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
;
576 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
577 slave_ops
->ndo_vlan_rx_register
)
578 slave_ops
->ndo_vlan_rx_register(slave_dev
, bond
->vlgrp
);
580 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
581 !(slave_ops
->ndo_vlan_rx_add_vid
))
584 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
)
585 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
588 static void bond_del_vlans_from_slave(struct bonding
*bond
,
589 struct net_device
*slave_dev
)
591 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
592 struct vlan_entry
*vlan
;
593 struct net_device
*vlan_dev
;
598 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
599 !(slave_ops
->ndo_vlan_rx_kill_vid
))
602 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
605 /* Save and then restore vlan_dev in the grp array,
606 * since the slave's driver might clear it.
608 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
609 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
610 vlan_group_set_device(bond
->vlgrp
, vlan
->vlan_id
, vlan_dev
);
614 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
615 slave_ops
->ndo_vlan_rx_register
)
616 slave_ops
->ndo_vlan_rx_register(slave_dev
, NULL
);
619 /*------------------------------- Link status -------------------------------*/
622 * Set the carrier state for the master according to the state of its
623 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
624 * do special 802.3ad magic.
626 * Returns zero if carrier state does not change, nonzero if it does.
628 static int bond_set_carrier(struct bonding
*bond
)
633 if (bond
->slave_cnt
== 0)
636 if (bond
->params
.mode
== BOND_MODE_8023AD
)
637 return bond_3ad_set_carrier(bond
);
639 bond_for_each_slave(bond
, slave
, i
) {
640 if (slave
->link
== BOND_LINK_UP
) {
641 if (!netif_carrier_ok(bond
->dev
)) {
642 netif_carrier_on(bond
->dev
);
650 if (netif_carrier_ok(bond
->dev
)) {
651 netif_carrier_off(bond
->dev
);
658 * Get link speed and duplex from the slave's base driver
659 * using ethtool. If for some reason the call fails or the
660 * values are invalid, fake speed and duplex to 100/Full
663 static int bond_update_speed_duplex(struct slave
*slave
)
665 struct net_device
*slave_dev
= slave
->dev
;
666 struct ethtool_cmd etool
;
669 /* Fake speed and duplex */
670 slave
->speed
= SPEED_100
;
671 slave
->duplex
= DUPLEX_FULL
;
673 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
676 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
680 switch (etool
.speed
) {
690 switch (etool
.duplex
) {
698 slave
->speed
= etool
.speed
;
699 slave
->duplex
= etool
.duplex
;
705 * if <dev> supports MII link status reporting, check its link status.
707 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
708 * depending upon the setting of the use_carrier parameter.
710 * Return either BMSR_LSTATUS, meaning that the link is up (or we
711 * can't tell and just pretend it is), or 0, meaning that the link is
714 * If reporting is non-zero, instead of faking link up, return -1 if
715 * both ETHTOOL and MII ioctls fail (meaning the device does not
716 * support them). If use_carrier is set, return whatever it says.
717 * It'd be nice if there was a good way to tell if a driver supports
718 * netif_carrier, but there really isn't.
720 static int bond_check_dev_link(struct bonding
*bond
,
721 struct net_device
*slave_dev
, int reporting
)
723 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
724 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
726 struct mii_ioctl_data
*mii
;
728 if (!reporting
&& !netif_running(slave_dev
))
731 if (bond
->params
.use_carrier
)
732 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
734 /* Try to get link status using Ethtool first. */
735 if (slave_dev
->ethtool_ops
) {
736 if (slave_dev
->ethtool_ops
->get_link
) {
739 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
741 return link
? BMSR_LSTATUS
: 0;
745 /* Ethtool can't be used, fallback to MII ioctls. */
746 ioctl
= slave_ops
->ndo_do_ioctl
;
748 /* TODO: set pointer to correct ioctl on a per team member */
749 /* bases to make this more efficient. that is, once */
750 /* we determine the correct ioctl, we will always */
751 /* call it and not the others for that team */
755 * We cannot assume that SIOCGMIIPHY will also read a
756 * register; not all network drivers (e.g., e100)
760 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
761 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
763 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
764 mii
->reg_num
= MII_BMSR
;
765 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
766 return mii
->val_out
& BMSR_LSTATUS
;
771 * If reporting, report that either there's no dev->do_ioctl,
772 * or both SIOCGMIIREG and get_link failed (meaning that we
773 * cannot report link status). If not reporting, pretend
776 return reporting
? -1 : BMSR_LSTATUS
;
779 /*----------------------------- Multicast list ------------------------------*/
782 * Push the promiscuity flag down to appropriate slaves
784 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
787 if (USES_PRIMARY(bond
->params
.mode
)) {
788 /* write lock already acquired */
789 if (bond
->curr_active_slave
) {
790 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
796 bond_for_each_slave(bond
, slave
, i
) {
797 err
= dev_set_promiscuity(slave
->dev
, inc
);
806 * Push the allmulti flag down to all slaves
808 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
811 if (USES_PRIMARY(bond
->params
.mode
)) {
812 /* write lock already acquired */
813 if (bond
->curr_active_slave
) {
814 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
820 bond_for_each_slave(bond
, slave
, i
) {
821 err
= dev_set_allmulti(slave
->dev
, inc
);
830 * Add a Multicast address to slaves
833 static void bond_mc_add(struct bonding
*bond
, void *addr
)
835 if (USES_PRIMARY(bond
->params
.mode
)) {
836 /* write lock already acquired */
837 if (bond
->curr_active_slave
)
838 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
843 bond_for_each_slave(bond
, slave
, i
)
844 dev_mc_add(slave
->dev
, addr
);
849 * Remove a multicast address from slave
852 static void bond_mc_del(struct bonding
*bond
, void *addr
)
854 if (USES_PRIMARY(bond
->params
.mode
)) {
855 /* write lock already acquired */
856 if (bond
->curr_active_slave
)
857 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
861 bond_for_each_slave(bond
, slave
, i
) {
862 dev_mc_del(slave
->dev
, addr
);
869 * Retrieve the list of registered multicast addresses for the bonding
870 * device and retransmit an IGMP JOIN request to the current active
873 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
875 struct in_device
*in_dev
;
876 struct ip_mc_list
*im
;
879 in_dev
= __in_dev_get_rcu(bond
->dev
);
881 for (im
= in_dev
->mc_list
; im
; im
= im
->next
)
882 ip_mc_rejoin_group(im
);
889 * flush all members of flush->mc_list from device dev->mc_list
891 static void bond_mc_list_flush(struct net_device
*bond_dev
,
892 struct net_device
*slave_dev
)
894 struct bonding
*bond
= netdev_priv(bond_dev
);
895 struct netdev_hw_addr
*ha
;
897 netdev_for_each_mc_addr(ha
, bond_dev
)
898 dev_mc_del(slave_dev
, ha
->addr
);
900 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
901 /* del lacpdu mc addr from mc list */
902 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
904 dev_mc_del(slave_dev
, lacpdu_multicast
);
908 /*--------------------------- Active slave change ---------------------------*/
911 * Update the mc list and multicast-related flags for the new and
912 * old active slaves (if any) according to the multicast mode, and
913 * promiscuous flags unconditionally.
915 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
916 struct slave
*old_active
)
918 struct netdev_hw_addr
*ha
;
920 if (!USES_PRIMARY(bond
->params
.mode
))
921 /* nothing to do - mc list is already up-to-date on
927 if (bond
->dev
->flags
& IFF_PROMISC
)
928 dev_set_promiscuity(old_active
->dev
, -1);
930 if (bond
->dev
->flags
& IFF_ALLMULTI
)
931 dev_set_allmulti(old_active
->dev
, -1);
933 netdev_for_each_mc_addr(ha
, bond
->dev
)
934 dev_mc_del(old_active
->dev
, ha
->addr
);
938 /* FIXME: Signal errors upstream. */
939 if (bond
->dev
->flags
& IFF_PROMISC
)
940 dev_set_promiscuity(new_active
->dev
, 1);
942 if (bond
->dev
->flags
& IFF_ALLMULTI
)
943 dev_set_allmulti(new_active
->dev
, 1);
945 netdev_for_each_mc_addr(ha
, bond
->dev
)
946 dev_mc_add(new_active
->dev
, ha
->addr
);
947 bond_resend_igmp_join_requests(bond
);
952 * bond_do_fail_over_mac
954 * Perform special MAC address swapping for fail_over_mac settings
956 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
958 static void bond_do_fail_over_mac(struct bonding
*bond
,
959 struct slave
*new_active
,
960 struct slave
*old_active
)
961 __releases(&bond
->curr_slave_lock
)
962 __releases(&bond
->lock
)
963 __acquires(&bond
->lock
)
964 __acquires(&bond
->curr_slave_lock
)
966 u8 tmp_mac
[ETH_ALEN
];
967 struct sockaddr saddr
;
970 switch (bond
->params
.fail_over_mac
) {
971 case BOND_FOM_ACTIVE
:
973 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
974 new_active
->dev
->addr_len
);
976 case BOND_FOM_FOLLOW
:
978 * if new_active && old_active, swap them
979 * if just old_active, do nothing (going to no active slave)
980 * if just new_active, set new_active to bond's MAC
985 write_unlock_bh(&bond
->curr_slave_lock
);
986 read_unlock(&bond
->lock
);
989 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
990 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
992 saddr
.sa_family
= new_active
->dev
->type
;
994 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
995 saddr
.sa_family
= bond
->dev
->type
;
998 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
1000 pr_err("%s: Error %d setting MAC of slave %s\n",
1001 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1008 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1009 saddr
.sa_family
= old_active
->dev
->type
;
1011 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
1013 pr_err("%s: Error %d setting MAC of slave %s\n",
1014 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1016 read_lock(&bond
->lock
);
1017 write_lock_bh(&bond
->curr_slave_lock
);
1020 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1021 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1027 static bool bond_should_change_active(struct bonding
*bond
)
1029 struct slave
*prim
= bond
->primary_slave
;
1030 struct slave
*curr
= bond
->curr_active_slave
;
1032 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
1034 if (bond
->force_primary
) {
1035 bond
->force_primary
= false;
1038 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
1039 (prim
->speed
< curr
->speed
||
1040 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
1042 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
1048 * find_best_interface - select the best available slave to be the active one
1049 * @bond: our bonding struct
1051 * Warning: Caller must hold curr_slave_lock for writing.
1053 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1055 struct slave
*new_active
, *old_active
;
1056 struct slave
*bestslave
= NULL
;
1057 int mintime
= bond
->params
.updelay
;
1060 new_active
= bond
->curr_active_slave
;
1062 if (!new_active
) { /* there were no active slaves left */
1063 if (bond
->slave_cnt
> 0) /* found one slave */
1064 new_active
= bond
->first_slave
;
1066 return NULL
; /* still no slave, return NULL */
1069 if ((bond
->primary_slave
) &&
1070 bond
->primary_slave
->link
== BOND_LINK_UP
&&
1071 bond_should_change_active(bond
)) {
1072 new_active
= bond
->primary_slave
;
1075 /* remember where to stop iterating over the slaves */
1076 old_active
= new_active
;
1078 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1079 if (new_active
->link
== BOND_LINK_UP
) {
1081 } else if (new_active
->link
== BOND_LINK_BACK
&&
1082 IS_UP(new_active
->dev
)) {
1083 /* link up, but waiting for stabilization */
1084 if (new_active
->delay
< mintime
) {
1085 mintime
= new_active
->delay
;
1086 bestslave
= new_active
;
1095 * change_active_interface - change the active slave into the specified one
1096 * @bond: our bonding struct
1097 * @new: the new slave to make the active one
1099 * Set the new slave to the bond's settings and unset them on the old
1100 * curr_active_slave.
1101 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1103 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1104 * because it is apparently the best available slave we have, even though its
1105 * updelay hasn't timed out yet.
1107 * If new_active is not NULL, caller must hold bond->lock for read and
1108 * curr_slave_lock for write_bh.
1110 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1112 struct slave
*old_active
= bond
->curr_active_slave
;
1114 if (old_active
== new_active
)
1118 new_active
->jiffies
= jiffies
;
1120 if (new_active
->link
== BOND_LINK_BACK
) {
1121 if (USES_PRIMARY(bond
->params
.mode
)) {
1122 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1123 bond
->dev
->name
, new_active
->dev
->name
,
1124 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1127 new_active
->delay
= 0;
1128 new_active
->link
= BOND_LINK_UP
;
1130 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1131 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1133 if (bond_is_lb(bond
))
1134 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1136 if (USES_PRIMARY(bond
->params
.mode
)) {
1137 pr_info("%s: making interface %s the new active one.\n",
1138 bond
->dev
->name
, new_active
->dev
->name
);
1143 if (USES_PRIMARY(bond
->params
.mode
))
1144 bond_mc_swap(bond
, new_active
, old_active
);
1146 if (bond_is_lb(bond
)) {
1147 bond_alb_handle_active_change(bond
, new_active
);
1149 bond_set_slave_inactive_flags(old_active
);
1151 bond_set_slave_active_flags(new_active
);
1153 bond
->curr_active_slave
= new_active
;
1156 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1158 bond_set_slave_inactive_flags(old_active
);
1161 bond_set_slave_active_flags(new_active
);
1163 if (bond
->params
.fail_over_mac
)
1164 bond_do_fail_over_mac(bond
, new_active
,
1167 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1168 bond_send_gratuitous_arp(bond
);
1170 bond
->send_unsol_na
= bond
->params
.num_unsol_na
;
1171 bond_send_unsolicited_na(bond
);
1173 write_unlock_bh(&bond
->curr_slave_lock
);
1174 read_unlock(&bond
->lock
);
1176 netdev_bonding_change(bond
->dev
, NETDEV_BONDING_FAILOVER
);
1178 read_lock(&bond
->lock
);
1179 write_lock_bh(&bond
->curr_slave_lock
);
1183 /* resend IGMP joins since all were sent on curr_active_slave */
1184 if (bond
->params
.mode
== BOND_MODE_ROUNDROBIN
) {
1185 bond_resend_igmp_join_requests(bond
);
1190 * bond_select_active_slave - select a new active slave, if needed
1191 * @bond: our bonding struct
1193 * This functions should be called when one of the following occurs:
1194 * - The old curr_active_slave has been released or lost its link.
