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/config.h>
37 #include <linux/kernel.h>
38 #include <linux/module.h>
39 #include <linux/sched.h>
40 #include <linux/types.h>
41 #include <linux/fcntl.h>
42 #include <linux/interrupt.h>
43 #include <linux/ptrace.h>
44 #include <linux/ioport.h>
48 #include <linux/tcp.h>
49 #include <linux/udp.h>
50 #include <linux/slab.h>
51 #include <linux/string.h>
52 #include <linux/init.h>
53 #include <linux/timer.h>
54 #include <linux/socket.h>
55 #include <linux/ctype.h>
56 #include <linux/inet.h>
57 #include <linux/bitops.h>
58 #include <asm/system.h>
61 #include <asm/uaccess.h>
62 #include <linux/errno.h>
63 #include <linux/netdevice.h>
64 #include <linux/inetdevice.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <net/route.h>
83 /*---------------------------- Module parameters ----------------------------*/
85 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
86 #define BOND_LINK_MON_INTERV 0
87 #define BOND_LINK_ARP_INTERV 0
89 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
90 static int miimon
= BOND_LINK_MON_INTERV
;
91 static int updelay
= 0;
92 static int downdelay
= 0;
93 static int use_carrier
= 1;
94 static char *mode
= NULL
;
95 static char *primary
= NULL
;
96 static char *lacp_rate
= NULL
;
97 static char *xmit_hash_policy
= NULL
;
98 static int arp_interval
= BOND_LINK_ARP_INTERV
;
99 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
] = { NULL
, };
100 struct bond_params bonding_defaults
;
102 module_param(max_bonds
, int, 0);
103 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
104 module_param(miimon
, int, 0);
105 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
106 module_param(updelay
, int, 0);
107 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
108 module_param(downdelay
, int, 0);
109 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
111 module_param(use_carrier
, int, 0);
112 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
113 "0 for off, 1 for on (default)");
114 module_param(mode
, charp
, 0);
115 MODULE_PARM_DESC(mode
, "Mode of operation : 0 for balance-rr, "
116 "1 for active-backup, 2 for balance-xor, "
117 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
118 "6 for balance-alb");
119 module_param(primary
, charp
, 0);
120 MODULE_PARM_DESC(primary
, "Primary network device to use");
121 module_param(lacp_rate
, charp
, 0);
122 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner "
124 module_param(xmit_hash_policy
, charp
, 0);
125 MODULE_PARM_DESC(xmit_hash_policy
, "XOR hashing method: 0 for layer 2 (default)"
126 ", 1 for layer 3+4");
127 module_param(arp_interval
, int, 0);
128 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
129 module_param_array(arp_ip_target
, charp
, NULL
, 0);
130 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
132 /*----------------------------- Global variables ----------------------------*/
134 static const char *version
=
135 DRV_DESCRIPTION
": v" DRV_VERSION
" (" DRV_RELDATE
")\n";
137 LIST_HEAD(bond_dev_list
);
139 #ifdef CONFIG_PROC_FS
140 static struct proc_dir_entry
*bond_proc_dir
= NULL
;
143 extern struct rw_semaphore bonding_rwsem
;
144 static u32 arp_target
[BOND_MAX_ARP_TARGETS
] = { 0, } ;
145 static int arp_ip_count
= 0;
146 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
147 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
148 static int lacp_fast
= 0;
151 struct bond_parm_tbl bond_lacp_tbl
[] = {
152 { "slow", AD_LACP_SLOW
},
153 { "fast", AD_LACP_FAST
},
157 struct bond_parm_tbl bond_mode_tbl
[] = {
158 { "balance-rr", BOND_MODE_ROUNDROBIN
},
159 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
160 { "balance-xor", BOND_MODE_XOR
},
161 { "broadcast", BOND_MODE_BROADCAST
},
162 { "802.3ad", BOND_MODE_8023AD
},
163 { "balance-tlb", BOND_MODE_TLB
},
164 { "balance-alb", BOND_MODE_ALB
},
168 struct bond_parm_tbl xmit_hashtype_tbl
[] = {
169 { "layer2", BOND_XMIT_POLICY_LAYER2
},
170 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
174 /*-------------------------- Forward declarations ---------------------------*/
176 static void bond_send_gratuitous_arp(struct bonding
*bond
);
178 /*---------------------------- General routines -----------------------------*/
180 const char *bond_mode_name(int mode
)
183 case BOND_MODE_ROUNDROBIN
:
184 return "load balancing (round-robin)";
185 case BOND_MODE_ACTIVEBACKUP
:
186 return "fault-tolerance (active-backup)";
188 return "load balancing (xor)";
189 case BOND_MODE_BROADCAST
:
190 return "fault-tolerance (broadcast)";
191 case BOND_MODE_8023AD
:
192 return "IEEE 802.3ad Dynamic link aggregation";
194 return "transmit load balancing";
196 return "adaptive load balancing";
202 /*---------------------------------- VLAN -----------------------------------*/
205 * bond_add_vlan - add a new vlan id on bond
206 * @bond: bond that got the notification
207 * @vlan_id: the vlan id to add
209 * Returns -ENOMEM if allocation failed.
211 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
213 struct vlan_entry
*vlan
;
215 dprintk("bond: %s, vlan id %d\n",
216 (bond
? bond
->dev
->name
: "None"), vlan_id
);
218 vlan
= kmalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
223 INIT_LIST_HEAD(&vlan
->vlan_list
);
224 vlan
->vlan_id
= vlan_id
;
227 write_lock_bh(&bond
->lock
);
229 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
231 write_unlock_bh(&bond
->lock
);
233 dprintk("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
239 * bond_del_vlan - delete a vlan id from bond
240 * @bond: bond that got the notification
241 * @vlan_id: the vlan id to delete
243 * returns -ENODEV if @vlan_id was not found in @bond.
245 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
247 struct vlan_entry
*vlan
, *next
;
250 dprintk("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
252 write_lock_bh(&bond
->lock
);
254 list_for_each_entry_safe(vlan
, next
, &bond
->vlan_list
, vlan_list
) {
255 if (vlan
->vlan_id
== vlan_id
) {
256 list_del(&vlan
->vlan_list
);
258 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
259 (bond
->params
.mode
== BOND_MODE_ALB
)) {
260 bond_alb_clear_vlan(bond
, vlan_id
);
263 dprintk("removed VLAN ID %d from bond %s\n", vlan_id
,
268 if (list_empty(&bond
->vlan_list
) &&
269 (bond
->slave_cnt
== 0)) {
270 /* Last VLAN removed and no slaves, so
271 * restore block on adding VLANs. This will
272 * be removed once new slaves that are not
273 * VLAN challenged will be added.
275 bond
->dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
283 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id
,
287 write_unlock_bh(&bond
->lock
);
292 * bond_has_challenged_slaves
293 * @bond: the bond we're working on
295 * Searches the slave list. Returns 1 if a vlan challenged slave
296 * was found, 0 otherwise.
298 * Assumes bond->lock is held.
300 static int bond_has_challenged_slaves(struct bonding
*bond
)
305 bond_for_each_slave(bond
, slave
, i
) {
306 if (slave
->dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
307 dprintk("found VLAN challenged slave - %s\n",
313 dprintk("no VLAN challenged slaves found\n");
318 * bond_next_vlan - safely skip to the next item in the vlans list.
319 * @bond: the bond we're working on
320 * @curr: item we're advancing from
322 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
323 * or @curr->next otherwise (even if it is @curr itself again).
325 * Caller must hold bond->lock
327 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
329 struct vlan_entry
*next
, *last
;
331 if (list_empty(&bond
->vlan_list
)) {
336 next
= list_entry(bond
->vlan_list
.next
,
337 struct vlan_entry
, vlan_list
);
339 last
= list_entry(bond
->vlan_list
.prev
,
340 struct vlan_entry
, vlan_list
);
342 next
= list_entry(bond
->vlan_list
.next
,
343 struct vlan_entry
, vlan_list
);
345 next
= list_entry(curr
->vlan_list
.next
,
346 struct vlan_entry
, vlan_list
);
354 * bond_dev_queue_xmit - Prepare skb for xmit.
356 * @bond: bond device that got this skb for tx.
357 * @skb: hw accel VLAN tagged skb to transmit
358 * @slave_dev: slave that is supposed to xmit this skbuff
360 * When the bond gets an skb to transmit that is
361 * already hardware accelerated VLAN tagged, and it
362 * needs to relay this skb to a slave that is not
363 * hw accel capable, the skb needs to be "unaccelerated",
364 * i.e. strip the hwaccel tag and re-insert it as part
367 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
)
369 unsigned short vlan_id
;
371 if (!list_empty(&bond
->vlan_list
) &&
372 !(slave_dev
->features
& NETIF_F_HW_VLAN_TX
) &&
373 vlan_get_tag(skb
, &vlan_id
) == 0) {
374 skb
->dev
= slave_dev
;
375 skb
= vlan_put_tag(skb
, vlan_id
);
377 /* vlan_put_tag() frees the skb in case of error,
378 * so return success here so the calling functions
379 * won't attempt to free is again.
384 skb
->dev
= slave_dev
;
394 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
395 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
397 * a. This operation is performed in IOCTL context,
398 * b. The operation is protected by the RTNL semaphore in the 8021q code,
399 * c. Holding a lock with BH disabled while directly calling a base driver
400 * entry point is generally a BAD idea.
402 * The design of synchronization/protection for this operation in the 8021q
403 * module is good for one or more VLAN devices over a single physical device
404 * and cannot be extended for a teaming solution like bonding, so there is a
405 * potential race condition here where a net device from the vlan group might
406 * be referenced (either by a base driver or the 8021q code) while it is being
407 * removed from the system. However, it turns out we're not making matters
408 * worse, and if it works for regular VLAN usage it will work here too.
412 * bond_vlan_rx_register - Propagates registration to slaves
413 * @bond_dev: bonding net device that got called
414 * @grp: vlan group being registered
416 static void bond_vlan_rx_register(struct net_device
*bond_dev
, struct vlan_group
*grp
)
418 struct bonding
*bond
= bond_dev
->priv
;
424 bond_for_each_slave(bond
, slave
, i
) {
425 struct net_device
*slave_dev
= slave
->dev
;
427 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
428 slave_dev
->vlan_rx_register
) {
429 slave_dev
->vlan_rx_register(slave_dev
, grp
);
435 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
436 * @bond_dev: bonding net device that got called
437 * @vid: vlan id being added
439 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
441 struct bonding
*bond
= bond_dev
->priv
;
445 bond_for_each_slave(bond
, slave
, i
) {
446 struct net_device
*slave_dev
= slave
->dev
;
448 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
449 slave_dev
->vlan_rx_add_vid
) {
450 slave_dev
->vlan_rx_add_vid(slave_dev
, vid
);
454 res
= bond_add_vlan(bond
, vid
);
456 printk(KERN_ERR DRV_NAME
457 ": %s: Error: Failed to add vlan id %d\n",
458 bond_dev
->name
, vid
);
463 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
464 * @bond_dev: bonding net device that got called
465 * @vid: vlan id being removed
467 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
469 struct bonding
*bond
= bond_dev
->priv
;
471 struct net_device
*vlan_dev
;
474 bond_for_each_slave(bond
, slave
, i
) {
475 struct net_device
*slave_dev
= slave
->dev
;
477 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
478 slave_dev
->vlan_rx_kill_vid
) {
479 /* Save and then restore vlan_dev in the grp array,
480 * since the slave's driver might clear it.
482 vlan_dev
= bond
->vlgrp
->vlan_devices
[vid
];
483 slave_dev
->vlan_rx_kill_vid(slave_dev
, vid
);
484 bond
->vlgrp
->vlan_devices
[vid
] = vlan_dev
;
488 res
= bond_del_vlan(bond
, vid
);
490 printk(KERN_ERR DRV_NAME
491 ": %s: Error: Failed to remove vlan id %d\n",
492 bond_dev
->name
, vid
);
496 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
498 struct vlan_entry
*vlan
;
500 write_lock_bh(&bond
->lock
);
502 if (list_empty(&bond
->vlan_list
)) {
506 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
507 slave_dev
->vlan_rx_register
) {
508 slave_dev
->vlan_rx_register(slave_dev
, bond
->vlgrp
);
511 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
512 !(slave_dev
->vlan_rx_add_vid
)) {
516 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
517 slave_dev
->vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
521 write_unlock_bh(&bond
->lock
);
524 static void bond_del_vlans_from_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
526 struct vlan_entry
*vlan
;
527 struct net_device
*vlan_dev
;
529 write_lock_bh(&bond
->lock
);
531 if (list_empty(&bond
->vlan_list
)) {
535 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
536 !(slave_dev
->vlan_rx_kill_vid
)) {
540 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
541 /* Save and then restore vlan_dev in the grp array,
542 * since the slave's driver might clear it.
544 vlan_dev
= bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
];
545 slave_dev
->vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
546 bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
] = vlan_dev
;
550 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
551 slave_dev
->vlan_rx_register
) {
552 slave_dev
->vlan_rx_register(slave_dev
, NULL
);
556 write_unlock_bh(&bond
->lock
);
559 /*------------------------------- Link status -------------------------------*/
562 * Get link speed and duplex from the slave's base driver
563 * using ethtool. If for some reason the call fails or the
564 * values are invalid, fake speed and duplex to 100/Full
567 static int bond_update_speed_duplex(struct slave
*slave
)
569 struct net_device
*slave_dev
= slave
->dev
;
570 static int (* ioctl
)(struct net_device
*, struct ifreq
*, int);
572 struct ethtool_cmd etool
;
574 /* Fake speed and duplex */
575 slave
->speed
= SPEED_100
;
576 slave
->duplex
= DUPLEX_FULL
;
578 if (slave_dev
->ethtool_ops
) {
581 if (!slave_dev
->ethtool_ops
->get_settings
) {
585 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
593 ioctl
= slave_dev
->do_ioctl
;
594 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
595 etool
.cmd
= ETHTOOL_GSET
;
596 ifr
.ifr_data
= (char*)&etool
;
597 if (!ioctl
|| (IOCTL(slave_dev
, &ifr
, SIOCETHTOOL
) < 0)) {
602 switch (etool
.speed
) {
611 switch (etool
.duplex
) {
619 slave
->speed
= etool
.speed
;
620 slave
->duplex
= etool
.duplex
;
626 * if <dev> supports MII link status reporting, check its link status.
628 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
629 * depening upon the setting of the use_carrier parameter.
631 * Return either BMSR_LSTATUS, meaning that the link is up (or we
632 * can't tell and just pretend it is), or 0, meaning that the link is
635 * If reporting is non-zero, instead of faking link up, return -1 if
636 * both ETHTOOL and MII ioctls fail (meaning the device does not
637 * support them). If use_carrier is set, return whatever it says.
638 * It'd be nice if there was a good way to tell if a driver supports
639 * netif_carrier, but there really isn't.
