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
;
181 const struct bond_parm_tbl bond_lacp_tbl
[] = {
182 { "slow", AD_LACP_SLOW
},
183 { "fast", AD_LACP_FAST
},
187 const struct bond_parm_tbl bond_mode_tbl
[] = {
188 { "balance-rr", BOND_MODE_ROUNDROBIN
},
189 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
190 { "balance-xor", BOND_MODE_XOR
},
191 { "broadcast", BOND_MODE_BROADCAST
},
192 { "802.3ad", BOND_MODE_8023AD
},
193 { "balance-tlb", BOND_MODE_TLB
},
194 { "balance-alb", BOND_MODE_ALB
},
198 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
199 { "layer2", BOND_XMIT_POLICY_LAYER2
},
200 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
201 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
205 const struct bond_parm_tbl arp_validate_tbl
[] = {
206 { "none", BOND_ARP_VALIDATE_NONE
},
207 { "active", BOND_ARP_VALIDATE_ACTIVE
},
208 { "backup", BOND_ARP_VALIDATE_BACKUP
},
209 { "all", BOND_ARP_VALIDATE_ALL
},
213 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
214 { "none", BOND_FOM_NONE
},
215 { "active", BOND_FOM_ACTIVE
},
216 { "follow", BOND_FOM_FOLLOW
},
220 const struct bond_parm_tbl pri_reselect_tbl
[] = {
221 { "always", BOND_PRI_RESELECT_ALWAYS
},
222 { "better", BOND_PRI_RESELECT_BETTER
},
223 { "failure", BOND_PRI_RESELECT_FAILURE
},
227 struct bond_parm_tbl ad_select_tbl
[] = {
228 { "stable", BOND_AD_STABLE
},
229 { "bandwidth", BOND_AD_BANDWIDTH
},
230 { "count", BOND_AD_COUNT
},
234 /*-------------------------- Forward declarations ---------------------------*/
236 static void bond_send_gratuitous_arp(struct bonding
*bond
);
237 static int bond_init(struct net_device
*bond_dev
);
238 static void bond_uninit(struct net_device
*bond_dev
);
240 /*---------------------------- General routines -----------------------------*/
242 static const char *bond_mode_name(int mode
)
244 static const char *names
[] = {
245 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
246 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
247 [BOND_MODE_XOR
] = "load balancing (xor)",
248 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
249 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
250 [BOND_MODE_TLB
] = "transmit load balancing",
251 [BOND_MODE_ALB
] = "adaptive load balancing",
254 if (mode
< 0 || mode
> BOND_MODE_ALB
)
260 /*---------------------------------- VLAN -----------------------------------*/
263 * bond_add_vlan - add a new vlan id on bond
264 * @bond: bond that got the notification
265 * @vlan_id: the vlan id to add
267 * Returns -ENOMEM if allocation failed.
269 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
271 struct vlan_entry
*vlan
;
273 pr_debug("bond: %s, vlan id %d\n",
274 (bond
? bond
->dev
->name
: "None"), vlan_id
);
276 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
280 INIT_LIST_HEAD(&vlan
->vlan_list
);
281 vlan
->vlan_id
= vlan_id
;
283 write_lock_bh(&bond
->lock
);
285 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
287 write_unlock_bh(&bond
->lock
);
289 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
295 * bond_del_vlan - delete a vlan id from bond
296 * @bond: bond that got the notification
297 * @vlan_id: the vlan id to delete
299 * returns -ENODEV if @vlan_id was not found in @bond.
301 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
303 struct vlan_entry
*vlan
;
306 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
308 write_lock_bh(&bond
->lock
);
310 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
311 if (vlan
->vlan_id
== vlan_id
) {
312 list_del(&vlan
->vlan_list
);
314 if (bond_is_lb(bond
))
315 bond_alb_clear_vlan(bond
, vlan_id
);
317 pr_debug("removed VLAN ID %d from bond %s\n",
318 vlan_id
, bond
->dev
->name
);
322 if (list_empty(&bond
->vlan_list
) &&
323 (bond
->slave_cnt
== 0)) {
324 /* Last VLAN removed and no slaves, so
325 * restore block on adding VLANs. This will
326 * be removed once new slaves that are not
327 * VLAN challenged will be added.
329 bond
->dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
337 pr_debug("couldn't find VLAN ID %d in bond %s\n",
338 vlan_id
, bond
->dev
->name
);
341 write_unlock_bh(&bond
->lock
);
346 * bond_has_challenged_slaves
347 * @bond: the bond we're working on
349 * Searches the slave list. Returns 1 if a vlan challenged slave
350 * was found, 0 otherwise.
352 * Assumes bond->lock is held.
354 static int bond_has_challenged_slaves(struct bonding
*bond
)
359 bond_for_each_slave(bond
, slave
, i
) {
360 if (slave
->dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
361 pr_debug("found VLAN challenged slave - %s\n",
367 pr_debug("no VLAN challenged slaves found\n");
372 * bond_next_vlan - safely skip to the next item in the vlans list.
373 * @bond: the bond we're working on
374 * @curr: item we're advancing from
376 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
377 * or @curr->next otherwise (even if it is @curr itself again).
379 * Caller must hold bond->lock
381 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
383 struct vlan_entry
*next
, *last
;
385 if (list_empty(&bond
->vlan_list
))
389 next
= list_entry(bond
->vlan_list
.next
,
390 struct vlan_entry
, vlan_list
);
392 last
= list_entry(bond
->vlan_list
.prev
,
393 struct vlan_entry
, vlan_list
);
395 next
= list_entry(bond
->vlan_list
.next
,
396 struct vlan_entry
, vlan_list
);
398 next
= list_entry(curr
->vlan_list
.next
,
399 struct vlan_entry
, vlan_list
);
407 * bond_dev_queue_xmit - Prepare skb for xmit.
409 * @bond: bond device that got this skb for tx.
410 * @skb: hw accel VLAN tagged skb to transmit
411 * @slave_dev: slave that is supposed to xmit this skbuff
413 * When the bond gets an skb to transmit that is
414 * already hardware accelerated VLAN tagged, and it
415 * needs to relay this skb to a slave that is not
416 * hw accel capable, the skb needs to be "unaccelerated",
417 * i.e. strip the hwaccel tag and re-insert it as part
420 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
421 struct net_device
*slave_dev
)
423 unsigned short uninitialized_var(vlan_id
);
425 if (!list_empty(&bond
->vlan_list
) &&
426 !(slave_dev
->features
& NETIF_F_HW_VLAN_TX
) &&
427 vlan_get_tag(skb
, &vlan_id
) == 0) {
428 skb
->dev
= slave_dev
;
429 skb
= vlan_put_tag(skb
, vlan_id
);
431 /* vlan_put_tag() frees the skb in case of error,
432 * so return success here so the calling functions
433 * won't attempt to free is again.
438 skb
->dev
= slave_dev
;
442 #ifdef CONFIG_NET_POLL_CONTROLLER
443 if (unlikely(bond
->dev
->priv_flags
& IFF_IN_NETPOLL
)) {
444 struct netpoll
*np
= bond
->dev
->npinfo
->netpoll
;
445 slave_dev
->npinfo
= bond
->dev
->npinfo
;
446 np
->real_dev
= np
->dev
= skb
->dev
;
447 slave_dev
->priv_flags
|= IFF_IN_NETPOLL
;
448 netpoll_send_skb(np
, skb
);
449 slave_dev
->priv_flags
&= ~IFF_IN_NETPOLL
;
459 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
460 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
462 * a. This operation is performed in IOCTL context,
463 * b. The operation is protected by the RTNL semaphore in the 8021q code,
464 * c. Holding a lock with BH disabled while directly calling a base driver
465 * entry point is generally a BAD idea.
467 * The design of synchronization/protection for this operation in the 8021q
468 * module is good for one or more VLAN devices over a single physical device
469 * and cannot be extended for a teaming solution like bonding, so there is a
470 * potential race condition here where a net device from the vlan group might
471 * be referenced (either by a base driver or the 8021q code) while it is being
472 * removed from the system. However, it turns out we're not making matters
473 * worse, and if it works for regular VLAN usage it will work here too.
477 * bond_vlan_rx_register - Propagates registration to slaves
478 * @bond_dev: bonding net device that got called
479 * @grp: vlan group being registered
481 static void bond_vlan_rx_register(struct net_device
*bond_dev
,
482 struct vlan_group
*grp
)
484 struct bonding
*bond
= netdev_priv(bond_dev
);
490 bond_for_each_slave(bond
, slave
, i
) {
491 struct net_device
*slave_dev
= slave
->dev
;
492 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
494 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
495 slave_ops
->ndo_vlan_rx_register
) {
496 slave_ops
->ndo_vlan_rx_register(slave_dev
, grp
);
502 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
503 * @bond_dev: bonding net device that got called
504 * @vid: vlan id being added
506 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
508 struct bonding
*bond
= netdev_priv(bond_dev
);
512 bond_for_each_slave(bond
, slave
, i
) {
513 struct net_device
*slave_dev
= slave
->dev
;
514 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
516 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
517 slave_ops
->ndo_vlan_rx_add_vid
) {
518 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vid
);
522 res
= bond_add_vlan(bond
, vid
);
524 pr_err("%s: Error: Failed to add vlan id %d\n",
525 bond_dev
->name
, vid
);
530 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
531 * @bond_dev: bonding net device that got called
532 * @vid: vlan id being removed
534 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
536 struct bonding
*bond
= netdev_priv(bond_dev
);
538 struct net_device
*vlan_dev
;
541 bond_for_each_slave(bond
, slave
, i
) {
542 struct net_device
*slave_dev
= slave
->dev
;
543 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
545 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
546 slave_ops
->ndo_vlan_rx_kill_vid
) {
547 /* Save and then restore vlan_dev in the grp array,
548 * since the slave's driver might clear it.
550 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vid
);
551 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vid
);
552 vlan_group_set_device(bond
->vlgrp
, vid
, vlan_dev
);
556 res
= bond_del_vlan(bond
, vid
);
558 pr_err("%s: Error: Failed to remove vlan id %d\n",
559 bond_dev
->name
, vid
);
563 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
565 struct vlan_entry
*vlan
;
566 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
568 write_lock_bh(&bond
->lock
);
570 if (list_empty(&bond
->vlan_list
))
573 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
574 slave_ops
->ndo_vlan_rx_register
)
575 slave_ops
->ndo_vlan_rx_register(slave_dev
, bond
->vlgrp
);
577 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
578 !(slave_ops
->ndo_vlan_rx_add_vid
))
581 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
)
582 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
585 write_unlock_bh(&bond
->lock
);
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
;
595 write_lock_bh(&bond
->lock
);
597 if (list_empty(&bond
->vlan_list
))
600 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
601 !(slave_ops
->ndo_vlan_rx_kill_vid
))
604 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 write_unlock_bh(&bond
->lock
);
622 /*------------------------------- Link status -------------------------------*/
625 * Set the carrier state for the master according to the state of its
626 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
627 * do special 802.3ad magic.
629 * Returns zero if carrier state does not change, nonzero if it does.
631 static int bond_set_carrier(struct bonding
*bond
)
636 if (bond
->slave_cnt
== 0)
639 if (bond
->params
.mode
== BOND_MODE_8023AD
)
640 return bond_3ad_set_carrier(bond
);
642 bond_for_each_slave(bond
, slave
, i
) {
643 if (slave
->link
== BOND_LINK_UP
) {
644 if (!netif_carrier_ok(bond
->dev
)) {
645 netif_carrier_on(bond
->dev
);
653 if (netif_carrier_ok(bond
->dev
)) {
654 netif_carrier_off(bond
->dev
);
661 * Get link speed and duplex from the slave's base driver
662 * using ethtool. If for some reason the call fails or the
663 * values are invalid, fake speed and duplex to 100/Full
666 static int bond_update_speed_duplex(struct slave
*slave
)
668 struct net_device
*slave_dev
= slave
->dev
;
669 struct ethtool_cmd etool
;
672 /* Fake speed and duplex */
673 slave
->speed
= SPEED_100
;
674 slave
->duplex
= DUPLEX_FULL
;
676 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
679 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
683 switch (etool
.speed
) {
693 switch (etool
.duplex
) {
701 slave
->speed
= etool
.speed
;
702 slave
->duplex
= etool
.duplex
;
708 * if <dev> supports MII link status reporting, check its link status.
710 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
711 * depending upon the setting of the use_carrier parameter.
713 * Return either BMSR_LSTATUS, meaning that the link is up (or we
714 * can't tell and just pretend it is), or 0, meaning that the link is
717 * If reporting is non-zero, instead of faking link up, return -1 if
718 * both ETHTOOL and MII ioctls fail (meaning the device does not
719 * support them). If use_carrier is set, return whatever it says.
720 * It'd be nice if there was a good way to tell if a driver supports
721 * netif_carrier, but there really isn't.
723 static int bond_check_dev_link(struct bonding
*bond
,
724 struct net_device
*slave_dev
, int reporting
)
726 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
727 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
729 struct mii_ioctl_data
*mii
;
731 if (!reporting
&& !netif_running(slave_dev
))
734 if (bond
->params
.use_carrier
)
735 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
737 /* Try to get link status using Ethtool first. */
738 if (slave_dev
->ethtool_ops
) {
739 if (slave_dev
->ethtool_ops
->get_link
) {
742 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
744 return link
? BMSR_LSTATUS
: 0;
748 /* Ethtool can't be used, fallback to MII ioctls. */
749 ioctl
= slave_ops
->ndo_do_ioctl
;
751 /* TODO: set pointer to correct ioctl on a per team member */
752 /* bases to make this more efficient. that is, once */
753 /* we determine the correct ioctl, we will always */
754 /* call it and not the others for that team */
758 * We cannot assume that SIOCGMIIPHY will also read a
759 * register; not all network drivers (e.g., e100)
763 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
764 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
766 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
767 mii
->reg_num
= MII_BMSR
;
768 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
769 return mii
->val_out
& BMSR_LSTATUS
;
774 * If reporting, report that either there's no dev->do_ioctl,
775 * or both SIOCGMIIREG and get_link failed (meaning that we
776 * cannot report link status). If not reporting, pretend
779 return reporting
? -1 : BMSR_LSTATUS
;
782 /*----------------------------- Multicast list ------------------------------*/
785 * Push the promiscuity flag down to appropriate slaves
787 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
790 if (USES_PRIMARY(bond
->params
.mode
)) {
791 /* write lock already acquired */
792 if (bond
->curr_active_slave
) {
793 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
799 bond_for_each_slave(bond
, slave
, i
) {
800 err
= dev_set_promiscuity(slave
->dev
, inc
);
809 * Push the allmulti flag down to all slaves
811 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
814 if (USES_PRIMARY(bond
->params
.mode
)) {
815 /* write lock already acquired */
816 if (bond
->curr_active_slave
) {
817 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
823 bond_for_each_slave(bond
, slave
, i
) {
824 err
= dev_set_allmulti(slave
->dev
, inc
);
833 * Add a Multicast address to slaves
836 static void bond_mc_add(struct bonding
*bond
, void *addr
)
838 if (USES_PRIMARY(bond
->params
.mode
)) {
839 /* write lock already acquired */
840 if (bond
->curr_active_slave
)
841 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
846 bond_for_each_slave(bond
, slave
, i
)
847 dev_mc_add(slave
->dev
, addr
);
852 * Remove a multicast address from slave
855 static void bond_mc_del(struct bonding
*bond
, void *addr
)
857 if (USES_PRIMARY(bond
->params
.mode
)) {
858 /* write lock already acquired */
859 if (bond
->curr_active_slave
)
860 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
864 bond_for_each_slave(bond
, slave
, i
) {
865 dev_mc_del(slave
->dev
, addr
);
872 * Retrieve the list of registered multicast addresses for the bonding
873 * device and retransmit an IGMP JOIN request to the current active
876 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
878 struct in_device
*in_dev
;
879 struct ip_mc_list
*im
;
882 in_dev
= __in_dev_get_rcu(bond
->dev
);
884 for (im
= in_dev
->mc_list
; im
; im
= im
->next
)
885 ip_mc_rejoin_group(im
);
892 * flush all members of flush->mc_list from device dev->mc_list
894 static void bond_mc_list_flush(struct net_device
*bond_dev
,
895 struct net_device
*slave_dev
)
897 struct bonding
*bond
= netdev_priv(bond_dev
);
898 struct netdev_hw_addr
*ha
;
900 netdev_for_each_mc_addr(ha
, bond_dev
)
901 dev_mc_del(slave_dev
, ha
->addr
);
903 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
904 /* del lacpdu mc addr from mc list */
905 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
907 dev_mc_del(slave_dev
, lacpdu_multicast
);
911 /*--------------------------- Active slave change ---------------------------*/
914 * Update the mc list and multicast-related flags for the new and
915 * old active slaves (if any) according to the multicast mode, and
916 * promiscuous flags unconditionally.