1195 * - The primary_slave has got its link back.
1196 * - A slave has got its link back and there's no old curr_active_slave.
1198 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1200 void bond_select_active_slave(struct bonding
*bond
)
1202 struct slave
*best_slave
;
1205 best_slave
= bond_find_best_slave(bond
);
1206 if (best_slave
!= bond
->curr_active_slave
) {
1207 bond_change_active_slave(bond
, best_slave
);
1208 rv
= bond_set_carrier(bond
);
1212 if (netif_carrier_ok(bond
->dev
)) {
1213 pr_info("%s: first active interface up!\n",
1216 pr_info("%s: now running without any active interface !\n",
1222 /*--------------------------- slave list handling ---------------------------*/
1225 * This function attaches the slave to the end of list.
1227 * bond->lock held for writing by caller.
1229 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1231 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1232 new_slave
->next
= new_slave
;
1233 new_slave
->prev
= new_slave
;
1234 bond
->first_slave
= new_slave
;
1236 new_slave
->next
= bond
->first_slave
;
1237 new_slave
->prev
= bond
->first_slave
->prev
;
1238 new_slave
->next
->prev
= new_slave
;
1239 new_slave
->prev
->next
= new_slave
;
1246 * This function detaches the slave from the list.
1247 * WARNING: no check is made to verify if the slave effectively
1248 * belongs to <bond>.
1249 * Nothing is freed on return, structures are just unchained.
1250 * If any slave pointer in bond was pointing to <slave>,
1251 * it should be changed by the calling function.
1253 * bond->lock held for writing by caller.
1255 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1258 slave
->next
->prev
= slave
->prev
;
1261 slave
->prev
->next
= slave
->next
;
1263 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1264 if (bond
->slave_cnt
> 1) { /* there are more slave */
1265 bond
->first_slave
= slave
->next
;
1267 bond
->first_slave
= NULL
; /* slave was the last one */
1276 #ifdef CONFIG_NET_POLL_CONTROLLER
1278 * You must hold read lock on bond->lock before calling this.
1280 static bool slaves_support_netpoll(struct net_device
*bond_dev
)
1282 struct bonding
*bond
= netdev_priv(bond_dev
);
1283 struct slave
*slave
;
1287 bond_for_each_slave(bond
, slave
, i
) {
1288 if ((slave
->dev
->priv_flags
& IFF_DISABLE_NETPOLL
) ||
1289 !slave
->dev
->netdev_ops
->ndo_poll_controller
)
1292 return i
!= 0 && ret
;
1295 static void bond_poll_controller(struct net_device
*bond_dev
)
1297 struct net_device
*dev
= bond_dev
->npinfo
->netpoll
->real_dev
;
1298 if (dev
!= bond_dev
)
1299 netpoll_poll_dev(dev
);
1302 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1304 struct bonding
*bond
= netdev_priv(bond_dev
);
1305 struct slave
*slave
;
1306 const struct net_device_ops
*ops
;
1309 read_lock(&bond
->lock
);
1310 bond_dev
->npinfo
= NULL
;
1311 bond_for_each_slave(bond
, slave
, i
) {
1313 ops
= slave
->dev
->netdev_ops
;
1314 if (ops
->ndo_netpoll_cleanup
)
1315 ops
->ndo_netpoll_cleanup(slave
->dev
);
1317 slave
->dev
->npinfo
= NULL
;
1320 read_unlock(&bond
->lock
);
1325 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1331 /*---------------------------------- IOCTL ----------------------------------*/
1333 static int bond_sethwaddr(struct net_device
*bond_dev
,
1334 struct net_device
*slave_dev
)
1336 pr_debug("bond_dev=%p\n", bond_dev
);
1337 pr_debug("slave_dev=%p\n", slave_dev
);
1338 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1339 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1343 #define BOND_VLAN_FEATURES \
1344 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1345 NETIF_F_HW_VLAN_FILTER)
1348 * Compute the common dev->feature set available to all slaves. Some
1349 * feature bits are managed elsewhere, so preserve those feature bits
1350 * on the master device.
1352 static int bond_compute_features(struct bonding
*bond
)
1354 struct slave
*slave
;
1355 struct net_device
*bond_dev
= bond
->dev
;
1356 unsigned long features
= bond_dev
->features
;
1357 unsigned long vlan_features
= 0;
1358 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1359 bond_dev
->hard_header_len
);
1362 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1363 features
|= NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1365 if (!bond
->first_slave
)
1368 features
&= ~NETIF_F_ONE_FOR_ALL
;
1370 vlan_features
= bond
->first_slave
->dev
->vlan_features
;
1371 bond_for_each_slave(bond
, slave
, i
) {
1372 features
= netdev_increment_features(features
,
1373 slave
->dev
->features
,
1374 NETIF_F_ONE_FOR_ALL
);
1375 vlan_features
= netdev_increment_features(vlan_features
,
1376 slave
->dev
->vlan_features
,
1377 NETIF_F_ONE_FOR_ALL
);
1378 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1379 max_hard_header_len
= slave
->dev
->hard_header_len
;
1383 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1384 bond_dev
->features
= netdev_fix_features(features
, NULL
);
1385 bond_dev
->vlan_features
= netdev_fix_features(vlan_features
, NULL
);
1386 bond_dev
->hard_header_len
= max_hard_header_len
;
1391 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1392 struct net_device
*slave_dev
)
1394 struct bonding
*bond
= netdev_priv(bond_dev
);
1396 bond_dev
->header_ops
= slave_dev
->header_ops
;
1398 bond_dev
->type
= slave_dev
->type
;
1399 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1400 bond_dev
->addr_len
= slave_dev
->addr_len
;
1402 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1403 slave_dev
->addr_len
);
1404 bond
->setup_by_slave
= 1;
1407 /* enslave device <slave> to bond device <master> */
1408 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1410 struct bonding
*bond
= netdev_priv(bond_dev
);
1411 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1412 struct slave
*new_slave
= NULL
;
1413 struct netdev_hw_addr
*ha
;
1414 struct sockaddr addr
;
1416 int old_features
= bond_dev
->features
;
1419 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1420 slave_ops
->ndo_do_ioctl
== NULL
) {
1421 pr_warning("%s: Warning: no link monitoring support for %s\n",
1422 bond_dev
->name
, slave_dev
->name
);
1425 /* bond must be initialized by bond_open() before enslaving */
1426 if (!(bond_dev
->flags
& IFF_UP
)) {
1427 pr_warning("%s: master_dev is not up in bond_enslave\n",
1431 /* already enslaved */
1432 if (slave_dev
->flags
& IFF_SLAVE
) {
1433 pr_debug("Error, Device was already enslaved\n");
1437 /* vlan challenged mutual exclusion */
1438 /* no need to lock since we're protected by rtnl_lock */
1439 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1440 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1442 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1443 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1446 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1447 bond_dev
->name
, slave_dev
->name
,
1448 slave_dev
->name
, bond_dev
->name
);
1449 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1452 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1453 if (bond
->slave_cnt
== 0) {
1454 /* First slave, and it is not VLAN challenged,
1455 * so remove the block of adding VLANs over the bond.
1457 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1462 * Old ifenslave binaries are no longer supported. These can
1463 * be identified with moderate accuracy by the state of the slave:
1464 * the current ifenslave will set the interface down prior to
1465 * enslaving it; the old ifenslave will not.
1467 if ((slave_dev
->flags
& IFF_UP
)) {
1468 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1471 goto err_undo_flags
;
1474 /* set bonding device ether type by slave - bonding netdevices are
1475 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1476 * there is a need to override some of the type dependent attribs/funcs.
1478 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1479 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1481 if (bond
->slave_cnt
== 0) {
1482 if (bond_dev
->type
!= slave_dev
->type
) {
1483 pr_debug("%s: change device type from %d to %d\n",
1485 bond_dev
->type
, slave_dev
->type
);
1487 res
= netdev_bonding_change(bond_dev
,
1488 NETDEV_PRE_TYPE_CHANGE
);
1489 res
= notifier_to_errno(res
);
1491 pr_err("%s: refused to change device type\n",
1494 goto err_undo_flags
;
1497 /* Flush unicast and multicast addresses */
1498 dev_uc_flush(bond_dev
);
1499 dev_mc_flush(bond_dev
);
1501 if (slave_dev
->type
!= ARPHRD_ETHER
)
1502 bond_setup_by_slave(bond_dev
, slave_dev
);
1504 ether_setup(bond_dev
);
1506 netdev_bonding_change(bond_dev
,
1507 NETDEV_POST_TYPE_CHANGE
);
1509 } else if (bond_dev
->type
!= slave_dev
->type
) {
1510 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1512 slave_dev
->type
, bond_dev
->type
);
1514 goto err_undo_flags
;
1517 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1518 if (bond
->slave_cnt
== 0) {
1519 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1521 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1522 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1523 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",
1526 goto err_undo_flags
;
1530 /* If this is the first slave, then we need to set the master's hardware
1531 * address to be the same as the slave's. */
1532 if (bond
->slave_cnt
== 0)
1533 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1534 slave_dev
->addr_len
);
1537 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1540 goto err_undo_flags
;
1544 * Set the new_slave's queue_id to be zero. Queue ID mapping
1545 * is set via sysfs or module option if desired.
1547 new_slave
->queue_id
= 0;
1549 /* Save slave's original mtu and then set it to match the bond */
1550 new_slave
->original_mtu
= slave_dev
->mtu
;
1551 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1553 pr_debug("Error %d calling dev_set_mtu\n", res
);
1558 * Save slave's original ("permanent") mac address for modes
1559 * that need it, and for restoring it upon release, and then
1560 * set it to the master's address
1562 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1564 if (!bond
->params
.fail_over_mac
) {
1566 * Set slave to master's mac address. The application already
1567 * set the master's mac address to that of the first slave
1569 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1570 addr
.sa_family
= slave_dev
->type
;
1571 res
= dev_set_mac_address(slave_dev
, &addr
);
1573 pr_debug("Error %d calling set_mac_address\n", res
);
1574 goto err_restore_mtu
;
1578 res
= netdev_set_master(slave_dev
, bond_dev
);
1580 pr_debug("Error %d calling netdev_set_master\n", res
);
1581 goto err_restore_mac
;
1583 /* open the slave since the application closed it */
1584 res
= dev_open(slave_dev
);
1586 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1587 goto err_unset_master
;
1590 new_slave
->dev
= slave_dev
;
1591 slave_dev
->priv_flags
|= IFF_BONDING
;
1593 if (bond_is_lb(bond
)) {
1594 /* bond_alb_init_slave() must be called before all other stages since
1595 * it might fail and we do not want to have to undo everything
1597 res
= bond_alb_init_slave(bond
, new_slave
);
1602 /* If the mode USES_PRIMARY, then the new slave gets the
1603 * master's promisc (and mc) settings only if it becomes the
1604 * curr_active_slave, and that is taken care of later when calling
1605 * bond_change_active()
1607 if (!USES_PRIMARY(bond
->params
.mode
)) {
1608 /* set promiscuity level to new slave */
1609 if (bond_dev
->flags
& IFF_PROMISC
) {
1610 res
= dev_set_promiscuity(slave_dev
, 1);
1615 /* set allmulti level to new slave */
1616 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1617 res
= dev_set_allmulti(slave_dev
, 1);
1622 netif_addr_lock_bh(bond_dev
);
1623 /* upload master's mc_list to new slave */
1624 netdev_for_each_mc_addr(ha
, bond_dev
)
1625 dev_mc_add(slave_dev
, ha
->addr
);
1626 netif_addr_unlock_bh(bond_dev
);
1629 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1630 /* add lacpdu mc addr to mc list */
1631 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1633 dev_mc_add(slave_dev
, lacpdu_multicast
);
1636 bond_add_vlans_on_slave(bond
, slave_dev
);
1638 write_lock_bh(&bond
->lock
);
1640 bond_attach_slave(bond
, new_slave
);
1642 new_slave
->delay
= 0;
1643 new_slave
->link_failure_count
= 0;
1645 bond_compute_features(bond
);
1647 write_unlock_bh(&bond
->lock
);
1649 read_lock(&bond
->lock
);
1651 new_slave
->last_arp_rx
= jiffies
;
1653 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1654 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1656 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1658 * miimon is set but a bonded network driver
1659 * does not support ETHTOOL/MII and
1660 * arp_interval is not set. Note: if
1661 * use_carrier is enabled, we will never go
1662 * here (because netif_carrier is always
1663 * supported); thus, we don't need to change
1664 * the messages for netif_carrier.
