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 pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
57 #include <asm/system.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/netpoll.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.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 <linux/jiffies.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV 0
90 #define BOND_LINK_ARP_INTERV 0
92 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
93 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
94 static int num_grat_arp
= 1;
95 static int num_unsol_na
= 1;
96 static int miimon
= BOND_LINK_MON_INTERV
;
99 static int use_carrier
= 1;
101 static char *primary
;
102 static char *primary_reselect
;
103 static char *lacp_rate
;
104 static char *ad_select
;
105 static char *xmit_hash_policy
;
106 static int arp_interval
= BOND_LINK_ARP_INTERV
;
107 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
108 static char *arp_validate
;
109 static char *fail_over_mac
;
110 static int all_slaves_active
= 0;
111 static struct bond_params bonding_defaults
;
113 module_param(max_bonds
, int, 0);
114 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
115 module_param(tx_queues
, int, 0);
116 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
117 module_param(num_grat_arp
, int, 0644);
118 MODULE_PARM_DESC(num_grat_arp
, "Number of gratuitous ARP packets to send on failover event");
119 module_param(num_unsol_na
, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na
, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
121 module_param(miimon
, int, 0);
122 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
123 module_param(updelay
, int, 0);
124 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
125 module_param(downdelay
, int, 0);
126 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
128 module_param(use_carrier
, int, 0);
129 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
130 "0 for off, 1 for on (default)");
131 module_param(mode
, charp
, 0);
132 MODULE_PARM_DESC(mode
, "Mode of operation : 0 for balance-rr, "
133 "1 for active-backup, 2 for balance-xor, "
134 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
135 "6 for balance-alb");
136 module_param(primary
, charp
, 0);
137 MODULE_PARM_DESC(primary
, "Primary network device to use");
138 module_param(primary_reselect
, charp
, 0);
139 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
141 "0 for always (default), "
142 "1 for only if speed of primary is "
144 "2 for only on active slave "
146 module_param(lacp_rate
, charp
, 0);
147 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner "
149 module_param(ad_select
, charp
, 0);
150 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
151 module_param(xmit_hash_policy
, charp
, 0);
152 MODULE_PARM_DESC(xmit_hash_policy
, "XOR hashing method: 0 for layer 2 (default)"
153 ", 1 for layer 3+4");
154 module_param(arp_interval
, int, 0);
155 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
156 module_param_array(arp_ip_target
, charp
, NULL
, 0);
157 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
158 module_param(arp_validate
, charp
, 0);
159 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes: none (default), active, backup or all");
160 module_param(fail_over_mac
, charp
, 0);
161 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
162 module_param(all_slaves_active
, int, 0);
163 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
164 "by setting active flag for all slaves. "
165 "0 for never (default), 1 for always.");
167 /*----------------------------- Global variables ----------------------------*/
169 static const char * const version
=
170 DRV_DESCRIPTION
": v" DRV_VERSION
" (" DRV_RELDATE
")\n";
172 int bond_net_id __read_mostly
;
174 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
175 static int arp_ip_count
;
176 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
177 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
178 static int lacp_fast
;
179 #ifdef CONFIG_NET_POLL_CONTROLLER
180 static int disable_netpoll
= 1;
183 const struct bond_parm_tbl bond_lacp_tbl
[] = {
184 { "slow", AD_LACP_SLOW
},
185 { "fast", AD_LACP_FAST
},
189 const struct bond_parm_tbl bond_mode_tbl
[] = {
190 { "balance-rr", BOND_MODE_ROUNDROBIN
},
191 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
192 { "balance-xor", BOND_MODE_XOR
},
193 { "broadcast", BOND_MODE_BROADCAST
},
194 { "802.3ad", BOND_MODE_8023AD
},
195 { "balance-tlb", BOND_MODE_TLB
},
196 { "balance-alb", BOND_MODE_ALB
},
200 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
201 { "layer2", BOND_XMIT_POLICY_LAYER2
},
202 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
203 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
207 const struct bond_parm_tbl arp_validate_tbl
[] = {
208 { "none", BOND_ARP_VALIDATE_NONE
},
209 { "active", BOND_ARP_VALIDATE_ACTIVE
},
210 { "backup", BOND_ARP_VALIDATE_BACKUP
},
211 { "all", BOND_ARP_VALIDATE_ALL
},
215 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
216 { "none", BOND_FOM_NONE
},
217 { "active", BOND_FOM_ACTIVE
},
218 { "follow", BOND_FOM_FOLLOW
},
222 const struct bond_parm_tbl pri_reselect_tbl
[] = {
223 { "always", BOND_PRI_RESELECT_ALWAYS
},
224 { "better", BOND_PRI_RESELECT_BETTER
},
225 { "failure", BOND_PRI_RESELECT_FAILURE
},
229 struct bond_parm_tbl ad_select_tbl
[] = {
230 { "stable", BOND_AD_STABLE
},
231 { "bandwidth", BOND_AD_BANDWIDTH
},
232 { "count", BOND_AD_COUNT
},
236 /*-------------------------- Forward declarations ---------------------------*/
238 static void bond_send_gratuitous_arp(struct bonding
*bond
);
239 static int bond_init(struct net_device
*bond_dev
);
240 static void bond_uninit(struct net_device
*bond_dev
);
242 /*---------------------------- General routines -----------------------------*/
244 static const char *bond_mode_name(int mode
)
246 static const char *names
[] = {
247 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
248 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
249 [BOND_MODE_XOR
] = "load balancing (xor)",
250 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
251 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
252 [BOND_MODE_TLB
] = "transmit load balancing",
253 [BOND_MODE_ALB
] = "adaptive load balancing",
256 if (mode
< 0 || mode
> BOND_MODE_ALB
)
262 /*---------------------------------- VLAN -----------------------------------*/
265 * bond_add_vlan - add a new vlan id on bond
266 * @bond: bond that got the notification
267 * @vlan_id: the vlan id to add
269 * Returns -ENOMEM if allocation failed.
271 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
273 struct vlan_entry
*vlan
;
275 pr_debug("bond: %s, vlan id %d\n",
276 (bond
? bond
->dev
->name
: "None"), vlan_id
);
278 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
282 INIT_LIST_HEAD(&vlan
->vlan_list
);
283 vlan
->vlan_id
= vlan_id
;
285 write_lock_bh(&bond
->lock
);
287 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
289 write_unlock_bh(&bond
->lock
);
291 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
297 * bond_del_vlan - delete a vlan id from bond
298 * @bond: bond that got the notification
299 * @vlan_id: the vlan id to delete
301 * returns -ENODEV if @vlan_id was not found in @bond.
303 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
305 struct vlan_entry
*vlan
;
308 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
310 write_lock_bh(&bond
->lock
);
312 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
313 if (vlan
->vlan_id
== vlan_id
) {
314 list_del(&vlan
->vlan_list
);
316 if (bond_is_lb(bond
))
317 bond_alb_clear_vlan(bond
, vlan_id
);
319 pr_debug("removed VLAN ID %d from bond %s\n",
320 vlan_id
, bond
->dev
->name
);
324 if (list_empty(&bond
->vlan_list
) &&
325 (bond
->slave_cnt
== 0)) {
326 /* Last VLAN removed and no slaves, so
327 * restore block on adding VLANs. This will
328 * be removed once new slaves that are not
329 * VLAN challenged will be added.
331 bond
->dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
339 pr_debug("couldn't find VLAN ID %d in bond %s\n",
340 vlan_id
, bond
->dev
->name
);
343 write_unlock_bh(&bond
->lock
);
348 * bond_has_challenged_slaves
349 * @bond: the bond we're working on
351 * Searches the slave list. Returns 1 if a vlan challenged slave
352 * was found, 0 otherwise.
354 * Assumes bond->lock is held.
356 static int bond_has_challenged_slaves(struct bonding
*bond
)
361 bond_for_each_slave(bond
, slave
, i
) {
362 if (slave
->dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
363 pr_debug("found VLAN challenged slave - %s\n",
369 pr_debug("no VLAN challenged slaves found\n");
374 * bond_next_vlan - safely skip to the next item in the vlans list.
375 * @bond: the bond we're working on
376 * @curr: item we're advancing from
378 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
379 * or @curr->next otherwise (even if it is @curr itself again).
381 * Caller must hold bond->lock
383 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
385 struct vlan_entry
*next
, *last
;
387 if (list_empty(&bond
->vlan_list
))
391 next
= list_entry(bond
->vlan_list
.next
,
392 struct vlan_entry
, vlan_list
);
394 last
= list_entry(bond
->vlan_list
.prev
,
395 struct vlan_entry
, vlan_list
);
397 next
= list_entry(bond
->vlan_list
.next
,
398 struct vlan_entry
, vlan_list
);
400 next
= list_entry(curr
->vlan_list
.next
,
401 struct vlan_entry
, vlan_list
);
409 * bond_dev_queue_xmit - Prepare skb for xmit.
411 * @bond: bond device that got this skb for tx.
412 * @skb: hw accel VLAN tagged skb to transmit
413 * @slave_dev: slave that is supposed to xmit this skbuff
415 * When the bond gets an skb to transmit that is
416 * already hardware accelerated VLAN tagged, and it
417 * needs to relay this skb to a slave that is not
418 * hw accel capable, the skb needs to be "unaccelerated",
419 * i.e. strip the hwaccel tag and re-insert it as part
422 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
423 struct net_device
*slave_dev
)
425 unsigned short uninitialized_var(vlan_id
);
427 /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
428 if (!list_empty(&bond
->vlan_list
) &&
429 !(slave_dev
->features
& NETIF_F_HW_VLAN_TX
) &&
430 vlan_get_tag(skb
, &vlan_id
) == 0) {
431 skb
->dev
= slave_dev
;
432 skb
= vlan_put_tag(skb
, vlan_id
);
434 /* vlan_put_tag() frees the skb in case of error,
435 * so return success here so the calling functions
436 * won't attempt to free is again.
441 skb
->dev
= slave_dev
;
445 #ifdef CONFIG_NET_POLL_CONTROLLER
446 if (unlikely(bond
->dev
->priv_flags
& IFF_IN_NETPOLL
)) {
447 struct netpoll
*np
= bond
->dev
->npinfo
->netpoll
;
448 slave_dev
->npinfo
= bond
->dev
->npinfo
;
449 np
->real_dev
= np
->dev
= skb
->dev
;
450 slave_dev
->priv_flags
|= IFF_IN_NETPOLL
;
451 netpoll_send_skb(np
, skb
);
452 slave_dev
->priv_flags
&= ~IFF_IN_NETPOLL
;
462 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
463 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
465 * a. This operation is performed in IOCTL context,
466 * b. The operation is protected by the RTNL semaphore in the 8021q code,
467 * c. Holding a lock with BH disabled while directly calling a base driver
468 * entry point is generally a BAD idea.
470 * The design of synchronization/protection for this operation in the 8021q
471 * module is good for one or more VLAN devices over a single physical device
472 * and cannot be extended for a teaming solution like bonding, so there is a
473 * potential race condition here where a net device from the vlan group might
474 * be referenced (either by a base driver or the 8021q code) while it is being
475 * removed from the system. However, it turns out we're not making matters
476 * worse, and if it works for regular VLAN usage it will work here too.
480 * bond_vlan_rx_register - Propagates registration to slaves
481 * @bond_dev: bonding net device that got called
482 * @grp: vlan group being registered
484 static void bond_vlan_rx_register(struct net_device
*bond_dev
,
485 struct vlan_group
*grp
)
487 struct bonding
*bond
= netdev_priv(bond_dev
);
491 write_lock(&bond
->lock
);
493 write_unlock(&bond
->lock
);
495 bond_for_each_slave(bond
, slave
, i
) {
496 struct net_device
*slave_dev
= slave
->dev
;
497 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
499 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
500 slave_ops
->ndo_vlan_rx_register
) {
501 slave_ops
->ndo_vlan_rx_register(slave_dev
, grp
);
507 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
508 * @bond_dev: bonding net device that got called
509 * @vid: vlan id being added
511 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
513 struct bonding
*bond
= netdev_priv(bond_dev
);
517 bond_for_each_slave(bond
, slave
, i
) {
518 struct net_device
*slave_dev
= slave
->dev
;
519 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
521 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
522 slave_ops
->ndo_vlan_rx_add_vid
) {
523 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vid
);
527 res
= bond_add_vlan(bond
, vid
);
529 pr_err("%s: Error: Failed to add vlan id %d\n",
530 bond_dev
->name
, vid
);
535 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
536 * @bond_dev: bonding net device that got called
537 * @vid: vlan id being removed
539 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
541 struct bonding
*bond
= netdev_priv(bond_dev
);
543 struct net_device
*vlan_dev
;
546 bond_for_each_slave(bond
, slave
, i
) {
547 struct net_device
*slave_dev
= slave
->dev
;
548 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
550 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
551 slave_ops
->ndo_vlan_rx_kill_vid
) {
552 /* Save and then restore vlan_dev in the grp array,
553 * since the slave's driver might clear it.
555 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vid
);
556 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vid
);
557 vlan_group_set_device(bond
->vlgrp
, vid
, vlan_dev
);
561 res
= bond_del_vlan(bond
, vid
);
563 pr_err("%s: Error: Failed to remove vlan id %d\n",
564 bond_dev
->name
, vid
);
568 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
570 struct vlan_entry
*vlan
;
571 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
573 write_lock_bh(&bond
->lock
);
578 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
579 slave_ops
->ndo_vlan_rx_register
)
580 slave_ops
->ndo_vlan_rx_register(slave_dev
, bond
->vlgrp
);
582 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
583 !(slave_ops
->ndo_vlan_rx_add_vid
))
586 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
)
587 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
590 write_unlock_bh(&bond
->lock
);
593 static void bond_del_vlans_from_slave(struct bonding
*bond
,
594 struct net_device
*slave_dev
)
596 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
597 struct vlan_entry
*vlan
;
598 struct net_device
*vlan_dev
;
600 write_lock_bh(&bond
->lock
);
605 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
606 !(slave_ops
->ndo_vlan_rx_kill_vid
))
609 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
612 /* Save and then restore vlan_dev in the grp array,
613 * since the slave's driver might clear it.
615 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
616 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
617 vlan_group_set_device(bond
->vlgrp
, vlan
->vlan_id
, vlan_dev
);
621 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
622 slave_ops
->ndo_vlan_rx_register
)
623 slave_ops
->ndo_vlan_rx_register(slave_dev
, NULL
);
626 write_unlock_bh(&bond
->lock
);
629 /*------------------------------- Link status -------------------------------*/
632 * Set the carrier state for the master according to the state of its
633 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
634 * do special 802.3ad magic.
636 * Returns zero if carrier state does not change, nonzero if it does.
638 static int bond_set_carrier(struct bonding
*bond
)
643 if (bond
->slave_cnt
== 0)
646 if (bond
->params
.mode
== BOND_MODE_8023AD
)
647 return bond_3ad_set_carrier(bond
);
649 bond_for_each_slave(bond
, slave
, i
) {
650 if (slave
->link
== BOND_LINK_UP
) {
651 if (!netif_carrier_ok(bond
->dev
)) {
652 netif_carrier_on(bond
->dev
);
660 if (netif_carrier_ok(bond
->dev
)) {
661 netif_carrier_off(bond
->dev
);
668 * Get link speed and duplex from the slave's base driver
669 * using ethtool. If for some reason the call fails or the
670 * values are invalid, fake speed and duplex to 100/Full
673 static int bond_update_speed_duplex(struct slave
*slave
)
675 struct net_device
*slave_dev
= slave
->dev
;
676 struct ethtool_cmd etool
;
679 /* Fake speed and duplex */
680 slave
->speed
= SPEED_100
;
681 slave
->duplex
= DUPLEX_FULL
;
683 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
686 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
690 switch (etool
.speed
) {
700 switch (etool
.duplex
) {
708 slave
->speed
= etool
.speed
;
709 slave
->duplex
= etool
.duplex
;
715 * if <dev> supports MII link status reporting, check its link status.