918 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
919 struct slave
*old_active
)
921 struct netdev_hw_addr
*ha
;
923 if (!USES_PRIMARY(bond
->params
.mode
))
924 /* nothing to do - mc list is already up-to-date on
930 if (bond
->dev
->flags
& IFF_PROMISC
)
931 dev_set_promiscuity(old_active
->dev
, -1);
933 if (bond
->dev
->flags
& IFF_ALLMULTI
)
934 dev_set_allmulti(old_active
->dev
, -1);
936 netdev_for_each_mc_addr(ha
, bond
->dev
)
937 dev_mc_del(old_active
->dev
, ha
->addr
);
941 /* FIXME: Signal errors upstream. */
942 if (bond
->dev
->flags
& IFF_PROMISC
)
943 dev_set_promiscuity(new_active
->dev
, 1);
945 if (bond
->dev
->flags
& IFF_ALLMULTI
)
946 dev_set_allmulti(new_active
->dev
, 1);
948 netdev_for_each_mc_addr(ha
, bond
->dev
)
949 dev_mc_add(new_active
->dev
, ha
->addr
);
950 bond_resend_igmp_join_requests(bond
);
955 * bond_do_fail_over_mac
957 * Perform special MAC address swapping for fail_over_mac settings
959 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
961 static void bond_do_fail_over_mac(struct bonding
*bond
,
962 struct slave
*new_active
,
963 struct slave
*old_active
)
964 __releases(&bond
->curr_slave_lock
)
965 __releases(&bond
->lock
)
966 __acquires(&bond
->lock
)
967 __acquires(&bond
->curr_slave_lock
)
969 u8 tmp_mac
[ETH_ALEN
];
970 struct sockaddr saddr
;
973 switch (bond
->params
.fail_over_mac
) {
974 case BOND_FOM_ACTIVE
:
976 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
977 new_active
->dev
->addr_len
);
979 case BOND_FOM_FOLLOW
:
981 * if new_active && old_active, swap them
982 * if just old_active, do nothing (going to no active slave)
983 * if just new_active, set new_active to bond's MAC
988 write_unlock_bh(&bond
->curr_slave_lock
);
989 read_unlock(&bond
->lock
);
992 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
993 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
995 saddr
.sa_family
= new_active
->dev
->type
;
997 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
998 saddr
.sa_family
= bond
->dev
->type
;
1001 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
1003 pr_err("%s: Error %d setting MAC of slave %s\n",
1004 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1011 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1012 saddr
.sa_family
= old_active
->dev
->type
;
1014 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
1016 pr_err("%s: Error %d setting MAC of slave %s\n",
1017 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1019 read_lock(&bond
->lock
);
1020 write_lock_bh(&bond
->curr_slave_lock
);
1023 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1024 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1030 static bool bond_should_change_active(struct bonding
*bond
)
1032 struct slave
*prim
= bond
->primary_slave
;
1033 struct slave
*curr
= bond
->curr_active_slave
;
1035 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
1037 if (bond
->force_primary
) {
1038 bond
->force_primary
= false;
1041 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
1042 (prim
->speed
< curr
->speed
||
1043 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
1045 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
1051 * find_best_interface - select the best available slave to be the active one
1052 * @bond: our bonding struct
1054 * Warning: Caller must hold curr_slave_lock for writing.
1056 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1058 struct slave
*new_active
, *old_active
;
1059 struct slave
*bestslave
= NULL
;
1060 int mintime
= bond
->params
.updelay
;
1063 new_active
= bond
->curr_active_slave
;
1065 if (!new_active
) { /* there were no active slaves left */
1066 if (bond
->slave_cnt
> 0) /* found one slave */
1067 new_active
= bond
->first_slave
;
1069 return NULL
; /* still no slave, return NULL */
1072 if ((bond
->primary_slave
) &&
1073 bond
->primary_slave
->link
== BOND_LINK_UP
&&
1074 bond_should_change_active(bond
)) {
1075 new_active
= bond
->primary_slave
;
1078 /* remember where to stop iterating over the slaves */
1079 old_active
= new_active
;
1081 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1082 if (new_active
->link
== BOND_LINK_UP
) {
1084 } else if (new_active
->link
== BOND_LINK_BACK
&&
1085 IS_UP(new_active
->dev
)) {
1086 /* link up, but waiting for stabilization */
1087 if (new_active
->delay
< mintime
) {
1088 mintime
= new_active
->delay
;
1089 bestslave
= new_active
;
1098 * change_active_interface - change the active slave into the specified one
1099 * @bond: our bonding struct
1100 * @new: the new slave to make the active one
1102 * Set the new slave to the bond's settings and unset them on the old
1103 * curr_active_slave.
1104 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1106 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1107 * because it is apparently the best available slave we have, even though its
1108 * updelay hasn't timed out yet.
1110 * If new_active is not NULL, caller must hold bond->lock for read and
1111 * curr_slave_lock for write_bh.
1113 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1115 struct slave
*old_active
= bond
->curr_active_slave
;
1117 if (old_active
== new_active
)
1121 new_active
->jiffies
= jiffies
;
1123 if (new_active
->link
== BOND_LINK_BACK
) {
1124 if (USES_PRIMARY(bond
->params
.mode
)) {
1125 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1126 bond
->dev
->name
, new_active
->dev
->name
,
1127 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1130 new_active
->delay
= 0;
1131 new_active
->link
= BOND_LINK_UP
;
1133 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1134 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1136 if (bond_is_lb(bond
))
1137 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1139 if (USES_PRIMARY(bond
->params
.mode
)) {
1140 pr_info("%s: making interface %s the new active one.\n",
1141 bond
->dev
->name
, new_active
->dev
->name
);
1146 if (USES_PRIMARY(bond
->params
.mode
))
1147 bond_mc_swap(bond
, new_active
, old_active
);
1149 if (bond_is_lb(bond
)) {
1150 bond_alb_handle_active_change(bond
, new_active
);
1152 bond_set_slave_inactive_flags(old_active
);
1154 bond_set_slave_active_flags(new_active
);
1156 bond
->curr_active_slave
= new_active
;
1159 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1161 bond_set_slave_inactive_flags(old_active
);
1164 bond_set_slave_active_flags(new_active
);
1166 if (bond
->params
.fail_over_mac
)
1167 bond_do_fail_over_mac(bond
, new_active
,
1170 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1171 bond_send_gratuitous_arp(bond
);
1173 bond
->send_unsol_na
= bond
->params
.num_unsol_na
;
1174 bond_send_unsolicited_na(bond
);
1176 write_unlock_bh(&bond
->curr_slave_lock
);
1177 read_unlock(&bond
->lock
);
1179 netdev_bonding_change(bond
->dev
, NETDEV_BONDING_FAILOVER
);
1181 read_lock(&bond
->lock
);
1182 write_lock_bh(&bond
->curr_slave_lock
);
1186 /* resend IGMP joins since all were sent on curr_active_slave */
1187 if (bond
->params
.mode
== BOND_MODE_ROUNDROBIN
) {
1188 bond_resend_igmp_join_requests(bond
);
1193 * bond_select_active_slave - select a new active slave, if needed
1194 * @bond: our bonding struct
1196 * This functions should be called when one of the following occurs:
1197 * - The old curr_active_slave has been released or lost its link.
1198 * - The primary_slave has got its link back.
1199 * - A slave has got its link back and there's no old curr_active_slave.
1201 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1203 void bond_select_active_slave(struct bonding
*bond
)
1205 struct slave
*best_slave
;
1208 best_slave
= bond_find_best_slave(bond
);
1209 if (best_slave
!= bond
->curr_active_slave
) {
1210 bond_change_active_slave(bond
, best_slave
);
1211 rv
= bond_set_carrier(bond
);
1215 if (netif_carrier_ok(bond
->dev
)) {
1216 pr_info("%s: first active interface up!\n",
1219 pr_info("%s: now running without any active interface !\n",
1225 /*--------------------------- slave list handling ---------------------------*/
1228 * This function attaches the slave to the end of list.
1230 * bond->lock held for writing by caller.
1232 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1234 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1235 new_slave
->next
= new_slave
;
1236 new_slave
->prev
= new_slave
;
1237 bond
->first_slave
= new_slave
;
1239 new_slave
->next
= bond
->first_slave
;
1240 new_slave
->prev
= bond
->first_slave
->prev
;
1241 new_slave
->next
->prev
= new_slave
;
1242 new_slave
->prev
->next
= new_slave
;
1249 * This function detaches the slave from the list.
1250 * WARNING: no check is made to verify if the slave effectively
1251 * belongs to <bond>.
1252 * Nothing is freed on return, structures are just unchained.
1253 * If any slave pointer in bond was pointing to <slave>,
1254 * it should be changed by the calling function.
1256 * bond->lock held for writing by caller.
1258 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1261 slave
->next
->prev
= slave
->prev
;
1264 slave
->prev
->next
= slave
->next
;
1266 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1267 if (bond
->slave_cnt
> 1) { /* there are more slave */
1268 bond
->first_slave
= slave
->next
;
1270 bond
->first_slave
= NULL
; /* slave was the last one */
1279 #ifdef CONFIG_NET_POLL_CONTROLLER
1281 * You must hold read lock on bond->lock before calling this.
1283 static bool slaves_support_netpoll(struct net_device
*bond_dev
)
1285 struct bonding
*bond
= netdev_priv(bond_dev
);
1286 struct slave
*slave
;
1290 bond_for_each_slave(bond
, slave
, i
) {
1291 if ((slave
->dev
->priv_flags
& IFF_DISABLE_NETPOLL
) ||
1292 !slave
->dev
->netdev_ops
->ndo_poll_controller
)
1295 return i
!= 0 && ret
;
1298 static void bond_poll_controller(struct net_device
*bond_dev
)
1300 struct net_device
*dev
= bond_dev
->npinfo
->netpoll
->real_dev
;
1301 if (dev
!= bond_dev
)
1302 netpoll_poll_dev(dev
);
1305 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1307 struct bonding
*bond
= netdev_priv(bond_dev
);
1308 struct slave
*slave
;
1309 const struct net_device_ops
*ops
;
1312 read_lock(&bond
->lock
);
1313 bond_dev
->npinfo
= NULL
;
1314 bond_for_each_slave(bond
, slave
, i
) {
1316 ops
= slave
->dev
->netdev_ops
;
1317 if (ops
->ndo_netpoll_cleanup
)
1318 ops
->ndo_netpoll_cleanup(slave
->dev
);
1320 slave
->dev
->npinfo
= NULL
;
1323 read_unlock(&bond
->lock
);
1328 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1334 /*---------------------------------- IOCTL ----------------------------------*/
1336 static int bond_sethwaddr(struct net_device
*bond_dev
,
1337 struct net_device
*slave_dev
)
1339 pr_debug("bond_dev=%p\n", bond_dev
);
1340 pr_debug("slave_dev=%p\n", slave_dev
);
1341 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1342 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1346 #define BOND_VLAN_FEATURES \
1347 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1348 NETIF_F_HW_VLAN_FILTER)
1351 * Compute the common dev->feature set available to all slaves. Some
1352 * feature bits are managed elsewhere, so preserve those feature bits
1353 * on the master device.
1355 static int bond_compute_features(struct bonding
*bond
)
1357 struct slave
*slave
;
1358 struct net_device
*bond_dev
= bond
->dev
;
1359 unsigned long features
= bond_dev
->features
;
1360 unsigned long vlan_features
= 0;
1361 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1362 bond_dev
->hard_header_len
);
1365 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1366 features
|= NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1368 if (!bond
->first_slave
)
1371 features
&= ~NETIF_F_ONE_FOR_ALL
;
1373 vlan_features
= bond
->first_slave
->dev
->vlan_features
;
1374 bond_for_each_slave(bond
, slave
, i
) {
1375 features
= netdev_increment_features(features
,
1376 slave
->dev
->features
,
1377 NETIF_F_ONE_FOR_ALL
);
1378 vlan_features
= netdev_increment_features(vlan_features
,
1379 slave
->dev
->vlan_features
,
1380 NETIF_F_ONE_FOR_ALL
);
1381 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1382 max_hard_header_len
= slave
->dev
->hard_header_len
;
1386 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1387 bond_dev
->features
= netdev_fix_features(features
, NULL
);
1388 bond_dev
->vlan_features
= netdev_fix_features(vlan_features
, NULL
);
1389 bond_dev
->hard_header_len
= max_hard_header_len
;
1394 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1395 struct net_device
*slave_dev
)
1397 struct bonding
*bond
= netdev_priv(bond_dev
);
1399 bond_dev
->header_ops
= slave_dev
->header_ops
;
1401 bond_dev
->type
= slave_dev
->type
;
1402 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1403 bond_dev
->addr_len
= slave_dev
->addr_len
;
1405 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1406 slave_dev
->addr_len
);
1407 bond
->setup_by_slave
= 1;
1410 /* enslave device <slave> to bond device <master> */
1411 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1413 struct bonding
*bond
= netdev_priv(bond_dev
);
1414 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1415 struct slave
*new_slave
= NULL
;
1416 struct netdev_hw_addr
*ha
;
1417 struct sockaddr addr
;
1419 int old_features
= bond_dev
->features
;
1422 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1423 slave_ops
->ndo_do_ioctl
== NULL
) {
1424 pr_warning("%s: Warning: no link monitoring support for %s\n",
1425 bond_dev
->name
, slave_dev
->name
);
1428 /* bond must be initialized by bond_open() before enslaving */
1429 if (!(bond_dev
->flags
& IFF_UP
)) {
1430 pr_warning("%s: master_dev is not up in bond_enslave\n",
1434 /* already enslaved */
1435 if (slave_dev
->flags
& IFF_SLAVE
) {
1436 pr_debug("Error, Device was already enslaved\n");
1440 /* vlan challenged mutual exclusion */
1441 /* no need to lock since we're protected by rtnl_lock */
1442 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1443 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1444 if (!list_empty(&bond
->vlan_list
)) {
1445 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1446 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1449 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1450 bond_dev
->name
, slave_dev
->name
,
1451 slave_dev
->name
, bond_dev
->name
);
1452 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1455 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1456 if (bond
->slave_cnt
== 0) {
1457 /* First slave, and it is not VLAN challenged,
1458 * so remove the block of adding VLANs over the bond.