1666 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",
1667 bond_dev
->name
, slave_dev
->name
);
1668 } else if (link_reporting
== -1) {
1669 /* unable get link status using mii/ethtool */
1670 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",
1671 bond_dev
->name
, slave_dev
->name
);
1675 /* check for initial state */
1676 if (!bond
->params
.miimon
||
1677 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1678 if (bond
->params
.updelay
) {
1679 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1680 new_slave
->link
= BOND_LINK_BACK
;
1681 new_slave
->delay
= bond
->params
.updelay
;
1683 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1684 new_slave
->link
= BOND_LINK_UP
;
1686 new_slave
->jiffies
= jiffies
;
1688 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1689 new_slave
->link
= BOND_LINK_DOWN
;
1692 if (bond_update_speed_duplex(new_slave
) &&
1693 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1694 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1695 bond_dev
->name
, new_slave
->dev
->name
);
1697 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1698 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1703 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1704 /* if there is a primary slave, remember it */
1705 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1706 bond
->primary_slave
= new_slave
;
1707 bond
->force_primary
= true;
1711 write_lock_bh(&bond
->curr_slave_lock
);
1713 switch (bond
->params
.mode
) {
1714 case BOND_MODE_ACTIVEBACKUP
:
1715 bond_set_slave_inactive_flags(new_slave
);
1716 bond_select_active_slave(bond
);
1718 case BOND_MODE_8023AD
:
1719 /* in 802.3ad mode, the internal mechanism
1720 * will activate the slaves in the selected
1723 bond_set_slave_inactive_flags(new_slave
);
1724 /* if this is the first slave */
1725 if (bond
->slave_cnt
== 1) {
1726 SLAVE_AD_INFO(new_slave
).id
= 1;
1727 /* Initialize AD with the number of times that the AD timer is called in 1 second
1728 * can be called only after the mac address of the bond is set
1730 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1731 bond
->params
.lacp_fast
);
1733 SLAVE_AD_INFO(new_slave
).id
=
1734 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1737 bond_3ad_bind_slave(new_slave
);
1741 new_slave
->state
= BOND_STATE_ACTIVE
;
1742 bond_set_slave_inactive_flags(new_slave
);
1743 bond_select_active_slave(bond
);
1746 pr_debug("This slave is always active in trunk mode\n");
1748 /* always active in trunk mode */
1749 new_slave
->state
= BOND_STATE_ACTIVE
;
1751 /* In trunking mode there is little meaning to curr_active_slave
1752 * anyway (it holds no special properties of the bond device),
1753 * so we can change it without calling change_active_interface()
1755 if (!bond
->curr_active_slave
)
1756 bond
->curr_active_slave
= new_slave
;
1759 } /* switch(bond_mode) */
1761 write_unlock_bh(&bond
->curr_slave_lock
);
1763 bond_set_carrier(bond
);
1765 #ifdef CONFIG_NET_POLL_CONTROLLER
1767 * Netpoll and bonding is broken, make sure it is not initialized
1768 * until it is fixed.
1770 if (disable_netpoll
) {
1771 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1773 if (slaves_support_netpoll(bond_dev
)) {
1774 bond_dev
->priv_flags
&= ~IFF_DISABLE_NETPOLL
;
1775 if (bond_dev
->npinfo
)
1776 slave_dev
->npinfo
= bond_dev
->npinfo
;
1777 } else if (!(bond_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)) {
1778 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1779 pr_info("New slave device %s does not support netpoll\n",
1781 pr_info("Disabling netpoll support for %s\n", bond_dev
->name
);
1785 read_unlock(&bond
->lock
);
1787 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1791 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1792 bond_dev
->name
, slave_dev
->name
,
1793 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1794 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1796 /* enslave is successful */
1799 /* Undo stages on error */
1801 dev_close(slave_dev
);
1804 netdev_set_master(slave_dev
, NULL
);
1807 if (!bond
->params
.fail_over_mac
) {
1808 /* XXX TODO - fom follow mode needs to change master's
1809 * MAC if this slave's MAC is in use by the bond, or at
1810 * least print a warning.
1812 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1813 addr
.sa_family
= slave_dev
->type
;
1814 dev_set_mac_address(slave_dev
, &addr
);
1818 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1824 bond_dev
->features
= old_features
;
1830 * Try to release the slave device <slave> from the bond device <master>
1831 * It is legal to access curr_active_slave without a lock because all the function
1834 * The rules for slave state should be:
1835 * for Active/Backup:
1836 * Active stays on all backups go down
1837 * for Bonded connections:
1838 * The first up interface should be left on and all others downed.
1840 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1842 struct bonding
*bond
= netdev_priv(bond_dev
);
1843 struct slave
*slave
, *oldcurrent
;
1844 struct sockaddr addr
;
1846 /* slave is not a slave or master is not master of this slave */
1847 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1848 (slave_dev
->master
!= bond_dev
)) {
1849 pr_err("%s: Error: cannot release %s.\n",
1850 bond_dev
->name
, slave_dev
->name
);
1854 netdev_bonding_change(bond_dev
, NETDEV_BONDING_DESLAVE
);
1855 write_lock_bh(&bond
->lock
);
1857 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1859 /* not a slave of this bond */
1860 pr_info("%s: %s not enslaved\n",
1861 bond_dev
->name
, slave_dev
->name
);
1862 write_unlock_bh(&bond
->lock
);
1866 if (!bond
->params
.fail_over_mac
) {
1867 if (!compare_ether_addr(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1868 bond
->slave_cnt
> 1)
1869 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",
1870 bond_dev
->name
, slave_dev
->name
,
1872 bond_dev
->name
, slave_dev
->name
);
1875 /* Inform AD package of unbinding of slave. */
1876 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1877 /* must be called before the slave is
1878 * detached from the list
1880 bond_3ad_unbind_slave(slave
);
1883 pr_info("%s: releasing %s interface %s\n",
1885 (slave
->state
== BOND_STATE_ACTIVE
) ? "active" : "backup",
1888 oldcurrent
= bond
->curr_active_slave
;
1890 bond
->current_arp_slave
= NULL
;
1892 /* release the slave from its bond */
1893 bond_detach_slave(bond
, slave
);
1895 bond_compute_features(bond
);
1897 if (bond
->primary_slave
== slave
)
1898 bond
->primary_slave
= NULL
;
1900 if (oldcurrent
== slave
)
1901 bond_change_active_slave(bond
, NULL
);
1903 if (bond_is_lb(bond
)) {
1904 /* Must be called only after the slave has been
1905 * detached from the list and the curr_active_slave
1906 * has been cleared (if our_slave == old_current),
1907 * but before a new active slave is selected.
1909 write_unlock_bh(&bond
->lock
);
1910 bond_alb_deinit_slave(bond
, slave
);
1911 write_lock_bh(&bond
->lock
);
1914 if (oldcurrent
== slave
) {
1916 * Note that we hold RTNL over this sequence, so there
1917 * is no concern that another slave add/remove event
1920 write_unlock_bh(&bond
->lock
);
1921 read_lock(&bond
->lock
);
1922 write_lock_bh(&bond
->curr_slave_lock
);
1924 bond_select_active_slave(bond
);
1926 write_unlock_bh(&bond
->curr_slave_lock
);
1927 read_unlock(&bond
->lock
);
1928 write_lock_bh(&bond
->lock
);
1931 if (bond
->slave_cnt
== 0) {
1932 bond_set_carrier(bond
);
1934 /* if the last slave was removed, zero the mac address
1935 * of the master so it will be set by the application
1936 * to the mac address of the first slave
1938 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1941 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1943 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1944 bond_dev
->name
, bond_dev
->name
);
1945 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1948 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1949 !bond_has_challenged_slaves(bond
)) {
1950 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1951 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1952 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1955 write_unlock_bh(&bond
->lock
);
1957 /* must do this from outside any spinlocks */
1958 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1960 bond_del_vlans_from_slave(bond
, slave_dev
);
1962 /* If the mode USES_PRIMARY, then we should only remove its
1963 * promisc and mc settings if it was the curr_active_slave, but that was
1964 * already taken care of above when we detached the slave
1966 if (!USES_PRIMARY(bond
->params
.mode
)) {
1967 /* unset promiscuity level from slave */
1968 if (bond_dev
->flags
& IFF_PROMISC
)
1969 dev_set_promiscuity(slave_dev
, -1);
1971 /* unset allmulti level from slave */
1972 if (bond_dev
->flags
& IFF_ALLMULTI
)
1973 dev_set_allmulti(slave_dev
, -1);
1975 /* flush master's mc_list from slave */
1976 netif_addr_lock_bh(bond_dev
);
1977 bond_mc_list_flush(bond_dev
, slave_dev
);
1978 netif_addr_unlock_bh(bond_dev
);
1981 netdev_set_master(slave_dev
, NULL
);
1983 #ifdef CONFIG_NET_POLL_CONTROLLER
1984 read_lock_bh(&bond
->lock
);
1986 /* Make sure netpoll over stays disabled until fixed. */
1987 if (!disable_netpoll
)
1988 if (slaves_support_netpoll(bond_dev
))
1989 bond_dev
->priv_flags
&= ~IFF_DISABLE_NETPOLL
;
1990 read_unlock_bh(&bond
->lock
);
1991 if (slave_dev
->netdev_ops
->ndo_netpoll_cleanup
)
1992 slave_dev
->netdev_ops
->ndo_netpoll_cleanup(slave_dev
);
1994 slave_dev
->npinfo
= NULL
;
1997 /* close slave before restoring its mac address */
1998 dev_close(slave_dev
);
2000 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2001 /* restore original ("permanent") mac address */
2002 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2003 addr
.sa_family
= slave_dev
->type
;
2004 dev_set_mac_address(slave_dev
, &addr
);
2007 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2009 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2010 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
2015 return 0; /* deletion OK */
2019 * First release a slave and than destroy the bond if no more slaves are left.
2020 * Must be under rtnl_lock when this function is called.
2022 int bond_release_and_destroy(struct net_device
*bond_dev
,
2023 struct net_device
*slave_dev
)
2025 struct bonding
*bond
= netdev_priv(bond_dev
);
2028 ret
= bond_release(bond_dev
, slave_dev
);
2029 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2030 pr_info("%s: destroying bond %s.\n",
2031 bond_dev
->name
, bond_dev
->name
);
2032 unregister_netdevice(bond_dev
);
2038 * This function releases all slaves.
2040 static int bond_release_all(struct net_device
*bond_dev
)
2042 struct bonding
*bond
= netdev_priv(bond_dev
);
2043 struct slave
*slave
;
2044 struct net_device
*slave_dev
;
2045 struct sockaddr addr
;
2047 write_lock_bh(&bond
->lock
);
2049 netif_carrier_off(bond_dev
);
2051 if (bond
->slave_cnt
== 0)
2054 bond
->current_arp_slave
= NULL
;
2055 bond
->primary_slave
= NULL
;
2056 bond_change_active_slave(bond
, NULL
);
2058 while ((slave
= bond
->first_slave
) != NULL
) {
2059 /* Inform AD package of unbinding of slave
2060 * before slave is detached from the list.
2062 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2063 bond_3ad_unbind_slave(slave
);
2065 slave_dev
= slave
->dev
;
2066 bond_detach_slave(bond
, slave
);
2068 /* now that the slave is detached, unlock and perform
2069 * all the undo steps that should not be called from
2072 write_unlock_bh(&bond
->lock
);
2074 if (bond_is_lb(bond
)) {
2075 /* must be called only after the slave
2076 * has been detached from the list
2078 bond_alb_deinit_slave(bond
, slave
);
2081 bond_compute_features(bond
);
2083 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2084 bond_del_vlans_from_slave(bond
, slave_dev
);
2086 /* If the mode USES_PRIMARY, then we should only remove its
2087 * promisc and mc settings if it was the curr_active_slave, but that was
2088 * already taken care of above when we detached the slave
2090 if (!USES_PRIMARY(bond
->params
.mode
)) {
2091 /* unset promiscuity level from slave */
2092 if (bond_dev
->flags
& IFF_PROMISC
)
2093 dev_set_promiscuity(slave_dev
, -1);
2095 /* unset allmulti level from slave */
2096 if (bond_dev
->flags
& IFF_ALLMULTI
)
2097 dev_set_allmulti(slave_dev
, -1);
2099 /* flush master's mc_list from slave */
2100 netif_addr_lock_bh(bond_dev
);
2101 bond_mc_list_flush(bond_dev
, slave_dev
);
2102 netif_addr_unlock_bh(bond_dev
);
2105 netdev_set_master(slave_dev
, NULL
);
2107 /* close slave before restoring its mac address */
2108 dev_close(slave_dev
);
2110 if (!bond
->params
.fail_over_mac
) {
2111 /* restore original ("permanent") mac address*/
2112 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2113 addr
.sa_family
= slave_dev
->type
;
2114 dev_set_mac_address(slave_dev
, &addr
);
2117 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2118 IFF_SLAVE_INACTIVE
);
2122 /* re-acquire the lock before getting the next slave */
2123 write_lock_bh(&bond
->lock
);
2126 /* zero the mac address of the master so it will be
2127 * set by the application to the mac address of the
2130 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2133 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2135 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2136 bond_dev
->name
, bond_dev
->name
);
2137 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2141 pr_info("%s: released all slaves\n", bond_dev
->name
);
2144 write_unlock_bh(&bond
->lock
);
2150 * This function changes the active slave to slave <slave_dev>.
2151 * It returns -EINVAL in the following cases.
2152 * - <slave_dev> is not found in the list.
2153 * - There is not active slave now.
2154 * - <slave_dev> is already active.
2155 * - The link state of <slave_dev> is not BOND_LINK_UP.
2156 * - <slave_dev> is not running.
2157 * In these cases, this function does nothing.
2158 * In the other cases, current_slave pointer is changed and 0 is returned.
2160 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2162 struct bonding
*bond
= netdev_priv(bond_dev
);
2163 struct slave
*old_active
= NULL
;
2164 struct slave
*new_active
= NULL
;
2167 if (!USES_PRIMARY(bond
->params
.mode
))
2170 /* Verify that master_dev is indeed the master of slave_dev */
2171 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2174 read_lock(&bond
->lock
);
2176 read_lock(&bond
->curr_slave_lock
);
2177 old_active
= bond
->curr_active_slave
;
2178 read_unlock(&bond
->curr_slave_lock
);
2180 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2183 * Changing to the current active: do nothing; return success.
2185 if (new_active
&& (new_active
== old_active
)) {
2186 read_unlock(&bond
->lock
);
2192 (new_active
->link
== BOND_LINK_UP
) &&
2193 IS_UP(new_active
->dev
)) {
2194 write_lock_bh(&bond
->curr_slave_lock
);
2195 bond_change_active_slave(bond
, new_active
);
2196 write_unlock_bh(&bond
->curr_slave_lock
);
2200 read_unlock(&bond
->lock
);
2205 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2207 struct bonding
*bond
= netdev_priv(bond_dev
);
2209 info
->bond_mode
= bond
->params
.mode
;
2210 info
->miimon
= bond
->params
.miimon
;
2212 read_lock(&bond
->lock
);
2213 info
->num_slaves
= bond
->slave_cnt
;
2214 read_unlock(&bond
->lock
);
2219 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2221 struct bonding
*bond
= netdev_priv(bond_dev
);
2222 struct slave
*slave
;
2223 int i
, res
= -ENODEV
;
2225 read_lock(&bond
->lock
);
2227 bond_for_each_slave(bond
, slave
, i
) {
2228 if (i
== (int)info
->slave_id
) {
2230 strcpy(info
->slave_name
, slave
->dev
->name
);
2231 info
->link
= slave
->link
;
2232 info
->state
= slave
->state
;
2233 info
->link_failure_count
= slave
->link_failure_count
;
2238 read_unlock(&bond
->lock
);
2243 /*-------------------------------- Monitoring -------------------------------*/
2246 static int bond_miimon_inspect(struct bonding
*bond
)
2248 struct slave
*slave
;
2249 int i
, link_state
, commit
= 0;
2250 bool ignore_updelay
;
2252 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2254 bond_for_each_slave(bond
, slave
, i
) {
2255 slave
->new_link
= BOND_LINK_NOCHANGE
;
2257 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2259 switch (slave
->link
) {
2264 slave
->link
= BOND_LINK_FAIL
;
2265 slave
->delay
= bond
->params
.downdelay
;
2267 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2269 (bond
->params
.mode
==
2270 BOND_MODE_ACTIVEBACKUP
) ?