717 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
718 * depending upon the setting of the use_carrier parameter.
720 * Return either BMSR_LSTATUS, meaning that the link is up (or we
721 * can't tell and just pretend it is), or 0, meaning that the link is
724 * If reporting is non-zero, instead of faking link up, return -1 if
725 * both ETHTOOL and MII ioctls fail (meaning the device does not
726 * support them). If use_carrier is set, return whatever it says.
727 * It'd be nice if there was a good way to tell if a driver supports
728 * netif_carrier, but there really isn't.
730 static int bond_check_dev_link(struct bonding
*bond
,
731 struct net_device
*slave_dev
, int reporting
)
733 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
734 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
736 struct mii_ioctl_data
*mii
;
738 if (!reporting
&& !netif_running(slave_dev
))
741 if (bond
->params
.use_carrier
)
742 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
744 /* Try to get link status using Ethtool first. */
745 if (slave_dev
->ethtool_ops
) {
746 if (slave_dev
->ethtool_ops
->get_link
) {
749 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
751 return link
? BMSR_LSTATUS
: 0;
755 /* Ethtool can't be used, fallback to MII ioctls. */
756 ioctl
= slave_ops
->ndo_do_ioctl
;
758 /* TODO: set pointer to correct ioctl on a per team member */
759 /* bases to make this more efficient. that is, once */
760 /* we determine the correct ioctl, we will always */
761 /* call it and not the others for that team */
765 * We cannot assume that SIOCGMIIPHY will also read a
766 * register; not all network drivers (e.g., e100)
770 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
771 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
773 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
774 mii
->reg_num
= MII_BMSR
;
775 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
776 return mii
->val_out
& BMSR_LSTATUS
;
781 * If reporting, report that either there's no dev->do_ioctl,
782 * or both SIOCGMIIREG and get_link failed (meaning that we
783 * cannot report link status). If not reporting, pretend
786 return reporting
? -1 : BMSR_LSTATUS
;
789 /*----------------------------- Multicast list ------------------------------*/
792 * Push the promiscuity flag down to appropriate slaves
794 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
797 if (USES_PRIMARY(bond
->params
.mode
)) {
798 /* write lock already acquired */
799 if (bond
->curr_active_slave
) {
800 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
806 bond_for_each_slave(bond
, slave
, i
) {
807 err
= dev_set_promiscuity(slave
->dev
, inc
);
816 * Push the allmulti flag down to all slaves
818 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
821 if (USES_PRIMARY(bond
->params
.mode
)) {
822 /* write lock already acquired */
823 if (bond
->curr_active_slave
) {
824 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
830 bond_for_each_slave(bond
, slave
, i
) {
831 err
= dev_set_allmulti(slave
->dev
, inc
);
840 * Add a Multicast address to slaves
843 static void bond_mc_add(struct bonding
*bond
, void *addr
)
845 if (USES_PRIMARY(bond
->params
.mode
)) {
846 /* write lock already acquired */
847 if (bond
->curr_active_slave
)
848 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
853 bond_for_each_slave(bond
, slave
, i
)
854 dev_mc_add(slave
->dev
, addr
);
859 * Remove a multicast address from slave
862 static void bond_mc_del(struct bonding
*bond
, void *addr
)
864 if (USES_PRIMARY(bond
->params
.mode
)) {
865 /* write lock already acquired */
866 if (bond
->curr_active_slave
)
867 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
871 bond_for_each_slave(bond
, slave
, i
) {
872 dev_mc_del(slave
->dev
, addr
);
879 * Retrieve the list of registered multicast addresses for the bonding
880 * device and retransmit an IGMP JOIN request to the current active
883 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
885 struct in_device
*in_dev
;
886 struct ip_mc_list
*im
;
889 in_dev
= __in_dev_get_rcu(bond
->dev
);
891 for (im
= in_dev
->mc_list
; im
; im
= im
->next
)
892 ip_mc_rejoin_group(im
);
899 * flush all members of flush->mc_list from device dev->mc_list
901 static void bond_mc_list_flush(struct net_device
*bond_dev
,
902 struct net_device
*slave_dev
)
904 struct bonding
*bond
= netdev_priv(bond_dev
);
905 struct netdev_hw_addr
*ha
;
907 netdev_for_each_mc_addr(ha
, bond_dev
)
908 dev_mc_del(slave_dev
, ha
->addr
);
910 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
911 /* del lacpdu mc addr from mc list */
912 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
914 dev_mc_del(slave_dev
, lacpdu_multicast
);
918 /*--------------------------- Active slave change ---------------------------*/
921 * Update the mc list and multicast-related flags for the new and
922 * old active slaves (if any) according to the multicast mode, and
923 * promiscuous flags unconditionally.
925 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
926 struct slave
*old_active
)
928 struct netdev_hw_addr
*ha
;
930 if (!USES_PRIMARY(bond
->params
.mode
))
931 /* nothing to do - mc list is already up-to-date on
937 if (bond
->dev
->flags
& IFF_PROMISC
)
938 dev_set_promiscuity(old_active
->dev
, -1);
940 if (bond
->dev
->flags
& IFF_ALLMULTI
)
941 dev_set_allmulti(old_active
->dev
, -1);
943 netdev_for_each_mc_addr(ha
, bond
->dev
)
944 dev_mc_del(old_active
->dev
, ha
->addr
);
948 /* FIXME: Signal errors upstream. */
949 if (bond
->dev
->flags
& IFF_PROMISC
)
950 dev_set_promiscuity(new_active
->dev
, 1);
952 if (bond
->dev
->flags
& IFF_ALLMULTI
)
953 dev_set_allmulti(new_active
->dev
, 1);
955 netdev_for_each_mc_addr(ha
, bond
->dev
)
956 dev_mc_add(new_active
->dev
, ha
->addr
);
957 bond_resend_igmp_join_requests(bond
);
962 * bond_do_fail_over_mac
964 * Perform special MAC address swapping for fail_over_mac settings
966 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
968 static void bond_do_fail_over_mac(struct bonding
*bond
,
969 struct slave
*new_active
,
970 struct slave
*old_active
)
971 __releases(&bond
->curr_slave_lock
)
972 __releases(&bond
->lock
)
973 __acquires(&bond
->lock
)
974 __acquires(&bond
->curr_slave_lock
)
976 u8 tmp_mac
[ETH_ALEN
];
977 struct sockaddr saddr
;
980 switch (bond
->params
.fail_over_mac
) {
981 case BOND_FOM_ACTIVE
:
983 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
984 new_active
->dev
->addr_len
);
986 case BOND_FOM_FOLLOW
:
988 * if new_active && old_active, swap them
989 * if just old_active, do nothing (going to no active slave)
990 * if just new_active, set new_active to bond's MAC
995 write_unlock_bh(&bond
->curr_slave_lock
);
996 read_unlock(&bond
->lock
);
999 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
1000 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
1002 saddr
.sa_family
= new_active
->dev
->type
;
1004 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
1005 saddr
.sa_family
= bond
->dev
->type
;
1008 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
1010 pr_err("%s: Error %d setting MAC of slave %s\n",
1011 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1018 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1019 saddr
.sa_family
= old_active
->dev
->type
;
1021 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
1023 pr_err("%s: Error %d setting MAC of slave %s\n",
1024 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1026 read_lock(&bond
->lock
);
1027 write_lock_bh(&bond
->curr_slave_lock
);
1030 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1031 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1037 static bool bond_should_change_active(struct bonding
*bond
)
1039 struct slave
*prim
= bond
->primary_slave
;
1040 struct slave
*curr
= bond
->curr_active_slave
;
1042 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
1044 if (bond
->force_primary
) {
1045 bond
->force_primary
= false;
1048 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
1049 (prim
->speed
< curr
->speed
||
1050 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
1052 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
1058 * find_best_interface - select the best available slave to be the active one
1059 * @bond: our bonding struct
1061 * Warning: Caller must hold curr_slave_lock for writing.
1063 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1065 struct slave
*new_active
, *old_active
;
1066 struct slave
*bestslave
= NULL
;
1067 int mintime
= bond
->params
.updelay
;
1070 new_active
= bond
->curr_active_slave
;
1072 if (!new_active
) { /* there were no active slaves left */
1073 if (bond
->slave_cnt
> 0) /* found one slave */
1074 new_active
= bond
->first_slave
;
1076 return NULL
; /* still no slave, return NULL */
1079 if ((bond
->primary_slave
) &&
1080 bond
->primary_slave
->link
== BOND_LINK_UP
&&
1081 bond_should_change_active(bond
)) {
1082 new_active
= bond
->primary_slave
;
1085 /* remember where to stop iterating over the slaves */
1086 old_active
= new_active
;
1088 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1089 if (new_active
->link
== BOND_LINK_UP
) {
1091 } else if (new_active
->link
== BOND_LINK_BACK
&&
1092 IS_UP(new_active
->dev
)) {
1093 /* link up, but waiting for stabilization */
1094 if (new_active
->delay
< mintime
) {
1095 mintime
= new_active
->delay
;
1096 bestslave
= new_active
;
1105 * change_active_interface - change the active slave into the specified one
1106 * @bond: our bonding struct
1107 * @new: the new slave to make the active one
1109 * Set the new slave to the bond's settings and unset them on the old
1110 * curr_active_slave.
1111 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1113 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1114 * because it is apparently the best available slave we have, even though its
1115 * updelay hasn't timed out yet.
1117 * If new_active is not NULL, caller must hold bond->lock for read and
1118 * curr_slave_lock for write_bh.
1120 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1122 struct slave
*old_active
= bond
->curr_active_slave
;
1124 if (old_active
== new_active
)
1128 new_active
->jiffies
= jiffies
;
1130 if (new_active
->link
== BOND_LINK_BACK
) {
1131 if (USES_PRIMARY(bond
->params
.mode
)) {
1132 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1133 bond
->dev
->name
, new_active
->dev
->name
,
1134 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1137 new_active
->delay
= 0;
1138 new_active
->link
= BOND_LINK_UP
;
1140 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1141 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1143 if (bond_is_lb(bond
))
1144 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1146 if (USES_PRIMARY(bond
->params
.mode
)) {
1147 pr_info("%s: making interface %s the new active one.\n",
1148 bond
->dev
->name
, new_active
->dev
->name
);
1153 if (USES_PRIMARY(bond
->params
.mode
))
1154 bond_mc_swap(bond
, new_active
, old_active
);
1156 if (bond_is_lb(bond
)) {
1157 bond_alb_handle_active_change(bond
, new_active
);
1159 bond_set_slave_inactive_flags(old_active
);
1161 bond_set_slave_active_flags(new_active
);
1163 bond
->curr_active_slave
= new_active
;
1166 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1168 bond_set_slave_inactive_flags(old_active
);
1171 bond_set_slave_active_flags(new_active
);
1173 if (bond
->params
.fail_over_mac
)
1174 bond_do_fail_over_mac(bond
, new_active
,
1177 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1178 bond_send_gratuitous_arp(bond
);
1180 bond
->send_unsol_na
= bond
->params
.num_unsol_na
;
1181 bond_send_unsolicited_na(bond
);
1183 write_unlock_bh(&bond
->curr_slave_lock
);
1184 read_unlock(&bond
->lock
);
1186 netdev_bonding_change(bond
->dev
, NETDEV_BONDING_FAILOVER
);
1188 read_lock(&bond
->lock
);
1189 write_lock_bh(&bond
->curr_slave_lock
);
1193 /* resend IGMP joins since all were sent on curr_active_slave */
1194 if (bond
->params
.mode
== BOND_MODE_ROUNDROBIN
) {
1195 bond_resend_igmp_join_requests(bond
);
1200 * bond_select_active_slave - select a new active slave, if needed
1201 * @bond: our bonding struct
1203 * This functions should be called when one of the following occurs:
1204 * - The old curr_active_slave has been released or lost its link.
1205 * - The primary_slave has got its link back.
1206 * - A slave has got its link back and there's no old curr_active_slave.
1208 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1210 void bond_select_active_slave(struct bonding
*bond
)
1212 struct slave
*best_slave
;
1215 best_slave
= bond_find_best_slave(bond
);
1216 if (best_slave
!= bond
->curr_active_slave
) {
1217 bond_change_active_slave(bond
, best_slave
);
1218 rv
= bond_set_carrier(bond
);
1222 if (netif_carrier_ok(bond
->dev
)) {
1223 pr_info("%s: first active interface up!\n",
1226 pr_info("%s: now running without any active interface !\n",
1232 /*--------------------------- slave list handling ---------------------------*/
1235 * This function attaches the slave to the end of list.
1237 * bond->lock held for writing by caller.
1239 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1241 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1242 new_slave
->next
= new_slave
;
1243 new_slave
->prev
= new_slave
;
1244 bond
->first_slave
= new_slave
;
1246 new_slave
->next
= bond
->first_slave
;
1247 new_slave
->prev
= bond
->first_slave
->prev
;
1248 new_slave
->next
->prev
= new_slave
;
1249 new_slave
->prev
->next
= new_slave
;
1256 * This function detaches the slave from the list.
1257 * WARNING: no check is made to verify if the slave effectively
1258 * belongs to <bond>.
1259 * Nothing is freed on return, structures are just unchained.
1260 * If any slave pointer in bond was pointing to <slave>,
1261 * it should be changed by the calling function.
1263 * bond->lock held for writing by caller.
1265 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1268 slave
->next
->prev
= slave
->prev
;
1271 slave
->prev
->next
= slave
->next
;
1273 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1274 if (bond
->slave_cnt
> 1) { /* there are more slave */
1275 bond
->first_slave
= slave
->next
;
1277 bond
->first_slave
= NULL
; /* slave was the last one */
1286 #ifdef CONFIG_NET_POLL_CONTROLLER
1288 * You must hold read lock on bond->lock before calling this.
1290 static bool slaves_support_netpoll(struct net_device
*bond_dev
)
1292 struct bonding
*bond
= netdev_priv(bond_dev
);
1293 struct slave
*slave
;
1297 bond_for_each_slave(bond
, slave
, i
) {
1298 if ((slave
->dev
->priv_flags
& IFF_DISABLE_NETPOLL
) ||
1299 !slave
->dev
->netdev_ops
->ndo_poll_controller
)
1302 return i
!= 0 && ret
;
1305 static void bond_poll_controller(struct net_device
*bond_dev
)
1307 struct net_device
*dev
= bond_dev
->npinfo
->netpoll
->real_dev
;
1308 if (dev
!= bond_dev
)
1309 netpoll_poll_dev(dev
);
1312 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1314 struct bonding
*bond
= netdev_priv(bond_dev
);
1315 struct slave
*slave
;
1316 const struct net_device_ops
*ops
;
1319 read_lock(&bond
->lock
);
1320 bond_dev
->npinfo
= NULL
;
1321 bond_for_each_slave(bond
, slave
, i
) {
1323 ops
= slave
->dev
->netdev_ops
;
1324 if (ops
->ndo_netpoll_cleanup
)
1325 ops
->ndo_netpoll_cleanup(slave
->dev
);
1327 slave
->dev
->npinfo
= NULL
;
1330 read_unlock(&bond
->lock
);
1335 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1341 /*---------------------------------- IOCTL ----------------------------------*/
1343 static int bond_sethwaddr(struct net_device
*bond_dev
,
1344 struct net_device
*slave_dev
)
1346 pr_debug("bond_dev=%p\n", bond_dev
);
1347 pr_debug("slave_dev=%p\n", slave_dev
);
1348 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1349 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1353 #define BOND_VLAN_FEATURES \
1354 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1355 NETIF_F_HW_VLAN_FILTER)
1358 * Compute the common dev->feature set available to all slaves. Some
1359 * feature bits are managed elsewhere, so preserve those feature bits
1360 * on the master device.