1460 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1465 * Old ifenslave binaries are no longer supported. These can
1466 * be identified with moderate accuracy by the state of the slave:
1467 * the current ifenslave will set the interface down prior to
1468 * enslaving it; the old ifenslave will not.
1470 if ((slave_dev
->flags
& IFF_UP
)) {
1471 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1474 goto err_undo_flags
;
1477 /* set bonding device ether type by slave - bonding netdevices are
1478 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1479 * there is a need to override some of the type dependent attribs/funcs.
1481 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1482 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1484 if (bond
->slave_cnt
== 0) {
1485 if (bond_dev
->type
!= slave_dev
->type
) {
1486 pr_debug("%s: change device type from %d to %d\n",
1488 bond_dev
->type
, slave_dev
->type
);
1490 res
= netdev_bonding_change(bond_dev
,
1491 NETDEV_PRE_TYPE_CHANGE
);
1492 res
= notifier_to_errno(res
);
1494 pr_err("%s: refused to change device type\n",
1497 goto err_undo_flags
;
1500 /* Flush unicast and multicast addresses */
1501 dev_uc_flush(bond_dev
);
1502 dev_mc_flush(bond_dev
);
1504 if (slave_dev
->type
!= ARPHRD_ETHER
)
1505 bond_setup_by_slave(bond_dev
, slave_dev
);
1507 ether_setup(bond_dev
);
1509 netdev_bonding_change(bond_dev
,
1510 NETDEV_POST_TYPE_CHANGE
);
1512 } else if (bond_dev
->type
!= slave_dev
->type
) {
1513 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1515 slave_dev
->type
, bond_dev
->type
);
1517 goto err_undo_flags
;
1520 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1521 if (bond
->slave_cnt
== 0) {
1522 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1524 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1525 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1526 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",
1529 goto err_undo_flags
;
1533 /* If this is the first slave, then we need to set the master's hardware
1534 * address to be the same as the slave's. */
1535 if (bond
->slave_cnt
== 0)
1536 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1537 slave_dev
->addr_len
);
1540 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1543 goto err_undo_flags
;
1547 * Set the new_slave's queue_id to be zero. Queue ID mapping
1548 * is set via sysfs or module option if desired.
1550 new_slave
->queue_id
= 0;
1552 /* Save slave's original mtu and then set it to match the bond */
1553 new_slave
->original_mtu
= slave_dev
->mtu
;
1554 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1556 pr_debug("Error %d calling dev_set_mtu\n", res
);
1561 * Save slave's original ("permanent") mac address for modes
1562 * that need it, and for restoring it upon release, and then
1563 * set it to the master's address
1565 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1567 if (!bond
->params
.fail_over_mac
) {
1569 * Set slave to master's mac address. The application already
1570 * set the master's mac address to that of the first slave
1572 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1573 addr
.sa_family
= slave_dev
->type
;
1574 res
= dev_set_mac_address(slave_dev
, &addr
);
1576 pr_debug("Error %d calling set_mac_address\n", res
);
1577 goto err_restore_mtu
;
1581 res
= netdev_set_master(slave_dev
, bond_dev
);
1583 pr_debug("Error %d calling netdev_set_master\n", res
);
1584 goto err_restore_mac
;
1586 /* open the slave since the application closed it */
1587 res
= dev_open(slave_dev
);
1589 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1590 goto err_unset_master
;
1593 new_slave
->dev
= slave_dev
;
1594 slave_dev
->priv_flags
|= IFF_BONDING
;
1596 if (bond_is_lb(bond
)) {
1597 /* bond_alb_init_slave() must be called before all other stages since
1598 * it might fail and we do not want to have to undo everything
1600 res
= bond_alb_init_slave(bond
, new_slave
);
1605 /* If the mode USES_PRIMARY, then the new slave gets the
1606 * master's promisc (and mc) settings only if it becomes the
1607 * curr_active_slave, and that is taken care of later when calling
1608 * bond_change_active()
1610 if (!USES_PRIMARY(bond
->params
.mode
)) {
1611 /* set promiscuity level to new slave */
1612 if (bond_dev
->flags
& IFF_PROMISC
) {
1613 res
= dev_set_promiscuity(slave_dev
, 1);
1618 /* set allmulti level to new slave */
1619 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1620 res
= dev_set_allmulti(slave_dev
, 1);
1625 netif_addr_lock_bh(bond_dev
);
1626 /* upload master's mc_list to new slave */
1627 netdev_for_each_mc_addr(ha
, bond_dev
)
1628 dev_mc_add(slave_dev
, ha
->addr
);
1629 netif_addr_unlock_bh(bond_dev
);
1632 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1633 /* add lacpdu mc addr to mc list */
1634 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1636 dev_mc_add(slave_dev
, lacpdu_multicast
);
1639 bond_add_vlans_on_slave(bond
, slave_dev
);
1641 write_lock_bh(&bond
->lock
);
1643 bond_attach_slave(bond
, new_slave
);
1645 new_slave
->delay
= 0;
1646 new_slave
->link_failure_count
= 0;
1648 bond_compute_features(bond
);
1650 write_unlock_bh(&bond
->lock
);
1652 read_lock(&bond
->lock
);
1654 new_slave
->last_arp_rx
= jiffies
;
1656 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1657 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1659 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1661 * miimon is set but a bonded network driver
1662 * does not support ETHTOOL/MII and
1663 * arp_interval is not set. Note: if
1664 * use_carrier is enabled, we will never go
1665 * here (because netif_carrier is always
1666 * supported); thus, we don't need to change
1667 * the messages for netif_carrier.
1669 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",
1670 bond_dev
->name
, slave_dev
->name
);
1671 } else if (link_reporting
== -1) {
1672 /* unable get link status using mii/ethtool */
1673 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",
1674 bond_dev
->name
, slave_dev
->name
);
1678 /* check for initial state */
1679 if (!bond
->params
.miimon
||
1680 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1681 if (bond
->params
.updelay
) {
1682 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1683 new_slave
->link
= BOND_LINK_BACK
;
1684 new_slave
->delay
= bond
->params
.updelay
;
1686 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1687 new_slave
->link
= BOND_LINK_UP
;
1689 new_slave
->jiffies
= jiffies
;
1691 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1692 new_slave
->link
= BOND_LINK_DOWN
;
1695 if (bond_update_speed_duplex(new_slave
) &&
1696 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1697 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1698 bond_dev
->name
, new_slave
->dev
->name
);
1700 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1701 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1706 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1707 /* if there is a primary slave, remember it */
1708 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1709 bond
->primary_slave
= new_slave
;
1710 bond
->force_primary
= true;
1714 write_lock_bh(&bond
->curr_slave_lock
);
1716 switch (bond
->params
.mode
) {
1717 case BOND_MODE_ACTIVEBACKUP
:
1718 bond_set_slave_inactive_flags(new_slave
);
1719 bond_select_active_slave(bond
);
1721 case BOND_MODE_8023AD
:
1722 /* in 802.3ad mode, the internal mechanism
1723 * will activate the slaves in the selected
1726 bond_set_slave_inactive_flags(new_slave
);
1727 /* if this is the first slave */
1728 if (bond
->slave_cnt
== 1) {
1729 SLAVE_AD_INFO(new_slave
).id
= 1;
1730 /* Initialize AD with the number of times that the AD timer is called in 1 second
1731 * can be called only after the mac address of the bond is set
1733 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1734 bond
->params
.lacp_fast
);
1736 SLAVE_AD_INFO(new_slave
).id
=
1737 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1740 bond_3ad_bind_slave(new_slave
);
1744 new_slave
->state
= BOND_STATE_ACTIVE
;
1745 bond_set_slave_inactive_flags(new_slave
);
1746 bond_select_active_slave(bond
);
1749 pr_debug("This slave is always active in trunk mode\n");
1751 /* always active in trunk mode */
1752 new_slave
->state
= BOND_STATE_ACTIVE
;
1754 /* In trunking mode there is little meaning to curr_active_slave
1755 * anyway (it holds no special properties of the bond device),
1756 * so we can change it without calling change_active_interface()
1758 if (!bond
->curr_active_slave
)
1759 bond
->curr_active_slave
= new_slave
;
1762 } /* switch(bond_mode) */
1764 write_unlock_bh(&bond
->curr_slave_lock
);
1766 bond_set_carrier(bond
);
1768 #ifdef CONFIG_NET_POLL_CONTROLLER
1769 if (slaves_support_netpoll(bond_dev
)) {
1770 bond_dev
->priv_flags
&= ~IFF_DISABLE_NETPOLL
;
1771 if (bond_dev
->npinfo
)
1772 slave_dev
->npinfo
= bond_dev
->npinfo
;
1773 } else if (!(bond_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)) {
1774 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1775 pr_info("New slave device %s does not support netpoll\n",
1777 pr_info("Disabling netpoll support for %s\n", bond_dev
->name
);
1780 read_unlock(&bond
->lock
);
1782 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1786 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1787 bond_dev
->name
, slave_dev
->name
,
1788 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1789 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1791 /* enslave is successful */
1794 /* Undo stages on error */
1796 dev_close(slave_dev
);
1799 netdev_set_master(slave_dev
, NULL
);
1802 if (!bond
->params
.fail_over_mac
) {
1803 /* XXX TODO - fom follow mode needs to change master's
1804 * MAC if this slave's MAC is in use by the bond, or at
1805 * least print a warning.
1807 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1808 addr
.sa_family
= slave_dev
->type
;
1809 dev_set_mac_address(slave_dev
, &addr
);
1813 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1819 bond_dev
->features
= old_features
;
1825 * Try to release the slave device <slave> from the bond device <master>
1826 * It is legal to access curr_active_slave without a lock because all the function
1829 * The rules for slave state should be:
1830 * for Active/Backup:
1831 * Active stays on all backups go down
1832 * for Bonded connections:
1833 * The first up interface should be left on and all others downed.
1835 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1837 struct bonding
*bond
= netdev_priv(bond_dev
);
1838 struct slave
*slave
, *oldcurrent
;
1839 struct sockaddr addr
;
1841 /* slave is not a slave or master is not master of this slave */
1842 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1843 (slave_dev
->master
!= bond_dev
)) {
1844 pr_err("%s: Error: cannot release %s.\n",
1845 bond_dev
->name
, slave_dev
->name
);
1849 netdev_bonding_change(bond_dev
, NETDEV_BONDING_DESLAVE
);
1850 write_lock_bh(&bond
->lock
);
1852 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1854 /* not a slave of this bond */
1855 pr_info("%s: %s not enslaved\n",
1856 bond_dev
->name
, slave_dev
->name
);
1857 write_unlock_bh(&bond
->lock
);
1861 if (!bond
->params
.fail_over_mac
) {
1862 if (!compare_ether_addr(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1863 bond
->slave_cnt
> 1)
1864 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",
1865 bond_dev
->name
, slave_dev
->name
,
1867 bond_dev
->name
, slave_dev
->name
);
1870 /* Inform AD package of unbinding of slave. */
1871 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1872 /* must be called before the slave is
1873 * detached from the list
1875 bond_3ad_unbind_slave(slave
);
1878 pr_info("%s: releasing %s interface %s\n",
1880 (slave
->state
== BOND_STATE_ACTIVE
) ? "active" : "backup",
1883 oldcurrent
= bond
->curr_active_slave
;
1885 bond
->current_arp_slave
= NULL
;
1887 /* release the slave from its bond */
1888 bond_detach_slave(bond
, slave
);
1890 bond_compute_features(bond
);
1892 if (bond
->primary_slave
== slave
)
1893 bond
->primary_slave
= NULL
;
1895 if (oldcurrent
== slave
)
1896 bond_change_active_slave(bond
, NULL
);
1898 if (bond_is_lb(bond
)) {
1899 /* Must be called only after the slave has been
1900 * detached from the list and the curr_active_slave
1901 * has been cleared (if our_slave == old_current),
1902 * but before a new active slave is selected.
1904 write_unlock_bh(&bond
->lock
);
1905 bond_alb_deinit_slave(bond
, slave
);
1906 write_lock_bh(&bond
->lock
);
1909 if (oldcurrent
== slave
) {
1911 * Note that we hold RTNL over this sequence, so there
1912 * is no concern that another slave add/remove event
1915 write_unlock_bh(&bond
->lock
);
1916 read_lock(&bond
->lock
);
1917 write_lock_bh(&bond
->curr_slave_lock
);
1919 bond_select_active_slave(bond
);
1921 write_unlock_bh(&bond
->curr_slave_lock
);
1922 read_unlock(&bond
->lock
);
1923 write_lock_bh(&bond
->lock
);
1926 if (bond
->slave_cnt
== 0) {
1927 bond_set_carrier(bond
);
1929 /* if the last slave was removed, zero the mac address
1930 * of the master so it will be set by the application
1931 * to the mac address of the first slave
1933 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
1935 if (list_empty(&bond
->vlan_list
)) {
1936 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1938 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1939 bond_dev
->name
, bond_dev
->name
);
1940 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1943 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1944 !bond_has_challenged_slaves(bond
)) {
1945 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1946 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1947 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1950 write_unlock_bh(&bond
->lock
);
1952 /* must do this from outside any spinlocks */
1953 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1955 bond_del_vlans_from_slave(bond
, slave_dev
);
1957 /* If the mode USES_PRIMARY, then we should only remove its
1958 * promisc and mc settings if it was the curr_active_slave, but that was
1959 * already taken care of above when we detached the slave
1961 if (!USES_PRIMARY(bond
->params
.mode
)) {
1962 /* unset promiscuity level from slave */
1963 if (bond_dev
->flags
& IFF_PROMISC
)
1964 dev_set_promiscuity(slave_dev
, -1);
1966 /* unset allmulti level from slave */
1967 if (bond_dev
->flags
& IFF_ALLMULTI
)
1968 dev_set_allmulti(slave_dev
, -1);
1970 /* flush master's mc_list from slave */
1971 netif_addr_lock_bh(bond_dev
);
1972 bond_mc_list_flush(bond_dev
, slave_dev
);
1973 netif_addr_unlock_bh(bond_dev
);
1976 netdev_set_master(slave_dev
, NULL
);
1978 #ifdef CONFIG_NET_POLL_CONTROLLER
1979 read_lock_bh(&bond
->lock
);
1980 if (slaves_support_netpoll(bond_dev
))
1981 bond_dev
->priv_flags
&= ~IFF_DISABLE_NETPOLL
;
1982 read_unlock_bh(&bond
->lock
);
1983 if (slave_dev
->netdev_ops
->ndo_netpoll_cleanup
)
1984 slave_dev
->netdev_ops
->ndo_netpoll_cleanup(slave_dev
);
1986 slave_dev
->npinfo
= NULL
;
1989 /* close slave before restoring its mac address */
1990 dev_close(slave_dev
);
1992 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1993 /* restore original ("permanent") mac address */
1994 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
1995 addr
.sa_family
= slave_dev
->type
;
1996 dev_set_mac_address(slave_dev
, &addr
);
1999 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2001 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2002 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
2007 return 0; /* deletion OK */
2011 * First release a slave and than destroy the bond if no more slaves are left.