2271 ((slave
->state
== BOND_STATE_ACTIVE
) ?
2272 "active " : "backup ") : "",
2274 bond
->params
.downdelay
* bond
->params
.miimon
);
2277 case BOND_LINK_FAIL
:
2280 * recovered before downdelay expired
2282 slave
->link
= BOND_LINK_UP
;
2283 slave
->jiffies
= jiffies
;
2284 pr_info("%s: link status up again after %d ms for interface %s.\n",
2286 (bond
->params
.downdelay
- slave
->delay
) *
2287 bond
->params
.miimon
,
2292 if (slave
->delay
<= 0) {
2293 slave
->new_link
= BOND_LINK_DOWN
;
2301 case BOND_LINK_DOWN
:
2305 slave
->link
= BOND_LINK_BACK
;
2306 slave
->delay
= bond
->params
.updelay
;
2309 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2310 bond
->dev
->name
, slave
->dev
->name
,
2311 ignore_updelay
? 0 :
2312 bond
->params
.updelay
*
2313 bond
->params
.miimon
);
2316 case BOND_LINK_BACK
:
2318 slave
->link
= BOND_LINK_DOWN
;
2319 pr_info("%s: link status down again after %d ms for interface %s.\n",
2321 (bond
->params
.updelay
- slave
->delay
) *
2322 bond
->params
.miimon
,
2331 if (slave
->delay
<= 0) {
2332 slave
->new_link
= BOND_LINK_UP
;
2334 ignore_updelay
= false;
2346 static void bond_miimon_commit(struct bonding
*bond
)
2348 struct slave
*slave
;
2351 bond_for_each_slave(bond
, slave
, i
) {
2352 switch (slave
->new_link
) {
2353 case BOND_LINK_NOCHANGE
:
2357 slave
->link
= BOND_LINK_UP
;
2358 slave
->jiffies
= jiffies
;
2360 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2361 /* prevent it from being the active one */
2362 slave
->state
= BOND_STATE_BACKUP
;
2363 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2364 /* make it immediately active */
2365 slave
->state
= BOND_STATE_ACTIVE
;
2366 } else if (slave
!= bond
->primary_slave
) {
2367 /* prevent it from being the active one */
2368 slave
->state
= BOND_STATE_BACKUP
;
2371 pr_info("%s: link status definitely up for interface %s.\n",
2372 bond
->dev
->name
, slave
->dev
->name
);
2374 /* notify ad that the link status has changed */
2375 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2376 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2378 if (bond_is_lb(bond
))
2379 bond_alb_handle_link_change(bond
, slave
,
2382 if (!bond
->curr_active_slave
||
2383 (slave
== bond
->primary_slave
))
2388 case BOND_LINK_DOWN
:
2389 if (slave
->link_failure_count
< UINT_MAX
)
2390 slave
->link_failure_count
++;
2392 slave
->link
= BOND_LINK_DOWN
;
2394 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2395 bond
->params
.mode
== BOND_MODE_8023AD
)
2396 bond_set_slave_inactive_flags(slave
);
2398 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2399 bond
->dev
->name
, slave
->dev
->name
);
2401 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2402 bond_3ad_handle_link_change(slave
,
2405 if (bond_is_lb(bond
))
2406 bond_alb_handle_link_change(bond
, slave
,
2409 if (slave
== bond
->curr_active_slave
)
2415 pr_err("%s: invalid new link %d on slave %s\n",
2416 bond
->dev
->name
, slave
->new_link
,
2418 slave
->new_link
= BOND_LINK_NOCHANGE
;
2425 write_lock_bh(&bond
->curr_slave_lock
);
2426 bond_select_active_slave(bond
);
2427 write_unlock_bh(&bond
->curr_slave_lock
);
2430 bond_set_carrier(bond
);
2436 * Really a wrapper that splits the mii monitor into two phases: an
2437 * inspection, then (if inspection indicates something needs to be done)
2438 * an acquisition of appropriate locks followed by a commit phase to
2439 * implement whatever link state changes are indicated.
2441 void bond_mii_monitor(struct work_struct
*work
)
2443 struct bonding
*bond
= container_of(work
, struct bonding
,
2446 read_lock(&bond
->lock
);
2447 if (bond
->kill_timers
)
2450 if (bond
->slave_cnt
== 0)
2453 if (bond
->send_grat_arp
) {
2454 read_lock(&bond
->curr_slave_lock
);
2455 bond_send_gratuitous_arp(bond
);
2456 read_unlock(&bond
->curr_slave_lock
);
2459 if (bond
->send_unsol_na
) {
2460 read_lock(&bond
->curr_slave_lock
);
2461 bond_send_unsolicited_na(bond
);
2462 read_unlock(&bond
->curr_slave_lock
);
2465 if (bond_miimon_inspect(bond
)) {
2466 read_unlock(&bond
->lock
);
2468 read_lock(&bond
->lock
);
2470 bond_miimon_commit(bond
);
2472 read_unlock(&bond
->lock
);
2473 rtnl_unlock(); /* might sleep, hold no other locks */
2474 read_lock(&bond
->lock
);
2478 if (bond
->params
.miimon
)
2479 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2480 msecs_to_jiffies(bond
->params
.miimon
));
2482 read_unlock(&bond
->lock
);
2485 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2487 struct in_device
*idev
;
2488 struct in_ifaddr
*ifa
;
2495 idev
= __in_dev_get_rcu(dev
);
2499 ifa
= idev
->ifa_list
;
2503 addr
= ifa
->ifa_local
;
2509 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2511 struct vlan_entry
*vlan
;
2513 if (ip
== bond
->master_ip
)
2516 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2517 if (ip
== vlan
->vlan_ip
)
2525 * We go to the (large) trouble of VLAN tagging ARP frames because
2526 * switches in VLAN mode (especially if ports are configured as
2527 * "native" to a VLAN) might not pass non-tagged frames.
2529 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2531 struct sk_buff
*skb
;
2533 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2534 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2536 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2537 NULL
, slave_dev
->dev_addr
, NULL
);
2540 pr_err("ARP packet allocation failed\n");
2544 skb
= vlan_put_tag(skb
, vlan_id
);
2546 pr_err("failed to insert VLAN tag\n");
2554 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2557 __be32
*targets
= bond
->params
.arp_targets
;
2558 struct vlan_entry
*vlan
;
2559 struct net_device
*vlan_dev
;
2563 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2566 pr_debug("basa: target %x\n", targets
[i
]);
2568 pr_debug("basa: empty vlan: arp_send\n");
2569 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2570 bond
->master_ip
, 0);
2575 * If VLANs are configured, we do a route lookup to
2576 * determine which VLAN interface would be used, so we
2577 * can tag the ARP with the proper VLAN tag.
2579 memset(&fl
, 0, sizeof(fl
));
2580 fl
.fl4_dst
= targets
[i
];
2581 fl
.fl4_tos
= RTO_ONLINK
;
2583 rv
= ip_route_output_key(dev_net(bond
->dev
), &rt
, &fl
);
2585 if (net_ratelimit()) {
2586 pr_warning("%s: no route to arp_ip_target %pI4\n",
2587 bond
->dev
->name
, &fl
.fl4_dst
);
2593 * This target is not on a VLAN
2595 if (rt
->dst
.dev
== bond
->dev
) {
2597 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2598 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2599 bond
->master_ip
, 0);
2604 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2605 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2606 if (vlan_dev
== rt
->dst
.dev
) {
2607 vlan_id
= vlan
->vlan_id
;
2608 pr_debug("basa: vlan match on %s %d\n",
2609 vlan_dev
->name
, vlan_id
);
2616 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2617 vlan
->vlan_ip
, vlan_id
);
2621 if (net_ratelimit()) {
2622 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2623 bond
->dev
->name
, &fl
.fl4_dst
,
2624 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2631 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2632 * for each VLAN above us.
2634 * Caller must hold curr_slave_lock for read or better
2636 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2638 struct slave
*slave
= bond
->curr_active_slave
;
2639 struct vlan_entry
*vlan
;
2640 struct net_device
*vlan_dev
;
2642 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2643 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
2645 if (!slave
|| !bond
->send_grat_arp
||
2646 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2649 bond
->send_grat_arp
--;
2651 if (bond
->master_ip
) {
2652 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2653 bond
->master_ip
, 0);
2659 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2660 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2661 if (vlan
->vlan_ip
) {
2662 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2663 vlan
->vlan_ip
, vlan
->vlan_id
);
2668 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2671 __be32
*targets
= bond
->params
.arp_targets
;
2673 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2674 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2675 &sip
, &tip
, i
, &targets
[i
],
2676 bond_has_this_ip(bond
, tip
));
2677 if (sip
== targets
[i
]) {
2678 if (bond_has_this_ip(bond
, tip
))
2679 slave
->last_arp_rx
= jiffies
;
2685 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2688 struct slave
*slave
;
2689 struct bonding
*bond
;
2690 unsigned char *arp_ptr
;
2693 if (dev
->priv_flags
& IFF_802_1Q_VLAN
) {
2695 * When using VLANS and bonding, dev and oriv_dev may be
2696 * incorrect if the physical interface supports VLAN
2697 * acceleration. With this change ARP validation now
2698 * works for hosts only reachable on the VLAN interface.
2700 dev
= vlan_dev_real_dev(dev
);
2701 orig_dev
= dev_get_by_index_rcu(dev_net(skb
->dev
),skb
->skb_iif
);
2704 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2707 bond
= netdev_priv(dev
);
2708 read_lock(&bond
->lock
);
2710 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2711 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2712 orig_dev
? orig_dev
->name
: "NULL");
2714 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2715 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2718 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2722 if (arp
->ar_hln
!= dev
->addr_len
||
2723 skb
->pkt_type
== PACKET_OTHERHOST
||
2724 skb
->pkt_type
== PACKET_LOOPBACK
||
2725 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2726 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2730 arp_ptr
= (unsigned char *)(arp
+ 1);
2731 arp_ptr
+= dev
->addr_len
;
2732 memcpy(&sip
, arp_ptr
, 4);
2733 arp_ptr
+= 4 + dev
->addr_len
;
2734 memcpy(&tip
, arp_ptr
, 4);
2736 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2737 bond
->dev
->name
, slave
->dev
->name
, slave
->state
,
2738 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2742 * Backup slaves won't see the ARP reply, but do come through
2743 * here for each ARP probe (so we swap the sip/tip to validate
2744 * the probe). In a "redundant switch, common router" type of
2745 * configuration, the ARP probe will (hopefully) travel from
2746 * the active, through one switch, the router, then the other
2747 * switch before reaching the backup.
2749 if (slave
->state
== BOND_STATE_ACTIVE
)
2750 bond_validate_arp(bond
, slave
, sip
, tip
);
2752 bond_validate_arp(bond
, slave
, tip
, sip
);
2755 read_unlock(&bond
->lock
);
2758 return NET_RX_SUCCESS
;
2762 * this function is called regularly to monitor each slave's link
2763 * ensuring that traffic is being sent and received when arp monitoring
2764 * is used in load-balancing mode. if the adapter has been dormant, then an
2765 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2766 * arp monitoring in active backup mode.
2768 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2770 struct bonding
*bond
= container_of(work
, struct bonding
,
2772 struct slave
*slave
, *oldcurrent
;
2773 int do_failover
= 0;
2777 read_lock(&bond
->lock
);
2779 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2781 if (bond
->kill_timers
)
2784 if (bond
->slave_cnt
== 0)
2787 read_lock(&bond
->curr_slave_lock
);
2788 oldcurrent
= bond
->curr_active_slave
;
2789 read_unlock(&bond
->curr_slave_lock
);
2791 /* see if any of the previous devices are up now (i.e. they have
2792 * xmt and rcv traffic). the curr_active_slave does not come into
2793 * the picture unless it is null. also, slave->jiffies is not needed
2794 * here because we send an arp on each slave and give a slave as
2795 * long as it needs to get the tx/rx within the delta.