1362 static int bond_compute_features(struct bonding
*bond
)
1364 struct slave
*slave
;
1365 struct net_device
*bond_dev
= bond
->dev
;
1366 unsigned long features
= bond_dev
->features
;
1367 unsigned long vlan_features
= 0;
1368 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1369 bond_dev
->hard_header_len
);
1372 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1373 features
|= NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1375 if (!bond
->first_slave
)
1378 features
&= ~NETIF_F_ONE_FOR_ALL
;
1380 vlan_features
= bond
->first_slave
->dev
->vlan_features
;
1381 bond_for_each_slave(bond
, slave
, i
) {
1382 features
= netdev_increment_features(features
,
1383 slave
->dev
->features
,
1384 NETIF_F_ONE_FOR_ALL
);
1385 vlan_features
= netdev_increment_features(vlan_features
,
1386 slave
->dev
->vlan_features
,
1387 NETIF_F_ONE_FOR_ALL
);
1388 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1389 max_hard_header_len
= slave
->dev
->hard_header_len
;
1393 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1394 bond_dev
->features
= netdev_fix_features(features
, NULL
);
1395 bond_dev
->vlan_features
= netdev_fix_features(vlan_features
, NULL
);
1396 bond_dev
->hard_header_len
= max_hard_header_len
;
1401 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1402 struct net_device
*slave_dev
)
1404 struct bonding
*bond
= netdev_priv(bond_dev
);
1406 bond_dev
->header_ops
= slave_dev
->header_ops
;
1408 bond_dev
->type
= slave_dev
->type
;
1409 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1410 bond_dev
->addr_len
= slave_dev
->addr_len
;
1412 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1413 slave_dev
->addr_len
);
1414 bond
->setup_by_slave
= 1;
1417 /* enslave device <slave> to bond device <master> */
1418 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1420 struct bonding
*bond
= netdev_priv(bond_dev
);
1421 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1422 struct slave
*new_slave
= NULL
;
1423 struct netdev_hw_addr
*ha
;
1424 struct sockaddr addr
;
1426 int old_features
= bond_dev
->features
;
1429 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1430 slave_ops
->ndo_do_ioctl
== NULL
) {
1431 pr_warning("%s: Warning: no link monitoring support for %s\n",
1432 bond_dev
->name
, slave_dev
->name
);
1435 /* bond must be initialized by bond_open() before enslaving */
1436 if (!(bond_dev
->flags
& IFF_UP
)) {
1437 pr_warning("%s: master_dev is not up in bond_enslave\n",
1441 /* already enslaved */
1442 if (slave_dev
->flags
& IFF_SLAVE
) {
1443 pr_debug("Error, Device was already enslaved\n");
1447 /* vlan challenged mutual exclusion */
1448 /* no need to lock since we're protected by rtnl_lock */
1449 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1450 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1452 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1453 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1456 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1457 bond_dev
->name
, slave_dev
->name
,
1458 slave_dev
->name
, bond_dev
->name
);
1459 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1462 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1463 if (bond
->slave_cnt
== 0) {
1464 /* First slave, and it is not VLAN challenged,
1465 * so remove the block of adding VLANs over the bond.
1467 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1472 * Old ifenslave binaries are no longer supported. These can
1473 * be identified with moderate accuracy by the state of the slave:
1474 * the current ifenslave will set the interface down prior to
1475 * enslaving it; the old ifenslave will not.
1477 if ((slave_dev
->flags
& IFF_UP
)) {
1478 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1481 goto err_undo_flags
;
1484 /* set bonding device ether type by slave - bonding netdevices are
1485 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1486 * there is a need to override some of the type dependent attribs/funcs.
1488 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1489 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1491 if (bond
->slave_cnt
== 0) {
1492 if (bond_dev
->type
!= slave_dev
->type
) {
1493 pr_debug("%s: change device type from %d to %d\n",
1495 bond_dev
->type
, slave_dev
->type
);
1497 res
= netdev_bonding_change(bond_dev
,
1498 NETDEV_PRE_TYPE_CHANGE
);
1499 res
= notifier_to_errno(res
);
1501 pr_err("%s: refused to change device type\n",
1504 goto err_undo_flags
;
1507 /* Flush unicast and multicast addresses */
1508 dev_uc_flush(bond_dev
);
1509 dev_mc_flush(bond_dev
);
1511 if (slave_dev
->type
!= ARPHRD_ETHER
)
1512 bond_setup_by_slave(bond_dev
, slave_dev
);
1514 ether_setup(bond_dev
);
1516 netdev_bonding_change(bond_dev
,
1517 NETDEV_POST_TYPE_CHANGE
);
1519 } else if (bond_dev
->type
!= slave_dev
->type
) {
1520 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1522 slave_dev
->type
, bond_dev
->type
);
1524 goto err_undo_flags
;
1527 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1528 if (bond
->slave_cnt
== 0) {
1529 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1531 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1532 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1533 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1536 goto err_undo_flags
;
1540 /* If this is the first slave, then we need to set the master's hardware
1541 * address to be the same as the slave's. */
1542 if (bond
->slave_cnt
== 0)
1543 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1544 slave_dev
->addr_len
);
1547 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1550 goto err_undo_flags
;
1554 * Set the new_slave's queue_id to be zero. Queue ID mapping
1555 * is set via sysfs or module option if desired.
1557 new_slave
->queue_id
= 0;
1559 /* Save slave's original mtu and then set it to match the bond */
1560 new_slave
->original_mtu
= slave_dev
->mtu
;
1561 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1563 pr_debug("Error %d calling dev_set_mtu\n", res
);
1568 * Save slave's original ("permanent") mac address for modes
1569 * that need it, and for restoring it upon release, and then
1570 * set it to the master's address
1572 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1574 if (!bond
->params
.fail_over_mac
) {
1576 * Set slave to master's mac address. The application already
1577 * set the master's mac address to that of the first slave
1579 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1580 addr
.sa_family
= slave_dev
->type
;
1581 res
= dev_set_mac_address(slave_dev
, &addr
);
1583 pr_debug("Error %d calling set_mac_address\n", res
);
1584 goto err_restore_mtu
;
1588 res
= netdev_set_master(slave_dev
, bond_dev
);
1590 pr_debug("Error %d calling netdev_set_master\n", res
);
1591 goto err_restore_mac
;
1593 /* open the slave since the application closed it */
1594 res
= dev_open(slave_dev
);
1596 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1597 goto err_unset_master
;
1600 new_slave
->dev
= slave_dev
;
1601 slave_dev
->priv_flags
|= IFF_BONDING
;
1603 if (bond_is_lb(bond
)) {
1604 /* bond_alb_init_slave() must be called before all other stages since
1605 * it might fail and we do not want to have to undo everything
1607 res
= bond_alb_init_slave(bond
, new_slave
);
1612 /* If the mode USES_PRIMARY, then the new slave gets the
1613 * master's promisc (and mc) settings only if it becomes the
1614 * curr_active_slave, and that is taken care of later when calling
1615 * bond_change_active()
1617 if (!USES_PRIMARY(bond
->params
.mode
)) {
1618 /* set promiscuity level to new slave */
1619 if (bond_dev
->flags
& IFF_PROMISC
) {
1620 res
= dev_set_promiscuity(slave_dev
, 1);
1625 /* set allmulti level to new slave */
1626 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1627 res
= dev_set_allmulti(slave_dev
, 1);
1632 netif_addr_lock_bh(bond_dev
);
1633 /* upload master's mc_list to new slave */
1634 netdev_for_each_mc_addr(ha
, bond_dev
)
1635 dev_mc_add(slave_dev
, ha
->addr
);
1636 netif_addr_unlock_bh(bond_dev
);
1639 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1640 /* add lacpdu mc addr to mc list */
1641 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1643 dev_mc_add(slave_dev
, lacpdu_multicast
);
1646 bond_add_vlans_on_slave(bond
, slave_dev
);
1648 write_lock_bh(&bond
->lock
);
1650 bond_attach_slave(bond
, new_slave
);
1652 new_slave
->delay
= 0;
1653 new_slave
->link_failure_count
= 0;
1655 bond_compute_features(bond
);
1657 write_unlock_bh(&bond
->lock
);
1659 read_lock(&bond
->lock
);
1661 new_slave
->last_arp_rx
= jiffies
;
1663 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1664 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1666 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1668 * miimon is set but a bonded network driver
1669 * does not support ETHTOOL/MII and
1670 * arp_interval is not set. Note: if
1671 * use_carrier is enabled, we will never go
1672 * here (because netif_carrier is always
1673 * supported); thus, we don't need to change
1674 * the messages for netif_carrier.
1676 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1677 bond_dev
->name
, slave_dev
->name
);
1678 } else if (link_reporting
== -1) {
1679 /* unable get link status using mii/ethtool */
1680 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1681 bond_dev
->name
, slave_dev
->name
);
1685 /* check for initial state */
1686 if (!bond
->params
.miimon
||
1687 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1688 if (bond
->params
.updelay
) {
1689 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1690 new_slave
->link
= BOND_LINK_BACK
;
1691 new_slave
->delay
= bond
->params
.updelay
;
1693 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1694 new_slave
->link
= BOND_LINK_UP
;
1696 new_slave
->jiffies
= jiffies
;
1698 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1699 new_slave
->link
= BOND_LINK_DOWN
;
1702 if (bond_update_speed_duplex(new_slave
) &&
1703 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1704 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1705 bond_dev
->name
, new_slave
->dev
->name
);
1707 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1708 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1713 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1714 /* if there is a primary slave, remember it */
1715 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1716 bond
->primary_slave
= new_slave
;
1717 bond
->force_primary
= true;
1721 write_lock_bh(&bond
->curr_slave_lock
);
1723 switch (bond
->params
.mode
) {
1724 case BOND_MODE_ACTIVEBACKUP
:
1725 bond_set_slave_inactive_flags(new_slave
);
1726 bond_select_active_slave(bond
);
1728 case BOND_MODE_8023AD
:
1729 /* in 802.3ad mode, the internal mechanism
1730 * will activate the slaves in the selected
1733 bond_set_slave_inactive_flags(new_slave
);
1734 /* if this is the first slave */
1735 if (bond
->slave_cnt
== 1) {
1736 SLAVE_AD_INFO(new_slave
).id
= 1;
1737 /* Initialize AD with the number of times that the AD timer is called in 1 second
1738 * can be called only after the mac address of the bond is set
1740 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1741 bond
->params
.lacp_fast
);
1743 SLAVE_AD_INFO(new_slave
).id
=
1744 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1747 bond_3ad_bind_slave(new_slave
);
1751 new_slave
->state
= BOND_STATE_ACTIVE
;
1752 bond_set_slave_inactive_flags(new_slave
);
1753 bond_select_active_slave(bond
);
1756 pr_debug("This slave is always active in trunk mode\n");
1758 /* always active in trunk mode */
1759 new_slave
->state
= BOND_STATE_ACTIVE
;
1761 /* In trunking mode there is little meaning to curr_active_slave
1762 * anyway (it holds no special properties of the bond device),
1763 * so we can change it without calling change_active_interface()
1765 if (!bond
->curr_active_slave
)
1766 bond
->curr_active_slave
= new_slave
;
1769 } /* switch(bond_mode) */
1771 write_unlock_bh(&bond
->curr_slave_lock
);
1773 bond_set_carrier(bond
);
1775 #ifdef CONFIG_NET_POLL_CONTROLLER
1777 * Netpoll and bonding is broken, make sure it is not initialized
1778 * until it is fixed.
1780 if (disable_netpoll
) {
1781 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1783 if (slaves_support_netpoll(bond_dev
)) {
1784 bond_dev
->priv_flags
&= ~IFF_DISABLE_NETPOLL
;
1785 if (bond_dev
->npinfo
)
1786 slave_dev
->npinfo
= bond_dev
->npinfo
;
1787 } else if (!(bond_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)) {
1788 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1789 pr_info("New slave device %s does not support netpoll\n",
1791 pr_info("Disabling netpoll support for %s\n", bond_dev
->name
);
1795 read_unlock(&bond
->lock
);
1797 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1801 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1802 bond_dev
->name
, slave_dev
->name
,
1803 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1804 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1806 /* enslave is successful */
1809 /* Undo stages on error */
1811 dev_close(slave_dev
);
1814 netdev_set_master(slave_dev
, NULL
);
1817 if (!bond
->params
.fail_over_mac
) {
1818 /* XXX TODO - fom follow mode needs to change master's
1819 * MAC if this slave's MAC is in use by the bond, or at
1820 * least print a warning.
1822 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1823 addr
.sa_family
= slave_dev
->type
;
1824 dev_set_mac_address(slave_dev
, &addr
);
1828 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1834 bond_dev
->features
= old_features
;
1840 * Try to release the slave device <slave> from the bond device <master>
1841 * It is legal to access curr_active_slave without a lock because all the function
1844 * The rules for slave state should be:
1845 * for Active/Backup:
1846 * Active stays on all backups go down
1847 * for Bonded connections:
1848 * The first up interface should be left on and all others downed.
1850 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1852 struct bonding
*bond
= netdev_priv(bond_dev
);
1853 struct slave
*slave
, *oldcurrent
;
1854 struct sockaddr addr
;
1856 /* slave is not a slave or master is not master of this slave */
1857 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1858 (slave_dev
->master
!= bond_dev
)) {
1859 pr_err("%s: Error: cannot release %s.\n",
1860 bond_dev
->name
, slave_dev
->name
);
1864 netdev_bonding_change(bond_dev
, NETDEV_BONDING_DESLAVE
);
1865 write_lock_bh(&bond
->lock
);
1867 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1869 /* not a slave of this bond */
1870 pr_info("%s: %s not enslaved\n",
1871 bond_dev
->name
, slave_dev
->name
);
1872 write_unlock_bh(&bond
->lock
);
1876 if (!bond
->params
.fail_over_mac
) {
1877 if (!compare_ether_addr(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1878 bond
->slave_cnt
> 1)
1879 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1880 bond_dev
->name
, slave_dev
->name
,
1882 bond_dev
->name
, slave_dev
->name
);
1885 /* Inform AD package of unbinding of slave. */
1886 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1887 /* must be called before the slave is
1888 * detached from the list
1890 bond_3ad_unbind_slave(slave
);
1893 pr_info("%s: releasing %s interface %s\n",
1895 (slave
->state
== BOND_STATE_ACTIVE
) ? "active" : "backup",
1898 oldcurrent
= bond
->curr_active_slave
;
1900 bond
->current_arp_slave
= NULL
;
1902 /* release the slave from its bond */
1903 bond_detach_slave(bond
, slave
);
1905 bond_compute_features(bond
);
1907 if (bond
->primary_slave
== slave
)
1908 bond
->primary_slave
= NULL
;
1910 if (oldcurrent
== slave
)
1911 bond_change_active_slave(bond
, NULL
);
1913 if (bond_is_lb(bond
)) {
1914 /* Must be called only after the slave has been
1915 * detached from the list and the curr_active_slave
1916 * has been cleared (if our_slave == old_current),
1917 * but before a new active slave is selected.