2012 * Must be under rtnl_lock when this function is called.
2014 int bond_release_and_destroy(struct net_device
*bond_dev
,
2015 struct net_device
*slave_dev
)
2017 struct bonding
*bond
= netdev_priv(bond_dev
);
2020 ret
= bond_release(bond_dev
, slave_dev
);
2021 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2022 pr_info("%s: destroying bond %s.\n",
2023 bond_dev
->name
, bond_dev
->name
);
2024 unregister_netdevice(bond_dev
);
2030 * This function releases all slaves.
2032 static int bond_release_all(struct net_device
*bond_dev
)
2034 struct bonding
*bond
= netdev_priv(bond_dev
);
2035 struct slave
*slave
;
2036 struct net_device
*slave_dev
;
2037 struct sockaddr addr
;
2039 write_lock_bh(&bond
->lock
);
2041 netif_carrier_off(bond_dev
);
2043 if (bond
->slave_cnt
== 0)
2046 bond
->current_arp_slave
= NULL
;
2047 bond
->primary_slave
= NULL
;
2048 bond_change_active_slave(bond
, NULL
);
2050 while ((slave
= bond
->first_slave
) != NULL
) {
2051 /* Inform AD package of unbinding of slave
2052 * before slave is detached from the list.
2054 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2055 bond_3ad_unbind_slave(slave
);
2057 slave_dev
= slave
->dev
;
2058 bond_detach_slave(bond
, slave
);
2060 /* now that the slave is detached, unlock and perform
2061 * all the undo steps that should not be called from
2064 write_unlock_bh(&bond
->lock
);
2066 if (bond_is_lb(bond
)) {
2067 /* must be called only after the slave
2068 * has been detached from the list
2070 bond_alb_deinit_slave(bond
, slave
);
2073 bond_compute_features(bond
);
2075 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2076 bond_del_vlans_from_slave(bond
, slave_dev
);
2078 /* If the mode USES_PRIMARY, then we should only remove its
2079 * promisc and mc settings if it was the curr_active_slave, but that was
2080 * already taken care of above when we detached the slave
2082 if (!USES_PRIMARY(bond
->params
.mode
)) {
2083 /* unset promiscuity level from slave */
2084 if (bond_dev
->flags
& IFF_PROMISC
)
2085 dev_set_promiscuity(slave_dev
, -1);
2087 /* unset allmulti level from slave */
2088 if (bond_dev
->flags
& IFF_ALLMULTI
)
2089 dev_set_allmulti(slave_dev
, -1);
2091 /* flush master's mc_list from slave */
2092 netif_addr_lock_bh(bond_dev
);
2093 bond_mc_list_flush(bond_dev
, slave_dev
);
2094 netif_addr_unlock_bh(bond_dev
);
2097 netdev_set_master(slave_dev
, NULL
);
2099 /* close slave before restoring its mac address */
2100 dev_close(slave_dev
);
2102 if (!bond
->params
.fail_over_mac
) {
2103 /* restore original ("permanent") mac address*/
2104 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2105 addr
.sa_family
= slave_dev
->type
;
2106 dev_set_mac_address(slave_dev
, &addr
);
2109 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2110 IFF_SLAVE_INACTIVE
);
2114 /* re-acquire the lock before getting the next slave */
2115 write_lock_bh(&bond
->lock
);
2118 /* zero the mac address of the master so it will be
2119 * set by the application to the mac address of the
2122 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2124 if (list_empty(&bond
->vlan_list
))
2125 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2127 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2128 bond_dev
->name
, bond_dev
->name
);
2129 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2133 pr_info("%s: released all slaves\n", bond_dev
->name
);
2136 write_unlock_bh(&bond
->lock
);
2142 * This function changes the active slave to slave <slave_dev>.
2143 * It returns -EINVAL in the following cases.
2144 * - <slave_dev> is not found in the list.
2145 * - There is not active slave now.
2146 * - <slave_dev> is already active.
2147 * - The link state of <slave_dev> is not BOND_LINK_UP.
2148 * - <slave_dev> is not running.
2149 * In these cases, this function does nothing.
2150 * In the other cases, current_slave pointer is changed and 0 is returned.
2152 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2154 struct bonding
*bond
= netdev_priv(bond_dev
);
2155 struct slave
*old_active
= NULL
;
2156 struct slave
*new_active
= NULL
;
2159 if (!USES_PRIMARY(bond
->params
.mode
))
2162 /* Verify that master_dev is indeed the master of slave_dev */
2163 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2166 read_lock(&bond
->lock
);
2168 read_lock(&bond
->curr_slave_lock
);
2169 old_active
= bond
->curr_active_slave
;
2170 read_unlock(&bond
->curr_slave_lock
);
2172 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2175 * Changing to the current active: do nothing; return success.
2177 if (new_active
&& (new_active
== old_active
)) {
2178 read_unlock(&bond
->lock
);
2184 (new_active
->link
== BOND_LINK_UP
) &&
2185 IS_UP(new_active
->dev
)) {
2186 write_lock_bh(&bond
->curr_slave_lock
);
2187 bond_change_active_slave(bond
, new_active
);
2188 write_unlock_bh(&bond
->curr_slave_lock
);
2192 read_unlock(&bond
->lock
);
2197 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2199 struct bonding
*bond
= netdev_priv(bond_dev
);
2201 info
->bond_mode
= bond
->params
.mode
;
2202 info
->miimon
= bond
->params
.miimon
;
2204 read_lock(&bond
->lock
);
2205 info
->num_slaves
= bond
->slave_cnt
;
2206 read_unlock(&bond
->lock
);
2211 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2213 struct bonding
*bond
= netdev_priv(bond_dev
);
2214 struct slave
*slave
;
2215 int i
, res
= -ENODEV
;
2217 read_lock(&bond
->lock
);
2219 bond_for_each_slave(bond
, slave
, i
) {
2220 if (i
== (int)info
->slave_id
) {
2222 strcpy(info
->slave_name
, slave
->dev
->name
);
2223 info
->link
= slave
->link
;
2224 info
->state
= slave
->state
;
2225 info
->link_failure_count
= slave
->link_failure_count
;
2230 read_unlock(&bond
->lock
);
2235 /*-------------------------------- Monitoring -------------------------------*/
2238 static int bond_miimon_inspect(struct bonding
*bond
)
2240 struct slave
*slave
;
2241 int i
, link_state
, commit
= 0;
2242 bool ignore_updelay
;
2244 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2246 bond_for_each_slave(bond
, slave
, i
) {
2247 slave
->new_link
= BOND_LINK_NOCHANGE
;
2249 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2251 switch (slave
->link
) {
2256 slave
->link
= BOND_LINK_FAIL
;
2257 slave
->delay
= bond
->params
.downdelay
;
2259 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2261 (bond
->params
.mode
==
2262 BOND_MODE_ACTIVEBACKUP
) ?
2263 ((slave
->state
== BOND_STATE_ACTIVE
) ?
2264 "active " : "backup ") : "",
2266 bond
->params
.downdelay
* bond
->params
.miimon
);
2269 case BOND_LINK_FAIL
:
2272 * recovered before downdelay expired
2274 slave
->link
= BOND_LINK_UP
;
2275 slave
->jiffies
= jiffies
;
2276 pr_info("%s: link status up again after %d ms for interface %s.\n",
2278 (bond
->params
.downdelay
- slave
->delay
) *
2279 bond
->params
.miimon
,
2284 if (slave
->delay
<= 0) {
2285 slave
->new_link
= BOND_LINK_DOWN
;
2293 case BOND_LINK_DOWN
:
2297 slave
->link
= BOND_LINK_BACK
;
2298 slave
->delay
= bond
->params
.updelay
;
2301 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2302 bond
->dev
->name
, slave
->dev
->name
,
2303 ignore_updelay
? 0 :
2304 bond
->params
.updelay
*
2305 bond
->params
.miimon
);
2308 case BOND_LINK_BACK
:
2310 slave
->link
= BOND_LINK_DOWN
;
2311 pr_info("%s: link status down again after %d ms for interface %s.\n",
2313 (bond
->params
.updelay
- slave
->delay
) *
2314 bond
->params
.miimon
,
2323 if (slave
->delay
<= 0) {
2324 slave
->new_link
= BOND_LINK_UP
;
2326 ignore_updelay
= false;
2338 static void bond_miimon_commit(struct bonding
*bond
)
2340 struct slave
*slave
;
2343 bond_for_each_slave(bond
, slave
, i
) {
2344 switch (slave
->new_link
) {
2345 case BOND_LINK_NOCHANGE
:
2349 slave
->link
= BOND_LINK_UP
;
2350 slave
->jiffies
= jiffies
;
2352 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2353 /* prevent it from being the active one */
2354 slave
->state
= BOND_STATE_BACKUP
;
2355 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2356 /* make it immediately active */
2357 slave
->state
= BOND_STATE_ACTIVE
;
2358 } else if (slave
!= bond
->primary_slave
) {
2359 /* prevent it from being the active one */
2360 slave
->state
= BOND_STATE_BACKUP
;
2363 pr_info("%s: link status definitely up for interface %s.\n",
2364 bond
->dev
->name
, slave
->dev
->name
);
2366 /* notify ad that the link status has changed */
2367 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2368 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2370 if (bond_is_lb(bond
))
2371 bond_alb_handle_link_change(bond
, slave
,
2374 if (!bond
->curr_active_slave
||
2375 (slave
== bond
->primary_slave
))
2380 case BOND_LINK_DOWN
:
2381 if (slave
->link_failure_count
< UINT_MAX
)
2382 slave
->link_failure_count
++;
2384 slave
->link
= BOND_LINK_DOWN
;
2386 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2387 bond
->params
.mode
== BOND_MODE_8023AD
)
2388 bond_set_slave_inactive_flags(slave
);
2390 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2391 bond
->dev
->name
, slave
->dev
->name
);
2393 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2394 bond_3ad_handle_link_change(slave
,
2397 if (bond_is_lb(bond
))
2398 bond_alb_handle_link_change(bond
, slave
,
2401 if (slave
== bond
->curr_active_slave
)
2407 pr_err("%s: invalid new link %d on slave %s\n",
2408 bond
->dev
->name
, slave
->new_link
,
2410 slave
->new_link
= BOND_LINK_NOCHANGE
;
2417 write_lock_bh(&bond
->curr_slave_lock
);
2418 bond_select_active_slave(bond
);
2419 write_unlock_bh(&bond
->curr_slave_lock
);
2422 bond_set_carrier(bond
);
2428 * Really a wrapper that splits the mii monitor into two phases: an
2429 * inspection, then (if inspection indicates something needs to be done)
2430 * an acquisition of appropriate locks followed by a commit phase to
2431 * implement whatever link state changes are indicated.
2433 void bond_mii_monitor(struct work_struct
*work
)
2435 struct bonding
*bond
= container_of(work
, struct bonding
,
2438 read_lock(&bond
->lock
);
2439 if (bond
->kill_timers
)
2442 if (bond
->slave_cnt
== 0)
2445 if (bond
->send_grat_arp
) {
2446 read_lock(&bond
->curr_slave_lock
);
2447 bond_send_gratuitous_arp(bond
);
2448 read_unlock(&bond
->curr_slave_lock
);
2451 if (bond
->send_unsol_na
) {
2452 read_lock(&bond
->curr_slave_lock
);
2453 bond_send_unsolicited_na(bond
);
2454 read_unlock(&bond
->curr_slave_lock
);
2457 if (bond_miimon_inspect(bond
)) {
2458 read_unlock(&bond
->lock
);
2460 read_lock(&bond
->lock
);
2462 bond_miimon_commit(bond
);
2464 read_unlock(&bond
->lock
);
2465 rtnl_unlock(); /* might sleep, hold no other locks */
2466 read_lock(&bond
->lock
);
2470 if (bond
->params
.miimon
)
2471 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2472 msecs_to_jiffies(bond
->params
.miimon
));
2474 read_unlock(&bond
->lock
);
2477 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2479 struct in_device
*idev
;
2480 struct in_ifaddr
*ifa
;
2487 idev
= __in_dev_get_rcu(dev
);
2491 ifa
= idev
->ifa_list
;
2495 addr
= ifa
->ifa_local
;
2501 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2503 struct vlan_entry
*vlan
;
2505 if (ip
== bond
->master_ip
)
2508 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2509 if (ip
== vlan
->vlan_ip
)
2517 * We go to the (large) trouble of VLAN tagging ARP frames because
2518 * switches in VLAN mode (especially if ports are configured as
2519 * "native" to a VLAN) might not pass non-tagged frames.
2521 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2523 struct sk_buff
*skb
;
2525 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2526 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2528 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2529 NULL
, slave_dev
->dev_addr
, NULL
);
2532 pr_err("ARP packet allocation failed\n");
2536 skb
= vlan_put_tag(skb
, vlan_id
);
2538 pr_err("failed to insert VLAN tag\n");
2546 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2549 __be32
*targets
= bond
->params
.arp_targets
;
2550 struct vlan_entry
*vlan
;
2551 struct net_device
*vlan_dev
;
2555 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2558 pr_debug("basa: target %x\n", targets
[i
]);
2559 if (list_empty(&bond
->vlan_list
)) {
2560 pr_debug("basa: empty vlan: arp_send\n");
2561 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2562 bond
->master_ip
, 0);
2567 * If VLANs are configured, we do a route lookup to
2568 * determine which VLAN interface would be used, so we
2569 * can tag the ARP with the proper VLAN tag.
2571 memset(&fl
, 0, sizeof(fl
));
2572 fl
.fl4_dst
= targets
[i
];
2573 fl
.fl4_tos
= RTO_ONLINK
;
2575 rv
= ip_route_output_key(dev_net(bond
->dev
), &rt
, &fl
);
2577 if (net_ratelimit()) {
2578 pr_warning("%s: no route to arp_ip_target %pI4\n",
2579 bond
->dev
->name
, &fl
.fl4_dst
);
2585 * This target is not on a VLAN
2587 if (rt
->dst
.dev
== bond
->dev
) {
2589 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2590 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2591 bond
->master_ip
, 0);
2596 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2597 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2598 if (vlan_dev
== rt
->dst
.dev
) {
2599 vlan_id
= vlan
->vlan_id
;
2600 pr_debug("basa: vlan match on %s %d\n",
2601 vlan_dev
->name
, vlan_id
);
2608 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2609 vlan
->vlan_ip
, vlan_id
);
2613 if (net_ratelimit()) {
2614 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2615 bond
->dev
->name
, &fl
.fl4_dst
,
2616 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2623 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2624 * for each VLAN above us.
2626 * Caller must hold curr_slave_lock for read or better
2628 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2630 struct slave
*slave
= bond
->curr_active_slave
;
2631 struct vlan_entry
*vlan
;
2632 struct net_device
*vlan_dev
;
2634 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2635 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
2637 if (!slave
|| !bond
->send_grat_arp
||
2638 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2641 bond
->send_grat_arp
--;
2643 if (bond
->master_ip
) {
2644 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2645 bond
->master_ip
, 0);
2648 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2649 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2650 if (vlan
->vlan_ip
) {
2651 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2652 vlan
->vlan_ip
, vlan
->vlan_id
);
2657 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2660 __be32
*targets
= bond
->params
.arp_targets
;
2662 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2663 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2664 &sip
, &tip
, i
, &targets
[i
],
2665 bond_has_this_ip(bond
, tip
));
2666 if (sip
== targets
[i
]) {
2667 if (bond_has_this_ip(bond
, tip
))
2668 slave
->last_arp_rx
= jiffies
;
2674 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2677 struct slave
*slave
;
2678 struct bonding
*bond
;
2679 unsigned char *arp_ptr
;
2682 if (dev
->priv_flags
& IFF_802_1Q_VLAN
) {
2684 * When using VLANS and bonding, dev and oriv_dev may be
2685 * incorrect if the physical interface supports VLAN
2686 * acceleration. With this change ARP validation now
2687 * works for hosts only reachable on the VLAN interface.