2796 * TODO: what about up/down delay in arp mode? it wasn't here before
2799 bond_for_each_slave(bond
, slave
, i
) {
2800 if (slave
->link
!= BOND_LINK_UP
) {
2801 if (time_before_eq(jiffies
, dev_trans_start(slave
->dev
) + delta_in_ticks
) &&
2802 time_before_eq(jiffies
, slave
->dev
->last_rx
+ delta_in_ticks
)) {
2804 slave
->link
= BOND_LINK_UP
;
2805 slave
->state
= BOND_STATE_ACTIVE
;
2807 /* primary_slave has no meaning in round-robin
2808 * mode. the window of a slave being up and
2809 * curr_active_slave being null after enslaving
2813 pr_info("%s: link status definitely up for interface %s, ",
2818 pr_info("%s: interface %s is now up\n",
2824 /* slave->link == BOND_LINK_UP */
2826 /* not all switches will respond to an arp request
2827 * when the source ip is 0, so don't take the link down
2828 * if we don't know our ip yet
2830 if (time_after_eq(jiffies
, dev_trans_start(slave
->dev
) + 2*delta_in_ticks
) ||
2831 (time_after_eq(jiffies
, slave
->dev
->last_rx
+ 2*delta_in_ticks
))) {
2833 slave
->link
= BOND_LINK_DOWN
;
2834 slave
->state
= BOND_STATE_BACKUP
;
2836 if (slave
->link_failure_count
< UINT_MAX
)
2837 slave
->link_failure_count
++;
2839 pr_info("%s: interface %s is now down.\n",
2843 if (slave
== oldcurrent
)
2848 /* note: if switch is in round-robin mode, all links
2849 * must tx arp to ensure all links rx an arp - otherwise
2850 * links may oscillate or not come up at all; if switch is
2851 * in something like xor mode, there is nothing we can
2852 * do - all replies will be rx'ed on same link causing slaves
2853 * to be unstable during low/no traffic periods
2855 if (IS_UP(slave
->dev
))
2856 bond_arp_send_all(bond
, slave
);
2860 write_lock_bh(&bond
->curr_slave_lock
);
2862 bond_select_active_slave(bond
);
2864 write_unlock_bh(&bond
->curr_slave_lock
);
2868 if (bond
->params
.arp_interval
)
2869 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2871 read_unlock(&bond
->lock
);
2875 * Called to inspect slaves for active-backup mode ARP monitor link state
2876 * changes. Sets new_link in slaves to specify what action should take
2877 * place for the slave. Returns 0 if no changes are found, >0 if changes
2878 * to link states must be committed.
2880 * Called with bond->lock held for read.
2882 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2884 struct slave
*slave
;
2887 bond_for_each_slave(bond
, slave
, i
) {
2888 slave
->new_link
= BOND_LINK_NOCHANGE
;
2890 if (slave
->link
!= BOND_LINK_UP
) {
2891 if (time_before_eq(jiffies
, slave_last_rx(bond
, slave
) +
2893 slave
->new_link
= BOND_LINK_UP
;
2901 * Give slaves 2*delta after being enslaved or made
2902 * active. This avoids bouncing, as the last receive
2903 * times need a full ARP monitor cycle to be updated.
2905 if (!time_after_eq(jiffies
, slave
->jiffies
+
2906 2 * delta_in_ticks
))
2910 * Backup slave is down if:
2911 * - No current_arp_slave AND
2912 * - more than 3*delta since last receive AND
2913 * - the bond has an IP address
2915 * Note: a non-null current_arp_slave indicates
2916 * the curr_active_slave went down and we are
2917 * searching for a new one; under this condition
2918 * we only take the curr_active_slave down - this
2919 * gives each slave a chance to tx/rx traffic
2920 * before being taken out
2922 if (slave
->state
== BOND_STATE_BACKUP
&&
2923 !bond
->current_arp_slave
&&
2924 time_after(jiffies
, slave_last_rx(bond
, slave
) +
2925 3 * delta_in_ticks
)) {
2926 slave
->new_link
= BOND_LINK_DOWN
;
2931 * Active slave is down if:
2932 * - more than 2*delta since transmitting OR
2933 * - (more than 2*delta since receive AND
2934 * the bond has an IP address)
2936 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
2937 (time_after_eq(jiffies
, dev_trans_start(slave
->dev
) +
2938 2 * delta_in_ticks
) ||
2939 (time_after_eq(jiffies
, slave_last_rx(bond
, slave
)
2940 + 2 * delta_in_ticks
)))) {
2941 slave
->new_link
= BOND_LINK_DOWN
;
2950 * Called to commit link state changes noted by inspection step of
2951 * active-backup mode ARP monitor.
2953 * Called with RTNL and bond->lock for read.
2955 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2957 struct slave
*slave
;
2960 bond_for_each_slave(bond
, slave
, i
) {
2961 switch (slave
->new_link
) {
2962 case BOND_LINK_NOCHANGE
:
2966 if ((!bond
->curr_active_slave
&&
2967 time_before_eq(jiffies
,
2968 dev_trans_start(slave
->dev
) +
2970 bond
->curr_active_slave
!= slave
) {
2971 slave
->link
= BOND_LINK_UP
;
2972 bond
->current_arp_slave
= NULL
;
2974 pr_info("%s: link status definitely up for interface %s.\n",
2975 bond
->dev
->name
, slave
->dev
->name
);
2977 if (!bond
->curr_active_slave
||
2978 (slave
== bond
->primary_slave
))
2985 case BOND_LINK_DOWN
:
2986 if (slave
->link_failure_count
< UINT_MAX
)
2987 slave
->link_failure_count
++;
2989 slave
->link
= BOND_LINK_DOWN
;
2990 bond_set_slave_inactive_flags(slave
);
2992 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2993 bond
->dev
->name
, slave
->dev
->name
);
2995 if (slave
== bond
->curr_active_slave
) {
2996 bond
->current_arp_slave
= NULL
;
3003 pr_err("%s: impossible: new_link %d on slave %s\n",
3004 bond
->dev
->name
, slave
->new_link
,
3011 write_lock_bh(&bond
->curr_slave_lock
);
3012 bond_select_active_slave(bond
);
3013 write_unlock_bh(&bond
->curr_slave_lock
);
3016 bond_set_carrier(bond
);
3020 * Send ARP probes for active-backup mode ARP monitor.
3022 * Called with bond->lock held for read.
3024 static void bond_ab_arp_probe(struct bonding
*bond
)
3026 struct slave
*slave
;
3029 read_lock(&bond
->curr_slave_lock
);
3031 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3032 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3033 bond
->current_arp_slave
->dev
->name
,
3034 bond
->curr_active_slave
->dev
->name
);
3036 if (bond
->curr_active_slave
) {
3037 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3038 read_unlock(&bond
->curr_slave_lock
);
3042 read_unlock(&bond
->curr_slave_lock
);
3044 /* if we don't have a curr_active_slave, search for the next available
3045 * backup slave from the current_arp_slave and make it the candidate
3046 * for becoming the curr_active_slave
3049 if (!bond
->current_arp_slave
) {
3050 bond
->current_arp_slave
= bond
->first_slave
;
3051 if (!bond
->current_arp_slave
)
3055 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3057 /* search for next candidate */
3058 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3059 if (IS_UP(slave
->dev
)) {
3060 slave
->link
= BOND_LINK_BACK
;
3061 bond_set_slave_active_flags(slave
);
3062 bond_arp_send_all(bond
, slave
);
3063 slave
->jiffies
= jiffies
;
3064 bond
->current_arp_slave
= slave
;
3068 /* if the link state is up at this point, we
3069 * mark it down - this can happen if we have
3070 * simultaneous link failures and
3071 * reselect_active_interface doesn't make this
3072 * one the current slave so it is still marked
3073 * up when it is actually down
3075 if (slave
->link
== BOND_LINK_UP
) {
3076 slave
->link
= BOND_LINK_DOWN
;
3077 if (slave
->link_failure_count
< UINT_MAX
)
3078 slave
->link_failure_count
++;
3080 bond_set_slave_inactive_flags(slave
);
3082 pr_info("%s: backup interface %s is now down.\n",
3083 bond
->dev
->name
, slave
->dev
->name
);
3088 void bond_activebackup_arp_mon(struct work_struct
*work
)
3090 struct bonding
*bond
= container_of(work
, struct bonding
,
3094 read_lock(&bond
->lock
);
3096 if (bond
->kill_timers
)
3099 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3101 if (bond
->slave_cnt
== 0)
3104 if (bond
->send_grat_arp
) {
3105 read_lock(&bond
->curr_slave_lock
);
3106 bond_send_gratuitous_arp(bond
);
3107 read_unlock(&bond
->curr_slave_lock
);
3110 if (bond
->send_unsol_na
) {
3111 read_lock(&bond
->curr_slave_lock
);
3112 bond_send_unsolicited_na(bond
);
3113 read_unlock(&bond
->curr_slave_lock
);
3116 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3117 read_unlock(&bond
->lock
);
3119 read_lock(&bond
->lock
);
3121 bond_ab_arp_commit(bond
, delta_in_ticks
);
3123 read_unlock(&bond
->lock
);
3125 read_lock(&bond
->lock
);
3128 bond_ab_arp_probe(bond
);
3131 if (bond
->params
.arp_interval
)
3132 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3134 read_unlock(&bond
->lock
);
3137 /*------------------------------ proc/seq_file-------------------------------*/
3139 #ifdef CONFIG_PROC_FS
3141 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3142 __acquires(&dev_base_lock
)
3143 __acquires(&bond
->lock
)
3145 struct bonding
*bond
= seq
->private;
3147 struct slave
*slave
;
3150 /* make sure the bond won't be taken away */
3151 read_lock(&dev_base_lock
);
3152 read_lock(&bond
->lock
);
3155 return SEQ_START_TOKEN
;
3157 bond_for_each_slave(bond
, slave
, i
) {
3165 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3167 struct bonding
*bond
= seq
->private;
3168 struct slave
*slave
= v
;
3171 if (v
== SEQ_START_TOKEN
)
3172 return bond
->first_slave
;
3174 slave
= slave
->next
;
3176 return (slave
== bond
->first_slave
) ? NULL
: slave
;
3179 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
3180 __releases(&bond
->lock
)
3181 __releases(&dev_base_lock
)
3183 struct bonding
*bond
= seq
->private;
3185 read_unlock(&bond
->lock
);
3186 read_unlock(&dev_base_lock
);
3189 static void bond_info_show_master(struct seq_file
*seq
)
3191 struct bonding
*bond
= seq
->private;
3195 read_lock(&bond
->curr_slave_lock
);
3196 curr
= bond
->curr_active_slave
;
3197 read_unlock(&bond
->curr_slave_lock
);
3199 seq_printf(seq
, "Bonding Mode: %s",
3200 bond_mode_name(bond
->params
.mode
));
3202 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
&&
3203 bond
->params
.fail_over_mac
)
3204 seq_printf(seq
, " (fail_over_mac %s)",
3205 fail_over_mac_tbl
[bond
->params
.fail_over_mac
].modename
);
3207 seq_printf(seq
, "\n");
3209 if (bond
->params
.mode
== BOND_MODE_XOR
||
3210 bond
->params
.mode
== BOND_MODE_8023AD
) {
3211 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
3212 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
3213 bond
->params
.xmit_policy
);
3216 if (USES_PRIMARY(bond
->params
.mode
)) {
3217 seq_printf(seq
, "Primary Slave: %s",
3218 (bond
->primary_slave
) ?
3219 bond
->primary_slave
->dev
->name
: "None");
3220 if (bond
->primary_slave
)
3221 seq_printf(seq
, " (primary_reselect %s)",
3222 pri_reselect_tbl
[bond
->params
.primary_reselect
].modename
);
3224 seq_printf(seq
, "\nCurrently Active Slave: %s\n",
3225 (curr
) ? curr
->dev
->name
: "None");
3228 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
3230 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
3231 seq_printf(seq
, "Up Delay (ms): %d\n",
3232 bond
->params
.updelay
* bond
->params
.miimon
);
3233 seq_printf(seq
, "Down Delay (ms): %d\n",
3234 bond
->params
.downdelay
* bond
->params
.miimon
);
3237 /* ARP information */
3238 if (bond
->params
.arp_interval
> 0) {
3240 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
3241 bond
->params
.arp_interval
);
3243 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
3245 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
3246 if (!bond
->params
.arp_targets
[i
])
3249 seq_printf(seq
, ",");
3250 seq_printf(seq
, " %pI4", &bond
->params
.arp_targets
[i
]);
3253 seq_printf(seq
, "\n");
3256 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3257 struct ad_info ad_info
;
3259 seq_puts(seq
, "\n802.3ad info\n");
3260 seq_printf(seq
, "LACP rate: %s\n",
3261 (bond
->params
.lacp_fast
) ? "fast" : "slow");
3262 seq_printf(seq
, "Aggregator selection policy (ad_select): %s\n",
3263 ad_select_tbl
[bond
->params
.ad_select
].modename
);
3265 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3266 seq_printf(seq
, "bond %s has no active aggregator\n",
3269 seq_printf(seq
, "Active Aggregator Info:\n");
3271 seq_printf(seq
, "\tAggregator ID: %d\n",
3272 ad_info
.aggregator_id
);
3273 seq_printf(seq
, "\tNumber of ports: %d\n",
3275 seq_printf(seq
, "\tActor Key: %d\n",
3277 seq_printf(seq
, "\tPartner Key: %d\n",
3278 ad_info
.partner_key
);
3279 seq_printf(seq
, "\tPartner Mac Address: %pM\n",
3280 ad_info
.partner_system
);
3285 static void bond_info_show_slave(struct seq_file
*seq
,
3286 const struct slave
*slave
)
3288 struct bonding
*bond
= seq
->private;
3290 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
3291 seq_printf(seq
, "MII Status: %s\n",
3292 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
3293 seq_printf(seq
, "Link Failure Count: %u\n",
3294 slave
->link_failure_count
);
3296 seq_printf(seq
, "Permanent HW addr: %pM\n", slave
->perm_hwaddr
);
3298 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3299 const struct aggregator
*agg
3300 = SLAVE_AD_INFO(slave
).port
.aggregator
;
3303 seq_printf(seq
, "Aggregator ID: %d\n",
3304 agg
->aggregator_identifier
);
3306 seq_puts(seq
, "Aggregator ID: N/A\n");
3308 seq_printf(seq
, "Slave queue ID: %d\n", slave
->queue_id
);
3311 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
3313 if (v
== SEQ_START_TOKEN
) {
3314 seq_printf(seq
, "%s\n", version
);
3315 bond_info_show_master(seq
);
3317 bond_info_show_slave(seq
, v
);
3322 static const struct seq_operations bond_info_seq_ops
= {
3323 .start
= bond_info_seq_start
,
3324 .next
= bond_info_seq_next
,
3325 .stop
= bond_info_seq_stop
,
3326 .show
= bond_info_seq_show
,
3329 static int bond_info_open(struct inode
*inode
, struct file
*file
)
3331 struct seq_file
*seq
;
3332 struct proc_dir_entry
*proc
;
3335 res
= seq_open(file
, &bond_info_seq_ops
);
3337 /* recover the pointer buried in proc_dir_entry data */
3338 seq
= file
->private_data
;
3340 seq
->private = proc
->data
;
3346 static const struct file_operations bond_info_fops
= {
3347 .owner
= THIS_MODULE
,
3348 .open
= bond_info_open
,
3350 .llseek
= seq_lseek
,
3351 .release
= seq_release
,
3354 static void bond_create_proc_entry(struct bonding
*bond
)
3356 struct net_device
*bond_dev
= bond
->dev
;
3357 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
3360 bond
->proc_entry
= proc_create_data(bond_dev
->name
,
3361 S_IRUGO
, bn
->proc_dir
,
3362 &bond_info_fops
, bond
);
3363 if (bond
->proc_entry
== NULL
)
3364 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3365 DRV_NAME
, bond_dev
->name
);
3367 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3371 static void bond_remove_proc_entry(struct bonding
*bond
)
3373 struct net_device
*bond_dev
= bond
->dev
;
3374 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
3376 if (bn
->proc_dir
&& bond
->proc_entry
) {
3377 remove_proc_entry(bond
->proc_file_name
, bn
->proc_dir
);
3378 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3379 bond
->proc_entry
= NULL
;
3383 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3384 * Caller must hold rtnl_lock.
