIOCBONDRELEASE
:
3615 res
= bond_release(bond_dev
, slave_dev
);
3617 case BOND_SETHWADDR_OLD
:
3618 case SIOCBONDSETHWADDR
:
3619 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3621 case BOND_CHANGE_ACTIVE_OLD
:
3622 case SIOCBONDCHANGEACTIVE
:
3623 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3632 up_write(&(bonding_rwsem
));
3636 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3638 struct bonding
*bond
= bond_dev
->priv
;
3639 struct dev_mc_list
*dmi
;
3641 write_lock_bh(&bond
->lock
);
3644 * Do promisc before checking multicast_mode
3646 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
)) {
3647 bond_set_promiscuity(bond
, 1);
3650 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
)) {
3651 bond_set_promiscuity(bond
, -1);
3654 /* set allmulti flag to slaves */
3655 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
)) {
3656 bond_set_allmulti(bond
, 1);
3659 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
)) {
3660 bond_set_allmulti(bond
, -1);
3663 bond
->flags
= bond_dev
->flags
;
3665 /* looking for addresses to add to slaves' mc list */
3666 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
3667 if (!bond_mc_list_find_dmi(dmi
, bond
->mc_list
)) {
3668 bond_mc_add(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
3672 /* looking for addresses to delete from slaves' list */
3673 for (dmi
= bond
->mc_list
; dmi
; dmi
= dmi
->next
) {
3674 if (!bond_mc_list_find_dmi(dmi
, bond_dev
->mc_list
)) {
3675 bond_mc_delete(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
3679 /* save master's multicast list */
3680 bond_mc_list_destroy(bond
);
3681 bond_mc_list_copy(bond_dev
->mc_list
, bond
, GFP_ATOMIC
);
3683 write_unlock_bh(&bond
->lock
);
3687 * Change the MTU of all of a master's slaves to match the master
3689 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3691 struct bonding
*bond
= bond_dev
->priv
;
3692 struct slave
*slave
, *stop_at
;
3696 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond
,
3697 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3699 /* Can't hold bond->lock with bh disabled here since
3700 * some base drivers panic. On the other hand we can't
3701 * hold bond->lock without bh disabled because we'll
3702 * deadlock. The only solution is to rely on the fact
3703 * that we're under rtnl_lock here, and the slaves
3704 * list won't change. This doesn't solve the problem
3705 * of setting the slave's MTU while it is
3706 * transmitting, but the assumption is that the base
3707 * driver can handle that.
3709 * TODO: figure out a way to safely iterate the slaves
3710 * list, but without holding a lock around the actual
3711 * call to the base driver.
3714 bond_for_each_slave(bond
, slave
, i
) {
3715 dprintk("s %p s->p %p c_m %p\n", slave
,
3716 slave
->prev
, slave
->dev
->change_mtu
);
3718 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3721 /* If we failed to set the slave's mtu to the new value
3722 * we must abort the operation even in ACTIVE_BACKUP
3723 * mode, because if we allow the backup slaves to have
3724 * different mtu values than the active slave we'll
3725 * need to change their mtu when doing a failover. That
3726 * means changing their mtu from timer context, which
3727 * is probably not a good idea.
3729 dprintk("err %d %s\n", res
, slave
->dev
->name
);
3734 bond_dev
->mtu
= new_mtu
;
3739 /* unwind from head to the slave that failed */
3741 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3744 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3746 dprintk("unwind err %d dev %s\n", tmp_res
,
3757 * Note that many devices must be down to change the HW address, and
3758 * downing the master releases all slaves. We can make bonds full of
3759 * bonding devices to test this, however.
3761 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3763 struct bonding
*bond
= bond_dev
->priv
;
3764 struct sockaddr
*sa
= addr
, tmp_sa
;
3765 struct slave
*slave
, *stop_at
;
3769 dprintk("bond=%p, name=%s\n", bond
, (bond_dev
? bond_dev
->name
: "None"));
3771 if (!is_valid_ether_addr(sa
->sa_data
)) {
3772 return -EADDRNOTAVAIL
;
3775 /* Can't hold bond->lock with bh disabled here since
3776 * some base drivers panic. On the other hand we can't
3777 * hold bond->lock without bh disabled because we'll
3778 * deadlock. The only solution is to rely on the fact
3779 * that we're under rtnl_lock here, and the slaves
3780 * list won't change. This doesn't solve the problem
3781 * of setting the slave's hw address while it is
3782 * transmitting, but the assumption is that the base
3783 * driver can handle that.