641 static int bond_check_dev_link(struct bonding
*bond
, struct net_device
*slave_dev
, int reporting
)
643 static int (* ioctl
)(struct net_device
*, struct ifreq
*, int);
645 struct mii_ioctl_data
*mii
;
646 struct ethtool_value etool
;
648 if (bond
->params
.use_carrier
) {
649 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
652 ioctl
= slave_dev
->do_ioctl
;
654 /* TODO: set pointer to correct ioctl on a per team member */
655 /* bases to make this more efficient. that is, once */
656 /* we determine the correct ioctl, we will always */
657 /* call it and not the others for that team */
661 * We cannot assume that SIOCGMIIPHY will also read a
662 * register; not all network drivers (e.g., e100)
666 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
667 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
669 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
670 mii
->reg_num
= MII_BMSR
;
671 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0) {
672 return (mii
->val_out
& BMSR_LSTATUS
);
677 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
678 /* for a period of time so we attempt to get link status */
679 /* from it last if the above MII ioctls fail... */
680 if (slave_dev
->ethtool_ops
) {
681 if (slave_dev
->ethtool_ops
->get_link
) {
684 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
686 return link
? BMSR_LSTATUS
: 0;
691 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
692 etool
.cmd
= ETHTOOL_GLINK
;
693 ifr
.ifr_data
= (char*)&etool
;
694 if (IOCTL(slave_dev
, &ifr
, SIOCETHTOOL
) == 0) {
695 if (etool
.data
== 1) {
698 dprintk("SIOCETHTOOL shows link down\n");
705 * If reporting, report that either there's no dev->do_ioctl,
706 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
707 * cannot report link status). If not reporting, pretend
710 return (reporting
? -1 : BMSR_LSTATUS
);
713 /*----------------------------- Multicast list ------------------------------*/
716 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
718 static inline int bond_is_dmi_same(struct dev_mc_list
*dmi1
, struct dev_mc_list
*dmi2
)
720 return memcmp(dmi1
->dmi_addr
, dmi2
->dmi_addr
, dmi1
->dmi_addrlen
) == 0 &&
721 dmi1
->dmi_addrlen
== dmi2
->dmi_addrlen
;
725 * returns dmi entry if found, NULL otherwise
727 static struct dev_mc_list
*bond_mc_list_find_dmi(struct dev_mc_list
*dmi
, struct dev_mc_list
*mc_list
)
729 struct dev_mc_list
*idmi
;
731 for (idmi
= mc_list
; idmi
; idmi
= idmi
->next
) {
732 if (bond_is_dmi_same(dmi
, idmi
)) {
741 * Push the promiscuity flag down to appropriate slaves
743 static void bond_set_promiscuity(struct bonding
*bond
, int inc
)
745 if (USES_PRIMARY(bond
->params
.mode
)) {
746 /* write lock already acquired */
747 if (bond
->curr_active_slave
) {
748 dev_set_promiscuity(bond
->curr_active_slave
->dev
, inc
);
753 bond_for_each_slave(bond
, slave
, i
) {
754 dev_set_promiscuity(slave
->dev
, inc
);
760 * Push the allmulti flag down to all slaves
762 static void bond_set_allmulti(struct bonding
*bond
, int inc
)
764 if (USES_PRIMARY(bond
->params
.mode
)) {
765 /* write lock already acquired */
766 if (bond
->curr_active_slave
) {
767 dev_set_allmulti(bond
->curr_active_slave
->dev
, inc
);
772 bond_for_each_slave(bond
, slave
, i
) {
773 dev_set_allmulti(slave
->dev
, inc
);
779 * Add a Multicast address to slaves
782 static void bond_mc_add(struct bonding
*bond
, void *addr
, int alen
)
784 if (USES_PRIMARY(bond
->params
.mode
)) {
785 /* write lock already acquired */
786 if (bond
->curr_active_slave
) {
787 dev_mc_add(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
792 bond_for_each_slave(bond
, slave
, i
) {
793 dev_mc_add(slave
->dev
, addr
, alen
, 0);
799 * Remove a multicast address from slave
802 static void bond_mc_delete(struct bonding
*bond
, void *addr
, int alen
)
804 if (USES_PRIMARY(bond
->params
.mode
)) {
805 /* write lock already acquired */
806 if (bond
->curr_active_slave
) {
807 dev_mc_delete(bond
->curr_active_slave
->dev
, addr
, alen
, 0);
812 bond_for_each_slave(bond
, slave
, i
) {
813 dev_mc_delete(slave
->dev
, addr
, alen
, 0);
819 * Totally destroys the mc_list in bond
821 static void bond_mc_list_destroy(struct bonding
*bond
)
823 struct dev_mc_list
*dmi
;
827 bond
->mc_list
= dmi
->next
;
834 * Copy all the Multicast addresses from src to the bonding device dst
836 static int bond_mc_list_copy(struct dev_mc_list
*mc_list
, struct bonding
*bond
,
839 struct dev_mc_list
*dmi
, *new_dmi
;
841 for (dmi
= mc_list
; dmi
; dmi
= dmi
->next
) {
842 new_dmi
= kmalloc(sizeof(struct dev_mc_list
), gfp_flag
);
845 /* FIXME: Potential memory leak !!! */
849 new_dmi
->next
= bond
->mc_list
;
850 bond
->mc_list
= new_dmi
;
851 new_dmi
->dmi_addrlen
= dmi
->dmi_addrlen
;
852 memcpy(new_dmi
->dmi_addr
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
853 new_dmi
->dmi_users
= dmi
->dmi_users
;
854 new_dmi
->dmi_gusers
= dmi
->dmi_gusers
;
861 * flush all members of flush->mc_list from device dev->mc_list
863 static void bond_mc_list_flush(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
865 struct bonding
*bond
= bond_dev
->priv
;
866 struct dev_mc_list
*dmi
;
868 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
869 dev_mc_delete(slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
872 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
873 /* del lacpdu mc addr from mc list */
874 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
876 dev_mc_delete(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
880 /*--------------------------- Active slave change ---------------------------*/
883 * Update the mc list and multicast-related flags for the new and
884 * old active slaves (if any) according to the multicast mode, and
885 * promiscuous flags unconditionally.
887 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
, struct slave
*old_active
)
889 struct dev_mc_list
*dmi
;
891 if (!USES_PRIMARY(bond
->params
.mode
)) {
892 /* nothing to do - mc list is already up-to-date on
899 if (bond
->dev
->flags
& IFF_PROMISC
) {
900 dev_set_promiscuity(old_active
->dev
, -1);
903 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
904 dev_set_allmulti(old_active
->dev
, -1);
907 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
908 dev_mc_delete(old_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
913 if (bond
->dev
->flags
& IFF_PROMISC
) {
914 dev_set_promiscuity(new_active
->dev
, 1);
917 if (bond
->dev
->flags
& IFF_ALLMULTI
) {
918 dev_set_allmulti(new_active
->dev
, 1);
921 for (dmi
= bond
->dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
922 dev_mc_add(new_active
->dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
928 * find_best_interface - select the best available slave to be the active one
929 * @bond: our bonding struct
931 * Warning: Caller must hold curr_slave_lock for writing.
933 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
935 struct slave
*new_active
, *old_active
;
936 struct slave
*bestslave
= NULL
;
937 int mintime
= bond
->params
.updelay
;
940 new_active
= old_active
= bond
->curr_active_slave
;
942 if (!new_active
) { /* there were no active slaves left */
943 if (bond
->slave_cnt
> 0) { /* found one slave */
944 new_active
= bond
->first_slave
;
946 return NULL
; /* still no slave, return NULL */
950 /* first try the primary link; if arping, a link must tx/rx traffic
951 * before it can be considered the curr_active_slave - also, we would skip
952 * slaves between the curr_active_slave and primary_slave that may be up
955 if ((bond
->primary_slave
) &&
956 (!bond
->params
.arp_interval
) &&
957 (IS_UP(bond
->primary_slave
->dev
))) {
958 new_active
= bond
->primary_slave
;
961 /* remember where to stop iterating over the slaves */
962 old_active
= new_active
;
964 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
965 if (IS_UP(new_active
->dev
)) {
966 if (new_active
->link
== BOND_LINK_UP
) {
968 } else if (new_active
->link
== BOND_LINK_BACK
) {
969 /* link up, but waiting for stabilization */
970 if (new_active
->delay
< mintime
) {
971 mintime
= new_active
->delay
;
972 bestslave
= new_active
;
982 * change_active_interface - change the active slave into the specified one
983 * @bond: our bonding struct
984 * @new: the new slave to make the active one
986 * Set the new slave to the bond's settings and unset them on the old
988 * Setting include flags, mc-list, promiscuity, allmulti, etc.
990 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
991 * because it is apparently the best available slave we have, even though its
992 * updelay hasn't timed out yet.
994 * Warning: Caller must hold curr_slave_lock for writing.
996 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
998 struct slave
*old_active
= bond
->curr_active_slave
;
1000 if (old_active
== new_active
) {
1005 if (new_active
->link
== BOND_LINK_BACK
) {
1006 if (USES_PRIMARY(bond
->params
.mode
)) {
1007 printk(KERN_INFO DRV_NAME
1008 ": %s: making interface %s the new "
1009 "active one %d ms earlier.\n",
1010 bond
->dev
->name
, new_active
->dev
->name
,
1011 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1014 new_active
->delay
= 0;
1015 new_active
->link
= BOND_LINK_UP
;
1016 new_active
->jiffies
= jiffies
;
1018 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1019 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1022 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1023 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1024 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1027 if (USES_PRIMARY(bond
->params
.mode
)) {
1028 printk(KERN_INFO DRV_NAME
1029 ": %s: making interface %s the new "
1031 bond
->dev
->name
, new_active
->dev
->name
);
1036 if (USES_PRIMARY(bond
->params
.mode
)) {
1037 bond_mc_swap(bond
, new_active
, old_active
);
1040 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1041 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1042 bond_alb_handle_active_change(bond
, new_active
);
1044 bond
->curr_active_slave
= new_active
;
1047 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1049 bond_set_slave_inactive_flags(old_active
);
1053 bond_set_slave_active_flags(new_active
);
1055 bond_send_gratuitous_arp(bond
);
1060 * bond_select_active_slave - select a new active slave, if needed
1061 * @bond: our bonding struct
1063 * This functions shoud be called when one of the following occurs:
1064 * - The old curr_active_slave has been released or lost its link.
1065 * - The primary_slave has got its link back.
1066 * - A slave has got its link back and there's no old curr_active_slave.
1068 * Warning: Caller must hold curr_slave_lock for writing.
1070 void bond_select_active_slave(struct bonding
*bond
)
1072 struct slave
*best_slave
;
1074 best_slave
= bond_find_best_slave(bond
);
1075 if (best_slave
!= bond
->curr_active_slave
) {
1076 bond_change_active_slave(bond
, best_slave
);
1080 /*--------------------------- slave list handling ---------------------------*/
1083 * This function attaches the slave to the end of list.
1085 * bond->lock held for writing by caller.
1087 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1089 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1090 new_slave
->next
= new_slave
;
1091 new_slave
->prev
= new_slave
;
1092 bond
->first_slave
= new_slave
;
1094 new_slave
->next
= bond
->first_slave
;
1095 new_slave
->prev
= bond
->first_slave
->prev
;
1096 new_slave
->next
->prev
= new_slave
;
1097 new_slave
->prev
->next
= new_slave
;
1104 * This function detaches the slave from the list.
1105 * WARNING: no check is made to verify if the slave effectively
1106 * belongs to <bond>.
1107 * Nothing is freed on return, structures are just unchained.
1108 * If any slave pointer in bond was pointing to <slave>,
1109 * it should be changed by the calling function.
1111 * bond->lock held for writing by caller.
1113 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1116 slave
->next
->prev
= slave
->prev
;
1120 slave
->prev
->next
= slave
->next
;
1123 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1124 if (bond
->slave_cnt
> 1) { /* there are more slave */
1125 bond
->first_slave
= slave
->next
;
1127 bond
->first_slave
= NULL
; /* slave was the last one */
1136 /*---------------------------------- IOCTL ----------------------------------*/
1138 int bond_sethwaddr(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1140 dprintk("bond_dev=%p\n", bond_dev
);
1141 dprintk("slave_dev=%p\n", slave_dev
);
1142 dprintk("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1143 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1147 #define BOND_INTERSECT_FEATURES \
1148 (NETIF_F_SG|NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM|\
1149 NETIF_F_TSO|NETIF_F_UFO)
1152 * Compute the common dev->feature set available to all slaves. Some
1153 * feature bits are managed elsewhere, so preserve feature bits set on
1154 * master device that are not part of the examined set.
1156 static int bond_compute_features(struct bonding
*bond
)
1158 unsigned long features
= BOND_INTERSECT_FEATURES
;
1159 struct slave
*slave
;
1160 struct net_device
*bond_dev
= bond
->dev
;
1163 bond_for_each_slave(bond
, slave
, i
)
1164 features
&= (slave
->dev
->features
& BOND_INTERSECT_FEATURES
);
1166 if ((features
& NETIF_F_SG
) &&
1167 !(features
& (NETIF_F_IP_CSUM
|
1170 features
&= ~NETIF_F_SG
;
1173 * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1174 * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1175 * implies that all slaves also support scatter-gather
1176 * (NETIF_F_SG), which implies that features also includes
1177 * NETIF_F_SG. So no need to check whether we have an
1178 * illegal combination of NETIF_F_{TSO,UFO} and
1182 features
|= (bond_dev
->features
& ~BOND_INTERSECT_FEATURES
);
1183 bond_dev
->features
= features
;
1188 /* enslave device <slave> to bond device <master> */
1189 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1191 struct bonding
*bond
= bond_dev
->priv
;
1192 struct slave
*new_slave
= NULL
;
1193 struct dev_mc_list
*dmi
;
1194 struct sockaddr addr
;
1196 int old_features
= bond_dev
->features
;
1199 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1200 slave_dev
->do_ioctl
== NULL
) {
1201 printk(KERN_WARNING DRV_NAME
1202 ": %s: Warning: no link monitoring support for %s\n",
1203 bond_dev
->name
, slave_dev
->name
);
1206 /* bond must be initialized by bond_open() before enslaving */
1207 if (!(bond_dev
->flags
& IFF_UP
)) {
1208 dprintk("Error, master_dev is not up\n");
1212 /* already enslaved */
1213 if (slave_dev
->flags
& IFF_SLAVE
) {
1214 dprintk("Error, Device was already enslaved\n");
1218 /* vlan challenged mutual exclusion */
1219 /* no need to lock since we're protected by rtnl_lock */
1220 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1221 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1222 if (!list_empty(&bond
->vlan_list
)) {
1223 printk(KERN_ERR DRV_NAME
1224 ": %s: Error: cannot enslave VLAN "
1225 "challenged slave %s on VLAN enabled "
1226 "bond %s\n", bond_dev
->name
, slave_dev
->name
,
1230 printk(KERN_WARNING DRV_NAME
1231 ": %s: Warning: enslaved VLAN challenged "
1232 "slave %s. Adding VLANs will be blocked as "
1233 "long as %s is part of bond %s\n",
1234 bond_dev
->name
, slave_dev
->name
, slave_dev
->name
,
1236 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1239 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1240 if (bond
->slave_cnt
== 0) {
1241 /* First slave, and it is not VLAN challenged,
1242 * so remove the block of adding VLANs over the bond.