3386 static void __net_init
bond_create_proc_dir(struct bond_net
*bn
)
3388 if (!bn
->proc_dir
) {
3389 bn
->proc_dir
= proc_mkdir(DRV_NAME
, bn
->net
->proc_net
);
3391 pr_warning("Warning: cannot create /proc/net/%s\n",
3396 /* Destroy the bonding directory under /proc/net, if empty.
3397 * Caller must hold rtnl_lock.
3399 static void __net_exit
bond_destroy_proc_dir(struct bond_net
*bn
)
3402 remove_proc_entry(DRV_NAME
, bn
->net
->proc_net
);
3403 bn
->proc_dir
= NULL
;
3407 #else /* !CONFIG_PROC_FS */
3409 static void bond_create_proc_entry(struct bonding
*bond
)
3413 static void bond_remove_proc_entry(struct bonding
*bond
)
3417 static inline void bond_create_proc_dir(struct bond_net
*bn
)
3421 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
3425 #endif /* CONFIG_PROC_FS */
3428 /*-------------------------- netdev event handling --------------------------*/
3431 * Change device name
3433 static int bond_event_changename(struct bonding
*bond
)
3435 bond_remove_proc_entry(bond
);
3436 bond_create_proc_entry(bond
);
3441 static int bond_master_netdev_event(unsigned long event
,
3442 struct net_device
*bond_dev
)
3444 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3447 case NETDEV_CHANGENAME
:
3448 return bond_event_changename(event_bond
);
3456 static int bond_slave_netdev_event(unsigned long event
,
3457 struct net_device
*slave_dev
)
3459 struct net_device
*bond_dev
= slave_dev
->master
;
3460 struct bonding
*bond
= netdev_priv(bond_dev
);
3463 case NETDEV_UNREGISTER
:
3465 if (bond
->setup_by_slave
)
3466 bond_release_and_destroy(bond_dev
, slave_dev
);
3468 bond_release(bond_dev
, slave_dev
);
3472 if (bond
->params
.mode
== BOND_MODE_8023AD
|| bond_is_lb(bond
)) {
3473 struct slave
*slave
;
3475 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3477 u16 old_speed
= slave
->speed
;
3478 u16 old_duplex
= slave
->duplex
;
3480 bond_update_speed_duplex(slave
);
3482 if (bond_is_lb(bond
))
3485 if (old_speed
!= slave
->speed
)
3486 bond_3ad_adapter_speed_changed(slave
);
3487 if (old_duplex
!= slave
->duplex
)
3488 bond_3ad_adapter_duplex_changed(slave
);
3495 * ... Or is it this?
3498 case NETDEV_CHANGEMTU
:
3500 * TODO: Should slaves be allowed to
3501 * independently alter their MTU? For
3502 * an active-backup bond, slaves need
3503 * not be the same type of device, so
3504 * MTUs may vary. For other modes,
3505 * slaves arguably should have the
3506 * same MTUs. To do this, we'd need to
3507 * take over the slave's change_mtu
3508 * function for the duration of their
3512 case NETDEV_CHANGENAME
:
3514 * TODO: handle changing the primary's name
3517 case NETDEV_FEAT_CHANGE
:
3518 bond_compute_features(bond
);
3528 * bond_netdev_event: handle netdev notifier chain events.
3530 * This function receives events for the netdev chain. The caller (an
3531 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3532 * locks for us to safely manipulate the slave devices (RTNL lock,
3535 static int bond_netdev_event(struct notifier_block
*this,
3536 unsigned long event
, void *ptr
)
3538 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3540 pr_debug("event_dev: %s, event: %lx\n",
3541 event_dev
? event_dev
->name
: "None",
3544 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3547 if (event_dev
->flags
& IFF_MASTER
) {
3548 pr_debug("IFF_MASTER\n");
3549 return bond_master_netdev_event(event
, event_dev
);
3552 if (event_dev
->flags
& IFF_SLAVE
) {
3553 pr_debug("IFF_SLAVE\n");
3554 return bond_slave_netdev_event(event
, event_dev
);
3561 * bond_inetaddr_event: handle inetaddr notifier chain events.
3563 * We keep track of device IPs primarily to use as source addresses in
3564 * ARP monitor probes (rather than spewing out broadcasts all the time).
3566 * We track one IP for the main device (if it has one), plus one per VLAN.
3568 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3570 struct in_ifaddr
*ifa
= ptr
;
3571 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3572 struct bond_net
*bn
= net_generic(dev_net(event_dev
), bond_net_id
);
3573 struct bonding
*bond
;
3574 struct vlan_entry
*vlan
;
3576 list_for_each_entry(bond
, &bn
->dev_list
, bond_list
) {
3577 if (bond
->dev
== event_dev
) {
3580 bond
->master_ip
= ifa
->ifa_local
;
3583 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3590 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3593 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3594 if (vlan_dev
== event_dev
) {
3597 vlan
->vlan_ip
= ifa
->ifa_local
;
3601 bond_glean_dev_ip(vlan_dev
);
3612 static struct notifier_block bond_netdev_notifier
= {
3613 .notifier_call
= bond_netdev_event
,
3616 static struct notifier_block bond_inetaddr_notifier
= {
3617 .notifier_call
= bond_inetaddr_event
,
3620 /*-------------------------- Packet type handling ---------------------------*/
3622 /* register to receive lacpdus on a bond */
3623 static void bond_register_lacpdu(struct bonding
*bond
)
3625 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3627 /* initialize packet type */
3628 pk_type
->type
= PKT_TYPE_LACPDU
;
3629 pk_type
->dev
= bond
->dev
;
3630 pk_type
->func
= bond_3ad_lacpdu_recv
;
3632 dev_add_pack(pk_type
);
3635 /* unregister to receive lacpdus on a bond */
3636 static void bond_unregister_lacpdu(struct bonding
*bond
)
3638 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3641 void bond_register_arp(struct bonding
*bond
)
3643 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3648 pt
->type
= htons(ETH_P_ARP
);
3649 pt
->dev
= bond
->dev
;
3650 pt
->func
= bond_arp_rcv
;
3654 void bond_unregister_arp(struct bonding
*bond
)
3656 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3658 dev_remove_pack(pt
);
3662 /*---------------------------- Hashing Policies -----------------------------*/
3665 * Hash for the output device based upon layer 2 and layer 3 data. If
3666 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3668 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3670 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3671 struct iphdr
*iph
= ip_hdr(skb
);
3673 if (skb
->protocol
== htons(ETH_P_IP
)) {
3674 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3675 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3678 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3682 * Hash for the output device based upon layer 3 and layer 4 data. If
3683 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3684 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3686 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3688 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3689 struct iphdr
*iph
= ip_hdr(skb
);
3690 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3693 if (skb
->protocol
== htons(ETH_P_IP
)) {
3694 if (!(iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) &&
3695 (iph
->protocol
== IPPROTO_TCP
||
3696 iph
->protocol
== IPPROTO_UDP
)) {
3697 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3699 return (layer4_xor
^
3700 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3704 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3708 * Hash for the output device based upon layer 2 data
3710 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3712 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3714 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3717 /*-------------------------- Device entry points ----------------------------*/
3719 static int bond_open(struct net_device
*bond_dev
)
3721 struct bonding
*bond
= netdev_priv(bond_dev
);
3723 bond
->kill_timers
= 0;
3725 if (bond_is_lb(bond
)) {
3726 /* bond_alb_initialize must be called before the timer
3729 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3730 /* something went wrong - fail the open operation */
3734 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3735 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3738 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3739 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3740 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3743 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3744 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3745 INIT_DELAYED_WORK(&bond
->arp_work
,
3746 bond_activebackup_arp_mon
);
3748 INIT_DELAYED_WORK(&bond
->arp_work
,
3749 bond_loadbalance_arp_mon
);
3751 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3752 if (bond
->params
.arp_validate
)
3753 bond_register_arp(bond
);
3756 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3757 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3758 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3759 /* register to receive LACPDUs */
3760 bond_register_lacpdu(bond
);
3761 bond_3ad_initiate_agg_selection(bond
, 1);
3767 static int bond_close(struct net_device
*bond_dev
)
3769 struct bonding
*bond
= netdev_priv(bond_dev
);
3771 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3772 /* Unregister the receive of LACPDUs */
3773 bond_unregister_lacpdu(bond
);
3776 if (bond
->params
.arp_validate
)
3777 bond_unregister_arp(bond
);
3779 write_lock_bh(&bond
->lock
);
3781 bond
->send_grat_arp
= 0;
3782 bond
->send_unsol_na
= 0;
3784 /* signal timers not to re-arm */
3785 bond
->kill_timers
= 1;
3787 write_unlock_bh(&bond
->lock
);
3789 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3790 cancel_delayed_work(&bond
->mii_work
);
3793 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3794 cancel_delayed_work(&bond
->arp_work
);
3797 switch (bond
->params
.mode
) {
3798 case BOND_MODE_8023AD
:
3799 cancel_delayed_work(&bond
->ad_work
);
3803 cancel_delayed_work(&bond
->alb_work
);
3810 if (bond_is_lb(bond
)) {
3811 /* Must be called only after all
3812 * slaves have been released
3814 bond_alb_deinitialize(bond
);
3820 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3821 struct rtnl_link_stats64
*stats
)
3823 struct bonding
*bond
= netdev_priv(bond_dev
);
3824 struct rtnl_link_stats64 temp
;
3825 struct slave
*slave
;
3828 memset(stats
, 0, sizeof(*stats
));
3830 read_lock_bh(&bond
->lock
);
3832 bond_for_each_slave(bond
, slave
, i
) {
3833 const struct rtnl_link_stats64
*sstats
=
3834 dev_get_stats(slave
->dev
, &temp
);
3836 stats
->rx_packets
+= sstats
->rx_packets
;
3837 stats
->rx_bytes
+= sstats
->rx_bytes
;
3838 stats
->rx_errors
+= sstats
->rx_errors
;
3839 stats
->rx_dropped
+= sstats
->rx_dropped
;
3841 stats
->tx_packets
+= sstats
->tx_packets
;
3842 stats
->tx_bytes
+= sstats
->tx_bytes
;
3843 stats
->tx_errors
+= sstats
->tx_errors
;
3844 stats
->tx_dropped
+= sstats
->tx_dropped
;
3846 stats
->multicast
+= sstats
->multicast
;
3847 stats
->collisions
+= sstats
->collisions
;
3849 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3850 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3851 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3852 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3853 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3854 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3856 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3857 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3858 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3859 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3860 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3863 read_unlock_bh(&bond
->lock
);
3868 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3870 struct net_device
*slave_dev
= NULL
;
3871 struct ifbond k_binfo
;
3872 struct ifbond __user
*u_binfo
= NULL
;
3873 struct ifslave k_sinfo
;
3874 struct ifslave __user
*u_sinfo
= NULL
;
3875 struct mii_ioctl_data
*mii
= NULL
;
3878 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3890 * We do this again just in case we were called by SIOCGMIIREG
3891 * instead of SIOCGMIIPHY.