3785 * TODO: figure out a way to safely iterate the slaves
3786 * list, but without holding a lock around the actual
3787 * call to the base driver.
3790 bond_for_each_slave(bond
, slave
, i
) {
3791 dprintk("slave %p %s\n", slave
, slave
->dev
->name
);
3793 if (slave
->dev
->set_mac_address
== NULL
) {
3795 dprintk("EOPNOTSUPP %s\n", slave
->dev
->name
);
3799 res
= dev_set_mac_address(slave
->dev
, addr
);
3801 /* TODO: consider downing the slave
3803 * User should expect communications
3804 * breakage anyway until ARP finish
3807 dprintk("err %d %s\n", res
, slave
->dev
->name
);
3813 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3817 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3818 tmp_sa
.sa_family
= bond_dev
->type
;
3820 /* unwind from head to the slave that failed */
3822 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3825 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3827 dprintk("unwind err %d dev %s\n", tmp_res
,
3835 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3837 struct bonding
*bond
= bond_dev
->priv
;
3838 struct slave
*slave
, *start_at
;
3842 read_lock(&bond
->lock
);
3844 if (!BOND_IS_OK(bond
)) {
3848 read_lock(&bond
->curr_slave_lock
);
3849 slave
= start_at
= bond
->curr_active_slave
;
3850 read_unlock(&bond
->curr_slave_lock
);
3856 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3857 if (IS_UP(slave
->dev
) &&
3858 (slave
->link
== BOND_LINK_UP
) &&
3859 (slave
->state
== BOND_STATE_ACTIVE
)) {
3860 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3862 write_lock(&bond
->curr_slave_lock
);
3863 bond
->curr_active_slave
= slave
->next
;
3864 write_unlock(&bond
->curr_slave_lock
);
3873 /* no suitable interface, frame not sent */
3876 read_unlock(&bond
->lock
);
3880 static void bond_activebackup_xmit_copy(struct sk_buff
*skb
,
3881 struct bonding
*bond
,
3882 struct slave
*slave
)
3884 struct sk_buff
*skb2
= skb_copy(skb
, GFP_ATOMIC
);
3885 struct ethhdr
*eth_data
;
3890 printk(KERN_ERR DRV_NAME
": Error: "
3891 "bond_activebackup_xmit_copy(): skb_copy() failed\n");
3895 skb2
->mac
.raw
= (unsigned char *)skb2
->data
;
3896 eth_data
= eth_hdr(skb2
);
3898 /* Pick an appropriate source MAC address
3899 * -- use slave's perm MAC addr, unless used by bond
3900 * -- otherwise, borrow active slave's perm MAC addr
3901 * since that will not be used
3903 hwaddr
= slave
->perm_hwaddr
;
3904 if (!memcmp(eth_data
->h_source
, hwaddr
, ETH_ALEN
))
3905 hwaddr
= bond
->curr_active_slave
->perm_hwaddr
;
3907 /* Set source MAC address appropriately */
3908 memcpy(eth_data
->h_source
, hwaddr
, ETH_ALEN
);
3910 res
= bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3912 dev_kfree_skb(skb2
);
3918 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3919 * the bond has a usable interface.
3921 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3923 struct bonding
*bond
= bond_dev
->priv
;
3926 read_lock(&bond
->lock
);
3927 read_lock(&bond
->curr_slave_lock
);
3929 if (!BOND_IS_OK(bond
)) {
3933 if (!bond
->curr_active_slave
)
3936 /* Xmit IGMP frames on all slaves to ensure rapid fail-over
3937 for multicast traffic on snooping switches */
3938 if (skb
->protocol
== __constant_htons(ETH_P_IP
) &&
3939 skb
->nh
.iph
->protocol
== IPPROTO_IGMP
) {
3940 struct slave
*slave
, *active_slave
;
3943 active_slave
= bond
->curr_active_slave
;
3944 bond_for_each_slave_from_to(bond
, slave
, i
, active_slave
->next
,
3946 if (IS_UP(slave
->dev
) &&
3947 (slave
->link
== BOND_LINK_UP
))
3948 bond_activebackup_xmit_copy(skb
, bond
, slave
);
3951 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
3955 /* no suitable interface, frame not sent */
3958 read_unlock(&bond
->curr_slave_lock
);
3959 read_unlock(&bond
->lock
);
3964 * In bond_xmit_xor() , we determine the output device by using a pre-
3965 * determined xmit_hash_policy(), If the selected device is not enabled,
3966 * find the next active slave.