1244 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1249 * Old ifenslave binaries are no longer supported. These can
1250 * be identified with moderate accurary by the state of the slave:
1251 * the current ifenslave will set the interface down prior to
1252 * enslaving it; the old ifenslave will not.
1254 if ((slave_dev
->flags
& IFF_UP
)) {
1255 printk(KERN_ERR DRV_NAME
": %s is up. "
1256 "This may be due to an out of date ifenslave.\n",
1259 goto err_undo_flags
;
1262 if (slave_dev
->set_mac_address
== NULL
) {
1263 printk(KERN_ERR DRV_NAME
1264 ": %s: Error: The slave device you specified does "
1265 "not support setting the MAC address. "
1266 "Your kernel likely does not support slave "
1267 "devices.\n", bond_dev
->name
);
1269 goto err_undo_flags
;
1272 new_slave
= kmalloc(sizeof(struct slave
), GFP_KERNEL
);
1275 goto err_undo_flags
;
1278 memset(new_slave
, 0, sizeof(struct slave
));
1280 /* save slave's original flags before calling
1281 * netdev_set_master and dev_open
1283 new_slave
->original_flags
= slave_dev
->flags
;
1286 * Save slave's original ("permanent") mac address for modes
1287 * that need it, and for restoring it upon release, and then
1288 * set it to the master's address
1290 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1293 * Set slave to master's mac address. The application already
1294 * set the master's mac address to that of the first slave
1296 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1297 addr
.sa_family
= slave_dev
->type
;
1298 res
= dev_set_mac_address(slave_dev
, &addr
);
1300 dprintk("Error %d calling set_mac_address\n", res
);
1304 /* open the slave since the application closed it */
1305 res
= dev_open(slave_dev
);
1307 dprintk("Openning slave %s failed\n", slave_dev
->name
);
1308 goto err_restore_mac
;
1311 res
= netdev_set_master(slave_dev
, bond_dev
);
1313 dprintk("Error %d calling netdev_set_master\n", res
);
1317 new_slave
->dev
= slave_dev
;
1319 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1320 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1321 /* bond_alb_init_slave() must be called before all other stages since
1322 * it might fail and we do not want to have to undo everything
1324 res
= bond_alb_init_slave(bond
, new_slave
);
1326 goto err_unset_master
;
1330 /* If the mode USES_PRIMARY, then the new slave gets the
1331 * master's promisc (and mc) settings only if it becomes the
1332 * curr_active_slave, and that is taken care of later when calling
1333 * bond_change_active()
1335 if (!USES_PRIMARY(bond
->params
.mode
)) {
1336 /* set promiscuity level to new slave */
1337 if (bond_dev
->flags
& IFF_PROMISC
) {
1338 dev_set_promiscuity(slave_dev
, 1);
1341 /* set allmulti level to new slave */
1342 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1343 dev_set_allmulti(slave_dev
, 1);
1346 /* upload master's mc_list to new slave */
1347 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
1348 dev_mc_add (slave_dev
, dmi
->dmi_addr
, dmi
->dmi_addrlen
, 0);
1352 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1353 /* add lacpdu mc addr to mc list */
1354 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1356 dev_mc_add(slave_dev
, lacpdu_multicast
, ETH_ALEN
, 0);
1359 bond_add_vlans_on_slave(bond
, slave_dev
);
1361 write_lock_bh(&bond
->lock
);
1363 bond_attach_slave(bond
, new_slave
);
1365 new_slave
->delay
= 0;
1366 new_slave
->link_failure_count
= 0;
1368 bond_compute_features(bond
);
1370 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1371 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1373 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1375 * miimon is set but a bonded network driver
1376 * does not support ETHTOOL/MII and
1377 * arp_interval is not set. Note: if
1378 * use_carrier is enabled, we will never go
1379 * here (because netif_carrier is always
1380 * supported); thus, we don't need to change
1381 * the messages for netif_carrier.
1383 printk(KERN_WARNING DRV_NAME
1384 ": %s: Warning: MII and ETHTOOL support not "
1385 "available for interface %s, and "
1386 "arp_interval/arp_ip_target module parameters "
1387 "not specified, thus bonding will not detect "
1388 "link failures! see bonding.txt for details.\n",
1389 bond_dev
->name
, slave_dev
->name
);
1390 } else if (link_reporting
== -1) {
1391 /* unable get link status using mii/ethtool */
1392 printk(KERN_WARNING DRV_NAME
1393 ": %s: Warning: can't get link status from "
1394 "interface %s; the network driver associated "
1395 "with this interface does not support MII or "
1396 "ETHTOOL link status reporting, thus miimon "
1397 "has no effect on this interface.\n",
1398 bond_dev
->name
, slave_dev
->name
);
1402 /* check for initial state */
1403 if (!bond
->params
.miimon
||
1404 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1405 if (bond
->params
.updelay
) {
1406 dprintk("Initial state of slave_dev is "
1407 "BOND_LINK_BACK\n");
1408 new_slave
->link
= BOND_LINK_BACK
;
1409 new_slave
->delay
= bond
->params
.updelay
;
1411 dprintk("Initial state of slave_dev is "
1413 new_slave
->link
= BOND_LINK_UP
;
1415 new_slave
->jiffies
= jiffies
;
1417 dprintk("Initial state of slave_dev is "
1418 "BOND_LINK_DOWN\n");
1419 new_slave
->link
= BOND_LINK_DOWN
;
1422 if (bond_update_speed_duplex(new_slave
) &&
1423 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1424 printk(KERN_WARNING DRV_NAME
1425 ": %s: Warning: failed to get speed and duplex from %s, "
1426 "assumed to be 100Mb/sec and Full.\n",
1427 bond_dev
->name
, new_slave
->dev
->name
);
1429 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1430 printk(KERN_WARNING DRV_NAME
1431 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1432 "support in base driver for proper aggregator "
1433 "selection.\n", bond_dev
->name
);
1437 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1438 /* if there is a primary slave, remember it */
1439 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1440 bond
->primary_slave
= new_slave
;
1444 switch (bond
->params
.mode
) {
1445 case BOND_MODE_ACTIVEBACKUP
:
1446 /* if we're in active-backup mode, we need one and only one active
1447 * interface. The backup interfaces will have their NOARP flag set
1448 * because we need them to be completely deaf and not to respond to
1449 * any ARP request on the network to avoid fooling a switch. Thus,
1450 * since we guarantee that curr_active_slave always point to the last
1451 * usable interface, we just have to verify this interface's flag.
1453 if (((!bond
->curr_active_slave
) ||
1454 (bond
->curr_active_slave
->dev
->flags
& IFF_NOARP
)) &&
1455 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1456 dprintk("This is the first active slave\n");
1457 /* first slave or no active slave yet, and this link
1458 is OK, so make this interface the active one */
1459 bond_change_active_slave(bond
, new_slave
);
1461 dprintk("This is just a backup slave\n");
1462 bond_set_slave_inactive_flags(new_slave
);
1465 case BOND_MODE_8023AD
:
1466 /* in 802.3ad mode, the internal mechanism
1467 * will activate the slaves in the selected
1470 bond_set_slave_inactive_flags(new_slave
);
1471 /* if this is the first slave */
1472 if (bond
->slave_cnt
== 1) {
1473 SLAVE_AD_INFO(new_slave
).id
= 1;
1474 /* Initialize AD with the number of times that the AD timer is called in 1 second
1475 * can be called only after the mac address of the bond is set
1477 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1478 bond
->params
.lacp_fast
);
1480 SLAVE_AD_INFO(new_slave
).id
=
1481 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1484 bond_3ad_bind_slave(new_slave
);
1488 new_slave
->state
= BOND_STATE_ACTIVE
;
1489 if ((!bond
->curr_active_slave
) &&
1490 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1491 /* first slave or no active slave yet, and this link
1492 * is OK, so make this interface the active one
1494 bond_change_active_slave(bond
, new_slave
);
1498 dprintk("This slave is always active in trunk mode\n");
1500 /* always active in trunk mode */
1501 new_slave
->state
= BOND_STATE_ACTIVE
;
1503 /* In trunking mode there is little meaning to curr_active_slave
1504 * anyway (it holds no special properties of the bond device),
1505 * so we can change it without calling change_active_interface()
1507 if (!bond
->curr_active_slave
) {
1508 bond
->curr_active_slave
= new_slave
;
1511 } /* switch(bond_mode) */
1513 write_unlock_bh(&bond
->lock
);
1515 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1517 goto err_unset_master
;
1519 printk(KERN_INFO DRV_NAME
1520 ": %s: enslaving %s as a%s interface with a%s link.\n",
1521 bond_dev
->name
, slave_dev
->name
,
1522 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1523 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1525 /* enslave is successful */
1528 /* Undo stages on error */
1530 netdev_set_master(slave_dev
, NULL
);
1533 dev_close(slave_dev
);
1536 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1537 addr
.sa_family
= slave_dev
->type
;
1538 dev_set_mac_address(slave_dev
, &addr
);
1544 bond_dev
->features
= old_features
;
1550 * Try to release the slave device <slave> from the bond device <master>
1551 * It is legal to access curr_active_slave without a lock because all the function
1554 * The rules for slave state should be:
1555 * for Active/Backup:
1556 * Active stays on all backups go down
1557 * for Bonded connections:
1558 * The first up interface should be left on and all others downed.
1560 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1562 struct bonding
*bond
= bond_dev
->priv
;
1563 struct slave
*slave
, *oldcurrent
;
1564 struct sockaddr addr
;
1565 int mac_addr_differ
;
1567 /* slave is not a slave or master is not master of this slave */
1568 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1569 (slave_dev
->master
!= bond_dev
)) {
1570 printk(KERN_ERR DRV_NAME
1571 ": %s: Error: cannot release %s.\n",
1572 bond_dev
->name
, slave_dev
->name
);
1576 write_lock_bh(&bond
->lock
);
1578 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1580 /* not a slave of this bond */
1581 printk(KERN_INFO DRV_NAME
1582 ": %s: %s not enslaved\n",
1583 bond_dev
->name
, slave_dev
->name
);
1584 write_unlock_bh(&bond
->lock
);
1588 mac_addr_differ
= memcmp(bond_dev
->dev_addr
,
1591 if (!mac_addr_differ
&& (bond
->slave_cnt
> 1)) {
1592 printk(KERN_WARNING DRV_NAME
1593 ": %s: Warning: the permanent HWaddr of %s "
1594 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1595 "still in use by %s. Set the HWaddr of "
1596 "%s to a different address to avoid "
1600 slave
->perm_hwaddr
[0],
1601 slave
->perm_hwaddr
[1],
1602 slave
->perm_hwaddr
[2],
1603 slave
->perm_hwaddr
[3],
1604 slave
->perm_hwaddr
[4],
1605 slave
->perm_hwaddr
[5],
1610 /* Inform AD package of unbinding of slave. */
1611 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1612 /* must be called before the slave is
1613 * detached from the list
1615 bond_3ad_unbind_slave(slave
);
1618 printk(KERN_INFO DRV_NAME
1619 ": %s: releasing %s interface %s\n",
1621 (slave
->state
== BOND_STATE_ACTIVE
)
1622 ? "active" : "backup",
1625 oldcurrent
= bond
->curr_active_slave
;
1627 bond
->current_arp_slave
= NULL
;
1629 /* release the slave from its bond */
1630 bond_detach_slave(bond
, slave
);
1632 bond_compute_features(bond
);
1634 if (bond
->primary_slave
== slave
) {
1635 bond
->primary_slave
= NULL
;
1638 if (oldcurrent
== slave
) {
1639 bond_change_active_slave(bond
, NULL
);
1642 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1643 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1644 /* Must be called only after the slave has been
1645 * detached from the list and the curr_active_slave
1646 * has been cleared (if our_slave == old_current),
1647 * but before a new active slave is selected.
1649 bond_alb_deinit_slave(bond
, slave
);
1652 if (oldcurrent
== slave
) {
1653 bond_select_active_slave(bond
);
1655 if (!bond
->curr_active_slave
) {
1656 printk(KERN_INFO DRV_NAME
1657 ": %s: now running without any active "
1663 if (bond
->slave_cnt
== 0) {
1664 /* if the last slave was removed, zero the mac address
1665 * of the master so it will be set by the application
1666 * to the mac address of the first slave
1668 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1670 if (list_empty(&bond
->vlan_list
)) {
1671 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1673 printk(KERN_WARNING DRV_NAME
1674 ": %s: Warning: clearing HW address of %s while it "
1675 "still has VLANs.\n",
1676 bond_dev
->name
, bond_dev
->name
);
1677 printk(KERN_WARNING DRV_NAME
1678 ": %s: When re-adding slaves, make sure the bond's "
1679 "HW address matches its VLANs'.\n",
1682 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1683 !bond_has_challenged_slaves(bond
)) {
1684 printk(KERN_INFO DRV_NAME
1685 ": %s: last VLAN challenged slave %s "
1686 "left bond %s. VLAN blocking is removed\n",
1687 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1688 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1691 write_unlock_bh(&bond
->lock
);
1693 /* must do this from outside any spinlocks */
1694 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1696 bond_del_vlans_from_slave(bond
, slave_dev
);
1698 /* If the mode USES_PRIMARY, then we should only remove its
1699 * promisc and mc settings if it was the curr_active_slave, but that was
1700 * already taken care of above when we detached the slave
1702 if (!USES_PRIMARY(bond
->params
.mode
)) {
1703 /* unset promiscuity level from slave */
1704 if (bond_dev
->flags
& IFF_PROMISC
) {
1705 dev_set_promiscuity(slave_dev
, -1);
1708 /* unset allmulti level from slave */
1709 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1710 dev_set_allmulti(slave_dev
, -1);
1713 /* flush master's mc_list from slave */
1714 bond_mc_list_flush(bond_dev
, slave_dev
);
1717 netdev_set_master(slave_dev
, NULL
);
1719 /* close slave before restoring its mac address */
1720 dev_close(slave_dev
);
1722 /* restore original ("permanent") mac address */
1723 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1724 addr
.sa_family
= slave_dev
->type
;
1725 dev_set_mac_address(slave_dev
, &addr
);
1727 /* restore the original state of the
1728 * IFF_NOARP flag that might have been
1729 * set by bond_set_slave_inactive_flags()
1731 if ((slave
->original_flags
& IFF_NOARP
) == 0) {
1732 slave_dev
->flags
&= ~IFF_NOARP
;
1737 return 0; /* deletion OK */
1741 * This function releases all slaves.
1743 static int bond_release_all(struct net_device
*bond_dev
)
1745 struct bonding
*bond
= bond_dev
->priv
;
1746 struct slave
*slave
;
1747 struct net_device
*slave_dev
;
1748 struct sockaddr addr
;
1750 write_lock_bh(&bond
->lock
);
1752 if (bond
->slave_cnt
== 0) {
1756 bond
->current_arp_slave
= NULL
;
1757 bond
->primary_slave
= NULL
;
1758 bond_change_active_slave(bond
, NULL
);
1760 while ((slave
= bond
->first_slave
) != NULL
) {
1761 /* Inform AD package of unbinding of slave
1762 * before slave is detached from the list.