2689 dev
= vlan_dev_real_dev(dev
);
2690 orig_dev
= dev_get_by_index_rcu(dev_net(skb
->dev
),skb
->skb_iif
);
2693 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2696 bond
= netdev_priv(dev
);
2697 read_lock(&bond
->lock
);
2699 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2700 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2701 orig_dev
? orig_dev
->name
: "NULL");
2703 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2704 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2707 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2711 if (arp
->ar_hln
!= dev
->addr_len
||
2712 skb
->pkt_type
== PACKET_OTHERHOST
||
2713 skb
->pkt_type
== PACKET_LOOPBACK
||
2714 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2715 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2719 arp_ptr
= (unsigned char *)(arp
+ 1);
2720 arp_ptr
+= dev
->addr_len
;
2721 memcpy(&sip
, arp_ptr
, 4);
2722 arp_ptr
+= 4 + dev
->addr_len
;
2723 memcpy(&tip
, arp_ptr
, 4);
2725 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2726 bond
->dev
->name
, slave
->dev
->name
, slave
->state
,
2727 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2731 * Backup slaves won't see the ARP reply, but do come through
2732 * here for each ARP probe (so we swap the sip/tip to validate
2733 * the probe). In a "redundant switch, common router" type of
2734 * configuration, the ARP probe will (hopefully) travel from
2735 * the active, through one switch, the router, then the other
2736 * switch before reaching the backup.
2738 if (slave
->state
== BOND_STATE_ACTIVE
)
2739 bond_validate_arp(bond
, slave
, sip
, tip
);
2741 bond_validate_arp(bond
, slave
, tip
, sip
);
2744 read_unlock(&bond
->lock
);
2747 return NET_RX_SUCCESS
;
2751 * this function is called regularly to monitor each slave's link
2752 * ensuring that traffic is being sent and received when arp monitoring
2753 * is used in load-balancing mode. if the adapter has been dormant, then an
2754 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2755 * arp monitoring in active backup mode.
2757 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2759 struct bonding
*bond
= container_of(work
, struct bonding
,
2761 struct slave
*slave
, *oldcurrent
;
2762 int do_failover
= 0;
2766 read_lock(&bond
->lock
);
2768 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2770 if (bond
->kill_timers
)
2773 if (bond
->slave_cnt
== 0)
2776 read_lock(&bond
->curr_slave_lock
);
2777 oldcurrent
= bond
->curr_active_slave
;
2778 read_unlock(&bond
->curr_slave_lock
);
2780 /* see if any of the previous devices are up now (i.e. they have
2781 * xmt and rcv traffic). the curr_active_slave does not come into
2782 * the picture unless it is null. also, slave->jiffies is not needed
2783 * here because we send an arp on each slave and give a slave as
2784 * long as it needs to get the tx/rx within the delta.
2785 * TODO: what about up/down delay in arp mode? it wasn't here before
2788 bond_for_each_slave(bond
, slave
, i
) {
2789 if (slave
->link
!= BOND_LINK_UP
) {
2790 if (time_before_eq(jiffies
, dev_trans_start(slave
->dev
) + delta_in_ticks
) &&
2791 time_before_eq(jiffies
, slave
->dev
->last_rx
+ delta_in_ticks
)) {
2793 slave
->link
= BOND_LINK_UP
;
2794 slave
->state
= BOND_STATE_ACTIVE
;
2796 /* primary_slave has no meaning in round-robin
2797 * mode. the window of a slave being up and
2798 * curr_active_slave being null after enslaving
2802 pr_info("%s: link status definitely up for interface %s, ",
2807 pr_info("%s: interface %s is now up\n",
2813 /* slave->link == BOND_LINK_UP */
2815 /* not all switches will respond to an arp request
2816 * when the source ip is 0, so don't take the link down
2817 * if we don't know our ip yet
2819 if (time_after_eq(jiffies
, dev_trans_start(slave
->dev
) + 2*delta_in_ticks
) ||
2820 (time_after_eq(jiffies
, slave
->dev
->last_rx
+ 2*delta_in_ticks
))) {
2822 slave
->link
= BOND_LINK_DOWN
;
2823 slave
->state
= BOND_STATE_BACKUP
;
2825 if (slave
->link_failure_count
< UINT_MAX
)
2826 slave
->link_failure_count
++;
2828 pr_info("%s: interface %s is now down.\n",
2832 if (slave
== oldcurrent
)
2837 /* note: if switch is in round-robin mode, all links
2838 * must tx arp to ensure all links rx an arp - otherwise
2839 * links may oscillate or not come up at all; if switch is
2840 * in something like xor mode, there is nothing we can
2841 * do - all replies will be rx'ed on same link causing slaves
2842 * to be unstable during low/no traffic periods
2844 if (IS_UP(slave
->dev
))
2845 bond_arp_send_all(bond
, slave
);
2849 write_lock_bh(&bond
->curr_slave_lock
);
2851 bond_select_active_slave(bond
);
2853 write_unlock_bh(&bond
->curr_slave_lock
);
2857 if (bond
->params
.arp_interval
)
2858 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2860 read_unlock(&bond
->lock
);
2864 * Called to inspect slaves for active-backup mode ARP monitor link state
2865 * changes. Sets new_link in slaves to specify what action should take
2866 * place for the slave. Returns 0 if no changes are found, >0 if changes
2867 * to link states must be committed.
2869 * Called with bond->lock held for read.
2871 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2873 struct slave
*slave
;
2876 bond_for_each_slave(bond
, slave
, i
) {
2877 slave
->new_link
= BOND_LINK_NOCHANGE
;
2879 if (slave
->link
!= BOND_LINK_UP
) {
2880 if (time_before_eq(jiffies
, slave_last_rx(bond
, slave
) +
2882 slave
->new_link
= BOND_LINK_UP
;
2890 * Give slaves 2*delta after being enslaved or made
2891 * active. This avoids bouncing, as the last receive
2892 * times need a full ARP monitor cycle to be updated.
2894 if (!time_after_eq(jiffies
, slave
->jiffies
+
2895 2 * delta_in_ticks
))
2899 * Backup slave is down if:
2900 * - No current_arp_slave AND
2901 * - more than 3*delta since last receive AND
2902 * - the bond has an IP address
2904 * Note: a non-null current_arp_slave indicates
2905 * the curr_active_slave went down and we are
2906 * searching for a new one; under this condition
2907 * we only take the curr_active_slave down - this
2908 * gives each slave a chance to tx/rx traffic
2909 * before being taken out
2911 if (slave
->state
== BOND_STATE_BACKUP
&&
2912 !bond
->current_arp_slave
&&
2913 time_after(jiffies
, slave_last_rx(bond
, slave
) +
2914 3 * delta_in_ticks
)) {
2915 slave
->new_link
= BOND_LINK_DOWN
;
2920 * Active slave is down if:
2921 * - more than 2*delta since transmitting OR
2922 * - (more than 2*delta since receive AND
2923 * the bond has an IP address)
2925 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
2926 (time_after_eq(jiffies
, dev_trans_start(slave
->dev
) +
2927 2 * delta_in_ticks
) ||
2928 (time_after_eq(jiffies
, slave_last_rx(bond
, slave
)
2929 + 2 * delta_in_ticks
)))) {
2930 slave
->new_link
= BOND_LINK_DOWN
;
2939 * Called to commit link state changes noted by inspection step of
2940 * active-backup mode ARP monitor.
2942 * Called with RTNL and bond->lock for read.
2944 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2946 struct slave
*slave
;
2949 bond_for_each_slave(bond
, slave
, i
) {
2950 switch (slave
->new_link
) {
2951 case BOND_LINK_NOCHANGE
:
2955 if ((!bond
->curr_active_slave
&&
2956 time_before_eq(jiffies
,
2957 dev_trans_start(slave
->dev
) +
2959 bond
->curr_active_slave
!= slave
) {
2960 slave
->link
= BOND_LINK_UP
;
2961 bond
->current_arp_slave
= NULL
;
2963 pr_info("%s: link status definitely up for interface %s.\n",
2964 bond
->dev
->name
, slave
->dev
->name
);
2966 if (!bond
->curr_active_slave
||
2967 (slave
== bond
->primary_slave
))
2974 case BOND_LINK_DOWN
:
2975 if (slave
->link_failure_count
< UINT_MAX
)
2976 slave
->link_failure_count
++;
2978 slave
->link
= BOND_LINK_DOWN
;
2979 bond_set_slave_inactive_flags(slave
);
2981 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2982 bond
->dev
->name
, slave
->dev
->name
);
2984 if (slave
== bond
->curr_active_slave
) {
2985 bond
->current_arp_slave
= NULL
;
2992 pr_err("%s: impossible: new_link %d on slave %s\n",
2993 bond
->dev
->name
, slave
->new_link
,
3000 write_lock_bh(&bond
->curr_slave_lock
);
3001 bond_select_active_slave(bond
);
3002 write_unlock_bh(&bond
->curr_slave_lock
);
3005 bond_set_carrier(bond
);
3009 * Send ARP probes for active-backup mode ARP monitor.
3011 * Called with bond->lock held for read.
3013 static void bond_ab_arp_probe(struct bonding
*bond
)
3015 struct slave
*slave
;
3018 read_lock(&bond
->curr_slave_lock
);
3020 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3021 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3022 bond
->current_arp_slave
->dev
->name
,
3023 bond
->curr_active_slave
->dev
->name
);
3025 if (bond
->curr_active_slave
) {
3026 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3027 read_unlock(&bond
->curr_slave_lock
);
3031 read_unlock(&bond
->curr_slave_lock
);
3033 /* if we don't have a curr_active_slave, search for the next available
3034 * backup slave from the current_arp_slave and make it the candidate
3035 * for becoming the curr_active_slave
3038 if (!bond
->current_arp_slave
) {
3039 bond
->current_arp_slave
= bond
->first_slave
;
3040 if (!bond
->current_arp_slave
)
3044 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3046 /* search for next candidate */
3047 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3048 if (IS_UP(slave
->dev
)) {
3049 slave
->link
= BOND_LINK_BACK
;
3050 bond_set_slave_active_flags(slave
);
3051 bond_arp_send_all(bond
, slave
);
3052 slave
->jiffies
= jiffies
;
3053 bond
->current_arp_slave
= slave
;
3057 /* if the link state is up at this point, we
3058 * mark it down - this can happen if we have
3059 * simultaneous link failures and
3060 * reselect_active_interface doesn't make this
3061 * one the current slave so it is still marked
3062 * up when it is actually down
3064 if (slave
->link
== BOND_LINK_UP
) {
3065 slave
->link
= BOND_LINK_DOWN
;
3066 if (slave
->link_failure_count
< UINT_MAX
)
3067 slave
->link_failure_count
++;
3069 bond_set_slave_inactive_flags(slave
);
3071 pr_info("%s: backup interface %s is now down.\n",
3072 bond
->dev
->name
, slave
->dev
->name
);
3077 void bond_activebackup_arp_mon(struct work_struct
*work
)
3079 struct bonding
*bond
= container_of(work
, struct bonding
,
3083 read_lock(&bond
->lock
);
3085 if (bond
->kill_timers
)
3088 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3090 if (bond
->slave_cnt
== 0)
3093 if (bond
->send_grat_arp
) {
3094 read_lock(&bond
->curr_slave_lock
);
3095 bond_send_gratuitous_arp(bond
);
3096 read_unlock(&bond
->curr_slave_lock
);
3099 if (bond
->send_unsol_na
) {
3100 read_lock(&bond
->curr_slave_lock
);
3101 bond_send_unsolicited_na(bond
);
3102 read_unlock(&bond
->curr_slave_lock
);
3105 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3106 read_unlock(&bond
->lock
);
3108 read_lock(&bond
->lock
);
3110 bond_ab_arp_commit(bond
, delta_in_ticks
);
3112 read_unlock(&bond
->lock
);
3114 read_lock(&bond
->lock
);
3117 bond_ab_arp_probe(bond
);
3120 if (bond
->params
.arp_interval
)
3121 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3123 read_unlock(&bond
->lock
);
3126 /*------------------------------ proc/seq_file-------------------------------*/
3128 #ifdef CONFIG_PROC_FS
3130 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3131 __acquires(&dev_base_lock
)
3132 __acquires(&bond
->lock
)
3134 struct bonding
*bond
= seq
->private;
3136 struct slave
*slave
;
3139 /* make sure the bond won't be taken away */
3140 read_lock(&dev_base_lock
);
3141 read_lock(&bond
->lock
);
3144 return SEQ_START_TOKEN
;
3146 bond_for_each_slave(bond
, slave
, i
) {
3154 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3156 struct bonding
*bond
= seq
->private;
3157 struct slave
*slave
= v
;
3160 if (v
== SEQ_START_TOKEN
)
3161 return bond
->first_slave
;
3163 slave
= slave
->next
;
3165 return (slave
== bond
->first_slave
) ? NULL
: slave
;
3168 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
3169 __releases(&bond
->lock
)
3170 __releases(&dev_base_lock
)
3172 struct bonding
*bond
= seq
->private;
3174 read_unlock(&bond
->lock
);
3175 read_unlock(&dev_base_lock
);
3178 static void bond_info_show_master(struct seq_file
*seq
)
3180 struct bonding
*bond
= seq
->private;
3184 read_lock(&bond
->curr_slave_lock
);
3185 curr
= bond
->curr_active_slave
;
3186 read_unlock(&bond
->curr_slave_lock
);
3188 seq_printf(seq
, "Bonding Mode: %s",
3189 bond_mode_name(bond
->params
.mode
));
3191 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
&&
3192 bond
->params
.fail_over_mac
)
3193 seq_printf(seq
, " (fail_over_mac %s)",
3194 fail_over_mac_tbl
[bond
->params
.fail_over_mac
].modename
);
3196 seq_printf(seq
, "\n");
3198 if (bond
->params
.mode
== BOND_MODE_XOR
||
3199 bond
->params
.mode
== BOND_MODE_8023AD
) {
3200 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
3201 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
3202 bond
->params
.xmit_policy
);
3205 if (USES_PRIMARY(bond
->params
.mode
)) {
3206 seq_printf(seq
, "Primary Slave: %s",
3207 (bond
->primary_slave
) ?