3968 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3970 struct bonding
*bond
= bond_dev
->priv
;
3971 struct slave
*slave
, *start_at
;
3976 read_lock(&bond
->lock
);
3978 if (!BOND_IS_OK(bond
)) {
3982 slave_no
= bond
->xmit_hash_policy(skb
, bond_dev
, bond
->slave_cnt
);
3984 bond_for_each_slave(bond
, slave
, i
) {
3993 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3994 if (IS_UP(slave
->dev
) &&
3995 (slave
->link
== BOND_LINK_UP
) &&
3996 (slave
->state
== BOND_STATE_ACTIVE
)) {
3997 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4004 /* no suitable interface, frame not sent */
4007 read_unlock(&bond
->lock
);
4012 * in broadcast mode, we send everything to all usable interfaces.
4014 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4016 struct bonding
*bond
= bond_dev
->priv
;
4017 struct slave
*slave
, *start_at
;
4018 struct net_device
*tx_dev
= NULL
;
4022 read_lock(&bond
->lock
);
4024 if (!BOND_IS_OK(bond
)) {
4028 read_lock(&bond
->curr_slave_lock
);
4029 start_at
= bond
->curr_active_slave
;
4030 read_unlock(&bond
->curr_slave_lock
);
4036 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4037 if (IS_UP(slave
->dev
) &&
4038 (slave
->link
== BOND_LINK_UP
) &&
4039 (slave
->state
== BOND_STATE_ACTIVE
)) {
4041 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4043 printk(KERN_ERR DRV_NAME
4044 ": %s: Error: bond_xmit_broadcast(): "
4045 "skb_clone() failed\n",
4050 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4052 dev_kfree_skb(skb2
);
4056 tx_dev
= slave
->dev
;
4061 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4066 /* no suitable interface, frame not sent */
4069 /* frame sent to all suitable interfaces */
4070 read_unlock(&bond
->lock
);
4074 /*------------------------- Device initialization ---------------------------*/
4077 * set bond mode specific net device operations
4079 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4081 struct net_device
*bond_dev
= bond
->dev
;
4084 case BOND_MODE_ROUNDROBIN
:
4085 bond_dev
->hard_start_xmit
= bond_xmit_roundrobin
;
4087 case BOND_MODE_ACTIVEBACKUP
:
4088 bond_dev
->hard_start_xmit
= bond_xmit_activebackup
;
4091 bond_dev
->hard_start_xmit
= bond_xmit_xor
;
4092 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
)
4093 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4095 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4097 case BOND_MODE_BROADCAST
:
4098 bond_dev
->hard_start_xmit
= bond_xmit_broadcast
;
4100 case BOND_MODE_8023AD
:
4101 bond_set_master_3ad_flags(bond
);
4102 bond_dev
->hard_start_xmit
= bond_3ad_xmit_xor
;
4103 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
)
4104 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4106 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4109 bond_set_master_alb_flags(bond
);
4112 bond_dev
->hard_start_xmit
= bond_alb_xmit
;
4113 bond_dev
->set_mac_address
= bond_alb_set_mac_address
;
4116 /* Should never happen, mode already checked */
4117 printk(KERN_ERR DRV_NAME
4118 ": %s: Error: Unknown bonding mode %d\n",
4125 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4126 struct ethtool_drvinfo
*drvinfo
)
4128 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4129 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4130 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4133 static struct ethtool_ops bond_ethtool_ops
= {
4134 .get_tx_csum
= ethtool_op_get_tx_csum
,
4135 .get_tso
= ethtool_op_get_tso
,
4136 .get_ufo
= ethtool_op_get_ufo
,
4137 .get_sg
= ethtool_op_get_sg
,
4138 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4142 * Does not allocate but creates a /proc entry.