1764 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1765 bond_3ad_unbind_slave(slave
);
1768 slave_dev
= slave
->dev
;
1769 bond_detach_slave(bond
, slave
);
1771 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
1772 (bond
->params
.mode
== BOND_MODE_ALB
)) {
1773 /* must be called only after the slave
1774 * has been detached from the list
1776 bond_alb_deinit_slave(bond
, slave
);
1779 bond_compute_features(bond
);
1781 /* now that the slave is detached, unlock and perform
1782 * all the undo steps that should not be called from
1785 write_unlock_bh(&bond
->lock
);
1787 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1788 bond_del_vlans_from_slave(bond
, slave_dev
);
1790 /* If the mode USES_PRIMARY, then we should only remove its
1791 * promisc and mc settings if it was the curr_active_slave, but that was
1792 * already taken care of above when we detached the slave
1794 if (!USES_PRIMARY(bond
->params
.mode
)) {
1795 /* unset promiscuity level from slave */
1796 if (bond_dev
->flags
& IFF_PROMISC
) {
1797 dev_set_promiscuity(slave_dev
, -1);
1800 /* unset allmulti level from slave */
1801 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1802 dev_set_allmulti(slave_dev
, -1);
1805 /* flush master's mc_list from slave */
1806 bond_mc_list_flush(bond_dev
, slave_dev
);
1809 netdev_set_master(slave_dev
, NULL
);
1811 /* close slave before restoring its mac address */
1812 dev_close(slave_dev
);
1814 /* restore original ("permanent") mac address*/
1815 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1816 addr
.sa_family
= slave_dev
->type
;
1817 dev_set_mac_address(slave_dev
, &addr
);
1819 /* restore the original state of the IFF_NOARP flag that might have
1820 * been set by bond_set_slave_inactive_flags()
1822 if ((slave
->original_flags
& IFF_NOARP
) == 0) {
1823 slave_dev
->flags
&= ~IFF_NOARP
;
1828 /* re-acquire the lock before getting the next slave */
1829 write_lock_bh(&bond
->lock
);
1832 /* zero the mac address of the master so it will be
1833 * set by the application to the mac address of the
1836 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1838 if (list_empty(&bond
->vlan_list
)) {
1839 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1841 printk(KERN_WARNING DRV_NAME
1842 ": %s: Warning: clearing HW address of %s while it "
1843 "still has VLANs.\n",
1844 bond_dev
->name
, bond_dev
->name
);
1845 printk(KERN_WARNING DRV_NAME
1846 ": %s: When re-adding slaves, make sure the bond's "
1847 "HW address matches its VLANs'.\n",
1851 printk(KERN_INFO DRV_NAME
1852 ": %s: released all slaves\n",
1856 write_unlock_bh(&bond
->lock
);
1862 * This function changes the active slave to slave <slave_dev>.
1863 * It returns -EINVAL in the following cases.
1864 * - <slave_dev> is not found in the list.
1865 * - There is not active slave now.
1866 * - <slave_dev> is already active.
1867 * - The link state of <slave_dev> is not BOND_LINK_UP.
1868 * - <slave_dev> is not running.
1869 * In these cases, this fuction does nothing.
1870 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1872 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1874 struct bonding
*bond
= bond_dev
->priv
;
1875 struct slave
*old_active
= NULL
;
1876 struct slave
*new_active
= NULL
;
1879 if (!USES_PRIMARY(bond
->params
.mode
)) {
1883 /* Verify that master_dev is indeed the master of slave_dev */
1884 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1885 (slave_dev
->master
!= bond_dev
)) {
1889 write_lock_bh(&bond
->lock
);
1891 old_active
= bond
->curr_active_slave
;
1892 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
1895 * Changing to the current active: do nothing; return success.
1897 if (new_active
&& (new_active
== old_active
)) {
1898 write_unlock_bh(&bond
->lock
);
1904 (new_active
->link
== BOND_LINK_UP
) &&
1905 IS_UP(new_active
->dev
)) {
1906 bond_change_active_slave(bond
, new_active
);
1911 write_unlock_bh(&bond
->lock
);
1916 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1918 struct bonding
*bond
= bond_dev
->priv
;
1920 info
->bond_mode
= bond
->params
.mode
;
1921 info
->miimon
= bond
->params
.miimon
;
1923 read_lock_bh(&bond
->lock
);
1924 info
->num_slaves
= bond
->slave_cnt
;
1925 read_unlock_bh(&bond
->lock
);
1930 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
1932 struct bonding
*bond
= bond_dev
->priv
;
1933 struct slave
*slave
;
1936 if (info
->slave_id
< 0) {
1940 read_lock_bh(&bond
->lock
);
1942 bond_for_each_slave(bond
, slave
, i
) {
1943 if (i
== (int)info
->slave_id
) {
1949 read_unlock_bh(&bond
->lock
);
1952 strcpy(info
->slave_name
, slave
->dev
->name
);
1953 info
->link
= slave
->link
;
1954 info
->state
= slave
->state
;
1955 info
->link_failure_count
= slave
->link_failure_count
;
1963 /*-------------------------------- Monitoring -------------------------------*/
1965 /* this function is called regularly to monitor each slave's link. */
1966 void bond_mii_monitor(struct net_device
*bond_dev
)
1968 struct bonding
*bond
= bond_dev
->priv
;
1969 struct slave
*slave
, *oldcurrent
;
1970 int do_failover
= 0;
1974 read_lock(&bond
->lock
);
1976 delta_in_ticks
= (bond
->params
.miimon
* HZ
) / 1000;
1978 if (bond
->kill_timers
) {
1982 if (bond
->slave_cnt
== 0) {
1986 /* we will try to read the link status of each of our slaves, and
1987 * set their IFF_RUNNING flag appropriately. For each slave not
1988 * supporting MII status, we won't do anything so that a user-space
1989 * program could monitor the link itself if needed.
1992 read_lock(&bond
->curr_slave_lock
);
1993 oldcurrent
= bond
->curr_active_slave
;
1994 read_unlock(&bond
->curr_slave_lock
);
1996 bond_for_each_slave(bond
, slave
, i
) {
1997 struct net_device
*slave_dev
= slave
->dev
;
1999 u16 old_speed
= slave
->speed
;
2000 u8 old_duplex
= slave
->duplex
;
2002 link_state
= bond_check_dev_link(bond
, slave_dev
, 0);
2004 switch (slave
->link
) {
2005 case BOND_LINK_UP
: /* the link was up */
2006 if (link_state
== BMSR_LSTATUS
) {
2007 /* link stays up, nothing more to do */
2009 } else { /* link going down */
2010 slave
->link
= BOND_LINK_FAIL
;
2011 slave
->delay
= bond
->params
.downdelay
;
2013 if (slave
->link_failure_count
< UINT_MAX
) {
2014 slave
->link_failure_count
++;
2017 if (bond
->params
.downdelay
) {
2018 printk(KERN_INFO DRV_NAME
2019 ": %s: link status down for %s "
2020 "interface %s, disabling it in "
2024 ? ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
2025 ? ((slave
== oldcurrent
)
2026 ? "active " : "backup ")
2030 bond
->params
.downdelay
* bond
->params
.miimon
);
2033 /* no break ! fall through the BOND_LINK_FAIL test to
2034 ensure proper action to be taken
2036 case BOND_LINK_FAIL
: /* the link has just gone down */
2037 if (link_state
!= BMSR_LSTATUS
) {
2038 /* link stays down */
2039 if (slave
->delay
<= 0) {
2040 /* link down for too long time */
2041 slave
->link
= BOND_LINK_DOWN
;
2043 /* in active/backup mode, we must
2044 * completely disable this interface
2046 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) ||
2047 (bond
->params
.mode
== BOND_MODE_8023AD
)) {
2048 bond_set_slave_inactive_flags(slave
);
2051 printk(KERN_INFO DRV_NAME
2052 ": %s: link status definitely "
2053 "down for interface %s, "
2058 /* notify ad that the link status has changed */
2059 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2060 bond_3ad_handle_link_change(slave
, BOND_LINK_DOWN
);
2063 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
2064 (bond
->params
.mode
== BOND_MODE_ALB
)) {
2065 bond_alb_handle_link_change(bond
, slave
, BOND_LINK_DOWN
);
2068 if (slave
== oldcurrent
) {
2076 slave
->link
= BOND_LINK_UP
;
2077 slave
->jiffies
= jiffies
;
2078 printk(KERN_INFO DRV_NAME
2079 ": %s: link status up again after %d "
2080 "ms for interface %s.\n",
2082 (bond
->params
.downdelay
- slave
->delay
) * bond
->params
.miimon
,
2086 case BOND_LINK_DOWN
: /* the link was down */
2087 if (link_state
!= BMSR_LSTATUS
) {
2088 /* the link stays down, nothing more to do */
2090 } else { /* link going up */
2091 slave
->link
= BOND_LINK_BACK
;
2092 slave
->delay
= bond
->params
.updelay
;
2094 if (bond
->params
.updelay
) {
2095 /* if updelay == 0, no need to
2096 advertise about a 0 ms delay */
2097 printk(KERN_INFO DRV_NAME
2098 ": %s: link status up for "
2099 "interface %s, enabling it "
2103 bond
->params
.updelay
* bond
->params
.miimon
);
2106 /* no break ! fall through the BOND_LINK_BACK state in
2107 case there's something to do.
2109 case BOND_LINK_BACK
: /* the link has just come back */
2110 if (link_state
!= BMSR_LSTATUS
) {
2111 /* link down again */
2112 slave
->link
= BOND_LINK_DOWN
;
2114 printk(KERN_INFO DRV_NAME
2115 ": %s: link status down again after %d "
2116 "ms for interface %s.\n",
2118 (bond
->params
.updelay
- slave
->delay
) * bond
->params
.miimon
,
2122 if (slave
->delay
== 0) {
2123 /* now the link has been up for long time enough */
2124 slave
->link
= BOND_LINK_UP
;
2125 slave
->jiffies
= jiffies
;
2127 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2128 /* prevent it from being the active one */
2129 slave
->state
= BOND_STATE_BACKUP
;
2130 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2131 /* make it immediately active */
2132 slave
->state
= BOND_STATE_ACTIVE
;
2133 } else if (slave
!= bond
->primary_slave
) {
2134 /* prevent it from being the active one */
2135 slave
->state
= BOND_STATE_BACKUP
;
2138 printk(KERN_INFO DRV_NAME
2139 ": %s: link status definitely "
2140 "up for interface %s.\n",
2144 /* notify ad that the link status has changed */
2145 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2146 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2149 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
2150 (bond
->params
.mode
== BOND_MODE_ALB
)) {
2151 bond_alb_handle_link_change(bond
, slave
, BOND_LINK_UP
);
2154 if ((!oldcurrent
) ||
2155 (slave
== bond
->primary_slave
)) {
2164 /* Should not happen */
2165 printk(KERN_ERR DRV_NAME
2166 ": %s: Error: %s Illegal value (link=%d)\n",
2171 } /* end of switch (slave->link) */
2173 bond_update_speed_duplex(slave
);
2175 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2176 if (old_speed
!= slave
->speed
) {
2177 bond_3ad_adapter_speed_changed(slave
);
2180 if (old_duplex
!= slave
->duplex
) {
2181 bond_3ad_adapter_duplex_changed(slave
);
2188 write_lock(&bond
->curr_slave_lock
);
2190 bond_select_active_slave(bond
);
2192 if (oldcurrent
&& !bond
->curr_active_slave
) {
2193 printk(KERN_INFO DRV_NAME
2194 ": %s: now running without any active "
2199 write_unlock(&bond
->curr_slave_lock
);
2203 if (bond
->params
.miimon
) {
2204 mod_timer(&bond
->mii_timer
, jiffies
+ delta_in_ticks
);
2207 read_unlock(&bond
->lock
);
2211 static u32
bond_glean_dev_ip(struct net_device
*dev
)
2213 struct in_device
*idev
;
2214 struct in_ifaddr
*ifa
;
2221 idev
= __in_dev_get_rcu(dev
);
2225 ifa
= idev
->ifa_list
;
2229 addr
= ifa
->ifa_local
;
2235 static int bond_has_ip(struct bonding
*bond
)
2237 struct vlan_entry
*vlan
, *vlan_next
;
2239 if (bond
->master_ip
)
2242 if (list_empty(&bond
->vlan_list
))
2245 list_for_each_entry_safe(vlan
, vlan_next
, &bond
->vlan_list
,
2255 * We go to the (large) trouble of VLAN tagging ARP frames because
2256 * switches in VLAN mode (especially if ports are configured as
2257 * "native" to a VLAN) might not pass non-tagged frames.
2259 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, u32 dest_ip
, u32 src_ip
, unsigned short vlan_id
)
2261 struct sk_buff
*skb
;
2263 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2264 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2266 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2267 NULL
, slave_dev
->dev_addr
, NULL
);
2270 printk(KERN_ERR DRV_NAME
": ARP packet allocation failed\n");
2274 skb
= vlan_put_tag(skb
, vlan_id
);
2276 printk(KERN_ERR DRV_NAME
": failed to insert VLAN tag\n");
2284 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2287 u32
*targets
= bond
->params
.arp_targets
;
2288 struct vlan_entry
*vlan
, *vlan_next
;
2289 struct net_device
*vlan_dev
;
2293 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2296 dprintk("basa: target %x\n", targets
[i
]);
2297 if (list_empty(&bond
->vlan_list
)) {
2298 dprintk("basa: empty vlan: arp_send\n");
2299 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2300 bond
->master_ip
, 0);
2305 * If VLANs are configured, we do a route lookup to
2306 * determine which VLAN interface would be used, so we
2307 * can tag the ARP with the proper VLAN tag.
2309 memset(&fl
, 0, sizeof(fl
));
2310 fl
.fl4_dst
= targets
[i
];
2311 fl
.fl4_tos
= RTO_ONLINK
;
2313 rv
= ip_route_output_key(&rt
, &fl
);
2315 if (net_ratelimit()) {
2316 printk(KERN_WARNING DRV_NAME
2317 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2318 bond
->dev
->name
, NIPQUAD(fl
.fl4_dst
));
2324 * This target is not on a VLAN
2326 if (rt
->u
.dst
.dev
== bond
->dev
) {
2328 dprintk("basa: rtdev == bond->dev: arp_send\n");
2329 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2330 bond
->master_ip
, 0);
2335 list_for_each_entry_safe(vlan
, vlan_next
, &bond
->vlan_list
,
2337 vlan_dev
= bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
];
2338 if (vlan_dev
== rt
->u
.dst
.dev
) {
2339 vlan_id
= vlan
->vlan_id
;
2340 dprintk("basa: vlan match on %s %d\n",
2341 vlan_dev
->name
, vlan_id
);
2348 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2349 vlan
->vlan_ip
, vlan_id
);
2353 if (net_ratelimit()) {
2354 printk(KERN_WARNING DRV_NAME
2355 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2356 bond
->dev
->name
, NIPQUAD(fl
.fl4_dst
),
2357 rt
->u
.dst
.dev
? rt
->u
.dst
.dev
->name
: "NULL");
2364 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2365 * for each VLAN above us.