3208 bond
->primary_slave
->dev
->name
: "None");
3209 if (bond
->primary_slave
)
3210 seq_printf(seq
, " (primary_reselect %s)",
3211 pri_reselect_tbl
[bond
->params
.primary_reselect
].modename
);
3213 seq_printf(seq
, "\nCurrently Active Slave: %s\n",
3214 (curr
) ? curr
->dev
->name
: "None");
3217 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
3219 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
3220 seq_printf(seq
, "Up Delay (ms): %d\n",
3221 bond
->params
.updelay
* bond
->params
.miimon
);
3222 seq_printf(seq
, "Down Delay (ms): %d\n",
3223 bond
->params
.downdelay
* bond
->params
.miimon
);
3226 /* ARP information */
3227 if (bond
->params
.arp_interval
> 0) {
3229 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
3230 bond
->params
.arp_interval
);
3232 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
3234 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
3235 if (!bond
->params
.arp_targets
[i
])
3238 seq_printf(seq
, ",");
3239 seq_printf(seq
, " %pI4", &bond
->params
.arp_targets
[i
]);
3242 seq_printf(seq
, "\n");
3245 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3246 struct ad_info ad_info
;
3248 seq_puts(seq
, "\n802.3ad info\n");
3249 seq_printf(seq
, "LACP rate: %s\n",
3250 (bond
->params
.lacp_fast
) ? "fast" : "slow");
3251 seq_printf(seq
, "Aggregator selection policy (ad_select): %s\n",
3252 ad_select_tbl
[bond
->params
.ad_select
].modename
);
3254 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3255 seq_printf(seq
, "bond %s has no active aggregator\n",
3258 seq_printf(seq
, "Active Aggregator Info:\n");
3260 seq_printf(seq
, "\tAggregator ID: %d\n",
3261 ad_info
.aggregator_id
);
3262 seq_printf(seq
, "\tNumber of ports: %d\n",
3264 seq_printf(seq
, "\tActor Key: %d\n",
3266 seq_printf(seq
, "\tPartner Key: %d\n",
3267 ad_info
.partner_key
);
3268 seq_printf(seq
, "\tPartner Mac Address: %pM\n",
3269 ad_info
.partner_system
);
3274 static void bond_info_show_slave(struct seq_file
*seq
,
3275 const struct slave
*slave
)
3277 struct bonding
*bond
= seq
->private;
3279 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
3280 seq_printf(seq
, "MII Status: %s\n",
3281 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
3282 seq_printf(seq
, "Link Failure Count: %u\n",
3283 slave
->link_failure_count
);
3285 seq_printf(seq
, "Permanent HW addr: %pM\n", slave
->perm_hwaddr
);
3287 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3288 const struct aggregator
*agg
3289 = SLAVE_AD_INFO(slave
).port
.aggregator
;
3292 seq_printf(seq
, "Aggregator ID: %d\n",
3293 agg
->aggregator_identifier
);
3295 seq_puts(seq
, "Aggregator ID: N/A\n");
3297 seq_printf(seq
, "Slave queue ID: %d\n", slave
->queue_id
);
3300 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
3302 if (v
== SEQ_START_TOKEN
) {
3303 seq_printf(seq
, "%s\n", version
);
3304 bond_info_show_master(seq
);
3306 bond_info_show_slave(seq
, v
);
3311 static const struct seq_operations bond_info_seq_ops
= {
3312 .start
= bond_info_seq_start
,
3313 .next
= bond_info_seq_next
,
3314 .stop
= bond_info_seq_stop
,
3315 .show
= bond_info_seq_show
,
3318 static int bond_info_open(struct inode
*inode
, struct file
*file
)
3320 struct seq_file
*seq
;
3321 struct proc_dir_entry
*proc
;
3324 res
= seq_open(file
, &bond_info_seq_ops
);
3326 /* recover the pointer buried in proc_dir_entry data */
3327 seq
= file
->private_data
;
3329 seq
->private = proc
->data
;
3335 static const struct file_operations bond_info_fops
= {
3336 .owner
= THIS_MODULE
,
3337 .open
= bond_info_open
,
3339 .llseek
= seq_lseek
,
3340 .release
= seq_release
,
3343 static void bond_create_proc_entry(struct bonding
*bond
)
3345 struct net_device
*bond_dev
= bond
->dev
;
3346 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
3349 bond
->proc_entry
= proc_create_data(bond_dev
->name
,
3350 S_IRUGO
, bn
->proc_dir
,
3351 &bond_info_fops
, bond
);
3352 if (bond
->proc_entry
== NULL
)
3353 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3354 DRV_NAME
, bond_dev
->name
);
3356 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3360 static void bond_remove_proc_entry(struct bonding
*bond
)
3362 struct net_device
*bond_dev
= bond
->dev
;
3363 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
3365 if (bn
->proc_dir
&& bond
->proc_entry
) {
3366 remove_proc_entry(bond
->proc_file_name
, bn
->proc_dir
);
3367 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3368 bond
->proc_entry
= NULL
;
3372 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3373 * Caller must hold rtnl_lock.
3375 static void __net_init
bond_create_proc_dir(struct bond_net
*bn
)
3377 if (!bn
->proc_dir
) {
3378 bn
->proc_dir
= proc_mkdir(DRV_NAME
, bn
->net
->proc_net
);
3380 pr_warning("Warning: cannot create /proc/net/%s\n",
3385 /* Destroy the bonding directory under /proc/net, if empty.
3386 * Caller must hold rtnl_lock.
3388 static void __net_exit
bond_destroy_proc_dir(struct bond_net
*bn
)
3391 remove_proc_entry(DRV_NAME
, bn
->net
->proc_net
);
3392 bn
->proc_dir
= NULL
;
3396 #else /* !CONFIG_PROC_FS */
3398 static void bond_create_proc_entry(struct bonding
*bond
)
3402 static void bond_remove_proc_entry(struct bonding
*bond
)
3406 static inline void bond_create_proc_dir(struct bond_net
*bn
)
3410 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
3414 #endif /* CONFIG_PROC_FS */
3417 /*-------------------------- netdev event handling --------------------------*/
3420 * Change device name
3422 static int bond_event_changename(struct bonding
*bond
)
3424 bond_remove_proc_entry(bond
);
3425 bond_create_proc_entry(bond
);
3430 static int bond_master_netdev_event(unsigned long event
,
3431 struct net_device
*bond_dev
)
3433 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3436 case NETDEV_CHANGENAME
:
3437 return bond_event_changename(event_bond
);
3445 static int bond_slave_netdev_event(unsigned long event
,
3446 struct net_device
*slave_dev
)
3448 struct net_device
*bond_dev
= slave_dev
->master
;
3449 struct bonding
*bond
= netdev_priv(bond_dev
);
3452 case NETDEV_UNREGISTER
:
3454 if (bond
->setup_by_slave
)
3455 bond_release_and_destroy(bond_dev
, slave_dev
);
3457 bond_release(bond_dev
, slave_dev
);
3461 if (bond
->params
.mode
== BOND_MODE_8023AD
|| bond_is_lb(bond
)) {
3462 struct slave
*slave
;
3464 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3466 u16 old_speed
= slave
->speed
;
3467 u16 old_duplex
= slave
->duplex
;
3469 bond_update_speed_duplex(slave
);
3471 if (bond_is_lb(bond
))
3474 if (old_speed
!= slave
->speed
)
3475 bond_3ad_adapter_speed_changed(slave
);
3476 if (old_duplex
!= slave
->duplex
)
3477 bond_3ad_adapter_duplex_changed(slave
);
3484 * ... Or is it this?
3487 case NETDEV_CHANGEMTU
:
3489 * TODO: Should slaves be allowed to
3490 * independently alter their MTU? For
3491 * an active-backup bond, slaves need
3492 * not be the same type of device, so
3493 * MTUs may vary. For other modes,
3494 * slaves arguably should have the
3495 * same MTUs. To do this, we'd need to
3496 * take over the slave's change_mtu
3497 * function for the duration of their
3501 case NETDEV_CHANGENAME
:
3503 * TODO: handle changing the primary's name
3506 case NETDEV_FEAT_CHANGE
:
3507 bond_compute_features(bond
);
3517 * bond_netdev_event: handle netdev notifier chain events.
3519 * This function receives events for the netdev chain. The caller (an
3520 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3521 * locks for us to safely manipulate the slave devices (RTNL lock,
3524 static int bond_netdev_event(struct notifier_block
*this,
3525 unsigned long event
, void *ptr
)
3527 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3529 pr_debug("event_dev: %s, event: %lx\n",
3530 event_dev
? event_dev
->name
: "None",
3533 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3536 if (event_dev
->flags
& IFF_MASTER
) {
3537 pr_debug("IFF_MASTER\n");
3538 return bond_master_netdev_event(event
, event_dev
);
3541 if (event_dev
->flags
& IFF_SLAVE
) {
3542 pr_debug("IFF_SLAVE\n");
3543 return bond_slave_netdev_event(event
, event_dev
);
3550 * bond_inetaddr_event: handle inetaddr notifier chain events.
3552 * We keep track of device IPs primarily to use as source addresses in
3553 * ARP monitor probes (rather than spewing out broadcasts all the time).
3555 * We track one IP for the main device (if it has one), plus one per VLAN.
3557 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3559 struct in_ifaddr
*ifa
= ptr
;
3560 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3561 struct bond_net
*bn
= net_generic(dev_net(event_dev
), bond_net_id
);
3562 struct bonding
*bond
;
3563 struct vlan_entry
*vlan
;
3565 list_for_each_entry(bond
, &bn
->dev_list
, bond_list
) {
3566 if (bond
->dev
== event_dev
) {
3569 bond
->master_ip
= ifa
->ifa_local
;
3572 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3579 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3580 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3581 if (vlan_dev
== event_dev
) {
3584 vlan
->vlan_ip
= ifa
->ifa_local
;
3588 bond_glean_dev_ip(vlan_dev
);
3599 static struct notifier_block bond_netdev_notifier
= {
3600 .notifier_call
= bond_netdev_event
,
3603 static struct notifier_block bond_inetaddr_notifier
= {
3604 .notifier_call
= bond_inetaddr_event
,
3607 /*-------------------------- Packet type handling ---------------------------*/
3609 /* register to receive lacpdus on a bond */
3610 static void bond_register_lacpdu(struct bonding
*bond
)
3612 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3614 /* initialize packet type */
3615 pk_type
->type
= PKT_TYPE_LACPDU
;
3616 pk_type
->dev
= bond
->dev
;
3617 pk_type
->func
= bond_3ad_lacpdu_recv
;
3619 dev_add_pack(pk_type
);
3622 /* unregister to receive lacpdus on a bond */
3623 static void bond_unregister_lacpdu(struct bonding
*bond
)
3625 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3628 void bond_register_arp(struct bonding
*bond
)
3630 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3635 pt
->type
= htons(ETH_P_ARP
);
3636 pt
->dev
= bond
->dev
;
3637 pt
->func
= bond_arp_rcv
;
3641 void bond_unregister_arp(struct bonding
*bond
)
3643 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3645 dev_remove_pack(pt
);
3649 /*---------------------------- Hashing Policies -----------------------------*/
3652 * Hash for the output device based upon layer 2 and layer 3 data. If
3653 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3655 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3657 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3658 struct iphdr
*iph
= ip_hdr(skb
);
3660 if (skb
->protocol
== htons(ETH_P_IP
)) {
3661 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3662 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3665 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3669 * Hash for the output device based upon layer 3 and layer 4 data. If
3670 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3671 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3673 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3675 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3676 struct iphdr
*iph
= ip_hdr(skb
);
3677 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3680 if (skb
->protocol
== htons(ETH_P_IP
)) {
3681 if (!(iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) &&
3682 (iph
->protocol
== IPPROTO_TCP
||
3683 iph
->protocol
== IPPROTO_UDP
)) {
3684 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3686 return (layer4_xor
^
3687 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3691 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3695 * Hash for the output device based upon layer 2 data
3697 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3699 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3701 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3704 /*-------------------------- Device entry points ----------------------------*/
3706 static int bond_open(struct net_device
*bond_dev
)
3708 struct bonding
*bond
= netdev_priv(bond_dev
);
3710 bond
->kill_timers
= 0;
3712 if (bond_is_lb(bond
)) {
3713 /* bond_alb_initialize must be called before the timer
3716 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3717 /* something went wrong - fail the open operation */
3721 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3722 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3725 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3726 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3727 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3730 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3731 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3732 INIT_DELAYED_WORK(&bond
->arp_work
,
3733 bond_activebackup_arp_mon
);
3735 INIT_DELAYED_WORK(&bond
->arp_work
,
3736 bond_loadbalance_arp_mon
);
3738 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3739 if (bond
->params
.arp_validate
)
3740 bond_register_arp(bond
);
3743 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3744 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3745 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3746 /* register to receive LACPDUs */
3747 bond_register_lacpdu(bond
);
3748 bond_3ad_initiate_agg_selection(bond
, 1);
3754 static int bond_close(struct net_device
*bond_dev
)
3756 struct bonding
*bond
= netdev_priv(bond_dev
);
3758 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3759 /* Unregister the receive of LACPDUs */
3760 bond_unregister_lacpdu(bond
);
3763 if (bond
->params
.arp_validate
)
3764 bond_unregister_arp(bond
);
3766 write_lock_bh(&bond
->lock
);
3768 bond
->send_grat_arp
= 0;
3769 bond
->send_unsol_na
= 0;
3771 /* signal timers not to re-arm */
3772 bond
->kill_timers
= 1;
3774 write_unlock_bh(&bond
->lock
);
3776 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3777 cancel_delayed_work(&bond
->mii_work
);
3780 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3781 cancel_delayed_work(&bond
->arp_work
);
3784 switch (bond
->params
.mode
) {
3785 case BOND_MODE_8023AD
:
3786 cancel_delayed_work(&bond
->ad_work
);
3790 cancel_delayed_work(&bond
->alb_work
);
3797 if (bond_is_lb(bond
)) {
3798 /* Must be called only after all
3799 * slaves have been released
3801 bond_alb_deinitialize(bond
);
3807 static struct net_device_stats
*bond_get_stats(struct net_device
*bond_dev
)
3809 struct bonding
*bond
= netdev_priv(bond_dev
);
3810 struct net_device_stats
*stats
= &bond_dev
->stats
;
3811 struct net_device_stats local_stats
;
3812 struct slave
*slave
;
3815 memset(&local_stats
, 0, sizeof(struct net_device_stats
));
3817 read_lock_bh(&bond
->lock
);
3819 bond_for_each_slave(bond
, slave
, i
) {
3820 const struct net_device_stats
*sstats
= dev_get_stats(slave
->dev
);
3822 local_stats
.rx_packets
+= sstats
->rx_packets
;
3823 local_stats
.rx_bytes
+= sstats
->rx_bytes
;
3824 local_stats
.rx_errors
+= sstats
->rx_errors
;
3825 local_stats
.rx_dropped
+= sstats
->rx_dropped
;
3827 local_stats
.tx_packets
+= sstats
->tx_packets
;
3828 local_stats
.tx_bytes
+= sstats
->tx_bytes
;
3829 local_stats
.tx_errors
+= sstats
->tx_errors
;
3830 local_stats
.tx_dropped
+= sstats
->tx_dropped
;
3832 local_stats
.multicast
+= sstats
->multicast
;
3833 local_stats
.collisions
+= sstats
->collisions
;
3835 local_stats
.rx_length_errors
+= sstats
->rx_length_errors
;
3836 local_stats
.rx_over_errors
+= sstats
->rx_over_errors
;
3837 local_stats
.rx_crc_errors
+= sstats
->rx_crc_errors
;
3838 local_stats
.rx_frame_errors
+= sstats
->rx_frame_errors
;
3839 local_stats
.rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3840 local_stats
.rx_missed_errors
+= sstats
->rx_missed_errors
;
3842 local_stats
.tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3843 local_stats
.tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3844 local_stats
.tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3845 local_stats
.tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3846 local_stats
.tx_window_errors
+= sstats
->tx_window_errors
;
3849 memcpy(stats
, &local_stats
, sizeof(struct net_device_stats
));
3851 read_unlock_bh(&bond
->lock
);
3856 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3858 struct net_device
*slave_dev
= NULL
;
3859 struct ifbond k_binfo
;
3860 struct ifbond __user
*u_binfo
= NULL
;
3861 struct ifslave k_sinfo
;
3862 struct ifslave __user
*u_sinfo
= NULL
;
3863 struct mii_ioctl_data
*mii
= NULL
;
3866 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3878 * We do this again just in case we were called by SIOCGMIIREG
3879 * instead of SIOCGMIIPHY.