1919 write_unlock_bh(&bond
->lock
);
1920 bond_alb_deinit_slave(bond
, slave
);
1921 write_lock_bh(&bond
->lock
);
1924 if (oldcurrent
== slave
) {
1926 * Note that we hold RTNL over this sequence, so there
1927 * is no concern that another slave add/remove event
1930 write_unlock_bh(&bond
->lock
);
1931 read_lock(&bond
->lock
);
1932 write_lock_bh(&bond
->curr_slave_lock
);
1934 bond_select_active_slave(bond
);
1936 write_unlock_bh(&bond
->curr_slave_lock
);
1937 read_unlock(&bond
->lock
);
1938 write_lock_bh(&bond
->lock
);
1941 if (bond
->slave_cnt
== 0) {
1942 bond_set_carrier(bond
);
1944 /* if the last slave was removed, zero the mac address
1945 * of the master so it will be set by the application
1946 * to the mac address of the first slave
1948 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1951 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1953 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1954 bond_dev
->name
, bond_dev
->name
);
1955 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1958 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1959 !bond_has_challenged_slaves(bond
)) {
1960 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1961 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1962 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1965 write_unlock_bh(&bond
->lock
);
1967 /* must do this from outside any spinlocks */
1968 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1970 bond_del_vlans_from_slave(bond
, slave_dev
);
1972 /* If the mode USES_PRIMARY, then we should only remove its
1973 * promisc and mc settings if it was the curr_active_slave, but that was
1974 * already taken care of above when we detached the slave
1976 if (!USES_PRIMARY(bond
->params
.mode
)) {
1977 /* unset promiscuity level from slave */
1978 if (bond_dev
->flags
& IFF_PROMISC
)
1979 dev_set_promiscuity(slave_dev
, -1);
1981 /* unset allmulti level from slave */
1982 if (bond_dev
->flags
& IFF_ALLMULTI
)
1983 dev_set_allmulti(slave_dev
, -1);
1985 /* flush master's mc_list from slave */
1986 netif_addr_lock_bh(bond_dev
);
1987 bond_mc_list_flush(bond_dev
, slave_dev
);
1988 netif_addr_unlock_bh(bond_dev
);
1991 netdev_set_master(slave_dev
, NULL
);
1993 #ifdef CONFIG_NET_POLL_CONTROLLER
1994 read_lock_bh(&bond
->lock
);
1996 /* Make sure netpoll over stays disabled until fixed. */
1997 if (!disable_netpoll
)
1998 if (slaves_support_netpoll(bond_dev
))
1999 bond_dev
->priv_flags
&= ~IFF_DISABLE_NETPOLL
;
2000 read_unlock_bh(&bond
->lock
);
2001 if (slave_dev
->netdev_ops
->ndo_netpoll_cleanup
)
2002 slave_dev
->netdev_ops
->ndo_netpoll_cleanup(slave_dev
);
2004 slave_dev
->npinfo
= NULL
;
2007 /* close slave before restoring its mac address */
2008 dev_close(slave_dev
);
2010 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2011 /* restore original ("permanent") mac address */
2012 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2013 addr
.sa_family
= slave_dev
->type
;
2014 dev_set_mac_address(slave_dev
, &addr
);
2017 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2019 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2020 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
2025 return 0; /* deletion OK */
2029 * First release a slave and than destroy the bond if no more slaves are left.
2030 * Must be under rtnl_lock when this function is called.
2032 int bond_release_and_destroy(struct net_device
*bond_dev
,
2033 struct net_device
*slave_dev
)
2035 struct bonding
*bond
= netdev_priv(bond_dev
);
2038 ret
= bond_release(bond_dev
, slave_dev
);
2039 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2040 pr_info("%s: destroying bond %s.\n",
2041 bond_dev
->name
, bond_dev
->name
);
2042 unregister_netdevice(bond_dev
);
2048 * This function releases all slaves.
2050 static int bond_release_all(struct net_device
*bond_dev
)
2052 struct bonding
*bond
= netdev_priv(bond_dev
);
2053 struct slave
*slave
;
2054 struct net_device
*slave_dev
;
2055 struct sockaddr addr
;
2057 write_lock_bh(&bond
->lock
);
2059 netif_carrier_off(bond_dev
);
2061 if (bond
->slave_cnt
== 0)
2064 bond
->current_arp_slave
= NULL
;
2065 bond
->primary_slave
= NULL
;
2066 bond_change_active_slave(bond
, NULL
);
2068 while ((slave
= bond
->first_slave
) != NULL
) {
2069 /* Inform AD package of unbinding of slave
2070 * before slave is detached from the list.
2072 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2073 bond_3ad_unbind_slave(slave
);
2075 slave_dev
= slave
->dev
;
2076 bond_detach_slave(bond
, slave
);
2078 /* now that the slave is detached, unlock and perform
2079 * all the undo steps that should not be called from
2082 write_unlock_bh(&bond
->lock
);
2084 if (bond_is_lb(bond
)) {
2085 /* must be called only after the slave
2086 * has been detached from the list
2088 bond_alb_deinit_slave(bond
, slave
);
2091 bond_compute_features(bond
);
2093 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2094 bond_del_vlans_from_slave(bond
, slave_dev
);
2096 /* If the mode USES_PRIMARY, then we should only remove its
2097 * promisc and mc settings if it was the curr_active_slave, but that was
2098 * already taken care of above when we detached the slave
2100 if (!USES_PRIMARY(bond
->params
.mode
)) {
2101 /* unset promiscuity level from slave */
2102 if (bond_dev
->flags
& IFF_PROMISC
)
2103 dev_set_promiscuity(slave_dev
, -1);
2105 /* unset allmulti level from slave */
2106 if (bond_dev
->flags
& IFF_ALLMULTI
)
2107 dev_set_allmulti(slave_dev
, -1);
2109 /* flush master's mc_list from slave */
2110 netif_addr_lock_bh(bond_dev
);
2111 bond_mc_list_flush(bond_dev
, slave_dev
);
2112 netif_addr_unlock_bh(bond_dev
);
2115 netdev_set_master(slave_dev
, NULL
);
2117 /* close slave before restoring its mac address */
2118 dev_close(slave_dev
);
2120 if (!bond
->params
.fail_over_mac
) {
2121 /* restore original ("permanent") mac address*/
2122 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2123 addr
.sa_family
= slave_dev
->type
;
2124 dev_set_mac_address(slave_dev
, &addr
);
2127 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2128 IFF_SLAVE_INACTIVE
);
2132 /* re-acquire the lock before getting the next slave */
2133 write_lock_bh(&bond
->lock
);
2136 /* zero the mac address of the master so it will be
2137 * set by the application to the mac address of the
2140 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2143 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2145 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2146 bond_dev
->name
, bond_dev
->name
);
2147 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2151 pr_info("%s: released all slaves\n", bond_dev
->name
);
2154 write_unlock_bh(&bond
->lock
);
2160 * This function changes the active slave to slave <slave_dev>.
2161 * It returns -EINVAL in the following cases.
2162 * - <slave_dev> is not found in the list.
2163 * - There is not active slave now.
2164 * - <slave_dev> is already active.
2165 * - The link state of <slave_dev> is not BOND_LINK_UP.
2166 * - <slave_dev> is not running.
2167 * In these cases, this function does nothing.
2168 * In the other cases, current_slave pointer is changed and 0 is returned.
2170 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2172 struct bonding
*bond
= netdev_priv(bond_dev
);
2173 struct slave
*old_active
= NULL
;
2174 struct slave
*new_active
= NULL
;
2177 if (!USES_PRIMARY(bond
->params
.mode
))
2180 /* Verify that master_dev is indeed the master of slave_dev */
2181 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2184 read_lock(&bond
->lock
);
2186 read_lock(&bond
->curr_slave_lock
);
2187 old_active
= bond
->curr_active_slave
;
2188 read_unlock(&bond
->curr_slave_lock
);
2190 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2193 * Changing to the current active: do nothing; return success.
2195 if (new_active
&& (new_active
== old_active
)) {
2196 read_unlock(&bond
->lock
);
2202 (new_active
->link
== BOND_LINK_UP
) &&
2203 IS_UP(new_active
->dev
)) {
2204 write_lock_bh(&bond
->curr_slave_lock
);
2205 bond_change_active_slave(bond
, new_active
);
2206 write_unlock_bh(&bond
->curr_slave_lock
);
2210 read_unlock(&bond
->lock
);
2215 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2217 struct bonding
*bond
= netdev_priv(bond_dev
);
2219 info
->bond_mode
= bond
->params
.mode
;
2220 info
->miimon
= bond
->params
.miimon
;
2222 read_lock(&bond
->lock
);
2223 info
->num_slaves
= bond
->slave_cnt
;
2224 read_unlock(&bond
->lock
);
2229 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2231 struct bonding
*bond
= netdev_priv(bond_dev
);
2232 struct slave
*slave
;
2233 int i
, res
= -ENODEV
;
2235 read_lock(&bond
->lock
);
2237 bond_for_each_slave(bond
, slave
, i
) {
2238 if (i
== (int)info
->slave_id
) {
2240 strcpy(info
->slave_name
, slave
->dev
->name
);
2241 info
->link
= slave
->link
;
2242 info
->state
= slave
->state
;
2243 info
->link_failure_count
= slave
->link_failure_count
;
2248 read_unlock(&bond
->lock
);
2253 /*-------------------------------- Monitoring -------------------------------*/
2256 static int bond_miimon_inspect(struct bonding
*bond
)
2258 struct slave
*slave
;
2259 int i
, link_state
, commit
= 0;
2260 bool ignore_updelay
;
2262 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2264 bond_for_each_slave(bond
, slave
, i
) {
2265 slave
->new_link
= BOND_LINK_NOCHANGE
;
2267 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2269 switch (slave
->link
) {
2274 slave
->link
= BOND_LINK_FAIL
;
2275 slave
->delay
= bond
->params
.downdelay
;
2277 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2279 (bond
->params
.mode
==
2280 BOND_MODE_ACTIVEBACKUP
) ?
2281 ((slave
->state
== BOND_STATE_ACTIVE
) ?
2282 "active " : "backup ") : "",
2284 bond
->params
.downdelay
* bond
->params
.miimon
);
2287 case BOND_LINK_FAIL
:
2290 * recovered before downdelay expired
2292 slave
->link
= BOND_LINK_UP
;
2293 slave
->jiffies
= jiffies
;
2294 pr_info("%s: link status up again after %d ms for interface %s.\n",
2296 (bond
->params
.downdelay
- slave
->delay
) *
2297 bond
->params
.miimon
,
2302 if (slave
->delay
<= 0) {
2303 slave
->new_link
= BOND_LINK_DOWN
;
2311 case BOND_LINK_DOWN
:
2315 slave
->link
= BOND_LINK_BACK
;
2316 slave
->delay
= bond
->params
.updelay
;
2319 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2320 bond
->dev
->name
, slave
->dev
->name
,
2321 ignore_updelay
? 0 :
2322 bond
->params
.updelay
*
2323 bond
->params
.miimon
);
2326 case BOND_LINK_BACK
:
2328 slave
->link
= BOND_LINK_DOWN
;
2329 pr_info("%s: link status down again after %d ms for interface %s.\n",
2331 (bond
->params
.updelay
- slave
->delay
) *
2332 bond
->params
.miimon
,
2341 if (slave
->delay
<= 0) {
2342 slave
->new_link
= BOND_LINK_UP
;
2344 ignore_updelay
= false;
2356 static void bond_miimon_commit(struct bonding
*bond
)
2358 struct slave
*slave
;
2361 bond_for_each_slave(bond
, slave
, i
) {
2362 switch (slave
->new_link
) {
2363 case BOND_LINK_NOCHANGE
:
2367 slave
->link
= BOND_LINK_UP
;
2368 slave
->jiffies
= jiffies
;
2370 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2371 /* prevent it from being the active one */
2372 slave
->state
= BOND_STATE_BACKUP
;
2373 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2374 /* make it immediately active */
2375 slave
->state
= BOND_STATE_ACTIVE
;
2376 } else if (slave
!= bond
->primary_slave
) {
2377 /* prevent it from being the active one */
2378 slave
->state
= BOND_STATE_BACKUP
;
2381 pr_info("%s: link status definitely up for interface %s.\n",
2382 bond
->dev
->name
, slave
->dev
->name
);
2384 /* notify ad that the link status has changed */
2385 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2386 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2388 if (bond_is_lb(bond
))
2389 bond_alb_handle_link_change(bond
, slave
,
2392 if (!bond
->curr_active_slave
||
2393 (slave
== bond
->primary_slave
))
2398 case BOND_LINK_DOWN
:
2399 if (slave
->link_failure_count
< UINT_MAX
)
2400 slave
->link_failure_count
++;
2402 slave
->link
= BOND_LINK_DOWN
;
2404 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2405 bond
->params
.mode
== BOND_MODE_8023AD
)
2406 bond_set_slave_inactive_flags(slave
);
2408 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2409 bond
->dev
->name
, slave
->dev
->name
);
2411 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2412 bond_3ad_handle_link_change(slave
,
2415 if (bond_is_lb(bond
))
2416 bond_alb_handle_link_change(bond
, slave
,
2419 if (slave
== bond
->curr_active_slave
)
2425 pr_err("%s: invalid new link %d on slave %s\n",
2426 bond
->dev
->name
, slave
->new_link
,
2428 slave
->new_link
= BOND_LINK_NOCHANGE
;
2435 write_lock_bh(&bond
->curr_slave_lock
);
2436 bond_select_active_slave(bond
);
2437 write_unlock_bh(&bond
->curr_slave_lock
);
2440 bond_set_carrier(bond
);
2446 * Really a wrapper that splits the mii monitor into two phases: an
2447 * inspection, then (if inspection indicates something needs to be done)
2448 * an acquisition of appropriate locks followed by a commit phase to
2449 * implement whatever link state changes are indicated.
2451 void bond_mii_monitor(struct work_struct
*work
)
2453 struct bonding
*bond
= container_of(work
, struct bonding
,
2456 read_lock(&bond
->lock
);
2457 if (bond
->kill_timers
)
2460 if (bond
->slave_cnt
== 0)
2463 if (bond
->send_grat_arp
) {
2464 read_lock(&bond
->curr_slave_lock
);
2465 bond_send_gratuitous_arp(bond
);
2466 read_unlock(&bond
->curr_slave_lock
);
2469 if (bond
->send_unsol_na
) {
2470 read_lock(&bond
->curr_slave_lock
);
2471 bond_send_unsolicited_na(bond
);
2472 read_unlock(&bond
->curr_slave_lock
);
2475 if (bond_miimon_inspect(bond
)) {
2476 read_unlock(&bond
->lock
);
2478 read_lock(&bond
->lock
);
2480 bond_miimon_commit(bond
);
2482 read_unlock(&bond
->lock
);
2483 rtnl_unlock(); /* might sleep, hold no other locks */
2484 read_lock(&bond
->lock
);
2488 if (bond
->params
.miimon
)
2489 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2490 msecs_to_jiffies(bond
->params
.miimon
));
2492 read_unlock(&bond
->lock
);
2495 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2497 struct in_device
*idev
;
2498 struct in_ifaddr
*ifa
;
2505 idev
= __in_dev_get_rcu(dev
);
2509 ifa
= idev
->ifa_list
;
2513 addr
= ifa
->ifa_local
;
2519 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2521 struct vlan_entry
*vlan
;
2523 if (ip
== bond
->master_ip
)
2526 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2527 if (ip
== vlan
->vlan_ip
)
2535 * We go to the (large) trouble of VLAN tagging ARP frames because
2536 * switches in VLAN mode (especially if ports are configured as
2537 * "native" to a VLAN) might not pass non-tagged frames.
2539 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2541 struct sk_buff
*skb
;
2543 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2544 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2546 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2547 NULL
, slave_dev
->dev_addr
, NULL
);
2550 pr_err("ARP packet allocation failed\n");
2554 skb
= vlan_put_tag(skb
, vlan_id
);
2556 pr_err("failed to insert VLAN tag\n");
2564 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2567 __be32
*targets
= bond
->params
.arp_targets
;
2568 struct vlan_entry
*vlan
;
2569 struct net_device
*vlan_dev
;
2573 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2576 pr_debug("basa: target %x\n", targets
[i
]);
2578 pr_debug("basa: empty vlan: arp_send\n");
2579 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2580 bond
->master_ip
, 0);
2585 * If VLANs are configured, we do a route lookup to
2586 * determine which VLAN interface would be used, so we
2587 * can tag the ARP with the proper VLAN tag.