2367 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2369 struct slave
*slave
= bond
->curr_active_slave
;
2370 struct vlan_entry
*vlan
;
2371 struct net_device
*vlan_dev
;
2373 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond
->dev
->name
,
2374 slave
? slave
->dev
->name
: "NULL");
2378 if (bond
->master_ip
) {
2379 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2380 bond
->master_ip
, 0);
2383 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2384 vlan_dev
= bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
];
2385 if (vlan
->vlan_ip
) {
2386 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2387 vlan
->vlan_ip
, vlan
->vlan_id
);
2393 * this function is called regularly to monitor each slave's link
2394 * ensuring that traffic is being sent and received when arp monitoring
2395 * is used in load-balancing mode. if the adapter has been dormant, then an
2396 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2397 * arp monitoring in active backup mode.
2399 void bond_loadbalance_arp_mon(struct net_device
*bond_dev
)
2401 struct bonding
*bond
= bond_dev
->priv
;
2402 struct slave
*slave
, *oldcurrent
;
2403 int do_failover
= 0;
2407 read_lock(&bond
->lock
);
2409 delta_in_ticks
= (bond
->params
.arp_interval
* HZ
) / 1000;
2411 if (bond
->kill_timers
) {
2415 if (bond
->slave_cnt
== 0) {
2419 read_lock(&bond
->curr_slave_lock
);
2420 oldcurrent
= bond
->curr_active_slave
;
2421 read_unlock(&bond
->curr_slave_lock
);
2423 /* see if any of the previous devices are up now (i.e. they have
2424 * xmt and rcv traffic). the curr_active_slave does not come into
2425 * the picture unless it is null. also, slave->jiffies is not needed
2426 * here because we send an arp on each slave and give a slave as
2427 * long as it needs to get the tx/rx within the delta.
2428 * TODO: what about up/down delay in arp mode? it wasn't here before
2431 bond_for_each_slave(bond
, slave
, i
) {
2432 if (slave
->link
!= BOND_LINK_UP
) {
2433 if (((jiffies
- slave
->dev
->trans_start
) <= delta_in_ticks
) &&
2434 ((jiffies
- slave
->dev
->last_rx
) <= delta_in_ticks
)) {
2436 slave
->link
= BOND_LINK_UP
;
2437 slave
->state
= BOND_STATE_ACTIVE
;
2439 /* primary_slave has no meaning in round-robin
2440 * mode. the window of a slave being up and
2441 * curr_active_slave being null after enslaving
2445 printk(KERN_INFO DRV_NAME
2446 ": %s: link status definitely "
2447 "up for interface %s, ",
2452 printk(KERN_INFO DRV_NAME
2453 ": %s: interface %s is now up\n",
2459 /* slave->link == BOND_LINK_UP */
2461 /* not all switches will respond to an arp request
2462 * when the source ip is 0, so don't take the link down
2463 * if we don't know our ip yet
2465 if (((jiffies
- slave
->dev
->trans_start
) >= (2*delta_in_ticks
)) ||
2466 (((jiffies
- slave
->dev
->last_rx
) >= (2*delta_in_ticks
)) &&
2467 bond_has_ip(bond
))) {
2469 slave
->link
= BOND_LINK_DOWN
;
2470 slave
->state
= BOND_STATE_BACKUP
;
2472 if (slave
->link_failure_count
< UINT_MAX
) {
2473 slave
->link_failure_count
++;
2476 printk(KERN_INFO DRV_NAME
2477 ": %s: interface %s is now down.\n",
2481 if (slave
== oldcurrent
) {
2487 /* note: if switch is in round-robin mode, all links
2488 * must tx arp to ensure all links rx an arp - otherwise
2489 * links may oscillate or not come up at all; if switch is
2490 * in something like xor mode, there is nothing we can
2491 * do - all replies will be rx'ed on same link causing slaves
2492 * to be unstable during low/no traffic periods
2494 if (IS_UP(slave
->dev
)) {
2495 bond_arp_send_all(bond
, slave
);
2500 write_lock(&bond
->curr_slave_lock
);
2502 bond_select_active_slave(bond
);
2504 if (oldcurrent
&& !bond
->curr_active_slave
) {
2505 printk(KERN_INFO DRV_NAME
2506 ": %s: now running without any active "
2511 write_unlock(&bond
->curr_slave_lock
);
2515 if (bond
->params
.arp_interval
) {
2516 mod_timer(&bond
->arp_timer
, jiffies
+ delta_in_ticks
);
2519 read_unlock(&bond
->lock
);
2523 * When using arp monitoring in active-backup mode, this function is
2524 * called to determine if any backup slaves have went down or a new
2525 * current slave needs to be found.
2526 * The backup slaves never generate traffic, they are considered up by merely
2527 * receiving traffic. If the current slave goes down, each backup slave will
2528 * be given the opportunity to tx/rx an arp before being taken down - this
2529 * prevents all slaves from being taken down due to the current slave not
2530 * sending any traffic for the backups to receive. The arps are not necessarily
2531 * necessary, any tx and rx traffic will keep the current slave up. While any
2532 * rx traffic will keep the backup slaves up, the current slave is responsible
2533 * for generating traffic to keep them up regardless of any other traffic they
2534 * may have received.
2535 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2537 void bond_activebackup_arp_mon(struct net_device
*bond_dev
)
2539 struct bonding
*bond
= bond_dev
->priv
;
2540 struct slave
*slave
;
2544 read_lock(&bond
->lock
);
2546 delta_in_ticks
= (bond
->params
.arp_interval
* HZ
) / 1000;
2548 if (bond
->kill_timers
) {
2552 if (bond
->slave_cnt
== 0) {
2556 /* determine if any slave has come up or any backup slave has
2558 * TODO: what about up/down delay in arp mode? it wasn't here before
2561 bond_for_each_slave(bond
, slave
, i
) {
2562 if (slave
->link
!= BOND_LINK_UP
) {
2563 if ((jiffies
- slave
->dev
->last_rx
) <= delta_in_ticks
) {
2565 slave
->link
= BOND_LINK_UP
;
2567 write_lock(&bond
->curr_slave_lock
);
2569 if ((!bond
->curr_active_slave
) &&
2570 ((jiffies
- slave
->dev
->trans_start
) <= delta_in_ticks
)) {
2571 bond_change_active_slave(bond
, slave
);
2572 bond
->current_arp_slave
= NULL
;
2573 } else if (bond
->curr_active_slave
!= slave
) {
2574 /* this slave has just come up but we
2575 * already have a current slave; this
2576 * can also happen if bond_enslave adds
2577 * a new slave that is up while we are
2578 * searching for a new slave
2580 bond_set_slave_inactive_flags(slave
);
2581 bond
->current_arp_slave
= NULL
;
2584 if (slave
== bond
->curr_active_slave
) {
2585 printk(KERN_INFO DRV_NAME
2586 ": %s: %s is up and now the "
2587 "active interface\n",
2591 printk(KERN_INFO DRV_NAME
2592 ": %s: backup interface %s is "
2598 write_unlock(&bond
->curr_slave_lock
);
2601 read_lock(&bond
->curr_slave_lock
);
2603 if ((slave
!= bond
->curr_active_slave
) &&
2604 (!bond
->current_arp_slave
) &&
2605 (((jiffies
- slave
->dev
->last_rx
) >= 3*delta_in_ticks
) &&
2606 bond_has_ip(bond
))) {
2607 /* a backup slave has gone down; three times
2608 * the delta allows the current slave to be
2609 * taken out before the backup slave.
2610 * note: a non-null current_arp_slave indicates
2611 * the curr_active_slave went down and we are
2612 * searching for a new one; under this
2613 * condition we only take the curr_active_slave
2614 * down - this gives each slave a chance to
2615 * tx/rx traffic before being taken out
2618 read_unlock(&bond
->curr_slave_lock
);
2620 slave
->link
= BOND_LINK_DOWN
;
2622 if (slave
->link_failure_count
< UINT_MAX
) {
2623 slave
->link_failure_count
++;
2626 bond_set_slave_inactive_flags(slave
);
2628 printk(KERN_INFO DRV_NAME
2629 ": %s: backup interface %s is now down\n",
2633 read_unlock(&bond
->curr_slave_lock
);
2638 read_lock(&bond
->curr_slave_lock
);
2639 slave
= bond
->curr_active_slave
;
2640 read_unlock(&bond
->curr_slave_lock
);
2643 /* if we have sent traffic in the past 2*arp_intervals but
2644 * haven't xmit and rx traffic in that time interval, select
2645 * a different slave. slave->jiffies is only updated when
2646 * a slave first becomes the curr_active_slave - not necessarily
2647 * after every arp; this ensures the slave has a full 2*delta
2648 * before being taken out. if a primary is being used, check
2649 * if it is up and needs to take over as the curr_active_slave
2651 if ((((jiffies
- slave
->dev
->trans_start
) >= (2*delta_in_ticks
)) ||
2652 (((jiffies
- slave
->dev
->last_rx
) >= (2*delta_in_ticks
)) &&
2653 bond_has_ip(bond
))) &&
2654 ((jiffies
- slave
->jiffies
) >= 2*delta_in_ticks
)) {
2656 slave
->link
= BOND_LINK_DOWN
;
2658 if (slave
->link_failure_count
< UINT_MAX
) {
2659 slave
->link_failure_count
++;
2662 printk(KERN_INFO DRV_NAME
2663 ": %s: link status down for active interface "
2664 "%s, disabling it\n",
2668 write_lock(&bond
->curr_slave_lock
);
2670 bond_select_active_slave(bond
);
2671 slave
= bond
->curr_active_slave
;
2673 write_unlock(&bond
->curr_slave_lock
);
2675 bond
->current_arp_slave
= slave
;
2678 slave
->jiffies
= jiffies
;
2680 } else if ((bond
->primary_slave
) &&
2681 (bond
->primary_slave
!= slave
) &&
2682 (bond
->primary_slave
->link
== BOND_LINK_UP
)) {
2683 /* at this point, slave is the curr_active_slave */
2684 printk(KERN_INFO DRV_NAME
2685 ": %s: changing from interface %s to primary "
2689 bond
->primary_slave
->dev
->name
);
2691 /* primary is up so switch to it */
2692 write_lock(&bond
->curr_slave_lock
);
2693 bond_change_active_slave(bond
, bond
->primary_slave
);
2694 write_unlock(&bond
->curr_slave_lock
);
2696 slave
= bond
->primary_slave
;
2697 slave
->jiffies
= jiffies
;
2699 bond
->current_arp_slave
= NULL
;
2702 /* the current slave must tx an arp to ensure backup slaves
2705 if (slave
&& bond_has_ip(bond
)) {
2706 bond_arp_send_all(bond
, slave
);
2710 /* if we don't have a curr_active_slave, search for the next available
2711 * backup slave from the current_arp_slave and make it the candidate
2712 * for becoming the curr_active_slave
2715 if (!bond
->current_arp_slave
) {
2716 bond
->current_arp_slave
= bond
->first_slave
;
2719 if (bond
->current_arp_slave
) {
2720 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
2722 /* search for next candidate */
2723 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
2724 if (IS_UP(slave
->dev
)) {
2725 slave
->link
= BOND_LINK_BACK
;
2726 bond_set_slave_active_flags(slave
);
2727 bond_arp_send_all(bond
, slave
);
2728 slave
->jiffies
= jiffies
;
2729 bond
->current_arp_slave
= slave
;
2733 /* if the link state is up at this point, we
2734 * mark it down - this can happen if we have
2735 * simultaneous link failures and
2736 * reselect_active_interface doesn't make this
2737 * one the current slave so it is still marked
2738 * up when it is actually down
2740 if (slave
->link
== BOND_LINK_UP
) {
2741 slave
->link
= BOND_LINK_DOWN
;
2742 if (slave
->link_failure_count
< UINT_MAX
) {
2743 slave
->link_failure_count
++;
2746 bond_set_slave_inactive_flags(slave
);
2748 printk(KERN_INFO DRV_NAME
2749 ": %s: backup interface %s is "
2759 if (bond
->params
.arp_interval
) {
2760 mod_timer(&bond
->arp_timer
, jiffies
+ delta_in_ticks
);
2763 read_unlock(&bond
->lock
);
2766 /*------------------------------ proc/seq_file-------------------------------*/
2768 #ifdef CONFIG_PROC_FS
2770 #define SEQ_START_TOKEN ((void *)1)
2772 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2774 struct bonding
*bond
= seq
->private;
2776 struct slave
*slave
;
2779 /* make sure the bond won't be taken away */
2780 read_lock(&dev_base_lock
);
2781 read_lock_bh(&bond
->lock
);
2784 return SEQ_START_TOKEN
;
2787 bond_for_each_slave(bond
, slave
, i
) {
2788 if (++off
== *pos
) {
2796 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2798 struct bonding
*bond
= seq
->private;
2799 struct slave
*slave
= v
;
2802 if (v
== SEQ_START_TOKEN
) {
2803 return bond
->first_slave
;
2806 slave
= slave
->next
;
2808 return (slave
== bond
->first_slave
) ? NULL
: slave
;
2811 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
2813 struct bonding
*bond
= seq
->private;
2815 read_unlock_bh(&bond
->lock
);
2816 read_unlock(&dev_base_lock
);
2819 static void bond_info_show_master(struct seq_file
*seq
)
2821 struct bonding
*bond
= seq
->private;
2826 read_lock(&bond
->curr_slave_lock
);
2827 curr
= bond
->curr_active_slave
;
2828 read_unlock(&bond
->curr_slave_lock
);
2830 seq_printf(seq
, "Bonding Mode: %s\n",
2831 bond_mode_name(bond
->params
.mode
));
2833 if (bond
->params
.mode
== BOND_MODE_XOR
||
2834 bond
->params
.mode
== BOND_MODE_8023AD
) {
2835 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
2836 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
2837 bond
->params
.xmit_policy
);
2840 if (USES_PRIMARY(bond
->params
.mode
)) {
2841 seq_printf(seq
, "Primary Slave: %s\n",
2842 (bond
->primary_slave
) ?