3886 if (mii
->reg_num
== 1) {
3887 struct bonding
*bond
= netdev_priv(bond_dev
);
3889 read_lock(&bond
->lock
);
3890 read_lock(&bond
->curr_slave_lock
);
3891 if (netif_carrier_ok(bond
->dev
))
3892 mii
->val_out
= BMSR_LSTATUS
;
3894 read_unlock(&bond
->curr_slave_lock
);
3895 read_unlock(&bond
->lock
);
3899 case BOND_INFO_QUERY_OLD
:
3900 case SIOCBONDINFOQUERY
:
3901 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3903 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3906 res
= bond_info_query(bond_dev
, &k_binfo
);
3908 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3912 case BOND_SLAVE_INFO_QUERY_OLD
:
3913 case SIOCBONDSLAVEINFOQUERY
:
3914 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3916 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3919 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3921 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3930 if (!capable(CAP_NET_ADMIN
))
3933 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
3935 pr_debug("slave_dev=%p:\n", slave_dev
);
3940 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3942 case BOND_ENSLAVE_OLD
:
3943 case SIOCBONDENSLAVE
:
3944 res
= bond_enslave(bond_dev
, slave_dev
);
3946 case BOND_RELEASE_OLD
:
3947 case SIOCBONDRELEASE
:
3948 res
= bond_release(bond_dev
, slave_dev
);
3950 case BOND_SETHWADDR_OLD
:
3951 case SIOCBONDSETHWADDR
:
3952 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3954 case BOND_CHANGE_ACTIVE_OLD
:
3955 case SIOCBONDCHANGEACTIVE
:
3956 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3968 static bool bond_addr_in_mc_list(unsigned char *addr
,
3969 struct netdev_hw_addr_list
*list
,
3972 struct netdev_hw_addr
*ha
;
3974 netdev_hw_addr_list_for_each(ha
, list
)
3975 if (!memcmp(ha
->addr
, addr
, addrlen
))
3981 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3983 struct bonding
*bond
= netdev_priv(bond_dev
);
3984 struct netdev_hw_addr
*ha
;
3988 * Do promisc before checking multicast_mode
3990 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
))
3992 * FIXME: Need to handle the error when one of the multi-slaves
3995 bond_set_promiscuity(bond
, 1);
3998 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
))
3999 bond_set_promiscuity(bond
, -1);
4002 /* set allmulti flag to slaves */
4003 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
))
4005 * FIXME: Need to handle the error when one of the multi-slaves
4008 bond_set_allmulti(bond
, 1);
4011 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
))
4012 bond_set_allmulti(bond
, -1);
4015 read_lock(&bond
->lock
);
4017 bond
->flags
= bond_dev
->flags
;
4019 /* looking for addresses to add to slaves' mc list */
4020 netdev_for_each_mc_addr(ha
, bond_dev
) {
4021 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
4022 bond_dev
->addr_len
);
4024 bond_mc_add(bond
, ha
->addr
);
4027 /* looking for addresses to delete from slaves' list */
4028 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
4029 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
4030 bond_dev
->addr_len
);
4032 bond_mc_del(bond
, ha
->addr
);
4035 /* save master's multicast list */
4036 __hw_addr_flush(&bond
->mc_list
);
4037 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
4038 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
4040 read_unlock(&bond
->lock
);
4043 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
4045 struct bonding
*bond
= netdev_priv(dev
);
4046 struct slave
*slave
= bond
->first_slave
;
4049 const struct net_device_ops
*slave_ops
4050 = slave
->dev
->netdev_ops
;
4051 if (slave_ops
->ndo_neigh_setup
)
4052 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
4058 * Change the MTU of all of a master's slaves to match the master
4060 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
4062 struct bonding
*bond
= netdev_priv(bond_dev
);
4063 struct slave
*slave
, *stop_at
;
4067 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
4068 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
4070 /* Can't hold bond->lock with bh disabled here since
4071 * some base drivers panic. On the other hand we can't
4072 * hold bond->lock without bh disabled because we'll
4073 * deadlock. The only solution is to rely on the fact
4074 * that we're under rtnl_lock here, and the slaves
4075 * list won't change. This doesn't solve the problem
4076 * of setting the slave's MTU while it is
4077 * transmitting, but the assumption is that the base
4078 * driver can handle that.
4080 * TODO: figure out a way to safely iterate the slaves
4081 * list, but without holding a lock around the actual
4082 * call to the base driver.
4085 bond_for_each_slave(bond
, slave
, i
) {
4086 pr_debug("s %p s->p %p c_m %p\n",
4089 slave
->dev
->netdev_ops
->ndo_change_mtu
);
4091 res
= dev_set_mtu(slave
->dev
, new_mtu
);
4094 /* If we failed to set the slave's mtu to the new value
4095 * we must abort the operation even in ACTIVE_BACKUP
4096 * mode, because if we allow the backup slaves to have
4097 * different mtu values than the active slave we'll
4098 * need to change their mtu when doing a failover. That
4099 * means changing their mtu from timer context, which
4100 * is probably not a good idea.
4102 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4107 bond_dev
->mtu
= new_mtu
;
4112 /* unwind from head to the slave that failed */
4114 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4117 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
4119 pr_debug("unwind err %d dev %s\n",
4120 tmp_res
, slave
->dev
->name
);
4130 * Note that many devices must be down to change the HW address, and
4131 * downing the master releases all slaves. We can make bonds full of
4132 * bonding devices to test this, however.
4134 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4136 struct bonding
*bond
= netdev_priv(bond_dev
);
4137 struct sockaddr
*sa
= addr
, tmp_sa
;
4138 struct slave
*slave
, *stop_at
;
4142 if (bond
->params
.mode
== BOND_MODE_ALB
)
4143 return bond_alb_set_mac_address(bond_dev
, addr
);
4146 pr_debug("bond=%p, name=%s\n",
4147 bond
, bond_dev
? bond_dev
->name
: "None");
4150 * If fail_over_mac is set to active, do nothing and return
4151 * success. Returning an error causes ifenslave to fail.
4153 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4156 if (!is_valid_ether_addr(sa
->sa_data
))
4157 return -EADDRNOTAVAIL
;
4159 /* Can't hold bond->lock with bh disabled here since
4160 * some base drivers panic. On the other hand we can't
4161 * hold bond->lock without bh disabled because we'll
4162 * deadlock. The only solution is to rely on the fact
4163 * that we're under rtnl_lock here, and the slaves
4164 * list won't change. This doesn't solve the problem
4165 * of setting the slave's hw address while it is
4166 * transmitting, but the assumption is that the base
4167 * driver can handle that.
4169 * TODO: figure out a way to safely iterate the slaves
4170 * list, but without holding a lock around the actual
4171 * call to the base driver.
4174 bond_for_each_slave(bond
, slave
, i
) {
4175 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
4176 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
4178 if (slave_ops
->ndo_set_mac_address
== NULL
) {
4180 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
4184 res
= dev_set_mac_address(slave
->dev
, addr
);
4186 /* TODO: consider downing the slave
4188 * User should expect communications
4189 * breakage anyway until ARP finish
4192 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4198 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4202 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4203 tmp_sa
.sa_family
= bond_dev
->type
;
4205 /* unwind from head to the slave that failed */
4207 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4210 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4212 pr_debug("unwind err %d dev %s\n",
4213 tmp_res
, slave
->dev
->name
);
4220 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4222 struct bonding
*bond
= netdev_priv(bond_dev
);
4223 struct slave
*slave
, *start_at
;
4224 int i
, slave_no
, res
= 1;
4225 struct iphdr
*iph
= ip_hdr(skb
);
4227 read_lock(&bond
->lock
);
4229 if (!BOND_IS_OK(bond
))
4232 * Start with the curr_active_slave that joined the bond as the
4233 * default for sending IGMP traffic. For failover purposes one
4234 * needs to maintain some consistency for the interface that will
4235 * send the join/membership reports. The curr_active_slave found
4236 * will send all of this type of traffic.
4238 if ((iph
->protocol
== IPPROTO_IGMP
) &&
4239 (skb
->protocol
== htons(ETH_P_IP
))) {
4241 read_lock(&bond
->curr_slave_lock
);
4242 slave
= bond
->curr_active_slave
;
4243 read_unlock(&bond
->curr_slave_lock
);
4249 * Concurrent TX may collide on rr_tx_counter; we accept
4250 * that as being rare enough not to justify using an
4253 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4255 bond_for_each_slave(bond
, slave
, i
) {
4263 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4264 if (IS_UP(slave
->dev
) &&
4265 (slave
->link
== BOND_LINK_UP
) &&
4266 (slave
->state
== BOND_STATE_ACTIVE
)) {
4267 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4274 /* no suitable interface, frame not sent */
4277 read_unlock(&bond
->lock
);
4278 return NETDEV_TX_OK
;
4283 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4284 * the bond has a usable interface.
4286 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4288 struct bonding
*bond
= netdev_priv(bond_dev
);
4291 read_lock(&bond
->lock
);
4292 read_lock(&bond
->curr_slave_lock
);
4294 if (!BOND_IS_OK(bond
))
4297 if (!bond
->curr_active_slave
)
4300 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4304 /* no suitable interface, frame not sent */
4307 read_unlock(&bond
->curr_slave_lock
);
4308 read_unlock(&bond
->lock
);
4309 return NETDEV_TX_OK
;
4313 * In bond_xmit_xor() , we determine the output device by using a pre-
4314 * determined xmit_hash_policy(), If the selected device is not enabled,
4315 * find the next active slave.
4317 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4319 struct bonding
*bond
= netdev_priv(bond_dev
);
4320 struct slave
*slave
, *start_at
;
4325 read_lock(&bond
->lock
);
4327 if (!BOND_IS_OK(bond
))
4330 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4332 bond_for_each_slave(bond
, slave
, i
) {
4340 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4341 if (IS_UP(slave
->dev
) &&
4342 (slave
->link
== BOND_LINK_UP
) &&
4343 (slave
->state
== BOND_STATE_ACTIVE
)) {
4344 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4351 /* no suitable interface, frame not sent */
4354 read_unlock(&bond
->lock
);
4355 return NETDEV_TX_OK
;
4359 * in broadcast mode, we send everything to all usable interfaces.
4361 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4363 struct bonding
*bond
= netdev_priv(bond_dev
);
4364 struct slave
*slave
, *start_at
;
4365 struct net_device
*tx_dev
= NULL
;
4369 read_lock(&bond
->lock
);
4371 if (!BOND_IS_OK(bond
))
4374 read_lock(&bond
->curr_slave_lock
);
4375 start_at
= bond
->curr_active_slave
;
4376 read_unlock(&bond
->curr_slave_lock
);
4381 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4382 if (IS_UP(slave
->dev
) &&
4383 (slave
->link
== BOND_LINK_UP
) &&
4384 (slave
->state
== BOND_STATE_ACTIVE
)) {
4386 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4388 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4393 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4395 dev_kfree_skb(skb2
);
4399 tx_dev
= slave
->dev
;
4404 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4408 /* no suitable interface, frame not sent */
4411 /* frame sent to all suitable interfaces */
4412 read_unlock(&bond
->lock
);
4413 return NETDEV_TX_OK
;
4416 /*------------------------- Device initialization ---------------------------*/
4418 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4420 switch (bond
->params
.xmit_policy
) {
4421 case BOND_XMIT_POLICY_LAYER23
:
4422 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4424 case BOND_XMIT_POLICY_LAYER34
:
4425 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4427 case BOND_XMIT_POLICY_LAYER2
:
4429 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4435 * Lookup the slave that corresponds to a qid
4437 static inline int bond_slave_override(struct bonding
*bond
,
4438 struct sk_buff
*skb
)
4441 struct slave
*slave
= NULL
;
4442 struct slave
*check_slave
;
4444 read_lock(&bond
->lock
);
4446 if (!BOND_IS_OK(bond
) || !skb
->queue_mapping
)
4449 /* Find out if any slaves have the same mapping as this skb. */
4450 bond_for_each_slave(bond
, check_slave
, i
) {
4451 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4452 slave
= check_slave
;
4457 /* If the slave isn't UP, use default transmit policy. */
4458 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4459 (slave
->link
== BOND_LINK_UP
)) {
4460 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4464 read_unlock(&bond
->lock
);
4468 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4471 * This helper function exists to help dev_pick_tx get the correct
4472 * destination queue. Using a helper function skips the a call to
4473 * skb_tx_hash and will put the skbs in the queue we expect on their
4474 * way down to the bonding driver.