2589 memset(&fl
, 0, sizeof(fl
));
2590 fl
.fl4_dst
= targets
[i
];
2591 fl
.fl4_tos
= RTO_ONLINK
;
2593 rv
= ip_route_output_key(dev_net(bond
->dev
), &rt
, &fl
);
2595 if (net_ratelimit()) {
2596 pr_warning("%s: no route to arp_ip_target %pI4\n",
2597 bond
->dev
->name
, &fl
.fl4_dst
);
2603 * This target is not on a VLAN
2605 if (rt
->dst
.dev
== bond
->dev
) {
2607 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2608 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2609 bond
->master_ip
, 0);
2614 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2615 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2616 if (vlan_dev
== rt
->dst
.dev
) {
2617 vlan_id
= vlan
->vlan_id
;
2618 pr_debug("basa: vlan match on %s %d\n",
2619 vlan_dev
->name
, vlan_id
);
2626 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2627 vlan
->vlan_ip
, vlan_id
);
2631 if (net_ratelimit()) {
2632 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2633 bond
->dev
->name
, &fl
.fl4_dst
,
2634 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2641 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2642 * for each VLAN above us.
2644 * Caller must hold curr_slave_lock for read or better
2646 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2648 struct slave
*slave
= bond
->curr_active_slave
;
2649 struct vlan_entry
*vlan
;
2650 struct net_device
*vlan_dev
;
2652 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2653 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
2655 if (!slave
|| !bond
->send_grat_arp
||
2656 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2659 bond
->send_grat_arp
--;
2661 if (bond
->master_ip
) {
2662 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2663 bond
->master_ip
, 0);
2669 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2670 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2671 if (vlan
->vlan_ip
) {
2672 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2673 vlan
->vlan_ip
, vlan
->vlan_id
);
2678 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2681 __be32
*targets
= bond
->params
.arp_targets
;
2683 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2684 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2685 &sip
, &tip
, i
, &targets
[i
],
2686 bond_has_this_ip(bond
, tip
));
2687 if (sip
== targets
[i
]) {
2688 if (bond_has_this_ip(bond
, tip
))
2689 slave
->last_arp_rx
= jiffies
;
2695 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2698 struct slave
*slave
;
2699 struct bonding
*bond
;
2700 unsigned char *arp_ptr
;
2703 if (dev
->priv_flags
& IFF_802_1Q_VLAN
) {
2705 * When using VLANS and bonding, dev and oriv_dev may be
2706 * incorrect if the physical interface supports VLAN
2707 * acceleration. With this change ARP validation now
2708 * works for hosts only reachable on the VLAN interface.
2710 dev
= vlan_dev_real_dev(dev
);
2711 orig_dev
= dev_get_by_index_rcu(dev_net(skb
->dev
),skb
->skb_iif
);
2714 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2717 bond
= netdev_priv(dev
);
2718 read_lock(&bond
->lock
);
2720 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2721 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2722 orig_dev
? orig_dev
->name
: "NULL");
2724 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2725 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2728 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2732 if (arp
->ar_hln
!= dev
->addr_len
||
2733 skb
->pkt_type
== PACKET_OTHERHOST
||
2734 skb
->pkt_type
== PACKET_LOOPBACK
||
2735 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2736 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2740 arp_ptr
= (unsigned char *)(arp
+ 1);
2741 arp_ptr
+= dev
->addr_len
;
2742 memcpy(&sip
, arp_ptr
, 4);
2743 arp_ptr
+= 4 + dev
->addr_len
;
2744 memcpy(&tip
, arp_ptr
, 4);
2746 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2747 bond
->dev
->name
, slave
->dev
->name
, slave
->state
,
2748 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2752 * Backup slaves won't see the ARP reply, but do come through
2753 * here for each ARP probe (so we swap the sip/tip to validate
2754 * the probe). In a "redundant switch, common router" type of
2755 * configuration, the ARP probe will (hopefully) travel from
2756 * the active, through one switch, the router, then the other
2757 * switch before reaching the backup.
2759 if (slave
->state
== BOND_STATE_ACTIVE
)
2760 bond_validate_arp(bond
, slave
, sip
, tip
);
2762 bond_validate_arp(bond
, slave
, tip
, sip
);
2765 read_unlock(&bond
->lock
);
2768 return NET_RX_SUCCESS
;
2772 * this function is called regularly to monitor each slave's link
2773 * ensuring that traffic is being sent and received when arp monitoring
2774 * is used in load-balancing mode. if the adapter has been dormant, then an
2775 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2776 * arp monitoring in active backup mode.
2778 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2780 struct bonding
*bond
= container_of(work
, struct bonding
,
2782 struct slave
*slave
, *oldcurrent
;
2783 int do_failover
= 0;
2787 read_lock(&bond
->lock
);
2789 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2791 if (bond
->kill_timers
)
2794 if (bond
->slave_cnt
== 0)
2797 read_lock(&bond
->curr_slave_lock
);
2798 oldcurrent
= bond
->curr_active_slave
;
2799 read_unlock(&bond
->curr_slave_lock
);
2801 /* see if any of the previous devices are up now (i.e. they have
2802 * xmt and rcv traffic). the curr_active_slave does not come into
2803 * the picture unless it is null. also, slave->jiffies is not needed
2804 * here because we send an arp on each slave and give a slave as
2805 * long as it needs to get the tx/rx within the delta.
2806 * TODO: what about up/down delay in arp mode? it wasn't here before
2809 bond_for_each_slave(bond
, slave
, i
) {
2810 if (slave
->link
!= BOND_LINK_UP
) {
2811 if (time_before_eq(jiffies
, dev_trans_start(slave
->dev
) + delta_in_ticks
) &&
2812 time_before_eq(jiffies
, slave
->dev
->last_rx
+ delta_in_ticks
)) {
2814 slave
->link
= BOND_LINK_UP
;
2815 slave
->state
= BOND_STATE_ACTIVE
;
2817 /* primary_slave has no meaning in round-robin
2818 * mode. the window of a slave being up and
2819 * curr_active_slave being null after enslaving
2823 pr_info("%s: link status definitely up for interface %s, ",
2828 pr_info("%s: interface %s is now up\n",
2834 /* slave->link == BOND_LINK_UP */
2836 /* not all switches will respond to an arp request
2837 * when the source ip is 0, so don't take the link down
2838 * if we don't know our ip yet
2840 if (time_after_eq(jiffies
, dev_trans_start(slave
->dev
) + 2*delta_in_ticks
) ||
2841 (time_after_eq(jiffies
, slave
->dev
->last_rx
+ 2*delta_in_ticks
))) {
2843 slave
->link
= BOND_LINK_DOWN
;
2844 slave
->state
= BOND_STATE_BACKUP
;
2846 if (slave
->link_failure_count
< UINT_MAX
)
2847 slave
->link_failure_count
++;
2849 pr_info("%s: interface %s is now down.\n",
2853 if (slave
== oldcurrent
)
2858 /* note: if switch is in round-robin mode, all links
2859 * must tx arp to ensure all links rx an arp - otherwise
2860 * links may oscillate or not come up at all; if switch is
2861 * in something like xor mode, there is nothing we can
2862 * do - all replies will be rx'ed on same link causing slaves
2863 * to be unstable during low/no traffic periods
2865 if (IS_UP(slave
->dev
))
2866 bond_arp_send_all(bond
, slave
);
2870 write_lock_bh(&bond
->curr_slave_lock
);
2872 bond_select_active_slave(bond
);
2874 write_unlock_bh(&bond
->curr_slave_lock
);
2878 if (bond
->params
.arp_interval
)
2879 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2881 read_unlock(&bond
->lock
);
2885 * Called to inspect slaves for active-backup mode ARP monitor link state
2886 * changes. Sets new_link in slaves to specify what action should take
2887 * place for the slave. Returns 0 if no changes are found, >0 if changes
2888 * to link states must be committed.
2890 * Called with bond->lock held for read.
2892 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2894 struct slave
*slave
;
2897 bond_for_each_slave(bond
, slave
, i
) {
2898 slave
->new_link
= BOND_LINK_NOCHANGE
;
2900 if (slave
->link
!= BOND_LINK_UP
) {
2901 if (time_before_eq(jiffies
, slave_last_rx(bond
, slave
) +
2903 slave
->new_link
= BOND_LINK_UP
;
2911 * Give slaves 2*delta after being enslaved or made
2912 * active. This avoids bouncing, as the last receive
2913 * times need a full ARP monitor cycle to be updated.
2915 if (!time_after_eq(jiffies
, slave
->jiffies
+
2916 2 * delta_in_ticks
))
2920 * Backup slave is down if:
2921 * - No current_arp_slave AND
2922 * - more than 3*delta since last receive AND
2923 * - the bond has an IP address
2925 * Note: a non-null current_arp_slave indicates
2926 * the curr_active_slave went down and we are
2927 * searching for a new one; under this condition
2928 * we only take the curr_active_slave down - this
2929 * gives each slave a chance to tx/rx traffic
2930 * before being taken out
2932 if (slave
->state
== BOND_STATE_BACKUP
&&
2933 !bond
->current_arp_slave
&&
2934 time_after(jiffies
, slave_last_rx(bond
, slave
) +
2935 3 * delta_in_ticks
)) {
2936 slave
->new_link
= BOND_LINK_DOWN
;
2941 * Active slave is down if:
2942 * - more than 2*delta since transmitting OR
2943 * - (more than 2*delta since receive AND
2944 * the bond has an IP address)
2946 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
2947 (time_after_eq(jiffies
, dev_trans_start(slave
->dev
) +
2948 2 * delta_in_ticks
) ||
2949 (time_after_eq(jiffies
, slave_last_rx(bond
, slave
)
2950 + 2 * delta_in_ticks
)))) {
2951 slave
->new_link
= BOND_LINK_DOWN
;
2960 * Called to commit link state changes noted by inspection step of
2961 * active-backup mode ARP monitor.
2963 * Called with RTNL and bond->lock for read.
2965 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2967 struct slave
*slave
;
2970 bond_for_each_slave(bond
, slave
, i
) {
2971 switch (slave
->new_link
) {
2972 case BOND_LINK_NOCHANGE
:
2976 if ((!bond
->curr_active_slave
&&
2977 time_before_eq(jiffies
,
2978 dev_trans_start(slave
->dev
) +
2980 bond
->curr_active_slave
!= slave
) {
2981 slave
->link
= BOND_LINK_UP
;
2982 bond
->current_arp_slave
= NULL
;
2984 pr_info("%s: link status definitely up for interface %s.\n",
2985 bond
->dev
->name
, slave
->dev
->name
);
2987 if (!bond
->curr_active_slave
||
2988 (slave
== bond
->primary_slave
))
2995 case BOND_LINK_DOWN
:
2996 if (slave
->link_failure_count
< UINT_MAX
)
2997 slave
->link_failure_count
++;
2999 slave
->link
= BOND_LINK_DOWN
;
3000 bond_set_slave_inactive_flags(slave
);
3002 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3003 bond
->dev
->name
, slave
->dev
->name
);
3005 if (slave
== bond
->curr_active_slave
) {
3006 bond
->current_arp_slave
= NULL
;
3013 pr_err("%s: impossible: new_link %d on slave %s\n",
3014 bond
->dev
->name
, slave
->new_link
,
3021 write_lock_bh(&bond
->curr_slave_lock
);
3022 bond_select_active_slave(bond
);
3023 write_unlock_bh(&bond
->curr_slave_lock
);
3026 bond_set_carrier(bond
);
3030 * Send ARP probes for active-backup mode ARP monitor.
3032 * Called with bond->lock held for read.
3034 static void bond_ab_arp_probe(struct bonding
*bond
)
3036 struct slave
*slave
;
3039 read_lock(&bond
->curr_slave_lock
);
3041 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3042 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3043 bond
->current_arp_slave
->dev
->name
,
3044 bond
->curr_active_slave
->dev
->name
);
3046 if (bond
->curr_active_slave
) {
3047 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3048 read_unlock(&bond
->curr_slave_lock
);
3052 read_unlock(&bond
->curr_slave_lock
);
3054 /* if we don't have a curr_active_slave, search for the next available
3055 * backup slave from the current_arp_slave and make it the candidate
3056 * for becoming the curr_active_slave
3059 if (!bond
->current_arp_slave
) {
3060 bond
->current_arp_slave
= bond
->first_slave
;
3061 if (!bond
->current_arp_slave
)
3065 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3067 /* search for next candidate */
3068 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3069 if (IS_UP(slave
->dev
)) {
3070 slave
->link
= BOND_LINK_BACK
;
3071 bond_set_slave_active_flags(slave
);
3072 bond_arp_send_all(bond
, slave
);
3073 slave
->jiffies
= jiffies
;
3074 bond
->current_arp_slave
= slave
;
3078 /* if the link state is up at this point, we
3079 * mark it down - this can happen if we have
3080 * simultaneous link failures and
3081 * reselect_active_interface doesn't make this
3082 * one the current slave so it is still marked
3083 * up when it is actually down
3085 if (slave
->link
== BOND_LINK_UP
) {
3086 slave
->link
= BOND_LINK_DOWN
;
3087 if (slave
->link_failure_count
< UINT_MAX
)
3088 slave
->link_failure_count
++;
3090 bond_set_slave_inactive_flags(slave
);
3092 pr_info("%s: backup interface %s is now down.\n",
3093 bond
->dev
->name
, slave
->dev
->name
);
3098 void bond_activebackup_arp_mon(struct work_struct
*work
)
3100 struct bonding
*bond
= container_of(work
, struct bonding
,
3104 read_lock(&bond
->lock
);
3106 if (bond
->kill_timers
)
3109 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3111 if (bond
->slave_cnt
== 0)
3114 if (bond
->send_grat_arp
) {
3115 read_lock(&bond
->curr_slave_lock
);
3116 bond_send_gratuitous_arp(bond
);
3117 read_unlock(&bond
->curr_slave_lock
);
3120 if (bond
->send_unsol_na
) {
3121 read_lock(&bond
->curr_slave_lock
);
3122 bond_send_unsolicited_na(bond
);
3123 read_unlock(&bond
->curr_slave_lock
);
3126 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3127 read_unlock(&bond
->lock
);
3129 read_lock(&bond
->lock
);
3131 bond_ab_arp_commit(bond
, delta_in_ticks
);
3133 read_unlock(&bond
->lock
);
3135 read_lock(&bond
->lock
);
3138 bond_ab_arp_probe(bond
);
3141 if (bond
->params
.arp_interval
)
3142 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3144 read_unlock(&bond
->lock
);
3147 /*------------------------------ proc/seq_file-------------------------------*/
3149 #ifdef CONFIG_PROC_FS
3151 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3152 __acquires(&dev_base_lock
)
3153 __acquires(&bond
->lock
)
3155 struct bonding
*bond
= seq
->private;
3157 struct slave
*slave
;
3160 /* make sure the bond won't be taken away */
3161 read_lock(&dev_base_lock
);
3162 read_lock(&bond
->lock
);
3165 return SEQ_START_TOKEN
;
3167 bond_for_each_slave(bond
, slave
, i
) {
3175 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3177 struct bonding
*bond
= seq
->private;
3178 struct slave
*slave
= v
;
3181 if (v
== SEQ_START_TOKEN
)
3182 return bond
->first_slave
;
3184 slave
= slave
->next
;
3186 return (slave
== bond
->first_slave
) ? NULL
: slave
;
3189 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
3190 __releases(&bond
->lock
)
3191 __releases(&dev_base_lock
)
3193 struct bonding
*bond
= seq
->private;
3195 read_unlock(&bond
->lock
);
3196 read_unlock(&dev_base_lock
);
3199 static void bond_info_show_master(struct seq_file
*seq
)
3201 struct bonding
*bond
= seq
->private;
3205 read_lock(&bond
->curr_slave_lock
);
3206 curr
= bond
->curr_active_slave
;
3207 read_unlock(&bond
->curr_slave_lock
);
3209 seq_printf(seq
, "Bonding Mode: %s",
3210 bond_mode_name(bond
->params
.mode
));
3212 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
&&
3213 bond
->params
.fail_over_mac
)
3214 seq_printf(seq
, " (fail_over_mac %s)",
3215 fail_over_mac_tbl
[bond
->params
.fail_over_mac
].modename
);
3217 seq_printf(seq
, "\n");
3219 if (bond
->params
.mode
== BOND_MODE_XOR
||
3220 bond
->params
.mode
== BOND_MODE_8023AD
) {
3221 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
3222 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
3223 bond
->params
.xmit_policy
);
3226 if (USES_PRIMARY(bond
->params
.mode
)) {
3227 seq_printf(seq
, "Primary Slave: %s",
3228 (bond
->primary_slave
) ?