3898 if (mii
->reg_num
== 1) {
3899 struct bonding
*bond
= netdev_priv(bond_dev
);
3901 read_lock(&bond
->lock
);
3902 read_lock(&bond
->curr_slave_lock
);
3903 if (netif_carrier_ok(bond
->dev
))
3904 mii
->val_out
= BMSR_LSTATUS
;
3906 read_unlock(&bond
->curr_slave_lock
);
3907 read_unlock(&bond
->lock
);
3911 case BOND_INFO_QUERY_OLD
:
3912 case SIOCBONDINFOQUERY
:
3913 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3915 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3918 res
= bond_info_query(bond_dev
, &k_binfo
);
3920 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3924 case BOND_SLAVE_INFO_QUERY_OLD
:
3925 case SIOCBONDSLAVEINFOQUERY
:
3926 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3928 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3931 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3933 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3942 if (!capable(CAP_NET_ADMIN
))
3945 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
3947 pr_debug("slave_dev=%p:\n", slave_dev
);
3952 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3954 case BOND_ENSLAVE_OLD
:
3955 case SIOCBONDENSLAVE
:
3956 res
= bond_enslave(bond_dev
, slave_dev
);
3958 case BOND_RELEASE_OLD
:
3959 case SIOCBONDRELEASE
:
3960 res
= bond_release(bond_dev
, slave_dev
);
3962 case BOND_SETHWADDR_OLD
:
3963 case SIOCBONDSETHWADDR
:
3964 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3966 case BOND_CHANGE_ACTIVE_OLD
:
3967 case SIOCBONDCHANGEACTIVE
:
3968 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3980 static bool bond_addr_in_mc_list(unsigned char *addr
,
3981 struct netdev_hw_addr_list
*list
,
3984 struct netdev_hw_addr
*ha
;
3986 netdev_hw_addr_list_for_each(ha
, list
)
3987 if (!memcmp(ha
->addr
, addr
, addrlen
))
3993 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3995 struct bonding
*bond
= netdev_priv(bond_dev
);
3996 struct netdev_hw_addr
*ha
;
4000 * Do promisc before checking multicast_mode
4002 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
))
4004 * FIXME: Need to handle the error when one of the multi-slaves
4007 bond_set_promiscuity(bond
, 1);
4010 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
))
4011 bond_set_promiscuity(bond
, -1);
4014 /* set allmulti flag to slaves */
4015 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
))
4017 * FIXME: Need to handle the error when one of the multi-slaves
4020 bond_set_allmulti(bond
, 1);
4023 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
))
4024 bond_set_allmulti(bond
, -1);
4027 read_lock(&bond
->lock
);
4029 bond
->flags
= bond_dev
->flags
;
4031 /* looking for addresses to add to slaves' mc list */
4032 netdev_for_each_mc_addr(ha
, bond_dev
) {
4033 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
4034 bond_dev
->addr_len
);
4036 bond_mc_add(bond
, ha
->addr
);
4039 /* looking for addresses to delete from slaves' list */
4040 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
4041 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
4042 bond_dev
->addr_len
);
4044 bond_mc_del(bond
, ha
->addr
);
4047 /* save master's multicast list */
4048 __hw_addr_flush(&bond
->mc_list
);
4049 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
4050 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
4052 read_unlock(&bond
->lock
);
4055 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
4057 struct bonding
*bond
= netdev_priv(dev
);
4058 struct slave
*slave
= bond
->first_slave
;
4061 const struct net_device_ops
*slave_ops
4062 = slave
->dev
->netdev_ops
;
4063 if (slave_ops
->ndo_neigh_setup
)
4064 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
4070 * Change the MTU of all of a master's slaves to match the master
4072 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
4074 struct bonding
*bond
= netdev_priv(bond_dev
);
4075 struct slave
*slave
, *stop_at
;
4079 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
4080 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
4082 /* Can't hold bond->lock with bh disabled here since
4083 * some base drivers panic. On the other hand we can't
4084 * hold bond->lock without bh disabled because we'll
4085 * deadlock. The only solution is to rely on the fact
4086 * that we're under rtnl_lock here, and the slaves
4087 * list won't change. This doesn't solve the problem
4088 * of setting the slave's MTU while it is
4089 * transmitting, but the assumption is that the base
4090 * driver can handle that.
4092 * TODO: figure out a way to safely iterate the slaves
4093 * list, but without holding a lock around the actual
4094 * call to the base driver.
4097 bond_for_each_slave(bond
, slave
, i
) {
4098 pr_debug("s %p s->p %p c_m %p\n",
4101 slave
->dev
->netdev_ops
->ndo_change_mtu
);
4103 res
= dev_set_mtu(slave
->dev
, new_mtu
);
4106 /* If we failed to set the slave's mtu to the new value
4107 * we must abort the operation even in ACTIVE_BACKUP
4108 * mode, because if we allow the backup slaves to have
4109 * different mtu values than the active slave we'll
4110 * need to change their mtu when doing a failover. That
4111 * means changing their mtu from timer context, which
4112 * is probably not a good idea.
4114 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4119 bond_dev
->mtu
= new_mtu
;
4124 /* unwind from head to the slave that failed */
4126 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4129 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
4131 pr_debug("unwind err %d dev %s\n",
4132 tmp_res
, slave
->dev
->name
);
4142 * Note that many devices must be down to change the HW address, and
4143 * downing the master releases all slaves. We can make bonds full of
4144 * bonding devices to test this, however.
4146 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4148 struct bonding
*bond
= netdev_priv(bond_dev
);
4149 struct sockaddr
*sa
= addr
, tmp_sa
;
4150 struct slave
*slave
, *stop_at
;
4154 if (bond
->params
.mode
== BOND_MODE_ALB
)
4155 return bond_alb_set_mac_address(bond_dev
, addr
);
4158 pr_debug("bond=%p, name=%s\n",
4159 bond
, bond_dev
? bond_dev
->name
: "None");
4162 * If fail_over_mac is set to active, do nothing and return
4163 * success. Returning an error causes ifenslave to fail.
4165 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4168 if (!is_valid_ether_addr(sa
->sa_data
))
4169 return -EADDRNOTAVAIL
;
4171 /* Can't hold bond->lock with bh disabled here since
4172 * some base drivers panic. On the other hand we can't
4173 * hold bond->lock without bh disabled because we'll
4174 * deadlock. The only solution is to rely on the fact
4175 * that we're under rtnl_lock here, and the slaves
4176 * list won't change. This doesn't solve the problem
4177 * of setting the slave's hw address while it is
4178 * transmitting, but the assumption is that the base
4179 * driver can handle that.
4181 * TODO: figure out a way to safely iterate the slaves
4182 * list, but without holding a lock around the actual
4183 * call to the base driver.
4186 bond_for_each_slave(bond
, slave
, i
) {
4187 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
4188 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
4190 if (slave_ops
->ndo_set_mac_address
== NULL
) {
4192 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
4196 res
= dev_set_mac_address(slave
->dev
, addr
);
4198 /* TODO: consider downing the slave
4200 * User should expect communications
4201 * breakage anyway until ARP finish
4204 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4210 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4214 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4215 tmp_sa
.sa_family
= bond_dev
->type
;
4217 /* unwind from head to the slave that failed */
4219 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4222 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4224 pr_debug("unwind err %d dev %s\n",
4225 tmp_res
, slave
->dev
->name
);
4232 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4234 struct bonding
*bond
= netdev_priv(bond_dev
);
4235 struct slave
*slave
, *start_at
;
4236 int i
, slave_no
, res
= 1;
4237 struct iphdr
*iph
= ip_hdr(skb
);
4239 read_lock(&bond
->lock
);
4241 if (!BOND_IS_OK(bond
))
4244 * Start with the curr_active_slave that joined the bond as the
4245 * default for sending IGMP traffic. For failover purposes one
4246 * needs to maintain some consistency for the interface that will
4247 * send the join/membership reports. The curr_active_slave found
4248 * will send all of this type of traffic.
4250 if ((iph
->protocol
== IPPROTO_IGMP
) &&
4251 (skb
->protocol
== htons(ETH_P_IP
))) {
4253 read_lock(&bond
->curr_slave_lock
);
4254 slave
= bond
->curr_active_slave
;
4255 read_unlock(&bond
->curr_slave_lock
);
4261 * Concurrent TX may collide on rr_tx_counter; we accept
4262 * that as being rare enough not to justify using an
4265 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4267 bond_for_each_slave(bond
, slave
, i
) {
4275 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4276 if (IS_UP(slave
->dev
) &&
4277 (slave
->link
== BOND_LINK_UP
) &&
4278 (slave
->state
== BOND_STATE_ACTIVE
)) {
4279 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4286 /* no suitable interface, frame not sent */
4289 read_unlock(&bond
->lock
);
4290 return NETDEV_TX_OK
;
4295 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4296 * the bond has a usable interface.
4298 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4300 struct bonding
*bond
= netdev_priv(bond_dev
);
4303 read_lock(&bond
->lock
);
4304 read_lock(&bond
->curr_slave_lock
);
4306 if (!BOND_IS_OK(bond
))
4309 if (!bond
->curr_active_slave
)
4312 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4316 /* no suitable interface, frame not sent */
4319 read_unlock(&bond
->curr_slave_lock
);
4320 read_unlock(&bond
->lock
);
4321 return NETDEV_TX_OK
;
4325 * In bond_xmit_xor() , we determine the output device by using a pre-
4326 * determined xmit_hash_policy(), If the selected device is not enabled,
4327 * find the next active slave.
4329 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4331 struct bonding
*bond
= netdev_priv(bond_dev
);
4332 struct slave
*slave
, *start_at
;
4337 read_lock(&bond
->lock
);
4339 if (!BOND_IS_OK(bond
))
4342 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4344 bond_for_each_slave(bond
, slave
, i
) {
4352 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4353 if (IS_UP(slave
->dev
) &&
4354 (slave
->link
== BOND_LINK_UP
) &&
4355 (slave
->state
== BOND_STATE_ACTIVE
)) {
4356 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4363 /* no suitable interface, frame not sent */
4366 read_unlock(&bond
->lock
);
4367 return NETDEV_TX_OK
;
4371 * in broadcast mode, we send everything to all usable interfaces.
4373 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4375 struct bonding
*bond
= netdev_priv(bond_dev
);
4376 struct slave
*slave
, *start_at
;
4377 struct net_device
*tx_dev
= NULL
;
4381 read_lock(&bond
->lock
);
4383 if (!BOND_IS_OK(bond
))
4386 read_lock(&bond
->curr_slave_lock
);
4387 start_at
= bond
->curr_active_slave
;
4388 read_unlock(&bond
->curr_slave_lock
);
4393 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4394 if (IS_UP(slave
->dev
) &&
4395 (slave
->link
== BOND_LINK_UP
) &&
4396 (slave
->state
== BOND_STATE_ACTIVE
)) {
4398 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4400 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4405 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4407 dev_kfree_skb(skb2
);
4411 tx_dev
= slave
->dev
;
4416 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4420 /* no suitable interface, frame not sent */
4423 /* frame sent to all suitable interfaces */
4424 read_unlock(&bond
->lock
);
4425 return NETDEV_TX_OK
;
4428 /*------------------------- Device initialization ---------------------------*/
4430 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4432 switch (bond
->params
.xmit_policy
) {
4433 case BOND_XMIT_POLICY_LAYER23
:
4434 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4436 case BOND_XMIT_POLICY_LAYER34
:
4437 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4439 case BOND_XMIT_POLICY_LAYER2
:
4441 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4447 * Lookup the slave that corresponds to a qid
4449 static inline int bond_slave_override(struct bonding
*bond
,
4450 struct sk_buff
*skb
)
4453 struct slave
*slave
= NULL
;
4454 struct slave
*check_slave
;
4456 read_lock(&bond
->lock
);
4458 if (!BOND_IS_OK(bond
) || !skb
->queue_mapping
)
4461 /* Find out if any slaves have the same mapping as this skb. */
4462 bond_for_each_slave(bond
, check_slave
, i
) {
4463 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4464 slave
= check_slave
;
4469 /* If the slave isn't UP, use default transmit policy. */
4470 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4471 (slave
->link
== BOND_LINK_UP
)) {
4472 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4476 read_unlock(&bond
->lock
);
4480 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4483 * This helper function exists to help dev_pick_tx get the correct
4484 * destination queue. Using a helper function skips the a call to
4485 * skb_tx_hash and will put the skbs in the queue we expect on their
4486 * way down to the bonding driver.
4488 return skb
->queue_mapping
;
4491 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4493 struct bonding
*bond
= netdev_priv(dev
);
4495 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4496 if (!bond_slave_override(bond
, skb
))
4497 return NETDEV_TX_OK
;
4500 switch (bond
->params
.mode
) {
4501 case BOND_MODE_ROUNDROBIN
:
4502 return bond_xmit_roundrobin(skb
, dev
);
4503 case BOND_MODE_ACTIVEBACKUP
:
4504 return bond_xmit_activebackup(skb
, dev
);
4506 return bond_xmit_xor(skb
, dev
);
4507 case BOND_MODE_BROADCAST
:
4508 return bond_xmit_broadcast(skb
, dev
);
4509 case BOND_MODE_8023AD
:
4510 return bond_3ad_xmit_xor(skb
, dev
);
4513 return bond_alb_xmit(skb
, dev
);
4515 /* Should never happen, mode already checked */
4516 pr_err("%s: Error: Unknown bonding mode %d\n",
4517 dev
->name
, bond
->params
.mode
);
4520 return NETDEV_TX_OK
;
4526 * set bond mode specific net device operations
4528 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4530 struct net_device
*bond_dev
= bond
->dev
;
4533 case BOND_MODE_ROUNDROBIN
:
4535 case BOND_MODE_ACTIVEBACKUP
:
4538 bond_set_xmit_hash_policy(bond
);
4540 case BOND_MODE_BROADCAST
:
4542 case BOND_MODE_8023AD
:
4543 bond_set_master_3ad_flags(bond
);
4544 bond_set_xmit_hash_policy(bond
);
4547 bond_set_master_alb_flags(bond
);
4552 /* Should never happen, mode already checked */
4553 pr_err("%s: Error: Unknown bonding mode %d\n",
4554 bond_dev
->name
, mode
);
4559 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4560 struct ethtool_drvinfo
*drvinfo
)
4562 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4563 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4564 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4567 static const struct ethtool_ops bond_ethtool_ops
= {
4568 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4569 .get_link
= ethtool_op_get_link
,
4570 .get_tx_csum
= ethtool_op_get_tx_csum
,
4571 .get_sg
= ethtool_op_get_sg
,
4572 .get_tso
= ethtool_op_get_tso
,
4573 .get_ufo
= ethtool_op_get_ufo
,
4574 .get_flags
= ethtool_op_get_flags
,
4577 static const struct net_device_ops bond_netdev_ops
= {
4578 .ndo_init
= bond_init
,
4579 .ndo_uninit
= bond_uninit
,
4580 .ndo_open
= bond_open
,
4581 .ndo_stop
= bond_close
,
4582 .ndo_start_xmit
= bond_start_xmit
,
4583 .ndo_select_queue
= bond_select_queue
,
4584 .ndo_get_stats64
= bond_get_stats
,
4585 .ndo_do_ioctl
= bond_do_ioctl
,
4586 .ndo_set_multicast_list
= bond_set_multicast_list
,
4587 .ndo_change_mtu
= bond_change_mtu
,
4588 .ndo_set_mac_address
= bond_set_mac_address
,
4589 .ndo_neigh_setup
= bond_neigh_setup
,
4590 .ndo_vlan_rx_register
= bond_vlan_rx_register
,
4591 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4592 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4593 #ifdef CONFIG_NET_POLL_CONTROLLER
4594 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4595 .ndo_poll_controller
= bond_poll_controller
,
4599 static void bond_destructor(struct net_device
*bond_dev
)
4601 struct bonding
*bond
= netdev_priv(bond_dev
);
4603 destroy_workqueue(bond
->wq
);
4604 free_netdev(bond_dev
);
4607 static void bond_setup(struct net_device
*bond_dev
)
4609 struct bonding
*bond
= netdev_priv(bond_dev
);
4611 /* initialize rwlocks */
4612 rwlock_init(&bond
->lock
);
4613 rwlock_init(&bond
->curr_slave_lock
);
4615 bond
->params
= bonding_defaults
;
4617 /* Initialize pointers */
4618 bond
->dev
= bond_dev
;
4619 INIT_LIST_HEAD(&bond
->vlan_list
);
4621 /* Initialize the device entry points */
4622 ether_setup(bond_dev
);
4623 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4624 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4625 bond_set_mode_ops(bond
, bond
->params
.mode
);
4627 bond_dev
->destructor
= bond_destructor
;
4629 /* Initialize the device options */
4630 bond_dev
->tx_queue_len
= 0;
4631 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4632 bond_dev
->priv_flags
|= IFF_BONDING
;
4633 bond_dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
4635 if (bond
->params
.arp_interval
)
4636 bond_dev
->priv_flags
|= IFF_MASTER_ARPMON
;
4638 /* At first, we block adding VLANs. That's the only way to
4639 * prevent problems that occur when adding VLANs over an
4640 * empty bond. The block will be removed once non-challenged
4641 * slaves are enslaved.