4145 static int bond_init(struct net_device
*bond_dev
, struct bond_params
*params
)
4147 struct bonding
*bond
= bond_dev
->priv
;
4149 dprintk("Begin bond_init for %s\n", bond_dev
->name
);
4151 /* initialize rwlocks */
4152 rwlock_init(&bond
->lock
);
4153 rwlock_init(&bond
->curr_slave_lock
);
4155 bond
->params
= *params
; /* copy params struct */
4157 /* Initialize pointers */
4158 bond
->first_slave
= NULL
;
4159 bond
->curr_active_slave
= NULL
;
4160 bond
->current_arp_slave
= NULL
;
4161 bond
->primary_slave
= NULL
;
4162 bond
->dev
= bond_dev
;
4163 INIT_LIST_HEAD(&bond
->vlan_list
);
4165 /* Initialize the device entry points */
4166 bond_dev
->open
= bond_open
;
4167 bond_dev
->stop
= bond_close
;
4168 bond_dev
->get_stats
= bond_get_stats
;
4169 bond_dev
->do_ioctl
= bond_do_ioctl
;
4170 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4171 bond_dev
->set_multicast_list
= bond_set_multicast_list
;
4172 bond_dev
->change_mtu
= bond_change_mtu
;
4173 bond_dev
->set_mac_address
= bond_set_mac_address
;
4175 bond_set_mode_ops(bond
, bond
->params
.mode
);
4177 bond_dev
->destructor
= free_netdev
;
4179 /* Initialize the device options */
4180 bond_dev
->tx_queue_len
= 0;
4181 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4183 /* At first, we block adding VLANs. That's the only way to
4184 * prevent problems that occur when adding VLANs over an
4185 * empty bond. The block will be removed once non-challenged
4186 * slaves are enslaved.
4188 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4190 /* don't acquire bond device's netif_tx_lock when
4192 bond_dev
->features
|= NETIF_F_LLTX
;
4194 /* By default, we declare the bond to be fully
4195 * VLAN hardware accelerated capable. Special
4196 * care is taken in the various xmit functions
4197 * when there are slaves that are not hw accel
4200 bond_dev
->vlan_rx_register
= bond_vlan_rx_register
;
4201 bond_dev
->vlan_rx_add_vid
= bond_vlan_rx_add_vid
;
4202 bond_dev
->vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
;
4203 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4204 NETIF_F_HW_VLAN_RX
|
4205 NETIF_F_HW_VLAN_FILTER
);
4207 #ifdef CONFIG_PROC_FS
4208 bond_create_proc_entry(bond
);
4211 list_add_tail(&bond
->bond_list
, &bond_dev_list
);
4216 /* De-initialize device specific data.
4217 * Caller must hold rtnl_lock.
4219 void bond_deinit(struct net_device
*bond_dev
)
4221 struct bonding
*bond
= bond_dev
->priv
;
4223 list_del(&bond
->bond_list
);
4225 #ifdef CONFIG_PROC_FS
4226 bond_remove_proc_entry(bond
);
4230 /* Unregister and free all bond devices.
4231 * Caller must hold rtnl_lock.
4233 static void bond_free_all(void)
4235 struct bonding
*bond
, *nxt
;
4237 list_for_each_entry_safe(bond
, nxt
, &bond_dev_list
, bond_list
) {
4238 struct net_device
*bond_dev
= bond
->dev
;
4240 unregister_netdevice(bond_dev
);
4241 bond_deinit(bond_dev
);
4244 #ifdef CONFIG_PROC_FS
4245 bond_destroy_proc_dir();
4249 /*------------------------- Module initialization ---------------------------*/
4252 * Convert string input module parms. Accept either the
4253 * number of the mode or its string name.
4255 int bond_parse_parm(char *mode_arg
, struct bond_parm_tbl
*tbl
)
4259 for (i
= 0; tbl
[i
].modename
; i
++) {
4260 if ((isdigit(*mode_arg
) &&
4261 tbl
[i
].mode
== simple_strtol(mode_arg
, NULL
, 0)) ||
4262 (strncmp(mode_arg
, tbl
[i
].modename
,
4263 strlen(tbl
[i
].modename
)) == 0)) {
4271 static int bond_check_params(struct bond_params
*params
)
4274 * Convert string parameters.