3229 bond
->primary_slave
->dev
->name
: "None");
3230 if (bond
->primary_slave
)
3231 seq_printf(seq
, " (primary_reselect %s)",
3232 pri_reselect_tbl
[bond
->params
.primary_reselect
].modename
);
3234 seq_printf(seq
, "\nCurrently Active Slave: %s\n",
3235 (curr
) ? curr
->dev
->name
: "None");
3238 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
3240 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
3241 seq_printf(seq
, "Up Delay (ms): %d\n",
3242 bond
->params
.updelay
* bond
->params
.miimon
);
3243 seq_printf(seq
, "Down Delay (ms): %d\n",
3244 bond
->params
.downdelay
* bond
->params
.miimon
);
3247 /* ARP information */
3248 if (bond
->params
.arp_interval
> 0) {
3250 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
3251 bond
->params
.arp_interval
);
3253 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
3255 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
3256 if (!bond
->params
.arp_targets
[i
])
3259 seq_printf(seq
, ",");
3260 seq_printf(seq
, " %pI4", &bond
->params
.arp_targets
[i
]);
3263 seq_printf(seq
, "\n");
3266 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3267 struct ad_info ad_info
;
3269 seq_puts(seq
, "\n802.3ad info\n");
3270 seq_printf(seq
, "LACP rate: %s\n",
3271 (bond
->params
.lacp_fast
) ? "fast" : "slow");
3272 seq_printf(seq
, "Aggregator selection policy (ad_select): %s\n",
3273 ad_select_tbl
[bond
->params
.ad_select
].modename
);
3275 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3276 seq_printf(seq
, "bond %s has no active aggregator\n",
3279 seq_printf(seq
, "Active Aggregator Info:\n");
3281 seq_printf(seq
, "\tAggregator ID: %d\n",
3282 ad_info
.aggregator_id
);
3283 seq_printf(seq
, "\tNumber of ports: %d\n",
3285 seq_printf(seq
, "\tActor Key: %d\n",
3287 seq_printf(seq
, "\tPartner Key: %d\n",
3288 ad_info
.partner_key
);
3289 seq_printf(seq
, "\tPartner Mac Address: %pM\n",
3290 ad_info
.partner_system
);
3295 static void bond_info_show_slave(struct seq_file
*seq
,
3296 const struct slave
*slave
)
3298 struct bonding
*bond
= seq
->private;
3300 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
3301 seq_printf(seq
, "MII Status: %s\n",
3302 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
3303 seq_printf(seq
, "Link Failure Count: %u\n",
3304 slave
->link_failure_count
);
3306 seq_printf(seq
, "Permanent HW addr: %pM\n", slave
->perm_hwaddr
);
3308 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3309 const struct aggregator
*agg
3310 = SLAVE_AD_INFO(slave
).port
.aggregator
;
3313 seq_printf(seq
, "Aggregator ID: %d\n",
3314 agg
->aggregator_identifier
);
3316 seq_puts(seq
, "Aggregator ID: N/A\n");
3318 seq_printf(seq
, "Slave queue ID: %d\n", slave
->queue_id
);
3321 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
3323 if (v
== SEQ_START_TOKEN
) {
3324 seq_printf(seq
, "%s\n", version
);
3325 bond_info_show_master(seq
);
3327 bond_info_show_slave(seq
, v
);
3332 static const struct seq_operations bond_info_seq_ops
= {
3333 .start
= bond_info_seq_start
,
3334 .next
= bond_info_seq_next
,
3335 .stop
= bond_info_seq_stop
,
3336 .show
= bond_info_seq_show
,
3339 static int bond_info_open(struct inode
*inode
, struct file
*file
)
3341 struct seq_file
*seq
;
3342 struct proc_dir_entry
*proc
;
3345 res
= seq_open(file
, &bond_info_seq_ops
);
3347 /* recover the pointer buried in proc_dir_entry data */
3348 seq
= file
->private_data
;
3350 seq
->private = proc
->data
;
3356 static const struct file_operations bond_info_fops
= {
3357 .owner
= THIS_MODULE
,
3358 .open
= bond_info_open
,
3360 .llseek
= seq_lseek
,
3361 .release
= seq_release
,
3364 static void bond_create_proc_entry(struct bonding
*bond
)
3366 struct net_device
*bond_dev
= bond
->dev
;
3367 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
3370 bond
->proc_entry
= proc_create_data(bond_dev
->name
,
3371 S_IRUGO
, bn
->proc_dir
,
3372 &bond_info_fops
, bond
);
3373 if (bond
->proc_entry
== NULL
)
3374 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3375 DRV_NAME
, bond_dev
->name
);
3377 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3381 static void bond_remove_proc_entry(struct bonding
*bond
)
3383 struct net_device
*bond_dev
= bond
->dev
;
3384 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
3386 if (bn
->proc_dir
&& bond
->proc_entry
) {
3387 remove_proc_entry(bond
->proc_file_name
, bn
->proc_dir
);
3388 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3389 bond
->proc_entry
= NULL
;
3393 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3394 * Caller must hold rtnl_lock.
3396 static void __net_init
bond_create_proc_dir(struct bond_net
*bn
)
3398 if (!bn
->proc_dir
) {
3399 bn
->proc_dir
= proc_mkdir(DRV_NAME
, bn
->net
->proc_net
);
3401 pr_warning("Warning: cannot create /proc/net/%s\n",
3406 /* Destroy the bonding directory under /proc/net, if empty.
3407 * Caller must hold rtnl_lock.
3409 static void __net_exit
bond_destroy_proc_dir(struct bond_net
*bn
)
3412 remove_proc_entry(DRV_NAME
, bn
->net
->proc_net
);
3413 bn
->proc_dir
= NULL
;
3417 #else /* !CONFIG_PROC_FS */
3419 static void bond_create_proc_entry(struct bonding
*bond
)
3423 static void bond_remove_proc_entry(struct bonding
*bond
)
3427 static inline void bond_create_proc_dir(struct bond_net
*bn
)
3431 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
3435 #endif /* CONFIG_PROC_FS */
3438 /*-------------------------- netdev event handling --------------------------*/
3441 * Change device name
3443 static int bond_event_changename(struct bonding
*bond
)
3445 bond_remove_proc_entry(bond
);
3446 bond_create_proc_entry(bond
);
3451 static int bond_master_netdev_event(unsigned long event
,
3452 struct net_device
*bond_dev
)
3454 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3457 case NETDEV_CHANGENAME
:
3458 return bond_event_changename(event_bond
);
3466 static int bond_slave_netdev_event(unsigned long event
,
3467 struct net_device
*slave_dev
)
3469 struct net_device
*bond_dev
= slave_dev
->master
;
3470 struct bonding
*bond
= netdev_priv(bond_dev
);
3473 case NETDEV_UNREGISTER
:
3475 if (bond
->setup_by_slave
)
3476 bond_release_and_destroy(bond_dev
, slave_dev
);
3478 bond_release(bond_dev
, slave_dev
);
3482 if (bond
->params
.mode
== BOND_MODE_8023AD
|| bond_is_lb(bond
)) {
3483 struct slave
*slave
;
3485 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3487 u16 old_speed
= slave
->speed
;
3488 u16 old_duplex
= slave
->duplex
;
3490 bond_update_speed_duplex(slave
);
3492 if (bond_is_lb(bond
))
3495 if (old_speed
!= slave
->speed
)
3496 bond_3ad_adapter_speed_changed(slave
);
3497 if (old_duplex
!= slave
->duplex
)
3498 bond_3ad_adapter_duplex_changed(slave
);
3505 * ... Or is it this?
3508 case NETDEV_CHANGEMTU
:
3510 * TODO: Should slaves be allowed to
3511 * independently alter their MTU? For
3512 * an active-backup bond, slaves need
3513 * not be the same type of device, so
3514 * MTUs may vary. For other modes,
3515 * slaves arguably should have the
3516 * same MTUs. To do this, we'd need to
3517 * take over the slave's change_mtu
3518 * function for the duration of their
3522 case NETDEV_CHANGENAME
:
3524 * TODO: handle changing the primary's name
3527 case NETDEV_FEAT_CHANGE
:
3528 bond_compute_features(bond
);
3538 * bond_netdev_event: handle netdev notifier chain events.
3540 * This function receives events for the netdev chain. The caller (an
3541 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3542 * locks for us to safely manipulate the slave devices (RTNL lock,
3545 static int bond_netdev_event(struct notifier_block
*this,
3546 unsigned long event
, void *ptr
)
3548 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3550 pr_debug("event_dev: %s, event: %lx\n",
3551 event_dev
? event_dev
->name
: "None",
3554 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3557 if (event_dev
->flags
& IFF_MASTER
) {
3558 pr_debug("IFF_MASTER\n");
3559 return bond_master_netdev_event(event
, event_dev
);
3562 if (event_dev
->flags
& IFF_SLAVE
) {
3563 pr_debug("IFF_SLAVE\n");
3564 return bond_slave_netdev_event(event
, event_dev
);
3571 * bond_inetaddr_event: handle inetaddr notifier chain events.
3573 * We keep track of device IPs primarily to use as source addresses in
3574 * ARP monitor probes (rather than spewing out broadcasts all the time).
3576 * We track one IP for the main device (if it has one), plus one per VLAN.
3578 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3580 struct in_ifaddr
*ifa
= ptr
;
3581 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3582 struct bond_net
*bn
= net_generic(dev_net(event_dev
), bond_net_id
);
3583 struct bonding
*bond
;
3584 struct vlan_entry
*vlan
;
3586 list_for_each_entry(bond
, &bn
->dev_list
, bond_list
) {
3587 if (bond
->dev
== event_dev
) {
3590 bond
->master_ip
= ifa
->ifa_local
;
3593 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3600 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3603 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3604 if (vlan_dev
== event_dev
) {
3607 vlan
->vlan_ip
= ifa
->ifa_local
;
3611 bond_glean_dev_ip(vlan_dev
);
3622 static struct notifier_block bond_netdev_notifier
= {
3623 .notifier_call
= bond_netdev_event
,
3626 static struct notifier_block bond_inetaddr_notifier
= {
3627 .notifier_call
= bond_inetaddr_event
,
3630 /*-------------------------- Packet type handling ---------------------------*/
3632 /* register to receive lacpdus on a bond */
3633 static void bond_register_lacpdu(struct bonding
*bond
)
3635 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3637 /* initialize packet type */
3638 pk_type
->type
= PKT_TYPE_LACPDU
;
3639 pk_type
->dev
= bond
->dev
;
3640 pk_type
->func
= bond_3ad_lacpdu_recv
;
3642 dev_add_pack(pk_type
);
3645 /* unregister to receive lacpdus on a bond */
3646 static void bond_unregister_lacpdu(struct bonding
*bond
)
3648 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3651 void bond_register_arp(struct bonding
*bond
)
3653 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3658 pt
->type
= htons(ETH_P_ARP
);
3659 pt
->dev
= bond
->dev
;
3660 pt
->func
= bond_arp_rcv
;
3664 void bond_unregister_arp(struct bonding
*bond
)
3666 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3668 dev_remove_pack(pt
);
3672 /*---------------------------- Hashing Policies -----------------------------*/
3675 * Hash for the output device based upon layer 2 and layer 3 data. If
3676 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3678 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3680 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3681 struct iphdr
*iph
= ip_hdr(skb
);
3683 if (skb
->protocol
== htons(ETH_P_IP
)) {
3684 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3685 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3688 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3692 * Hash for the output device based upon layer 3 and layer 4 data. If
3693 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3694 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3696 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3698 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3699 struct iphdr
*iph
= ip_hdr(skb
);
3700 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3703 if (skb
->protocol
== htons(ETH_P_IP
)) {
3704 if (!(iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) &&
3705 (iph
->protocol
== IPPROTO_TCP
||
3706 iph
->protocol
== IPPROTO_UDP
)) {
3707 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3709 return (layer4_xor
^
3710 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3714 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3718 * Hash for the output device based upon layer 2 data
3720 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3722 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3724 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3727 /*-------------------------- Device entry points ----------------------------*/
3729 static int bond_open(struct net_device
*bond_dev
)
3731 struct bonding
*bond
= netdev_priv(bond_dev
);
3733 bond
->kill_timers
= 0;
3735 if (bond_is_lb(bond
)) {
3736 /* bond_alb_initialize must be called before the timer
3739 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3740 /* something went wrong - fail the open operation */
3744 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3745 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3748 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3749 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3750 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3753 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3754 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3755 INIT_DELAYED_WORK(&bond
->arp_work
,
3756 bond_activebackup_arp_mon
);
3758 INIT_DELAYED_WORK(&bond
->arp_work
,
3759 bond_loadbalance_arp_mon
);
3761 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3762 if (bond
->params
.arp_validate
)
3763 bond_register_arp(bond
);
3766 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3767 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3768 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3769 /* register to receive LACPDUs */
3770 bond_register_lacpdu(bond
);
3771 bond_3ad_initiate_agg_selection(bond
, 1);
3777 static int bond_close(struct net_device
*bond_dev
)
3779 struct bonding
*bond
= netdev_priv(bond_dev
);
3781 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3782 /* Unregister the receive of LACPDUs */
3783 bond_unregister_lacpdu(bond
);
3786 if (bond
->params
.arp_validate
)
3787 bond_unregister_arp(bond
);
3789 write_lock_bh(&bond
->lock
);
3791 bond
->send_grat_arp
= 0;
3792 bond
->send_unsol_na
= 0;
3794 /* signal timers not to re-arm */
3795 bond
->kill_timers
= 1;
3797 write_unlock_bh(&bond
->lock
);
3799 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3800 cancel_delayed_work(&bond
->mii_work
);
3803 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3804 cancel_delayed_work(&bond
->arp_work
);
3807 switch (bond
->params
.mode
) {
3808 case BOND_MODE_8023AD
:
3809 cancel_delayed_work(&bond
->ad_work
);
3813 cancel_delayed_work(&bond
->alb_work
);
3820 if (bond_is_lb(bond
)) {
3821 /* Must be called only after all
3822 * slaves have been released
3824 bond_alb_deinitialize(bond
);
3830 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3831 struct rtnl_link_stats64
*stats
)
3833 struct bonding
*bond
= netdev_priv(bond_dev
);
3834 struct rtnl_link_stats64 temp
;
3835 struct slave
*slave
;
3838 memset(stats
, 0, sizeof(*stats
));
3840 read_lock_bh(&bond
->lock
);
3842 bond_for_each_slave(bond
, slave
, i
) {
3843 const struct rtnl_link_stats64
*sstats
=
3844 dev_get_stats(slave
->dev
, &temp
);
3846 stats
->rx_packets
+= sstats
->rx_packets
;
3847 stats
->rx_bytes
+= sstats
->rx_bytes
;
3848 stats
->rx_errors
+= sstats
->rx_errors
;
3849 stats
->rx_dropped
+= sstats
->rx_dropped
;
3851 stats
->tx_packets
+= sstats
->tx_packets
;
3852 stats
->tx_bytes
+= sstats
->tx_bytes
;
3853 stats
->tx_errors
+= sstats
->tx_errors
;
3854 stats
->tx_dropped
+= sstats
->tx_dropped
;
3856 stats
->multicast
+= sstats
->multicast
;
3857 stats
->collisions
+= sstats
->collisions
;
3859 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3860 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3861 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3862 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3863 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3864 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3866 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3867 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3868 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3869 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3870 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3873 read_unlock_bh(&bond
->lock
);
3878 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3880 struct net_device
*slave_dev
= NULL
;
3881 struct ifbond k_binfo
;
3882 struct ifbond __user
*u_binfo
= NULL
;
3883 struct ifslave k_sinfo
;
3884 struct ifslave __user
*u_sinfo
= NULL
;
3885 struct mii_ioctl_data
*mii
= NULL
;
3888 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3900 * We do this again just in case we were called by SIOCGMIIREG
3901 * instead of SIOCGMIIPHY.