4643 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4645 /* don't acquire bond device's netif_tx_lock when
4647 bond_dev
->features
|= NETIF_F_LLTX
;
4649 /* By default, we declare the bond to be fully
4650 * VLAN hardware accelerated capable. Special
4651 * care is taken in the various xmit functions
4652 * when there are slaves that are not hw accel
4655 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4656 NETIF_F_HW_VLAN_RX
|
4657 NETIF_F_HW_VLAN_FILTER
);
4661 static void bond_work_cancel_all(struct bonding
*bond
)
4663 write_lock_bh(&bond
->lock
);
4664 bond
->kill_timers
= 1;
4665 write_unlock_bh(&bond
->lock
);
4667 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4668 cancel_delayed_work(&bond
->mii_work
);
4670 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4671 cancel_delayed_work(&bond
->arp_work
);
4673 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4674 delayed_work_pending(&bond
->alb_work
))
4675 cancel_delayed_work(&bond
->alb_work
);
4677 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4678 delayed_work_pending(&bond
->ad_work
))
4679 cancel_delayed_work(&bond
->ad_work
);
4683 * Destroy a bonding device.
4684 * Must be under rtnl_lock when this function is called.
4686 static void bond_uninit(struct net_device
*bond_dev
)
4688 struct bonding
*bond
= netdev_priv(bond_dev
);
4689 struct vlan_entry
*vlan
, *tmp
;
4691 bond_netpoll_cleanup(bond_dev
);
4693 /* Release the bonded slaves */
4694 bond_release_all(bond_dev
);
4696 list_del(&bond
->bond_list
);
4698 bond_work_cancel_all(bond
);
4700 bond_remove_proc_entry(bond
);
4702 __hw_addr_flush(&bond
->mc_list
);
4704 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4705 list_del(&vlan
->vlan_list
);
4710 /*------------------------- Module initialization ---------------------------*/
4713 * Convert string input module parms. Accept either the
4714 * number of the mode or its string name. A bit complicated because
4715 * some mode names are substrings of other names, and calls from sysfs
4716 * may have whitespace in the name (trailing newlines, for example).
4718 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4720 int modeint
= -1, i
, rv
;
4721 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4723 for (p
= (char *)buf
; *p
; p
++)
4724 if (!(isdigit(*p
) || isspace(*p
)))
4728 rv
= sscanf(buf
, "%20s", modestr
);
4730 rv
= sscanf(buf
, "%d", &modeint
);
4735 for (i
= 0; tbl
[i
].modename
; i
++) {
4736 if (modeint
== tbl
[i
].mode
)
4738 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4745 static int bond_check_params(struct bond_params
*params
)
4747 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4750 * Convert string parameters.
4753 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4754 if (bond_mode
== -1) {
4755 pr_err("Error: Invalid bonding mode \"%s\"\n",
4756 mode
== NULL
? "NULL" : mode
);
4761 if (xmit_hash_policy
) {
4762 if ((bond_mode
!= BOND_MODE_XOR
) &&
4763 (bond_mode
!= BOND_MODE_8023AD
)) {
4764 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4765 bond_mode_name(bond_mode
));
4767 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4769 if (xmit_hashtype
== -1) {
4770 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4771 xmit_hash_policy
== NULL
? "NULL" :
4779 if (bond_mode
!= BOND_MODE_8023AD
) {
4780 pr_info("lacp_rate param is irrelevant in mode %s\n",
4781 bond_mode_name(bond_mode
));
4783 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4784 if (lacp_fast
== -1) {
4785 pr_err("Error: Invalid lacp rate \"%s\"\n",
4786 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4793 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4794 if (params
->ad_select
== -1) {
4795 pr_err("Error: Invalid ad_select \"%s\"\n",
4796 ad_select
== NULL
? "NULL" : ad_select
);
4800 if (bond_mode
!= BOND_MODE_8023AD
) {
4801 pr_warning("ad_select param only affects 802.3ad mode\n");
4804 params
->ad_select
= BOND_AD_STABLE
;
4807 if (max_bonds
< 0) {
4808 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4809 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4810 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4814 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4815 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4816 miimon
= BOND_LINK_MON_INTERV
;
4820 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4825 if (downdelay
< 0) {
4826 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4827 downdelay
, INT_MAX
);
4831 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4832 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4837 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
4838 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4843 if (num_unsol_na
< 0 || num_unsol_na
> 255) {
4844 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4849 /* reset values for 802.3ad */
4850 if (bond_mode
== BOND_MODE_8023AD
) {
4852 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");
4853 pr_warning("Forcing miimon to 100msec\n");
4858 if (tx_queues
< 1 || tx_queues
> 255) {
4859 pr_warning("Warning: tx_queues (%d) should be between "
4860 "1 and 255, resetting to %d\n",
4861 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4862 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4865 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4866 pr_warning("Warning: all_slaves_active module parameter (%d), "
4867 "not of valid value (0/1), so it was set to "
4868 "0\n", all_slaves_active
);
4869 all_slaves_active
= 0;
4872 /* reset values for TLB/ALB */
4873 if ((bond_mode
== BOND_MODE_TLB
) ||
4874 (bond_mode
== BOND_MODE_ALB
)) {
4876 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");
4877 pr_warning("Forcing miimon to 100msec\n");
4882 if (bond_mode
== BOND_MODE_ALB
) {
4883 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",
4888 if (updelay
|| downdelay
) {
4889 /* just warn the user the up/down delay will have
4890 * no effect since miimon is zero...
4892 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",
4893 updelay
, downdelay
);
4896 /* don't allow arp monitoring */
4898 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4899 miimon
, arp_interval
);
4903 if ((updelay
% miimon
) != 0) {
4904 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4906 (updelay
/ miimon
) * miimon
);
4911 if ((downdelay
% miimon
) != 0) {
4912 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4914 (downdelay
/ miimon
) * miimon
);
4917 downdelay
/= miimon
;
4920 if (arp_interval
< 0) {
4921 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4922 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4923 arp_interval
= BOND_LINK_ARP_INTERV
;
4926 for (arp_ip_count
= 0;
4927 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4929 /* not complete check, but should be good enough to
4931 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4932 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4933 arp_ip_target
[arp_ip_count
]);
4936 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4937 arp_target
[arp_ip_count
] = ip
;
4941 if (arp_interval
&& !arp_ip_count
) {
4942 /* don't allow arping if no arp_ip_target given... */
4943 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4949 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4950 pr_err("arp_validate only supported in active-backup mode\n");
4953 if (!arp_interval
) {
4954 pr_err("arp_validate requires arp_interval\n");
4958 arp_validate_value
= bond_parse_parm(arp_validate
,
4960 if (arp_validate_value
== -1) {
4961 pr_err("Error: invalid arp_validate \"%s\"\n",
4962 arp_validate
== NULL
? "NULL" : arp_validate
);
4966 arp_validate_value
= 0;
4969 pr_info("MII link monitoring set to %d ms\n", miimon
);
4970 } else if (arp_interval
) {
4973 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4975 arp_validate_tbl
[arp_validate_value
].modename
,
4978 for (i
= 0; i
< arp_ip_count
; i
++)
4979 pr_info(" %s", arp_ip_target
[i
]);
4983 } else if (max_bonds
) {
4984 /* miimon and arp_interval not set, we need one so things
4985 * work as expected, see bonding.txt for details
4987 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");
4990 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4991 /* currently, using a primary only makes sense
4992 * in active backup, TLB or ALB modes
4994 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4995 primary
, bond_mode_name(bond_mode
));
4999 if (primary
&& primary_reselect
) {
5000 primary_reselect_value
= bond_parse_parm(primary_reselect
,
5002 if (primary_reselect_value
== -1) {
5003 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5005 NULL
? "NULL" : primary_reselect
);
5009 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
5012 if (fail_over_mac
) {
5013 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
5015 if (fail_over_mac_value
== -1) {
5016 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5017 arp_validate
== NULL
? "NULL" : arp_validate
);
5021 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
5022 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5024 fail_over_mac_value
= BOND_FOM_NONE
;
5027 /* fill params struct with the proper values */
5028 params
->mode
= bond_mode
;
5029 params
->xmit_policy
= xmit_hashtype
;
5030 params
->miimon
= miimon
;
5031 params
->num_grat_arp
= num_grat_arp
;
5032 params
->num_unsol_na
= num_unsol_na
;
5033 params
->arp_interval
= arp_interval
;
5034 params
->arp_validate
= arp_validate_value
;
5035 params
->updelay
= updelay
;
5036 params
->downdelay
= downdelay
;
5037 params
->use_carrier
= use_carrier
;
5038 params
->lacp_fast
= lacp_fast
;
5039 params
->primary
[0] = 0;
5040 params
->primary_reselect
= primary_reselect_value
;
5041 params
->fail_over_mac
= fail_over_mac_value
;
5042 params
->tx_queues
= tx_queues
;
5043 params
->all_slaves_active
= all_slaves_active
;
5046 strncpy(params
->primary
, primary
, IFNAMSIZ
);
5047 params
->primary
[IFNAMSIZ
- 1] = 0;
5050 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
5055 static struct lock_class_key bonding_netdev_xmit_lock_key
;
5056 static struct lock_class_key bonding_netdev_addr_lock_key
;
5058 static void bond_set_lockdep_class_one(struct net_device
*dev
,
5059 struct netdev_queue
*txq
,
5062 lockdep_set_class(&txq
->_xmit_lock
,
5063 &bonding_netdev_xmit_lock_key
);
5066 static void bond_set_lockdep_class(struct net_device
*dev
)
5068 lockdep_set_class(&dev
->addr_list_lock
,
5069 &bonding_netdev_addr_lock_key
);
5070 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
5074 * Called from registration process
5076 static int bond_init(struct net_device
*bond_dev
)
5078 struct bonding
*bond
= netdev_priv(bond_dev
);
5079 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
5081 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
5083 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
5087 bond_set_lockdep_class(bond_dev
);
5089 netif_carrier_off(bond_dev
);
5091 bond_create_proc_entry(bond
);
5092 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
5094 bond_prepare_sysfs_group(bond
);
5096 __hw_addr_init(&bond
->mc_list
);
5100 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
5102 if (tb
[IFLA_ADDRESS
]) {
5103 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
5105 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
5106 return -EADDRNOTAVAIL
;
5111 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
5113 .priv_size
= sizeof(struct bonding
),
5114 .setup
= bond_setup
,
5115 .validate
= bond_validate
,
5118 /* Create a new bond based on the specified name and bonding parameters.
5119 * If name is NULL, obtain a suitable "bond%d" name for us.
5120 * Caller must NOT hold rtnl_lock; we need to release it here before we
5121 * set up our sysfs entries.
5123 int bond_create(struct net
*net
, const char *name
)
5125 struct net_device
*bond_dev
;
5130 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
), name
? name
: "",
5131 bond_setup
, tx_queues
);
5133 pr_err("%s: eek! can't alloc netdev!\n", name
);
5138 dev_net_set(bond_dev
, net
);
5139 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
5142 res
= dev_alloc_name(bond_dev
, "bond%d");
5147 res
= register_netdevice(bond_dev
);
5152 bond_destructor(bond_dev
);
5156 static int __net_init
bond_net_init(struct net
*net
)
5158 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5161 INIT_LIST_HEAD(&bn
->dev_list
);
5163 bond_create_proc_dir(bn
);
5168 static void __net_exit
bond_net_exit(struct net
*net
)
5170 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5172 bond_destroy_proc_dir(bn
);
5175 static struct pernet_operations bond_net_ops
= {
5176 .init
= bond_net_init
,
5177 .exit
= bond_net_exit
,
5179 .size
= sizeof(struct bond_net
),
5182 static int __init
bonding_init(void)
5187 pr_info("%s", version
);
5189 res
= bond_check_params(&bonding_defaults
);
5193 res
= register_pernet_subsys(&bond_net_ops
);
5197 res
= rtnl_link_register(&bond_link_ops
);
5201 for (i
= 0; i
< max_bonds
; i
++) {
5202 res
= bond_create(&init_net
, NULL
);
5207 res
= bond_create_sysfs();
5211 register_netdevice_notifier(&bond_netdev_notifier
);
5212 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5213 bond_register_ipv6_notifier();
5217 rtnl_link_unregister(&bond_link_ops
);
5219 unregister_pernet_subsys(&bond_net_ops
);
5224 static void __exit
bonding_exit(void)
5226 unregister_netdevice_notifier(&bond_netdev_notifier
);
5227 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5228 bond_unregister_ipv6_notifier();
5230 bond_destroy_sysfs();
5232 rtnl_link_unregister(&bond_link_ops
);
5233 unregister_pernet_subsys(&bond_net_ops
);
5236 module_init(bonding_init
);
5237 module_exit(bonding_exit
);
5238 MODULE_LICENSE("GPL");
5239 MODULE_VERSION(DRV_VERSION
);
5240 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5241 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5242 MODULE_ALIAS_RTNL_LINK("bond");