4277 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4278 if (bond_mode
== -1) {
4279 printk(KERN_ERR DRV_NAME
4280 ": Error: Invalid bonding mode \"%s\"\n",
4281 mode
== NULL
? "NULL" : mode
);
4286 if (xmit_hash_policy
) {
4287 if ((bond_mode
!= BOND_MODE_XOR
) &&
4288 (bond_mode
!= BOND_MODE_8023AD
)) {
4289 printk(KERN_INFO DRV_NAME
4290 ": xor_mode param is irrelevant in mode %s\n",
4291 bond_mode_name(bond_mode
));
4293 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4295 if (xmit_hashtype
== -1) {
4296 printk(KERN_ERR DRV_NAME
4297 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4298 xmit_hash_policy
== NULL
? "NULL" :
4306 if (bond_mode
!= BOND_MODE_8023AD
) {
4307 printk(KERN_INFO DRV_NAME
4308 ": lacp_rate param is irrelevant in mode %s\n",
4309 bond_mode_name(bond_mode
));
4311 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4312 if (lacp_fast
== -1) {
4313 printk(KERN_ERR DRV_NAME
4314 ": Error: Invalid lacp rate \"%s\"\n",
4315 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4321 if (max_bonds
< 1 || max_bonds
> INT_MAX
) {
4322 printk(KERN_WARNING DRV_NAME
4323 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4324 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4325 max_bonds
, 1, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4326 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4330 printk(KERN_WARNING DRV_NAME
4331 ": Warning: miimon module parameter (%d), "
4332 "not in range 0-%d, so it was reset to %d\n",
4333 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4334 miimon
= BOND_LINK_MON_INTERV
;
4338 printk(KERN_WARNING DRV_NAME
4339 ": Warning: updelay module parameter (%d), "
4340 "not in range 0-%d, so it was reset to 0\n",
4345 if (downdelay
< 0) {
4346 printk(KERN_WARNING DRV_NAME
4347 ": Warning: downdelay module parameter (%d), "
4348 "not in range 0-%d, so it was reset to 0\n",
4349 downdelay
, INT_MAX
);
4353 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4354 printk(KERN_WARNING DRV_NAME
4355 ": Warning: use_carrier module parameter (%d), "
4356 "not of valid value (0/1), so it was set to 1\n",
4361 /* reset values for 802.3ad */
4362 if (bond_mode
== BOND_MODE_8023AD
) {
4364 printk(KERN_WARNING DRV_NAME
4365 ": Warning: miimon must be specified, "
4366 "otherwise bonding will not detect link "
4367 "failure, speed and duplex which are "
4368 "essential for 802.3ad operation\n");
4369 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4374 /* reset values for TLB/ALB */
4375 if ((bond_mode
== BOND_MODE_TLB
) ||
4376 (bond_mode
== BOND_MODE_ALB
)) {
4378 printk(KERN_WARNING DRV_NAME
4379 ": Warning: miimon must be specified, "
4380 "otherwise bonding will not detect link "
4381 "failure and link speed which are essential "
4382 "for TLB/ALB load balancing\n");
4383 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4388 if (bond_mode
== BOND_MODE_ALB
) {
4389 printk(KERN_NOTICE DRV_NAME
4390 ": In ALB mode you might experience client "
4391 "disconnections upon reconnection of a link if the "
4392 "bonding module updelay parameter (%d msec) is "
4393 "incompatible with the forwarding delay time of the "
4399 if (updelay
|| downdelay
) {
4400 /* just warn the user the up/down delay will have
4401 * no effect since miimon is zero...
4403 printk(KERN_WARNING DRV_NAME
4404 ": Warning: miimon module parameter not set "
4405 "and updelay (%d) or downdelay (%d) module "
4406 "parameter is set; updelay and downdelay have "
4407 "no effect unless miimon is set\n",
4408 updelay
, downdelay
);
4411 /* don't allow arp monitoring */
4413 printk(KERN_WARNING DRV_NAME
4414 ": Warning: miimon (%d) and arp_interval (%d) "
4415 "can't be used simultaneously, disabling ARP "
4417 miimon
, arp_interval
);
4421 if ((updelay
% miimon
) != 0) {
4422 printk(KERN_WARNING DRV_NAME
4423 ": Warning: updelay (%d) is not a multiple "
4424 "of miimon (%d), updelay rounded to %d ms\n",
4425 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4430 if ((downdelay
% miimon
) != 0) {
4431 printk(KERN_WARNING DRV_NAME
4432 ": Warning: downdelay (%d) is not a multiple "
4433 "of miimon (%d), downdelay rounded to %d ms\n",
4435 (downdelay
/ miimon
) * miimon
);
4438 downdelay
/= miimon
;
4441 if (arp_interval
< 0) {
4442 printk(KERN_WARNING DRV_NAME
4443 ": Warning: arp_interval module parameter (%d) "
4444 ", not in range 0-%d, so it was reset to %d\n",
4445 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4446 arp_interval