3908 if (mii
->reg_num
== 1) {
3909 struct bonding
*bond
= netdev_priv(bond_dev
);
3911 read_lock(&bond
->lock
);
3912 read_lock(&bond
->curr_slave_lock
);
3913 if (netif_carrier_ok(bond
->dev
))
3914 mii
->val_out
= BMSR_LSTATUS
;
3916 read_unlock(&bond
->curr_slave_lock
);
3917 read_unlock(&bond
->lock
);
3921 case BOND_INFO_QUERY_OLD
:
3922 case SIOCBONDINFOQUERY
:
3923 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3925 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3928 res
= bond_info_query(bond_dev
, &k_binfo
);
3930 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3934 case BOND_SLAVE_INFO_QUERY_OLD
:
3935 case SIOCBONDSLAVEINFOQUERY
:
3936 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3938 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3941 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3943 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3952 if (!capable(CAP_NET_ADMIN
))
3955 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
3957 pr_debug("slave_dev=%p:\n", slave_dev
);
3962 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3964 case BOND_ENSLAVE_OLD
:
3965 case SIOCBONDENSLAVE
:
3966 res
= bond_enslave(bond_dev
, slave_dev
);
3968 case BOND_RELEASE_OLD
:
3969 case SIOCBONDRELEASE
:
3970 res
= bond_release(bond_dev
, slave_dev
);
3972 case BOND_SETHWADDR_OLD
:
3973 case SIOCBONDSETHWADDR
:
3974 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3976 case BOND_CHANGE_ACTIVE_OLD
:
3977 case SIOCBONDCHANGEACTIVE
:
3978 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3990 static bool bond_addr_in_mc_list(unsigned char *addr
,
3991 struct netdev_hw_addr_list
*list
,
3994 struct netdev_hw_addr
*ha
;
3996 netdev_hw_addr_list_for_each(ha
, list
)
3997 if (!memcmp(ha
->addr
, addr
, addrlen
))
4003 static void bond_set_multicast_list(struct net_device
*bond_dev
)
4005 struct bonding
*bond
= netdev_priv(bond_dev
);
4006 struct netdev_hw_addr
*ha
;
4010 * Do promisc before checking multicast_mode
4012 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
))
4014 * FIXME: Need to handle the error when one of the multi-slaves
4017 bond_set_promiscuity(bond
, 1);
4020 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
))
4021 bond_set_promiscuity(bond
, -1);
4024 /* set allmulti flag to slaves */
4025 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
))
4027 * FIXME: Need to handle the error when one of the multi-slaves
4030 bond_set_allmulti(bond
, 1);
4033 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
))
4034 bond_set_allmulti(bond
, -1);
4037 read_lock(&bond
->lock
);
4039 bond
->flags
= bond_dev
->flags
;
4041 /* looking for addresses to add to slaves' mc list */
4042 netdev_for_each_mc_addr(ha
, bond_dev
) {
4043 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
4044 bond_dev
->addr_len
);
4046 bond_mc_add(bond
, ha
->addr
);
4049 /* looking for addresses to delete from slaves' list */
4050 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
4051 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
4052 bond_dev
->addr_len
);
4054 bond_mc_del(bond
, ha
->addr
);
4057 /* save master's multicast list */
4058 __hw_addr_flush(&bond
->mc_list
);
4059 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
4060 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
4062 read_unlock(&bond
->lock
);
4065 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
4067 struct bonding
*bond
= netdev_priv(dev
);
4068 struct slave
*slave
= bond
->first_slave
;
4071 const struct net_device_ops
*slave_ops
4072 = slave
->dev
->netdev_ops
;
4073 if (slave_ops
->ndo_neigh_setup
)
4074 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
4080 * Change the MTU of all of a master's slaves to match the master
4082 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
4084 struct bonding
*bond
= netdev_priv(bond_dev
);
4085 struct slave
*slave
, *stop_at
;
4089 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
4090 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
4092 /* Can't hold bond->lock with bh disabled here since
4093 * some base drivers panic. On the other hand we can't
4094 * hold bond->lock without bh disabled because we'll
4095 * deadlock. The only solution is to rely on the fact
4096 * that we're under rtnl_lock here, and the slaves
4097 * list won't change. This doesn't solve the problem
4098 * of setting the slave's MTU while it is
4099 * transmitting, but the assumption is that the base
4100 * driver can handle that.
4102 * TODO: figure out a way to safely iterate the slaves
4103 * list, but without holding a lock around the actual
4104 * call to the base driver.
4107 bond_for_each_slave(bond
, slave
, i
) {
4108 pr_debug("s %p s->p %p c_m %p\n",
4111 slave
->dev
->netdev_ops
->ndo_change_mtu
);
4113 res
= dev_set_mtu(slave
->dev
, new_mtu
);
4116 /* If we failed to set the slave's mtu to the new value
4117 * we must abort the operation even in ACTIVE_BACKUP
4118 * mode, because if we allow the backup slaves to have
4119 * different mtu values than the active slave we'll
4120 * need to change their mtu when doing a failover. That
4121 * means changing their mtu from timer context, which
4122 * is probably not a good idea.
4124 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4129 bond_dev
->mtu
= new_mtu
;
4134 /* unwind from head to the slave that failed */
4136 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4139 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
4141 pr_debug("unwind err %d dev %s\n",
4142 tmp_res
, slave
->dev
->name
);
4152 * Note that many devices must be down to change the HW address, and
4153 * downing the master releases all slaves. We can make bonds full of
4154 * bonding devices to test this, however.
4156 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4158 struct bonding
*bond
= netdev_priv(bond_dev
);
4159 struct sockaddr
*sa
= addr
, tmp_sa
;
4160 struct slave
*slave
, *stop_at
;
4164 if (bond
->params
.mode
== BOND_MODE_ALB
)
4165 return bond_alb_set_mac_address(bond_dev
, addr
);
4168 pr_debug("bond=%p, name=%s\n",
4169 bond
, bond_dev
? bond_dev
->name
: "None");
4172 * If fail_over_mac is set to active, do nothing and return
4173 * success. Returning an error causes ifenslave to fail.
4175 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4178 if (!is_valid_ether_addr(sa
->sa_data
))
4179 return -EADDRNOTAVAIL
;
4181 /* Can't hold bond->lock with bh disabled here since
4182 * some base drivers panic. On the other hand we can't
4183 * hold bond->lock without bh disabled because we'll
4184 * deadlock. The only solution is to rely on the fact
4185 * that we're under rtnl_lock here, and the slaves
4186 * list won't change. This doesn't solve the problem
4187 * of setting the slave's hw address while it is
4188 * transmitting, but the assumption is that the base
4189 * driver can handle that.
4191 * TODO: figure out a way to safely iterate the slaves
4192 * list, but without holding a lock around the actual
4193 * call to the base driver.
4196 bond_for_each_slave(bond
, slave
, i
) {
4197 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
4198 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
4200 if (slave_ops
->ndo_set_mac_address
== NULL
) {
4202 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
4206 res
= dev_set_mac_address(slave
->dev
, addr
);
4208 /* TODO: consider downing the slave
4210 * User should expect communications
4211 * breakage anyway until ARP finish
4214 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4220 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4224 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4225 tmp_sa
.sa_family
= bond_dev
->type
;
4227 /* unwind from head to the slave that failed */
4229 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4232 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4234 pr_debug("unwind err %d dev %s\n",
4235 tmp_res
, slave
->dev
->name
);
4242 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4244 struct bonding
*bond
= netdev_priv(bond_dev
);
4245 struct slave
*slave
, *start_at
;
4246 int i
, slave_no
, res
= 1;
4247 struct iphdr
*iph
= ip_hdr(skb
);
4249 read_lock(&bond
->lock
);
4251 if (!BOND_IS_OK(bond
))
4254 * Start with the curr_active_slave that joined the bond as the
4255 * default for sending IGMP traffic. For failover purposes one
4256 * needs to maintain some consistency for the interface that will
4257 * send the join/membership reports. The curr_active_slave found
4258 * will send all of this type of traffic.
4260 if ((iph
->protocol
== IPPROTO_IGMP
) &&
4261 (skb
->protocol
== htons(ETH_P_IP
))) {
4263 read_lock(&bond
->curr_slave_lock
);
4264 slave
= bond
->curr_active_slave
;
4265 read_unlock(&bond
->curr_slave_lock
);
4271 * Concurrent TX may collide on rr_tx_counter; we accept
4272 * that as being rare enough not to justify using an
4275 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4277 bond_for_each_slave(bond
, slave
, i
) {
4285 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4286 if (IS_UP(slave
->dev
) &&
4287 (slave
->link
== BOND_LINK_UP
) &&
4288 (slave
->state
== BOND_STATE_ACTIVE
)) {
4289 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4296 /* no suitable interface, frame not sent */
4299 read_unlock(&bond
->lock
);
4300 return NETDEV_TX_OK
;
4305 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4306 * the bond has a usable interface.
4308 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4310 struct bonding
*bond
= netdev_priv(bond_dev
);
4313 read_lock(&bond
->lock
);
4314 read_lock(&bond
->curr_slave_lock
);
4316 if (!BOND_IS_OK(bond
))
4319 if (!bond
->curr_active_slave
)
4322 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4326 /* no suitable interface, frame not sent */
4329 read_unlock(&bond
->curr_slave_lock
);
4330 read_unlock(&bond
->lock
);
4331 return NETDEV_TX_OK
;
4335 * In bond_xmit_xor() , we determine the output device by using a pre-
4336 * determined xmit_hash_policy(), If the selected device is not enabled,
4337 * find the next active slave.
4339 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4341 struct bonding
*bond
= netdev_priv(bond_dev
);
4342 struct slave
*slave
, *start_at
;
4347 read_lock(&bond
->lock
);
4349 if (!BOND_IS_OK(bond
))
4352 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4354 bond_for_each_slave(bond
, slave
, i
) {
4362 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4363 if (IS_UP(slave
->dev
) &&
4364 (slave
->link
== BOND_LINK_UP
) &&
4365 (slave
->state
== BOND_STATE_ACTIVE
)) {
4366 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4373 /* no suitable interface, frame not sent */
4376 read_unlock(&bond
->lock
);
4377 return NETDEV_TX_OK
;
4381 * in broadcast mode, we send everything to all usable interfaces.
4383 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4385 struct bonding
*bond
= netdev_priv(bond_dev
);
4386 struct slave
*slave
, *start_at
;
4387 struct net_device
*tx_dev
= NULL
;
4391 read_lock(&bond
->lock
);
4393 if (!BOND_IS_OK(bond
))
4396 read_lock(&bond
->curr_slave_lock
);
4397 start_at
= bond
->curr_active_slave
;
4398 read_unlock(&bond
->curr_slave_lock
);
4403 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4404 if (IS_UP(slave
->dev
) &&
4405 (slave
->link
== BOND_LINK_UP
) &&
4406 (slave
->state
== BOND_STATE_ACTIVE
)) {
4408 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4410 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4415 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4417 dev_kfree_skb(skb2
);
4421 tx_dev
= slave
->dev
;
4426 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4430 /* no suitable interface, frame not sent */
4433 /* frame sent to all suitable interfaces */
4434 read_unlock(&bond
->lock
);
4435 return NETDEV_TX_OK
;
4438 /*------------------------- Device initialization ---------------------------*/
4440 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4442 switch (bond
->params
.xmit_policy
) {
4443 case BOND_XMIT_POLICY_LAYER23
:
4444 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4446 case BOND_XMIT_POLICY_LAYER34
:
4447 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4449 case BOND_XMIT_POLICY_LAYER2
:
4451 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4457 * Lookup the slave that corresponds to a qid
4459 static inline int bond_slave_override(struct bonding
*bond
,
4460 struct sk_buff
*skb
)
4463 struct slave
*slave
= NULL
;
4464 struct slave
*check_slave
;
4466 read_lock(&bond
->lock
);
4468 if (!BOND_IS_OK(bond
) || !skb
->queue_mapping
)
4471 /* Find out if any slaves have the same mapping as this skb. */
4472 bond_for_each_slave(bond
, check_slave
, i
) {
4473 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4474 slave
= check_slave
;
4479 /* If the slave isn't UP, use default transmit policy. */
4480 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4481 (slave
->link
== BOND_LINK_UP
)) {
4482 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4486 read_unlock(&bond
->lock
);
4490 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4493 * This helper function exists to help dev_pick_tx get the correct
4494 * destination queue. Using a helper function skips the a call to
4495 * skb_tx_hash and will put the skbs in the queue we expect on their
4496 * way down to the bonding driver.