2843 bond
->primary_slave
->dev
->name
: "None");
2845 seq_printf(seq
, "Currently Active Slave: %s\n",
2846 (curr
) ? curr
->dev
->name
: "None");
2849 seq_printf(seq
, "MII Status: %s\n", (curr
) ? "up" : "down");
2850 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
2851 seq_printf(seq
, "Up Delay (ms): %d\n",
2852 bond
->params
.updelay
* bond
->params
.miimon
);
2853 seq_printf(seq
, "Down Delay (ms): %d\n",
2854 bond
->params
.downdelay
* bond
->params
.miimon
);
2857 /* ARP information */
2858 if(bond
->params
.arp_interval
> 0) {
2860 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
2861 bond
->params
.arp_interval
);
2863 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
2865 for(i
= 0; (i
< BOND_MAX_ARP_TARGETS
) ;i
++) {
2866 if (!bond
->params
.arp_targets
[i
])
2869 seq_printf(seq
, ",");
2870 target
= ntohl(bond
->params
.arp_targets
[i
]);
2871 seq_printf(seq
, " %d.%d.%d.%d", HIPQUAD(target
));
2874 seq_printf(seq
, "\n");
2877 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2878 struct ad_info ad_info
;
2880 seq_puts(seq
, "\n802.3ad info\n");
2881 seq_printf(seq
, "LACP rate: %s\n",
2882 (bond
->params
.lacp_fast
) ? "fast" : "slow");
2884 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
2885 seq_printf(seq
, "bond %s has no active aggregator\n",
2888 seq_printf(seq
, "Active Aggregator Info:\n");
2890 seq_printf(seq
, "\tAggregator ID: %d\n",
2891 ad_info
.aggregator_id
);
2892 seq_printf(seq
, "\tNumber of ports: %d\n",
2894 seq_printf(seq
, "\tActor Key: %d\n",
2896 seq_printf(seq
, "\tPartner Key: %d\n",
2897 ad_info
.partner_key
);
2898 seq_printf(seq
, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2899 ad_info
.partner_system
[0],
2900 ad_info
.partner_system
[1],
2901 ad_info
.partner_system
[2],
2902 ad_info
.partner_system
[3],
2903 ad_info
.partner_system
[4],
2904 ad_info
.partner_system
[5]);
2909 static void bond_info_show_slave(struct seq_file
*seq
, const struct slave
*slave
)
2911 struct bonding
*bond
= seq
->private;
2913 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
2914 seq_printf(seq
, "MII Status: %s\n",
2915 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
2916 seq_printf(seq
, "Link Failure Count: %d\n",
2917 slave
->link_failure_count
);
2920 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
2921 slave
->perm_hwaddr
[0], slave
->perm_hwaddr
[1],
2922 slave
->perm_hwaddr
[2], slave
->perm_hwaddr
[3],
2923 slave
->perm_hwaddr
[4], slave
->perm_hwaddr
[5]);
2925 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2926 const struct aggregator
*agg
2927 = SLAVE_AD_INFO(slave
).port
.aggregator
;
2930 seq_printf(seq
, "Aggregator ID: %d\n",
2931 agg
->aggregator_identifier
);
2933 seq_puts(seq
, "Aggregator ID: N/A\n");
2938 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
2940 if (v
== SEQ_START_TOKEN
) {
2941 seq_printf(seq
, "%s\n", version
);
2942 bond_info_show_master(seq
);
2944 bond_info_show_slave(seq
, v
);
2950 static struct seq_operations bond_info_seq_ops
= {
2951 .start
= bond_info_seq_start
,
2952 .next
= bond_info_seq_next
,
2953 .stop
= bond_info_seq_stop
,
2954 .show
= bond_info_seq_show
,
2957 static int bond_info_open(struct inode
*inode
, struct file
*file
)
2959 struct seq_file
*seq
;
2960 struct proc_dir_entry
*proc
;
2963 res
= seq_open(file
, &bond_info_seq_ops
);
2965 /* recover the pointer buried in proc_dir_entry data */
2966 seq
= file
->private_data
;
2968 seq
->private = proc
->data
;
2974 static struct file_operations bond_info_fops
= {
2975 .owner
= THIS_MODULE
,
2976 .open
= bond_info_open
,
2978 .llseek
= seq_lseek
,
2979 .release
= seq_release
,
2982 static int bond_create_proc_entry(struct bonding
*bond
)
2984 struct net_device
*bond_dev
= bond
->dev
;
2986 if (bond_proc_dir
) {
2987 bond
->proc_entry
= create_proc_entry(bond_dev
->name
,
2990 if (bond
->proc_entry
== NULL
) {
2991 printk(KERN_WARNING DRV_NAME
2992 ": Warning: Cannot create /proc/net/%s/%s\n",
2993 DRV_NAME
, bond_dev
->name
);
2995 bond
->proc_entry
->data
= bond
;
2996 bond
->proc_entry
->proc_fops
= &bond_info_fops
;
2997 bond
->proc_entry
->owner
= THIS_MODULE
;
2998 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3005 static void bond_remove_proc_entry(struct bonding
*bond
)
3007 if (bond_proc_dir
&& bond
->proc_entry
) {
3008 remove_proc_entry(bond
->proc_file_name
, bond_proc_dir
);
3009 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3010 bond
->proc_entry
= NULL
;
3014 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3015 * Caller must hold rtnl_lock.
3017 static void bond_create_proc_dir(void)
3019 int len
= strlen(DRV_NAME
);
3021 for (bond_proc_dir
= proc_net
->subdir
; bond_proc_dir
;
3022 bond_proc_dir
= bond_proc_dir
->next
) {
3023 if ((bond_proc_dir
->namelen
== len
) &&
3024 !memcmp(bond_proc_dir
->name
, DRV_NAME
, len
)) {
3029 if (!bond_proc_dir
) {
3030 bond_proc_dir
= proc_mkdir(DRV_NAME
, proc_net
);
3031 if (bond_proc_dir
) {
3032 bond_proc_dir
->owner
= THIS_MODULE
;
3034 printk(KERN_WARNING DRV_NAME
3035 ": Warning: cannot create /proc/net/%s\n",
3041 /* Destroy the bonding directory under /proc/net, if empty.
3042 * Caller must hold rtnl_lock.
3044 static void bond_destroy_proc_dir(void)
3046 struct proc_dir_entry
*de
;
3048 if (!bond_proc_dir
) {
3052 /* verify that the /proc dir is empty */
3053 for (de
= bond_proc_dir
->subdir
; de
; de
= de
->next
) {
3054 /* ignore . and .. */
3055 if (*(de
->name
) != '.') {
3061 if (bond_proc_dir
->owner
== THIS_MODULE
) {
3062 bond_proc_dir
->owner
= NULL
;
3065 remove_proc_entry(DRV_NAME
, proc_net
);
3066 bond_proc_dir
= NULL
;
3069 #endif /* CONFIG_PROC_FS */
3071 /*-------------------------- netdev event handling --------------------------*/
3074 * Change device name
3076 static int bond_event_changename(struct bonding
*bond
)
3078 #ifdef CONFIG_PROC_FS
3079 bond_remove_proc_entry(bond
);
3080 bond_create_proc_entry(bond
);
3082 down_write(&(bonding_rwsem
));
3083 bond_destroy_sysfs_entry(bond
);
3084 bond_create_sysfs_entry(bond
);
3085 up_write(&(bonding_rwsem
));
3089 static int bond_master_netdev_event(unsigned long event
, struct net_device
*bond_dev
)
3091 struct bonding
*event_bond
= bond_dev
->priv
;
3094 case NETDEV_CHANGENAME
:
3095 return bond_event_changename(event_bond
);
3096 case NETDEV_UNREGISTER
:
3098 * TODO: remove a bond from the list?
3108 static int bond_slave_netdev_event(unsigned long event
, struct net_device
*slave_dev
)
3110 struct net_device
*bond_dev
= slave_dev
->master
;
3111 struct bonding
*bond
= bond_dev
->priv
;
3114 case NETDEV_UNREGISTER
:
3116 bond_release(bond_dev
, slave_dev
);
3121 * TODO: is this what we get if somebody
3122 * sets up a hierarchical bond, then rmmod's
3123 * one of the slave bonding devices?
3128 * ... Or is it this?
3131 case NETDEV_CHANGEMTU
:
3133 * TODO: Should slaves be allowed to
3134 * independently alter their MTU? For
3135 * an active-backup bond, slaves need
3136 * not be the same type of device, so
3137 * MTUs may vary. For other modes,
3138 * slaves arguably should have the
3139 * same MTUs. To do this, we'd need to
3140 * take over the slave's change_mtu
3141 * function for the duration of their
3145 case NETDEV_CHANGENAME
:
3147 * TODO: handle changing the primary's name
3150 case NETDEV_FEAT_CHANGE
:
3151 bond_compute_features(bond
);
3161 * bond_netdev_event: handle netdev notifier chain events.
3163 * This function receives events for the netdev chain. The caller (an
3164 * ioctl handler calling notifier_call_chain) holds the necessary
3165 * locks for us to safely manipulate the slave devices (RTNL lock,
3168 static int bond_netdev_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3170 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3172 dprintk("event_dev: %s, event: %lx\n",
3173 (event_dev
? event_dev
->name
: "None"),
3176 if (event_dev
->flags
& IFF_MASTER
) {
3177 dprintk("IFF_MASTER\n");
3178 return bond_master_netdev_event(event
, event_dev
);
3181 if (event_dev
->flags
& IFF_SLAVE
) {
3182 dprintk("IFF_SLAVE\n");
3183 return bond_slave_netdev_event(event
, event_dev
);
3190 * bond_inetaddr_event: handle inetaddr notifier chain events.
3192 * We keep track of device IPs primarily to use as source addresses in
3193 * ARP monitor probes (rather than spewing out broadcasts all the time).
3195 * We track one IP for the main device (if it has one), plus one per VLAN.
3197 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3199 struct in_ifaddr
*ifa
= ptr
;
3200 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3201 struct bonding
*bond
, *bond_next
;
3202 struct vlan_entry
*vlan
, *vlan_next
;
3204 list_for_each_entry_safe(bond
, bond_next
, &bond_dev_list
, bond_list
) {
3205 if (bond
->dev
== event_dev
) {
3208 bond
->master_ip
= ifa
->ifa_local
;
3211 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3218 if (list_empty(&bond
->vlan_list
))
3221 list_for_each_entry_safe(vlan
, vlan_next
, &bond
->vlan_list
,
3223 vlan_dev
= bond
->vlgrp
->vlan_devices
[vlan
->vlan_id
];
3224 if (vlan_dev
== event_dev
) {
3227 vlan
->vlan_ip
= ifa
->ifa_local
;
3231 bond_glean_dev_ip(vlan_dev
);
3242 static struct notifier_block bond_netdev_notifier
= {
3243 .notifier_call
= bond_netdev_event
,
3246 static struct notifier_block bond_inetaddr_notifier
= {
3247 .notifier_call
= bond_inetaddr_event
,
3250 /*-------------------------- Packet type handling ---------------------------*/
3252 /* register to receive lacpdus on a bond */
3253 static void bond_register_lacpdu(struct bonding
*bond
)
3255 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3257 /* initialize packet type */
3258 pk_type
->type
= PKT_TYPE_LACPDU
;
3259 pk_type
->dev
= bond
->dev
;
3260 pk_type
->func
= bond_3ad_lacpdu_recv
;
3262 dev_add_pack(pk_type
);
3265 /* unregister to receive lacpdus on a bond */
3266 static void bond_unregister_lacpdu(struct bonding
*bond
)
3268 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3271 /*---------------------------- Hashing Policies -----------------------------*/
3274 * Hash for the the output device based upon layer 3 and layer 4 data. If
3275 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3276 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3278 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
,
3279 struct net_device
*bond_dev
, int count
)
3281 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3282 struct iphdr
*iph
= skb
->nh
.iph
;
3283 u16
*layer4hdr
= (u16
*)((u32
*)iph
+ iph
->ihl
);
3286 if (skb
->protocol
== __constant_htons(ETH_P_IP
)) {
3287 if (!(iph
->frag_off
& __constant_htons(IP_MF
|IP_OFFSET
)) &&
3288 (iph
->protocol
== IPPROTO_TCP
||
3289 iph
->protocol
== IPPROTO_UDP
)) {
3290 layer4_xor
= htons((*layer4hdr
^ *(layer4hdr
+ 1)));
3292 return (layer4_xor
^
3293 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3297 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3301 * Hash for the output device based upon layer 2 data
3303 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
,
3304 struct net_device
*bond_dev
, int count
)
3306 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3308 return (data
->h_dest
[5] ^ bond_dev
->dev_addr
[5]) % count
;
3311 /*-------------------------- Device entry points ----------------------------*/
3313 static int bond_open(struct net_device
*bond_dev
)
3315 struct bonding
*bond
= bond_dev
->priv
;
3316 struct timer_list
*mii_timer
= &bond
->mii_timer
;
3317 struct timer_list
*arp_timer
= &bond
->arp_timer
;
3319 bond
->kill_timers
= 0;
3321 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
3322 (bond
->params
.mode
== BOND_MODE_ALB
)) {
3323 struct timer_list
*alb_timer
= &(BOND_ALB_INFO(bond
).alb_timer
);
3325 /* bond_alb_initialize must be called before the timer
3328 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3329 /* something went wrong - fail the open operation */
3333 init_timer(alb_timer
);
3334 alb_timer
->expires
= jiffies
+ 1;
3335 alb_timer
->data
= (unsigned long)bond
;
3336 alb_timer
->function
= (void *)&bond_alb_monitor
;
3337 add_timer(alb_timer
);
3340 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3341 init_timer(mii_timer
);
3342 mii_timer
->expires
= jiffies
+ 1;
3343 mii_timer
->data
= (unsigned long)bond_dev
;
3344 mii_timer
->function
= (void *)&bond_mii_monitor
;
3345 add_timer(mii_timer
);
3348 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3349 init_timer(arp_timer
);
3350 arp_timer
->expires
= jiffies
+ 1;
3351 arp_timer
->data
= (unsigned long)bond_dev
;
3352 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
3353 arp_timer
->function
= (void *)&bond_activebackup_arp_mon
;
3355 arp_timer
->function
= (void *)&bond_loadbalance_arp_mon
;
3357 add_timer(arp_timer
);
3360 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3361 struct timer_list
*ad_timer
= &(BOND_AD_INFO(bond
).ad_timer
);
3362 init_timer(ad_timer
);
3363 ad_timer
->expires
= jiffies
+ 1;
3364 ad_timer
->data
= (unsigned long)bond
;
3365 ad_timer
->function
= (void *)&bond_3ad_state_machine_handler
;
3366 add_timer(ad_timer
);
3368 /* register to receive LACPDUs */
3369 bond_register_lacpdu(bond
);
3375 static int bond_close(struct net_device
*bond_dev
)
3377 struct bonding
*bond
= bond_dev
->priv
;
3379 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3380 /* Unregister the receive of LACPDUs */
3381 bond_unregister_lacpdu(bond
);
3384 write_lock_bh(&bond
->lock
);
3386 bond_mc_list_destroy(bond
);
3388 /* signal timers not to re-arm */
3389 bond
->kill_timers
= 1;
3391 write_unlock_bh(&bond
->lock
);
3393 /* del_timer_sync must run without holding the bond->lock
3394 * because a running timer might be trying to hold it too
3397 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3398 del_timer_sync(&bond
->mii_timer
);
3401 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3402 del_timer_sync(&bond
->arp_timer
);
3405 switch (bond
->params
.mode
) {
3406 case BOND_MODE_8023AD
:
3407 del_timer_sync(&(BOND_AD_INFO(bond
).ad_timer
));
3411 del_timer_sync(&(BOND_ALB_INFO(bond
).alb_timer
));
3417 /* Release the bonded slaves */
3418 bond_release_all(bond_dev
);
3420 if ((bond
->params
.mode
== BOND_MODE_TLB
) ||
3421 (bond
->params
.mode
== BOND_MODE_ALB
)) {
3422 /* Must be called only after all
3423 * slaves have been released
3425 bond_alb_deinitialize(bond
);
3431 static struct net_device_stats
*bond_get_stats(struct net_device
*bond_dev
)
3433 struct bonding
*bond
= bond_dev
->priv
;
3434 struct net_device_stats
*stats
= &(bond
->stats
), *sstats
;
3435 struct slave
*slave
;
3438 memset(stats
, 0, sizeof(struct net_device_stats
));
3440 read_lock_bh(&bond
->lock
);
3442 bond_for_each_slave(bond
, slave
, i
) {
3443 sstats
= slave
->dev
->get_stats(slave
->dev
);
3445 stats
->rx_packets
+= sstats
->rx_packets
;
3446 stats
->rx_bytes
+= sstats
->rx_bytes
;
3447 stats
->rx_errors
+= sstats
->rx_errors
;
3448 stats
->rx_dropped
+= sstats
->rx_dropped
;
3450 stats
->tx_packets
+= sstats
->tx_packets
;
3451 stats
->tx_bytes
+= sstats
->tx_bytes
;
3452 stats
->tx_errors
+= sstats
->tx_errors
;
3453 stats
->tx_dropped
+= sstats
->tx_dropped
;
3455 stats
->multicast
+= sstats
->multicast
;
3456 stats
->collisions
+= sstats
->collisions
;
3458 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3459 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3460 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3461 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3462 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3463 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3465 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3466 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3467 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3468 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3469 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3472 read_unlock_bh(&bond
->lock
);
3477 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3479 struct net_device
*slave_dev
= NULL
;
3480 struct ifbond k_binfo
;
3481 struct ifbond __user
*u_binfo
= NULL
;
3482 struct ifslave k_sinfo
;
3483 struct ifslave __user
*u_sinfo
= NULL
;
3484 struct mii_ioctl_data
*mii
= NULL
;
3487 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3488 bond_dev
->name
, cmd
);
3500 * We do this again just in case we were called by SIOCGMIIREG
3501 * instead of SIOCGMIIPHY.