= BOND_LINK_ARP_INTERV
;
4449 for (arp_ip_count
= 0;
4450 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4452 /* not complete check, but should be good enough to
4454 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4455 printk(KERN_WARNING DRV_NAME
4456 ": Warning: bad arp_ip_target module parameter "
4457 "(%s), ARP monitoring will not be performed\n",
4458 arp_ip_target
[arp_ip_count
]);
4461 u32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4462 arp_target
[arp_ip_count
] = ip
;
4466 if (arp_interval
&& !arp_ip_count
) {
4467 /* don't allow arping if no arp_ip_target given... */
4468 printk(KERN_WARNING DRV_NAME
4469 ": Warning: arp_interval module parameter (%d) "
4470 "specified without providing an arp_ip_target "
4471 "parameter, arp_interval was reset to 0\n",
4477 printk(KERN_INFO DRV_NAME
4478 ": MII link monitoring set to %d ms\n",
4480 } else if (arp_interval
) {
4483 printk(KERN_INFO DRV_NAME
4484 ": ARP monitoring set to %d ms with %d target(s):",
4485 arp_interval
, arp_ip_count
);
4487 for (i
= 0; i
< arp_ip_count
; i
++)
4488 printk (" %s", arp_ip_target
[i
]);
4493 /* miimon and arp_interval not set, we need one so things
4494 * work as expected, see bonding.txt for details
4496 printk(KERN_WARNING DRV_NAME
4497 ": Warning: either miimon or arp_interval and "
4498 "arp_ip_target module parameters must be specified, "
4499 "otherwise bonding will not detect link failures! see "
4500 "bonding.txt for details.\n");
4503 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4504 /* currently, using a primary only makes sense
4505 * in active backup, TLB or ALB modes
4507 printk(KERN_WARNING DRV_NAME
4508 ": Warning: %s primary device specified but has no "
4509 "effect in %s mode\n",
4510 primary
, bond_mode_name(bond_mode
));
4514 /* fill params struct with the proper values */
4515 params
->mode
= bond_mode
;
4516 params
->xmit_policy
= xmit_hashtype
;
4517 params
->miimon
= miimon
;
4518 params
->arp_interval
= arp_interval
;
4519 params
->updelay
= updelay
;
4520 params
->downdelay
= downdelay
;
4521 params
->use_carrier
= use_carrier
;
4522 params
->lacp_fast
= lacp_fast
;
4523 params
->primary
[0] = 0;
4526 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4527 params
->primary
[IFNAMSIZ
- 1] = 0;
4530 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4535 /* Create a new bond based on the specified name and bonding parameters.
4536 * Caller must NOT hold rtnl_lock; we need to release it here before we
4537 * set up our sysfs entries.
4539 int bond_create(char *name
, struct bond_params
*params
, struct bonding
**newbond
)
4541 struct net_device
*bond_dev
;
4545 bond_dev
= alloc_netdev(sizeof(struct bonding
), name
, ether_setup
);
4547 printk(KERN_ERR DRV_NAME
4548 ": %s: eek! can't alloc netdev!\n",
4554 /* bond_init() must be called after dev_alloc_name() (for the
4555 * /proc files), but before register_netdevice(), because we
4556 * need to set function pointers.
4559 res
= bond_init(bond_dev
, params
);
4564 SET_MODULE_OWNER(bond_dev
);
4566 res
= register_netdevice(bond_dev
);
4571 *newbond
= bond_dev
->priv
;
4573 netif_carrier_off(bond_dev
);
4575 rtnl_unlock(); /* allows sysfs registration of net device */
4576 res
= bond_create_sysfs_entry(bond_dev
->priv
);
4579 bond_deinit(bond_dev
);
4581 free_netdev(bond_dev
);
4588 static int __init
bonding_init(void)
4592 char new_bond_name
[8]; /* Enough room for 999 bonds at init. */
4594 printk(KERN_INFO
"%s", version
);
4596 res
= bond_check_params(&bonding_defaults
);
4601 #ifdef CONFIG_PROC_FS
4602 bond_create_proc_dir();
4604 for (i
= 0; i
< max_bonds
; i
++) {
4605 sprintf(new_bond_name
, "bond%d",i
);
4606 res
= bond_create(new_bond_name
,&bonding_defaults
, NULL
);
4611 res
= bond_create_sysfs();
4615 register_netdevice_notifier(&bond_netdev_notifier
);
4616 register_inetaddr_notifier(&bond_inetaddr_notifier
);
4622 bond_destroy_sysfs();
4629 static void __exit
bonding_exit(void)
4631 unregister_netdevice_notifier(&bond_netdev_notifier
);
4632 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
4636 bond_destroy_sysfs();
4640 module_init(bonding_init
);
4641 module_exit(bonding_exit
);
4642 MODULE_LICENSE("GPL");
4643 MODULE_VERSION(DRV_VERSION
);
4644 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4645 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4646 MODULE_SUPPORTED_DEVICE("most ethernet devices");