4498 return skb
->queue_mapping
;
4501 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4503 struct bonding
*bond
= netdev_priv(dev
);
4505 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4506 if (!bond_slave_override(bond
, skb
))
4507 return NETDEV_TX_OK
;
4510 switch (bond
->params
.mode
) {
4511 case BOND_MODE_ROUNDROBIN
:
4512 return bond_xmit_roundrobin(skb
, dev
);
4513 case BOND_MODE_ACTIVEBACKUP
:
4514 return bond_xmit_activebackup(skb
, dev
);
4516 return bond_xmit_xor(skb
, dev
);
4517 case BOND_MODE_BROADCAST
:
4518 return bond_xmit_broadcast(skb
, dev
);
4519 case BOND_MODE_8023AD
:
4520 return bond_3ad_xmit_xor(skb
, dev
);
4523 return bond_alb_xmit(skb
, dev
);
4525 /* Should never happen, mode already checked */
4526 pr_err("%s: Error: Unknown bonding mode %d\n",
4527 dev
->name
, bond
->params
.mode
);
4530 return NETDEV_TX_OK
;
4536 * set bond mode specific net device operations
4538 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4540 struct net_device
*bond_dev
= bond
->dev
;
4543 case BOND_MODE_ROUNDROBIN
:
4545 case BOND_MODE_ACTIVEBACKUP
:
4548 bond_set_xmit_hash_policy(bond
);
4550 case BOND_MODE_BROADCAST
:
4552 case BOND_MODE_8023AD
:
4553 bond_set_master_3ad_flags(bond
);
4554 bond_set_xmit_hash_policy(bond
);
4557 bond_set_master_alb_flags(bond
);
4562 /* Should never happen, mode already checked */
4563 pr_err("%s: Error: Unknown bonding mode %d\n",
4564 bond_dev
->name
, mode
);
4569 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4570 struct ethtool_drvinfo
*drvinfo
)
4572 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4573 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4574 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4577 static const struct ethtool_ops bond_ethtool_ops
= {
4578 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4579 .get_link
= ethtool_op_get_link
,
4580 .get_tx_csum
= ethtool_op_get_tx_csum
,
4581 .get_sg
= ethtool_op_get_sg
,
4582 .get_tso
= ethtool_op_get_tso
,
4583 .get_ufo
= ethtool_op_get_ufo
,
4584 .get_flags
= ethtool_op_get_flags
,
4587 static const struct net_device_ops bond_netdev_ops
= {
4588 .ndo_init
= bond_init
,
4589 .ndo_uninit
= bond_uninit
,
4590 .ndo_open
= bond_open
,
4591 .ndo_stop
= bond_close
,
4592 .ndo_start_xmit
= bond_start_xmit
,
4593 .ndo_select_queue
= bond_select_queue
,
4594 .ndo_get_stats64
= bond_get_stats
,
4595 .ndo_do_ioctl
= bond_do_ioctl
,
4596 .ndo_set_multicast_list
= bond_set_multicast_list
,
4597 .ndo_change_mtu
= bond_change_mtu
,
4598 .ndo_set_mac_address
= bond_set_mac_address
,
4599 .ndo_neigh_setup
= bond_neigh_setup
,
4600 .ndo_vlan_rx_register
= bond_vlan_rx_register
,
4601 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4602 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4603 #ifdef CONFIG_NET_POLL_CONTROLLER
4604 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4605 .ndo_poll_controller
= bond_poll_controller
,
4609 static void bond_destructor(struct net_device
*bond_dev
)
4611 struct bonding
*bond
= netdev_priv(bond_dev
);
4613 destroy_workqueue(bond
->wq
);
4614 free_netdev(bond_dev
);
4617 static void bond_setup(struct net_device
*bond_dev
)
4619 struct bonding
*bond
= netdev_priv(bond_dev
);
4621 /* initialize rwlocks */
4622 rwlock_init(&bond
->lock
);
4623 rwlock_init(&bond
->curr_slave_lock
);
4625 bond
->params
= bonding_defaults
;
4627 /* Initialize pointers */
4628 bond
->dev
= bond_dev
;
4629 INIT_LIST_HEAD(&bond
->vlan_list
);
4631 /* Initialize the device entry points */
4632 ether_setup(bond_dev
);
4633 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4634 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4635 bond_set_mode_ops(bond
, bond
->params
.mode
);
4637 bond_dev
->destructor
= bond_destructor
;
4639 /* Initialize the device options */
4640 bond_dev
->tx_queue_len
= 0;
4641 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4642 bond_dev
->priv_flags
|= IFF_BONDING
;
4643 bond_dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
4645 if (bond
->params
.arp_interval
)
4646 bond_dev
->priv_flags
|= IFF_MASTER_ARPMON
;
4648 /* At first, we block adding VLANs. That's the only way to
4649 * prevent problems that occur when adding VLANs over an
4650 * empty bond. The block will be removed once non-challenged
4651 * slaves are enslaved.
4653 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4655 /* don't acquire bond device's netif_tx_lock when
4657 bond_dev
->features
|= NETIF_F_LLTX
;
4659 /* By default, we declare the bond to be fully
4660 * VLAN hardware accelerated capable. Special
4661 * care is taken in the various xmit functions
4662 * when there are slaves that are not hw accel
4665 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4666 NETIF_F_HW_VLAN_RX
|
4667 NETIF_F_HW_VLAN_FILTER
);
4671 static void bond_work_cancel_all(struct bonding
*bond
)
4673 write_lock_bh(&bond
->lock
);
4674 bond
->kill_timers
= 1;
4675 write_unlock_bh(&bond
->lock
);
4677 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4678 cancel_delayed_work(&bond
->mii_work
);
4680 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4681 cancel_delayed_work(&bond
->arp_work
);
4683 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4684 delayed_work_pending(&bond
->alb_work
))
4685 cancel_delayed_work(&bond
->alb_work
);
4687 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4688 delayed_work_pending(&bond
->ad_work
))
4689 cancel_delayed_work(&bond
->ad_work
);
4693 * Destroy a bonding device.
4694 * Must be under rtnl_lock when this function is called.
4696 static void bond_uninit(struct net_device
*bond_dev
)
4698 struct bonding
*bond
= netdev_priv(bond_dev
);
4699 struct vlan_entry
*vlan
, *tmp
;
4701 bond_netpoll_cleanup(bond_dev
);
4703 /* Release the bonded slaves */
4704 bond_release_all(bond_dev
);
4706 list_del(&bond
->bond_list
);
4708 bond_work_cancel_all(bond
);
4710 bond_remove_proc_entry(bond
);
4712 __hw_addr_flush(&bond
->mc_list
);
4714 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4715 list_del(&vlan
->vlan_list
);
4720 /*------------------------- Module initialization ---------------------------*/
4723 * Convert string input module parms. Accept either the
4724 * number of the mode or its string name. A bit complicated because
4725 * some mode names are substrings of other names, and calls from sysfs
4726 * may have whitespace in the name (trailing newlines, for example).
4728 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4730 int modeint
= -1, i
, rv
;
4731 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4733 for (p
= (char *)buf
; *p
; p
++)
4734 if (!(isdigit(*p
) || isspace(*p
)))
4738 rv
= sscanf(buf
, "%20s", modestr
);
4740 rv
= sscanf(buf
, "%d", &modeint
);
4745 for (i
= 0; tbl
[i
].modename
; i
++) {
4746 if (modeint
== tbl
[i
].mode
)
4748 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4755 static int bond_check_params(struct bond_params
*params
)
4757 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4760 * Convert string parameters.
4763 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4764 if (bond_mode
== -1) {
4765 pr_err("Error: Invalid bonding mode \"%s\"\n",
4766 mode
== NULL
? "NULL" : mode
);
4771 if (xmit_hash_policy
) {
4772 if ((bond_mode
!= BOND_MODE_XOR
) &&
4773 (bond_mode
!= BOND_MODE_8023AD
)) {
4774 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4775 bond_mode_name(bond_mode
));
4777 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4779 if (xmit_hashtype
== -1) {
4780 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4781 xmit_hash_policy
== NULL
? "NULL" :
4789 if (bond_mode
!= BOND_MODE_8023AD
) {
4790 pr_info("lacp_rate param is irrelevant in mode %s\n",
4791 bond_mode_name(bond_mode
));
4793 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4794 if (lacp_fast
== -1) {
4795 pr_err("Error: Invalid lacp rate \"%s\"\n",
4796 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4803 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4804 if (params
->ad_select
== -1) {
4805 pr_err("Error: Invalid ad_select \"%s\"\n",
4806 ad_select
== NULL
? "NULL" : ad_select
);
4810 if (bond_mode
!= BOND_MODE_8023AD
) {
4811 pr_warning("ad_select param only affects 802.3ad mode\n");
4814 params
->ad_select
= BOND_AD_STABLE
;
4817 if (max_bonds
< 0) {
4818 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4819 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4820 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4824 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4825 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4826 miimon
= BOND_LINK_MON_INTERV
;
4830 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4835 if (downdelay
< 0) {
4836 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4837 downdelay
, INT_MAX
);
4841 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4842 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4847 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
4848 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4853 if (num_unsol_na
< 0 || num_unsol_na
> 255) {
4854 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4859 /* reset values for 802.3ad */
4860 if (bond_mode
== BOND_MODE_8023AD
) {
4862 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4863 pr_warning("Forcing miimon to 100msec\n");
4868 if (tx_queues
< 1 || tx_queues
> 255) {
4869 pr_warning("Warning: tx_queues (%d) should be between "
4870 "1 and 255, resetting to %d\n",
4871 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4872 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4875 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4876 pr_warning("Warning: all_slaves_active module parameter (%d), "
4877 "not of valid value (0/1), so it was set to "
4878 "0\n", all_slaves_active
);
4879 all_slaves_active
= 0;
4882 /* reset values for TLB/ALB */
4883 if ((bond_mode
== BOND_MODE_TLB
) ||
4884 (bond_mode
== BOND_MODE_ALB
)) {
4886 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4887 pr_warning("Forcing miimon to 100msec\n");
4892 if (bond_mode
== BOND_MODE_ALB
) {
4893 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4898 if (updelay
|| downdelay
) {
4899 /* just warn the user the up/down delay will have
4900 * no effect since miimon is zero...
4902 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4903 updelay
, downdelay
);
4906 /* don't allow arp monitoring */
4908 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4909 miimon
, arp_interval
);
4913 if ((updelay
% miimon
) != 0) {
4914 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4916 (updelay
/ miimon
) * miimon
);
4921 if ((downdelay
% miimon
) != 0) {
4922 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4924 (downdelay
/ miimon
) * miimon
);
4927 downdelay
/= miimon
;
4930 if (arp_interval
< 0) {
4931 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4932 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4933 arp_interval
= BOND_LINK_ARP_INTERV
;
4936 for (arp_ip_count
= 0;
4937 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4939 /* not complete check, but should be good enough to
4941 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4942 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4943 arp_ip_target
[arp_ip_count
]);
4946 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4947 arp_target
[arp_ip_count
] = ip
;
4951 if (arp_interval
&& !arp_ip_count
) {
4952 /* don't allow arping if no arp_ip_target given... */
4953 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4959 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4960 pr_err("arp_validate only supported in active-backup mode\n");
4963 if (!arp_interval
) {
4964 pr_err("arp_validate requires arp_interval\n");
4968 arp_validate_value
= bond_parse_parm(arp_validate
,
4970 if (arp_validate_value
== -1) {
4971 pr_err("Error: invalid arp_validate \"%s\"\n",
4972 arp_validate
== NULL
? "NULL" : arp_validate
);
4976 arp_validate_value
= 0;
4979 pr_info("MII link monitoring set to %d ms\n", miimon
);
4980 } else if (arp_interval
) {
4983 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4985 arp_validate_tbl
[arp_validate_value
].modename
,
4988 for (i
= 0; i
< arp_ip_count
; i
++)
4989 pr_info(" %s", arp_ip_target
[i
]);
4993 } else if (max_bonds
) {
4994 /* miimon and arp_interval not set, we need one so things
4995 * work as expected, see bonding.txt for details
4997 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
5000 if (primary
&& !USES_PRIMARY(bond_mode
)) {
5001 /* currently, using a primary only makes sense
5002 * in active backup, TLB or ALB modes
5004 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5005 primary
, bond_mode_name(bond_mode
));
5009 if (primary
&& primary_reselect
) {
5010 primary_reselect_value
= bond_parse_parm(primary_reselect
,
5012 if (primary_reselect_value
== -1) {
5013 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5015 NULL
? "NULL" : primary_reselect
);
5019 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
5022 if (fail_over_mac
) {
5023 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
5025 if (fail_over_mac_value
== -1) {
5026 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5027 arp_validate
== NULL
? "NULL" : arp_validate
);
5031 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
5032 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5034 fail_over_mac_value
= BOND_FOM_NONE
;
5037 /* fill params struct with the proper values */
5038 params
->mode
= bond_mode
;
5039 params
->xmit_policy
= xmit_hashtype
;
5040 params
->miimon
= miimon
;
5041 params
->num_grat_arp
= num_grat_arp
;
5042 params
->num_unsol_na
= num_unsol_na
;
5043 params
->arp_interval
= arp_interval
;
5044 params
->arp_validate
= arp_validate_value
;
5045 params
->updelay
= updelay
;
5046 params
->downdelay
= downdelay
;
5047 params
->use_carrier
= use_carrier
;
5048 params
->lacp_fast
= lacp_fast
;
5049 params
->primary
[0] = 0;
5050 params
->primary_reselect
= primary_reselect_value
;
5051 params
->fail_over_mac
= fail_over_mac_value
;
5052 params
->tx_queues
= tx_queues
;
5053 params
->all_slaves_active
= all_slaves_active
;
5056 strncpy(params
->primary
, primary
, IFNAMSIZ
);
5057 params
->primary
[IFNAMSIZ
- 1] = 0;
5060 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
5065 static struct lock_class_key bonding_netdev_xmit_lock_key
;
5066 static struct lock_class_key bonding_netdev_addr_lock_key
;
5068 static void bond_set_lockdep_class_one(struct net_device
*dev
,
5069 struct netdev_queue
*txq
,
5072 lockdep_set_class(&txq
->_xmit_lock
,
5073 &bonding_netdev_xmit_lock_key
);
5076 static void bond_set_lockdep_class(struct net_device
*dev
)
5078 lockdep_set_class(&dev
->addr_list_lock
,
5079 &bonding_netdev_addr_lock_key
);
5080 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
5084 * Called from registration process
5086 static int bond_init(struct net_device
*bond_dev
)
5088 struct bonding
*bond
= netdev_priv(bond_dev
);
5089 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
5091 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
5093 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
5097 bond_set_lockdep_class(bond_dev
);
5099 netif_carrier_off(bond_dev
);
5101 bond_create_proc_entry(bond
);
5102 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
5104 bond_prepare_sysfs_group(bond
);
5106 __hw_addr_init(&bond
->mc_list
);
5110 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
5112 if (tb
[IFLA_ADDRESS
]) {
5113 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
5115 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
5116 return -EADDRNOTAVAIL
;
5121 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
5123 .priv_size
= sizeof(struct bonding
),
5124 .setup
= bond_setup
,
5125 .validate
= bond_validate
,
5128 /* Create a new bond based on the specified name and bonding parameters.
5129 * If name is NULL, obtain a suitable "bond%d" name for us.
5130 * Caller must NOT hold rtnl_lock; we need to release it here before we
5131 * set up our sysfs entries.
5133 int bond_create(struct net
*net
, const char *name
)
5135 struct net_device
*bond_dev
;
5140 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
), name
? name
: "",
5141 bond_setup
, tx_queues
);
5143 pr_err("%s: eek! can't alloc netdev!\n", name
);
5148 dev_net_set(bond_dev
, net
);
5149 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
5152 res
= dev_alloc_name(bond_dev
, "bond%d");
5157 res
= register_netdevice(bond_dev
);
5162 bond_destructor(bond_dev
);
5166 static int __net_init
bond_net_init(struct net
*net
)
5168 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5171 INIT_LIST_HEAD(&bn
->dev_list
);
5173 bond_create_proc_dir(bn
);
5178 static void __net_exit
bond_net_exit(struct net
*net
)
5180 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5182 bond_destroy_proc_dir(bn
);
5185 static struct pernet_operations bond_net_ops
= {
5186 .init
= bond_net_init
,
5187 .exit
= bond_net_exit
,
5189 .size
= sizeof(struct bond_net
),
5192 static int __init
bonding_init(void)
5197 pr_info("%s", version
);
5199 res
= bond_check_params(&bonding_defaults
);
5203 res
= register_pernet_subsys(&bond_net_ops
);
5207 res
= rtnl_link_register(&bond_link_ops
);
5211 for (i
= 0; i
< max_bonds
; i
++) {
5212 res
= bond_create(&init_net
, NULL
);
5217 res
= bond_create_sysfs();
5221 register_netdevice_notifier(&bond_netdev_notifier
);
5222 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5223 bond_register_ipv6_notifier();
5227 rtnl_link_unregister(&bond_link_ops
);
5229 unregister_pernet_subsys(&bond_net_ops
);
5234 static void __exit
bonding_exit(void)
5236 unregister_netdevice_notifier(&bond_netdev_notifier
);
5237 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5238 bond_unregister_ipv6_notifier();
5240 bond_destroy_sysfs();
5242 rtnl_link_unregister(&bond_link_ops
);
5243 unregister_pernet_subsys(&bond_net_ops
);
5246 module_init(bonding_init
);
5247 module_exit(bonding_exit
);
5248 MODULE_LICENSE("GPL");
5249 MODULE_VERSION(DRV_VERSION
);
5250 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5251 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5252 MODULE_ALIAS_RTNL_LINK("bond");