3508 if (mii
->reg_num
== 1) {
3509 struct bonding
*bond
= bond_dev
->priv
;
3511 read_lock_bh(&bond
->lock
);
3512 read_lock(&bond
->curr_slave_lock
);
3513 if (bond
->curr_active_slave
) {
3514 mii
->val_out
= BMSR_LSTATUS
;
3516 read_unlock(&bond
->curr_slave_lock
);
3517 read_unlock_bh(&bond
->lock
);
3521 case BOND_INFO_QUERY_OLD
:
3522 case SIOCBONDINFOQUERY
:
3523 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3525 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
))) {
3529 res
= bond_info_query(bond_dev
, &k_binfo
);
3531 if (copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
))) {
3537 case BOND_SLAVE_INFO_QUERY_OLD
:
3538 case SIOCBONDSLAVEINFOQUERY
:
3539 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3541 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
))) {
3545 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3547 if (copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
))) {
3558 if (!capable(CAP_NET_ADMIN
)) {
3562 down_write(&(bonding_rwsem
));
3563 slave_dev
= dev_get_by_name(ifr
->ifr_slave
);
3565 dprintk("slave_dev=%p: \n", slave_dev
);
3570 dprintk("slave_dev->name=%s: \n", slave_dev
->name
);
3572 case BOND_ENSLAVE_OLD
:
3573 case SIOCBONDENSLAVE
:
3574 res
= bond_enslave(bond_dev
, slave_dev
);
3576 case BOND_RELEASE_OLD
:
3577 case SIOCBONDRELEASE
:
3578 res
= bond_release(bond_dev
, slave_dev
);
3580 case BOND_SETHWADDR_OLD
:
3581 case SIOCBONDSETHWADDR
:
3582 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3584 case BOND_CHANGE_ACTIVE_OLD
:
3585 case SIOCBONDCHANGEACTIVE
:
3586 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3595 up_write(&(bonding_rwsem
));
3599 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3601 struct bonding
*bond
= bond_dev
->priv
;
3602 struct dev_mc_list
*dmi
;
3604 write_lock_bh(&bond
->lock
);
3607 * Do promisc before checking multicast_mode
3609 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
)) {
3610 bond_set_promiscuity(bond
, 1);
3613 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
)) {
3614 bond_set_promiscuity(bond
, -1);
3617 /* set allmulti flag to slaves */
3618 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
)) {
3619 bond_set_allmulti(bond
, 1);
3622 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
)) {
3623 bond_set_allmulti(bond
, -1);
3626 bond
->flags
= bond_dev
->flags
;
3628 /* looking for addresses to add to slaves' mc list */
3629 for (dmi
= bond_dev
->mc_list
; dmi
; dmi
= dmi
->next
) {
3630 if (!bond_mc_list_find_dmi(dmi
, bond
->mc_list
)) {
3631 bond_mc_add(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
3635 /* looking for addresses to delete from slaves' list */
3636 for (dmi
= bond
->mc_list
; dmi
; dmi
= dmi
->next
) {
3637 if (!bond_mc_list_find_dmi(dmi
, bond_dev
->mc_list
)) {
3638 bond_mc_delete(bond
, dmi
->dmi_addr
, dmi
->dmi_addrlen
);
3642 /* save master's multicast list */
3643 bond_mc_list_destroy(bond
);
3644 bond_mc_list_copy(bond_dev
->mc_list
, bond
, GFP_ATOMIC
);
3646 write_unlock_bh(&bond
->lock
);
3650 * Change the MTU of all of a master's slaves to match the master
3652 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3654 struct bonding
*bond
= bond_dev
->priv
;
3655 struct slave
*slave
, *stop_at
;
3659 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond
,
3660 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3662 /* Can't hold bond->lock with bh disabled here since
3663 * some base drivers panic. On the other hand we can't
3664 * hold bond->lock without bh disabled because we'll
3665 * deadlock. The only solution is to rely on the fact
3666 * that we're under rtnl_lock here, and the slaves
3667 * list won't change. This doesn't solve the problem
3668 * of setting the slave's MTU while it is
3669 * transmitting, but the assumption is that the base
3670 * driver can handle that.
3672 * TODO: figure out a way to safely iterate the slaves
3673 * list, but without holding a lock around the actual
3674 * call to the base driver.
3677 bond_for_each_slave(bond
, slave
, i
) {
3678 dprintk("s %p s->p %p c_m %p\n", slave
,
3679 slave
->prev
, slave
->dev
->change_mtu
);
3681 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3684 /* If we failed to set the slave's mtu to the new value
3685 * we must abort the operation even in ACTIVE_BACKUP
3686 * mode, because if we allow the backup slaves to have
3687 * different mtu values than the active slave we'll
3688 * need to change their mtu when doing a failover. That
3689 * means changing their mtu from timer context, which
3690 * is probably not a good idea.
3692 dprintk("err %d %s\n", res
, slave
->dev
->name
);
3697 bond_dev
->mtu
= new_mtu
;
3702 /* unwind from head to the slave that failed */
3704 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3707 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3709 dprintk("unwind err %d dev %s\n", tmp_res
,
3720 * Note that many devices must be down to change the HW address, and
3721 * downing the master releases all slaves. We can make bonds full of
3722 * bonding devices to test this, however.
3724 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3726 struct bonding
*bond
= bond_dev
->priv
;
3727 struct sockaddr
*sa
= addr
, tmp_sa
;
3728 struct slave
*slave
, *stop_at
;
3732 dprintk("bond=%p, name=%s\n", bond
, (bond_dev
? bond_dev
->name
: "None"));
3734 if (!is_valid_ether_addr(sa
->sa_data
)) {
3735 return -EADDRNOTAVAIL
;
3738 /* Can't hold bond->lock with bh disabled here since
3739 * some base drivers panic. On the other hand we can't
3740 * hold bond->lock without bh disabled because we'll
3741 * deadlock. The only solution is to rely on the fact
3742 * that we're under rtnl_lock here, and the slaves
3743 * list won't change. This doesn't solve the problem
3744 * of setting the slave's hw address while it is
3745 * transmitting, but the assumption is that the base
3746 * driver can handle that.
3748 * TODO: figure out a way to safely iterate the slaves
3749 * list, but without holding a lock around the actual
3750 * call to the base driver.
3753 bond_for_each_slave(bond
, slave
, i
) {
3754 dprintk("slave %p %s\n", slave
, slave
->dev
->name
);
3756 if (slave
->dev
->set_mac_address
== NULL
) {
3758 dprintk("EOPNOTSUPP %s\n", slave
->dev
->name
);
3762 res
= dev_set_mac_address(slave
->dev
, addr
);
3764 /* TODO: consider downing the slave
3766 * User should expect communications
3767 * breakage anyway until ARP finish
3770 dprintk("err %d %s\n", res
, slave
->dev
->name
);
3776 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3780 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3781 tmp_sa
.sa_family
= bond_dev
->type
;
3783 /* unwind from head to the slave that failed */
3785 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3788 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3790 dprintk("unwind err %d dev %s\n", tmp_res
,
3798 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3800 struct bonding
*bond
= bond_dev
->priv
;
3801 struct slave
*slave
, *start_at
;
3805 read_lock(&bond
->lock
);
3807 if (!BOND_IS_OK(bond
)) {
3811 read_lock(&bond
->curr_slave_lock
);
3812 slave
= start_at
= bond
->curr_active_slave
;
3813 read_unlock(&bond
->curr_slave_lock
);
3819 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3820 if (IS_UP(slave
->dev
) &&
3821 (slave
->link
== BOND_LINK_UP
) &&
3822 (slave
->state
== BOND_STATE_ACTIVE
)) {
3823 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3825 write_lock(&bond
->curr_slave_lock
);
3826 bond
->curr_active_slave
= slave
->next
;
3827 write_unlock(&bond
->curr_slave_lock
);
3836 /* no suitable interface, frame not sent */
3839 read_unlock(&bond
->lock
);
3843 static void bond_activebackup_xmit_copy(struct sk_buff
*skb
,
3844 struct bonding
*bond
,
3845 struct slave
*slave
)
3847 struct sk_buff
*skb2
= skb_copy(skb
, GFP_ATOMIC
);
3848 struct ethhdr
*eth_data
;
3853 printk(KERN_ERR DRV_NAME
": Error: "
3854 "bond_activebackup_xmit_copy(): skb_copy() failed\n");
3858 skb2
->mac
.raw
= (unsigned char *)skb2
->data
;
3859 eth_data
= eth_hdr(skb2
);
3861 /* Pick an appropriate source MAC address
3862 * -- use slave's perm MAC addr, unless used by bond
3863 * -- otherwise, borrow active slave's perm MAC addr
3864 * since that will not be used
3866 hwaddr
= slave
->perm_hwaddr
;
3867 if (!memcmp(eth_data
->h_source
, hwaddr
, ETH_ALEN
))
3868 hwaddr
= bond
->curr_active_slave
->perm_hwaddr
;
3870 /* Set source MAC address appropriately */
3871 memcpy(eth_data
->h_source
, hwaddr
, ETH_ALEN
);
3873 res
= bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3875 dev_kfree_skb(skb2
);
3881 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3882 * the bond has a usable interface.
3884 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3886 struct bonding
*bond
= bond_dev
->priv
;
3889 read_lock(&bond
->lock
);
3890 read_lock(&bond
->curr_slave_lock
);
3892 if (!BOND_IS_OK(bond
)) {
3896 if (!bond
->curr_active_slave
)
3899 /* Xmit IGMP frames on all slaves to ensure rapid fail-over
3900 for multicast traffic on snooping switches */
3901 if (skb
->protocol
== __constant_htons(ETH_P_IP
) &&
3902 skb
->nh
.iph
->protocol
== IPPROTO_IGMP
) {
3903 struct slave
*slave
, *active_slave
;
3906 active_slave
= bond
->curr_active_slave
;
3907 bond_for_each_slave_from_to(bond
, slave
, i
, active_slave
->next
,
3909 if (IS_UP(slave
->dev
) &&
3910 (slave
->link
== BOND_LINK_UP
))
3911 bond_activebackup_xmit_copy(skb
, bond
, slave
);
3914 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
3918 /* no suitable interface, frame not sent */
3921 read_unlock(&bond
->curr_slave_lock
);
3922 read_unlock(&bond
->lock
);
3927 * In bond_xmit_xor() , we determine the output device by using a pre-
3928 * determined xmit_hash_policy(), If the selected device is not enabled,
3929 * find the next active slave.
3931 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3933 struct bonding
*bond
= bond_dev
->priv
;
3934 struct slave
*slave
, *start_at
;
3939 read_lock(&bond
->lock
);
3941 if (!BOND_IS_OK(bond
)) {
3945 slave_no
= bond
->xmit_hash_policy(skb
, bond_dev
, bond
->slave_cnt
);
3947 bond_for_each_slave(bond
, slave
, i
) {
3956 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3957 if (IS_UP(slave
->dev
) &&
3958 (slave
->link
== BOND_LINK_UP
) &&
3959 (slave
->state
== BOND_STATE_ACTIVE
)) {
3960 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3967 /* no suitable interface, frame not sent */
3970 read_unlock(&bond
->lock
);
3975 * in broadcast mode, we send everything to all usable interfaces.
3977 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3979 struct bonding
*bond
= bond_dev
->priv
;
3980 struct slave
*slave
, *start_at
;
3981 struct net_device
*tx_dev
= NULL
;
3985 read_lock(&bond
->lock
);
3987 if (!BOND_IS_OK(bond
)) {
3991 read_lock(&bond
->curr_slave_lock
);
3992 start_at
= bond
->curr_active_slave
;
3993 read_unlock(&bond
->curr_slave_lock
);
3999 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4000 if (IS_UP(slave
->dev
) &&
4001 (slave
->link
== BOND_LINK_UP
) &&
4002 (slave
->state
== BOND_STATE_ACTIVE
)) {
4004 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4006 printk(KERN_ERR DRV_NAME
4007 ": %s: Error: bond_xmit_broadcast(): "
4008 "skb_clone() failed\n",
4013 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4015 dev_kfree_skb(skb2
);
4019 tx_dev
= slave
->dev
;
4024 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4029 /* no suitable interface, frame not sent */
4032 /* frame sent to all suitable interfaces */
4033 read_unlock(&bond
->lock
);
4037 /*------------------------- Device initialization ---------------------------*/
4040 * set bond mode specific net device operations
4042 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4044 struct net_device
*bond_dev
= bond
->dev
;
4047 case BOND_MODE_ROUNDROBIN
:
4048 bond_dev
->hard_start_xmit
= bond_xmit_roundrobin
;
4050 case BOND_MODE_ACTIVEBACKUP
:
4051 bond_dev
->hard_start_xmit
= bond_xmit_activebackup
;
4054 bond_dev
->hard_start_xmit
= bond_xmit_xor
;
4055 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
)
4056 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4058 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4060 case BOND_MODE_BROADCAST
:
4061 bond_dev
->hard_start_xmit
= bond_xmit_broadcast
;
4063 case BOND_MODE_8023AD
:
4064 bond_dev
->hard_start_xmit
= bond_3ad_xmit_xor
;
4065 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
)
4066 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4068 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4072 bond_dev
->hard_start_xmit
= bond_alb_xmit
;
4073 bond_dev
->set_mac_address
= bond_alb_set_mac_address
;
4076 /* Should never happen, mode already checked */
4077 printk(KERN_ERR DRV_NAME
4078 ": %s: Error: Unknown bonding mode %d\n",
4085 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4086 struct ethtool_drvinfo
*drvinfo
)
4088 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4089 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4090 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4093 static struct ethtool_ops bond_ethtool_ops
= {
4094 .get_tx_csum
= ethtool_op_get_tx_csum
,
4095 .get_tso
= ethtool_op_get_tso
,
4096 .get_ufo
= ethtool_op_get_ufo
,
4097 .get_sg
= ethtool_op_get_sg
,
4098 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4102 * Does not allocate but creates a /proc entry.