4476 return skb
->queue_mapping
;
4479 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4481 struct bonding
*bond
= netdev_priv(dev
);
4483 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4484 if (!bond_slave_override(bond
, skb
))
4485 return NETDEV_TX_OK
;
4488 switch (bond
->params
.mode
) {
4489 case BOND_MODE_ROUNDROBIN
:
4490 return bond_xmit_roundrobin(skb
, dev
);
4491 case BOND_MODE_ACTIVEBACKUP
:
4492 return bond_xmit_activebackup(skb
, dev
);
4494 return bond_xmit_xor(skb
, dev
);
4495 case BOND_MODE_BROADCAST
:
4496 return bond_xmit_broadcast(skb
, dev
);
4497 case BOND_MODE_8023AD
:
4498 return bond_3ad_xmit_xor(skb
, dev
);
4501 return bond_alb_xmit(skb
, dev
);
4503 /* Should never happen, mode already checked */
4504 pr_err("%s: Error: Unknown bonding mode %d\n",
4505 dev
->name
, bond
->params
.mode
);
4508 return NETDEV_TX_OK
;
4514 * set bond mode specific net device operations
4516 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4518 struct net_device
*bond_dev
= bond
->dev
;
4521 case BOND_MODE_ROUNDROBIN
:
4523 case BOND_MODE_ACTIVEBACKUP
:
4526 bond_set_xmit_hash_policy(bond
);
4528 case BOND_MODE_BROADCAST
:
4530 case BOND_MODE_8023AD
:
4531 bond_set_master_3ad_flags(bond
);
4532 bond_set_xmit_hash_policy(bond
);
4535 bond_set_master_alb_flags(bond
);
4540 /* Should never happen, mode already checked */
4541 pr_err("%s: Error: Unknown bonding mode %d\n",
4542 bond_dev
->name
, mode
);
4547 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4548 struct ethtool_drvinfo
*drvinfo
)
4550 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4551 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4552 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4555 static const struct ethtool_ops bond_ethtool_ops
= {
4556 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4557 .get_link
= ethtool_op_get_link
,
4558 .get_tx_csum
= ethtool_op_get_tx_csum
,
4559 .get_sg
= ethtool_op_get_sg
,
4560 .get_tso
= ethtool_op_get_tso
,
4561 .get_ufo
= ethtool_op_get_ufo
,
4562 .get_flags
= ethtool_op_get_flags
,
4565 static const struct net_device_ops bond_netdev_ops
= {
4566 .ndo_init
= bond_init
,
4567 .ndo_uninit
= bond_uninit
,
4568 .ndo_open
= bond_open
,
4569 .ndo_stop
= bond_close
,
4570 .ndo_start_xmit
= bond_start_xmit
,
4571 .ndo_select_queue
= bond_select_queue
,
4572 .ndo_get_stats
= bond_get_stats
,
4573 .ndo_do_ioctl
= bond_do_ioctl
,
4574 .ndo_set_multicast_list
= bond_set_multicast_list
,
4575 .ndo_change_mtu
= bond_change_mtu
,
4576 .ndo_set_mac_address
= bond_set_mac_address
,
4577 .ndo_neigh_setup
= bond_neigh_setup
,
4578 .ndo_vlan_rx_register
= bond_vlan_rx_register
,
4579 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4580 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4581 #ifdef CONFIG_NET_POLL_CONTROLLER
4582 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4583 .ndo_poll_controller
= bond_poll_controller
,
4587 static void bond_destructor(struct net_device
*bond_dev
)
4589 struct bonding
*bond
= netdev_priv(bond_dev
);
4591 destroy_workqueue(bond
->wq
);
4592 free_netdev(bond_dev
);
4595 static void bond_setup(struct net_device
*bond_dev
)
4597 struct bonding
*bond
= netdev_priv(bond_dev
);
4599 /* initialize rwlocks */
4600 rwlock_init(&bond
->lock
);
4601 rwlock_init(&bond
->curr_slave_lock
);
4603 bond
->params
= bonding_defaults
;
4605 /* Initialize pointers */
4606 bond
->dev
= bond_dev
;
4607 INIT_LIST_HEAD(&bond
->vlan_list
);
4609 /* Initialize the device entry points */
4610 ether_setup(bond_dev
);
4611 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4612 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4613 bond_set_mode_ops(bond
, bond
->params
.mode
);
4615 bond_dev
->destructor
= bond_destructor
;
4617 /* Initialize the device options */
4618 bond_dev
->tx_queue_len
= 0;
4619 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4620 bond_dev
->priv_flags
|= IFF_BONDING
;
4621 bond_dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
4623 if (bond
->params
.arp_interval
)
4624 bond_dev
->priv_flags
|= IFF_MASTER_ARPMON
;
4626 /* At first, we block adding VLANs. That's the only way to
4627 * prevent problems that occur when adding VLANs over an
4628 * empty bond. The block will be removed once non-challenged
4629 * slaves are enslaved.
4631 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4633 /* don't acquire bond device's netif_tx_lock when
4635 bond_dev
->features
|= NETIF_F_LLTX
;
4637 /* By default, we declare the bond to be fully
4638 * VLAN hardware accelerated capable. Special
4639 * care is taken in the various xmit functions
4640 * when there are slaves that are not hw accel
4643 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4644 NETIF_F_HW_VLAN_RX
|
4645 NETIF_F_HW_VLAN_FILTER
);
4649 static void bond_work_cancel_all(struct bonding
*bond
)
4651 write_lock_bh(&bond
->lock
);
4652 bond
->kill_timers
= 1;
4653 write_unlock_bh(&bond
->lock
);
4655 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4656 cancel_delayed_work(&bond
->mii_work
);
4658 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4659 cancel_delayed_work(&bond
->arp_work
);
4661 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4662 delayed_work_pending(&bond
->alb_work
))
4663 cancel_delayed_work(&bond
->alb_work
);
4665 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4666 delayed_work_pending(&bond
->ad_work
))
4667 cancel_delayed_work(&bond
->ad_work
);
4671 * Destroy a bonding device.
4672 * Must be under rtnl_lock when this function is called.
4674 static void bond_uninit(struct net_device
*bond_dev
)
4676 struct bonding
*bond
= netdev_priv(bond_dev
);
4678 bond_netpoll_cleanup(bond_dev
);
4680 /* Release the bonded slaves */
4681 bond_release_all(bond_dev
);
4683 list_del(&bond
->bond_list
);
4685 bond_work_cancel_all(bond
);
4687 bond_remove_proc_entry(bond
);
4689 __hw_addr_flush(&bond
->mc_list
);
4692 /*------------------------- Module initialization ---------------------------*/
4695 * Convert string input module parms. Accept either the
4696 * number of the mode or its string name. A bit complicated because
4697 * some mode names are substrings of other names, and calls from sysfs
4698 * may have whitespace in the name (trailing newlines, for example).
4700 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4702 int modeint
= -1, i
, rv
;
4703 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4705 for (p
= (char *)buf
; *p
; p
++)
4706 if (!(isdigit(*p
) || isspace(*p
)))
4710 rv
= sscanf(buf
, "%20s", modestr
);
4712 rv
= sscanf(buf
, "%d", &modeint
);
4717 for (i
= 0; tbl
[i
].modename
; i
++) {
4718 if (modeint
== tbl
[i
].mode
)
4720 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4727 static int bond_check_params(struct bond_params
*params
)
4729 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4732 * Convert string parameters.
4735 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4736 if (bond_mode
== -1) {
4737 pr_err("Error: Invalid bonding mode \"%s\"\n",
4738 mode
== NULL
? "NULL" : mode
);
4743 if (xmit_hash_policy
) {
4744 if ((bond_mode
!= BOND_MODE_XOR
) &&
4745 (bond_mode
!= BOND_MODE_8023AD
)) {
4746 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4747 bond_mode_name(bond_mode
));
4749 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4751 if (xmit_hashtype
== -1) {
4752 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4753 xmit_hash_policy
== NULL
? "NULL" :
4761 if (bond_mode
!= BOND_MODE_8023AD
) {
4762 pr_info("lacp_rate param is irrelevant in mode %s\n",
4763 bond_mode_name(bond_mode
));
4765 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4766 if (lacp_fast
== -1) {
4767 pr_err("Error: Invalid lacp rate \"%s\"\n",
4768 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4775 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4776 if (params
->ad_select
== -1) {
4777 pr_err("Error: Invalid ad_select \"%s\"\n",
4778 ad_select
== NULL
? "NULL" : ad_select
);
4782 if (bond_mode
!= BOND_MODE_8023AD
) {
4783 pr_warning("ad_select param only affects 802.3ad mode\n");
4786 params
->ad_select
= BOND_AD_STABLE
;
4789 if (max_bonds
< 0) {
4790 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4791 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4792 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4796 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4797 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4798 miimon
= BOND_LINK_MON_INTERV
;
4802 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4807 if (downdelay
< 0) {
4808 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4809 downdelay
, INT_MAX
);
4813 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4814 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4819 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
4820 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4825 if (num_unsol_na
< 0 || num_unsol_na
> 255) {
4826 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4831 /* reset values for 802.3ad */
4832 if (bond_mode
== BOND_MODE_8023AD
) {
4834 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");
4835 pr_warning("Forcing miimon to 100msec\n");
4840 if (tx_queues
< 1 || tx_queues
> 255) {
4841 pr_warning("Warning: tx_queues (%d) should be between "
4842 "1 and 255, resetting to %d\n",
4843 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4844 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4847 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4848 pr_warning("Warning: all_slaves_active module parameter (%d), "
4849 "not of valid value (0/1), so it was set to "
4850 "0\n", all_slaves_active
);
4851 all_slaves_active
= 0;
4854 /* reset values for TLB/ALB */
4855 if ((bond_mode
== BOND_MODE_TLB
) ||
4856 (bond_mode
== BOND_MODE_ALB
)) {
4858 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");
4859 pr_warning("Forcing miimon to 100msec\n");
4864 if (bond_mode
== BOND_MODE_ALB
) {
4865 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",
4870 if (updelay
|| downdelay
) {
4871 /* just warn the user the up/down delay will have
4872 * no effect since miimon is zero...
4874 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",
4875 updelay
, downdelay
);
4878 /* don't allow arp monitoring */
4880 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4881 miimon
, arp_interval
);
4885 if ((updelay
% miimon
) != 0) {
4886 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4888 (updelay
/ miimon
) * miimon
);
4893 if ((downdelay
% miimon
) != 0) {
4894 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4896 (downdelay
/ miimon
) * miimon
);
4899 downdelay
/= miimon
;
4902 if (arp_interval
< 0) {
4903 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4904 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4905 arp_interval
= BOND_LINK_ARP_INTERV
;
4908 for (arp_ip_count
= 0;
4909 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4911 /* not complete check, but should be good enough to
4913 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4914 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4915 arp_ip_target
[arp_ip_count
]);
4918 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4919 arp_target
[arp_ip_count
] = ip
;
4923 if (arp_interval
&& !arp_ip_count
) {
4924 /* don't allow arping if no arp_ip_target given... */
4925 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4931 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4932 pr_err("arp_validate only supported in active-backup mode\n");
4935 if (!arp_interval
) {
4936 pr_err("arp_validate requires arp_interval\n");
4940 arp_validate_value
= bond_parse_parm(arp_validate
,
4942 if (arp_validate_value
== -1) {
4943 pr_err("Error: invalid arp_validate \"%s\"\n",
4944 arp_validate
== NULL
? "NULL" : arp_validate
);
4948 arp_validate_value
= 0;
4951 pr_info("MII link monitoring set to %d ms\n", miimon
);
4952 } else if (arp_interval
) {
4955 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4957 arp_validate_tbl
[arp_validate_value
].modename
,
4960 for (i
= 0; i
< arp_ip_count
; i
++)
4961 pr_info(" %s", arp_ip_target
[i
]);
4965 } else if (max_bonds
) {
4966 /* miimon and arp_interval not set, we need one so things
4967 * work as expected, see bonding.txt for details
4969 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");
4972 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4973 /* currently, using a primary only makes sense
4974 * in active backup, TLB or ALB modes
4976 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4977 primary
, bond_mode_name(bond_mode
));
4981 if (primary
&& primary_reselect
) {
4982 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4984 if (primary_reselect_value
== -1) {
4985 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4987 NULL
? "NULL" : primary_reselect
);
4991 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4994 if (fail_over_mac
) {
4995 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4997 if (fail_over_mac_value
== -1) {
4998 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4999 arp_validate
== NULL
? "NULL" : arp_validate
);
5003 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
5004 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5006 fail_over_mac_value
= BOND_FOM_NONE
;
5009 /* fill params struct with the proper values */
5010 params
->mode
= bond_mode
;
5011 params
->xmit_policy
= xmit_hashtype
;
5012 params
->miimon
= miimon
;
5013 params
->num_grat_arp
= num_grat_arp
;
5014 params
->num_unsol_na
= num_unsol_na
;
5015 params
->arp_interval
= arp_interval
;
5016 params
->arp_validate
= arp_validate_value
;
5017 params
->updelay
= updelay
;
5018 params
->downdelay
= downdelay
;
5019 params
->use_carrier
= use_carrier
;
5020 params
->lacp_fast
= lacp_fast
;
5021 params
->primary
[0] = 0;
5022 params
->primary_reselect
= primary_reselect_value
;
5023 params
->fail_over_mac
= fail_over_mac_value
;
5024 params
->tx_queues
= tx_queues
;
5025 params
->all_slaves_active
= all_slaves_active
;
5028 strncpy(params
->primary
, primary
, IFNAMSIZ
);
5029 params
->primary
[IFNAMSIZ
- 1] = 0;
5032 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
5037 static struct lock_class_key bonding_netdev_xmit_lock_key
;
5038 static struct lock_class_key bonding_netdev_addr_lock_key
;
5040 static void bond_set_lockdep_class_one(struct net_device
*dev
,
5041 struct netdev_queue
*txq
,
5044 lockdep_set_class(&txq
->_xmit_lock
,
5045 &bonding_netdev_xmit_lock_key
);
5048 static void bond_set_lockdep_class(struct net_device
*dev
)
5050 lockdep_set_class(&dev
->addr_list_lock
,
5051 &bonding_netdev_addr_lock_key
);
5052 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
5056 * Called from registration process
5058 static int bond_init(struct net_device
*bond_dev
)
5060 struct bonding
*bond
= netdev_priv(bond_dev
);
5061 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
5063 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
5065 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
5069 bond_set_lockdep_class(bond_dev
);
5071 netif_carrier_off(bond_dev
);
5073 bond_create_proc_entry(bond
);
5074 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
5076 bond_prepare_sysfs_group(bond
);
5078 __hw_addr_init(&bond
->mc_list
);
5082 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
5084 if (tb
[IFLA_ADDRESS
]) {
5085 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
5087 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
5088 return -EADDRNOTAVAIL
;
5093 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
5095 .priv_size
= sizeof(struct bonding
),
5096 .setup
= bond_setup
,
5097 .validate
= bond_validate
,
5100 /* Create a new bond based on the specified name and bonding parameters.
5101 * If name is NULL, obtain a suitable "bond%d" name for us.
5102 * Caller must NOT hold rtnl_lock; we need to release it here before we
5103 * set up our sysfs entries.
5105 int bond_create(struct net
*net
, const char *name
)
5107 struct net_device
*bond_dev
;
5112 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
), name
? name
: "",
5113 bond_setup
, tx_queues
);
5115 pr_err("%s: eek! can't alloc netdev!\n", name
);
5120 dev_net_set(bond_dev
, net
);
5121 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
5124 res
= dev_alloc_name(bond_dev
, "bond%d");
5129 res
= register_netdevice(bond_dev
);
5134 bond_destructor(bond_dev
);
5138 static int __net_init
bond_net_init(struct net
*net
)
5140 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5143 INIT_LIST_HEAD(&bn
->dev_list
);
5145 bond_create_proc_dir(bn
);
5150 static void __net_exit
bond_net_exit(struct net
*net
)
5152 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5154 bond_destroy_proc_dir(bn
);
5157 static struct pernet_operations bond_net_ops
= {
5158 .init
= bond_net_init
,
5159 .exit
= bond_net_exit
,
5161 .size
= sizeof(struct bond_net
),
5164 static int __init
bonding_init(void)
5169 pr_info("%s", version
);
5171 res
= bond_check_params(&bonding_defaults
);
5175 res
= register_pernet_subsys(&bond_net_ops
);
5179 res
= rtnl_link_register(&bond_link_ops
);
5183 for (i
= 0; i
< max_bonds
; i
++) {
5184 res
= bond_create(&init_net
, NULL
);
5189 res
= bond_create_sysfs();
5193 register_netdevice_notifier(&bond_netdev_notifier
);
5194 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5195 bond_register_ipv6_notifier();
5199 rtnl_link_unregister(&bond_link_ops
);
5201 unregister_pernet_subsys(&bond_net_ops
);
5206 static void __exit
bonding_exit(void)
5208 unregister_netdevice_notifier(&bond_netdev_notifier
);
5209 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5210 bond_unregister_ipv6_notifier();
5212 bond_destroy_sysfs();
5214 rtnl_link_unregister(&bond_link_ops
);
5215 unregister_pernet_subsys(&bond_net_ops
);
5218 module_init(bonding_init
);
5219 module_exit(bonding_exit
);
5220 MODULE_LICENSE("GPL");
5221 MODULE_VERSION(DRV_VERSION
);
5222 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5223 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5224 MODULE_ALIAS_RTNL_LINK("bond");