4105 static int bond_init(struct net_device
*bond_dev
, struct bond_params
*params
)
4107 struct bonding
*bond
= bond_dev
->priv
;
4109 dprintk("Begin bond_init for %s\n", bond_dev
->name
);
4111 /* initialize rwlocks */
4112 rwlock_init(&bond
->lock
);
4113 rwlock_init(&bond
->curr_slave_lock
);
4115 bond
->params
= *params
; /* copy params struct */
4117 /* Initialize pointers */
4118 bond
->first_slave
= NULL
;
4119 bond
->curr_active_slave
= NULL
;
4120 bond
->current_arp_slave
= NULL
;
4121 bond
->primary_slave
= NULL
;
4122 bond
->dev
= bond_dev
;
4123 INIT_LIST_HEAD(&bond
->vlan_list
);
4125 /* Initialize the device entry points */
4126 bond_dev
->open
= bond_open
;
4127 bond_dev
->stop
= bond_close
;
4128 bond_dev
->get_stats
= bond_get_stats
;
4129 bond_dev
->do_ioctl
= bond_do_ioctl
;
4130 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4131 bond_dev
->set_multicast_list
= bond_set_multicast_list
;
4132 bond_dev
->change_mtu
= bond_change_mtu
;
4133 bond_dev
->set_mac_address
= bond_set_mac_address
;
4135 bond_set_mode_ops(bond
, bond
->params
.mode
);
4137 bond_dev
->destructor
= free_netdev
;
4139 /* Initialize the device options */
4140 bond_dev
->tx_queue_len
= 0;
4141 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4143 /* At first, we block adding VLANs. That's the only way to
4144 * prevent problems that occur when adding VLANs over an
4145 * empty bond. The block will be removed once non-challenged
4146 * slaves are enslaved.
4148 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4150 /* don't acquire bond device's xmit_lock when
4152 bond_dev
->features
|= NETIF_F_LLTX
;
4154 /* By default, we declare the bond to be fully
4155 * VLAN hardware accelerated capable. Special
4156 * care is taken in the various xmit functions
4157 * when there are slaves that are not hw accel
4160 bond_dev
->vlan_rx_register
= bond_vlan_rx_register
;
4161 bond_dev
->vlan_rx_add_vid
= bond_vlan_rx_add_vid
;
4162 bond_dev
->vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
;
4163 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4164 NETIF_F_HW_VLAN_RX
|
4165 NETIF_F_HW_VLAN_FILTER
);
4167 #ifdef CONFIG_PROC_FS
4168 bond_create_proc_entry(bond
);
4171 list_add_tail(&bond
->bond_list
, &bond_dev_list
);
4176 /* De-initialize device specific data.
4177 * Caller must hold rtnl_lock.
4179 void bond_deinit(struct net_device
*bond_dev
)
4181 struct bonding
*bond
= bond_dev
->priv
;
4183 list_del(&bond
->bond_list
);
4185 #ifdef CONFIG_PROC_FS
4186 bond_remove_proc_entry(bond
);
4190 /* Unregister and free all bond devices.
4191 * Caller must hold rtnl_lock.
4193 static void bond_free_all(void)
4195 struct bonding
*bond
, *nxt
;
4197 list_for_each_entry_safe(bond
, nxt
, &bond_dev_list
, bond_list
) {
4198 struct net_device
*bond_dev
= bond
->dev
;
4200 unregister_netdevice(bond_dev
);
4201 bond_deinit(bond_dev
);
4204 #ifdef CONFIG_PROC_FS
4205 bond_destroy_proc_dir();
4209 /*------------------------- Module initialization ---------------------------*/
4212 * Convert string input module parms. Accept either the
4213 * number of the mode or its string name.
4215 int bond_parse_parm(char *mode_arg
, struct bond_parm_tbl
*tbl
)
4219 for (i
= 0; tbl
[i
].modename
; i
++) {
4220 if ((isdigit(*mode_arg
) &&
4221 tbl
[i
].mode
== simple_strtol(mode_arg
, NULL
, 0)) ||
4222 (strncmp(mode_arg
, tbl
[i
].modename
,
4223 strlen(tbl
[i
].modename
)) == 0)) {
4231 static int bond_check_params(struct bond_params
*params
)
4234 * Convert string parameters.
4237 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4238 if (bond_mode
== -1) {
4239 printk(KERN_ERR DRV_NAME
4240 ": Error: Invalid bonding mode \"%s\"\n",
4241 mode
== NULL
? "NULL" : mode
);
4246 if (xmit_hash_policy
) {
4247 if ((bond_mode
!= BOND_MODE_XOR
) &&
4248 (bond_mode
!= BOND_MODE_8023AD
)) {
4249 printk(KERN_INFO DRV_NAME
4250 ": xor_mode param is irrelevant in mode %s\n",
4251 bond_mode_name(bond_mode
));
4253 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4255 if (xmit_hashtype
== -1) {
4256 printk(KERN_ERR DRV_NAME
4257 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4258 xmit_hash_policy
== NULL
? "NULL" :
4266 if (bond_mode
!= BOND_MODE_8023AD
) {
4267 printk(KERN_INFO DRV_NAME
4268 ": lacp_rate param is irrelevant in mode %s\n",
4269 bond_mode_name(bond_mode
));
4271 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4272 if (lacp_fast
== -1) {
4273 printk(KERN_ERR DRV_NAME
4274 ": Error: Invalid lacp rate \"%s\"\n",
4275 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4281 if (max_bonds
< 1 || max_bonds
> INT_MAX
) {
4282 printk(KERN_WARNING DRV_NAME
4283 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4284 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4285 max_bonds
, 1, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4286 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4290 printk(KERN_WARNING DRV_NAME
4291 ": Warning: miimon module parameter (%d), "
4292 "not in range 0-%d, so it was reset to %d\n",
4293 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4294 miimon
= BOND_LINK_MON_INTERV
;
4298 printk(KERN_WARNING DRV_NAME
4299 ": Warning: updelay module parameter (%d), "
4300 "not in range 0-%d, so it was reset to 0\n",
4305 if (downdelay
< 0) {
4306 printk(KERN_WARNING DRV_NAME
4307 ": Warning: downdelay module parameter (%d), "
4308 "not in range 0-%d, so it was reset to 0\n",
4309 downdelay
, INT_MAX
);
4313 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4314 printk(KERN_WARNING DRV_NAME
4315 ": Warning: use_carrier module parameter (%d), "
4316 "not of valid value (0/1), so it was set to 1\n",
4321 /* reset values for 802.3ad */
4322 if (bond_mode
== BOND_MODE_8023AD
) {
4324 printk(KERN_WARNING DRV_NAME
4325 ": Warning: miimon must be specified, "
4326 "otherwise bonding will not detect link "
4327 "failure, speed and duplex which are "
4328 "essential for 802.3ad operation\n");
4329 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4334 /* reset values for TLB/ALB */
4335 if ((bond_mode
== BOND_MODE_TLB
) ||
4336 (bond_mode
== BOND_MODE_ALB
)) {
4338 printk(KERN_WARNING DRV_NAME
4339 ": Warning: miimon must be specified, "
4340 "otherwise bonding will not detect link "
4341 "failure and link speed which are essential "
4342 "for TLB/ALB load balancing\n");
4343 printk(KERN_WARNING
"Forcing miimon to 100msec\n");
4348 if (bond_mode
== BOND_MODE_ALB
) {
4349 printk(KERN_NOTICE DRV_NAME
4350 ": In ALB mode you might experience client "
4351 "disconnections upon reconnection of a link if the "
4352 "bonding module updelay parameter (%d msec) is "
4353 "incompatible with the forwarding delay time of the "
4359 if (updelay
|| downdelay
) {
4360 /* just warn the user the up/down delay will have
4361 * no effect since miimon is zero...
4363 printk(KERN_WARNING DRV_NAME
4364 ": Warning: miimon module parameter not set "
4365 "and updelay (%d) or downdelay (%d) module "
4366 "parameter is set; updelay and downdelay have "
4367 "no effect unless miimon is set\n",
4368 updelay
, downdelay
);
4371 /* don't allow arp monitoring */
4373 printk(KERN_WARNING DRV_NAME
4374 ": Warning: miimon (%d) and arp_interval (%d) "
4375 "can't be used simultaneously, disabling ARP "
4377 miimon
, arp_interval
);
4381 if ((updelay
% miimon
) != 0) {
4382 printk(KERN_WARNING DRV_NAME
4383 ": Warning: updelay (%d) is not a multiple "
4384 "of miimon (%d), updelay rounded to %d ms\n",
4385 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4390 if ((downdelay
% miimon
) != 0) {
4391 printk(KERN_WARNING DRV_NAME
4392 ": Warning: downdelay (%d) is not a multiple "
4393 "of miimon (%d), downdelay rounded to %d ms\n",
4395 (downdelay
/ miimon
) * miimon
);
4398 downdelay
/= miimon
;
4401 if (arp_interval
< 0) {
4402 printk(KERN_WARNING DRV_NAME
4403 ": Warning: arp_interval module parameter (%d) "
4404 ", not in range 0-%d, so it was reset to %d\n",
4405 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4406 arp_interval
= BOND_LINK_ARP_INTERV
;
4409 for (arp_ip_count
= 0;
4410 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4412 /* not complete check, but should be good enough to
4414 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4415 printk(KERN_WARNING DRV_NAME
4416 ": Warning: bad arp_ip_target module parameter "
4417 "(%s), ARP monitoring will not be performed\n",
4418 arp_ip_target
[arp_ip_count
]);
4421 u32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4422 arp_target
[arp_ip_count
] = ip
;
4426 if (arp_interval
&& !arp_ip_count
) {
4427 /* don't allow arping if no arp_ip_target given... */
4428 printk(KERN_WARNING DRV_NAME
4429 ": Warning: arp_interval module parameter (%d) "
4430 "specified without providing an arp_ip_target "
4431 "parameter, arp_interval was reset to 0\n",
4437 printk(KERN_INFO DRV_NAME
4438 ": MII link monitoring set to %d ms\n",
4440 } else if (arp_interval
) {
4443 printk(KERN_INFO DRV_NAME
4444 ": ARP monitoring set to %d ms with %d target(s):",
4445 arp_interval
, arp_ip_count
);
4447 for (i
= 0; i
< arp_ip_count
; i
++)
4448 printk (" %s", arp_ip_target
[i
]);
4453 /* miimon and arp_interval not set, we need one so things
4454 * work as expected, see bonding.txt for details
4456 printk(KERN_WARNING DRV_NAME
4457 ": Warning: either miimon or arp_interval and "
4458 "arp_ip_target module parameters must be specified, "
4459 "otherwise bonding will not detect link failures! see "
4460 "bonding.txt for details.\n");
4463 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4464 /* currently, using a primary only makes sense
4465 * in active backup, TLB or ALB modes
4467 printk(KERN_WARNING DRV_NAME
4468 ": Warning: %s primary device specified but has no "
4469 "effect in %s mode\n",
4470 primary
, bond_mode_name(bond_mode
));
4474 /* fill params struct with the proper values */
4475 params
->mode
= bond_mode
;
4476 params
->xmit_policy
= xmit_hashtype
;
4477 params
->miimon
= miimon
;
4478 params
->arp_interval
= arp_interval
;
4479 params
->updelay
= updelay
;
4480 params
->downdelay
= downdelay
;
4481 params
->use_carrier
= use_carrier
;
4482 params
->lacp_fast
= lacp_fast
;
4483 params
->primary
[0] = 0;
4486 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4487 params
->primary
[IFNAMSIZ
- 1] = 0;
4490 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4495 /* Create a new bond based on the specified name and bonding parameters.
4496 * Caller must NOT hold rtnl_lock; we need to release it here before we
4497 * set up our sysfs entries.
4499 int bond_create(char *name
, struct bond_params
*params
, struct bonding
**newbond
)
4501 struct net_device
*bond_dev
;
4505 bond_dev
= alloc_netdev(sizeof(struct bonding
), name
, ether_setup
);
4507 printk(KERN_ERR DRV_NAME
4508 ": %s: eek! can't alloc netdev!\n",
4514 /* bond_init() must be called after dev_alloc_name() (for the
4515 * /proc files), but before register_netdevice(), because we
4516 * need to set function pointers.
4519 res
= bond_init(bond_dev
, params
);
4524 SET_MODULE_OWNER(bond_dev
);
4526 res
= register_netdevice(bond_dev
);
4531 *newbond
= bond_dev
->priv
;
4533 rtnl_unlock(); /* allows sysfs registration of net device */
4534 res
= bond_create_sysfs_entry(bond_dev
->priv
);
4537 bond_deinit(bond_dev
);
4539 free_netdev(bond_dev
);
4546 static int __init
bonding_init(void)
4550 char new_bond_name
[8]; /* Enough room for 999 bonds at init. */
4552 printk(KERN_INFO
"%s", version
);
4554 res
= bond_check_params(&bonding_defaults
);
4559 #ifdef CONFIG_PROC_FS
4560 bond_create_proc_dir();
4562 for (i
= 0; i
< max_bonds
; i
++) {
4563 sprintf(new_bond_name
, "bond%d",i
);
4564 res
= bond_create(new_bond_name
,&bonding_defaults
, NULL
);
4569 res
= bond_create_sysfs();
4573 register_netdevice_notifier(&bond_netdev_notifier
);
4574 register_inetaddr_notifier(&bond_inetaddr_notifier
);
4580 bond_destroy_sysfs();
4587 static void __exit
bonding_exit(void)
4589 unregister_netdevice_notifier(&bond_netdev_notifier
);
4590 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
4594 bond_destroy_sysfs();
4598 module_init(bonding_init
);
4599 module_exit(bonding_exit
);
4600 MODULE_LICENSE("GPL");
4601 MODULE_VERSION(DRV_VERSION
);
4602 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4603 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4604 MODULE_SUPPORTED_DEVICE("most ethernet devices");