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/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <linux/preempt.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
;
112 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
114 module_param(max_bonds
, int, 0);
115 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
116 module_param(tx_queues
, int, 0);
117 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
118 module_param(num_grat_arp
, int, 0644);
119 MODULE_PARM_DESC(num_grat_arp
, "Number of gratuitous ARP packets to send on failover event");
120 module_param(num_unsol_na
, int, 0644);
121 MODULE_PARM_DESC(num_unsol_na
, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
122 module_param(miimon
, int, 0);
123 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
124 module_param(updelay
, int, 0);
125 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
126 module_param(downdelay
, int, 0);
127 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
129 module_param(use_carrier
, int, 0);
130 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132 module_param(mode
, charp
, 0);
133 MODULE_PARM_DESC(mode
, "Mode of operation : 0 for balance-rr, "
134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
137 module_param(primary
, charp
, 0);
138 MODULE_PARM_DESC(primary
, "Primary network device to use");
139 module_param(primary_reselect
, charp
, 0);
140 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
145 "2 for only on active slave "
147 module_param(lacp_rate
, charp
, 0);
148 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner "
150 module_param(ad_select
, charp
, 0);
151 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
152 module_param(xmit_hash_policy
, charp
, 0);
153 MODULE_PARM_DESC(xmit_hash_policy
, "XOR hashing method: 0 for layer 2 (default)"
154 ", 1 for layer 3+4");
155 module_param(arp_interval
, int, 0);
156 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
157 module_param_array(arp_ip_target
, charp
, NULL
, 0);
158 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
159 module_param(arp_validate
, charp
, 0);
160 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes: none (default), active, backup or all");
161 module_param(fail_over_mac
, charp
, 0);
162 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
163 module_param(all_slaves_active
, int, 0);
164 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
165 "by setting active flag for all slaves. "
166 "0 for never (default), 1 for always.");
167 module_param(resend_igmp
, int, 0);
168 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on link failure");
170 /*----------------------------- Global variables ----------------------------*/
172 #ifdef CONFIG_NET_POLL_CONTROLLER
173 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
176 static const char * const version
=
177 DRV_DESCRIPTION
": v" DRV_VERSION
" (" DRV_RELDATE
")\n";
179 int bond_net_id __read_mostly
;
181 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
182 static int arp_ip_count
;
183 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
184 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
185 static int lacp_fast
;
187 const struct bond_parm_tbl bond_lacp_tbl
[] = {
188 { "slow", AD_LACP_SLOW
},
189 { "fast", AD_LACP_FAST
},
193 const struct bond_parm_tbl bond_mode_tbl
[] = {
194 { "balance-rr", BOND_MODE_ROUNDROBIN
},
195 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
196 { "balance-xor", BOND_MODE_XOR
},
197 { "broadcast", BOND_MODE_BROADCAST
},
198 { "802.3ad", BOND_MODE_8023AD
},
199 { "balance-tlb", BOND_MODE_TLB
},
200 { "balance-alb", BOND_MODE_ALB
},
204 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
205 { "layer2", BOND_XMIT_POLICY_LAYER2
},
206 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
207 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
211 const struct bond_parm_tbl arp_validate_tbl
[] = {
212 { "none", BOND_ARP_VALIDATE_NONE
},
213 { "active", BOND_ARP_VALIDATE_ACTIVE
},
214 { "backup", BOND_ARP_VALIDATE_BACKUP
},
215 { "all", BOND_ARP_VALIDATE_ALL
},
219 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
220 { "none", BOND_FOM_NONE
},
221 { "active", BOND_FOM_ACTIVE
},
222 { "follow", BOND_FOM_FOLLOW
},
226 const struct bond_parm_tbl pri_reselect_tbl
[] = {
227 { "always", BOND_PRI_RESELECT_ALWAYS
},
228 { "better", BOND_PRI_RESELECT_BETTER
},
229 { "failure", BOND_PRI_RESELECT_FAILURE
},
233 struct bond_parm_tbl ad_select_tbl
[] = {
234 { "stable", BOND_AD_STABLE
},
235 { "bandwidth", BOND_AD_BANDWIDTH
},
236 { "count", BOND_AD_COUNT
},
240 /*-------------------------- Forward declarations ---------------------------*/
242 static void bond_send_gratuitous_arp(struct bonding
*bond
);
243 static int bond_init(struct net_device
*bond_dev
);
244 static void bond_uninit(struct net_device
*bond_dev
);
246 /*---------------------------- General routines -----------------------------*/
248 static const char *bond_mode_name(int mode
)
250 static const char *names
[] = {
251 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
252 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
253 [BOND_MODE_XOR
] = "load balancing (xor)",
254 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
255 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
256 [BOND_MODE_TLB
] = "transmit load balancing",
257 [BOND_MODE_ALB
] = "adaptive load balancing",
260 if (mode
< 0 || mode
> BOND_MODE_ALB
)
266 /*---------------------------------- VLAN -----------------------------------*/
269 * bond_add_vlan - add a new vlan id on bond
270 * @bond: bond that got the notification
271 * @vlan_id: the vlan id to add
273 * Returns -ENOMEM if allocation failed.
275 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
277 struct vlan_entry
*vlan
;
279 pr_debug("bond: %s, vlan id %d\n",
280 (bond
? bond
->dev
->name
: "None"), vlan_id
);
282 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
286 INIT_LIST_HEAD(&vlan
->vlan_list
);
287 vlan
->vlan_id
= vlan_id
;
289 write_lock_bh(&bond
->lock
);
291 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
293 write_unlock_bh(&bond
->lock
);
295 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
301 * bond_del_vlan - delete a vlan id from bond
302 * @bond: bond that got the notification
303 * @vlan_id: the vlan id to delete
305 * returns -ENODEV if @vlan_id was not found in @bond.
307 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
309 struct vlan_entry
*vlan
;
312 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
315 write_lock_bh(&bond
->lock
);
317 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
318 if (vlan
->vlan_id
== vlan_id
) {
319 list_del(&vlan
->vlan_list
);
321 if (bond_is_lb(bond
))
322 bond_alb_clear_vlan(bond
, vlan_id
);
324 pr_debug("removed VLAN ID %d from bond %s\n",
325 vlan_id
, bond
->dev
->name
);
329 if (list_empty(&bond
->vlan_list
) &&
330 (bond
->slave_cnt
== 0)) {
331 /* Last VLAN removed and no slaves, so
332 * restore block on adding VLANs. This will
333 * be removed once new slaves that are not
334 * VLAN challenged will be added.
336 bond
->dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
344 pr_debug("couldn't find VLAN ID %d in bond %s\n",
345 vlan_id
, bond
->dev
->name
);
348 write_unlock_bh(&bond
->lock
);
349 unblock_netpoll_tx();
354 * bond_has_challenged_slaves
355 * @bond: the bond we're working on
357 * Searches the slave list. Returns 1 if a vlan challenged slave
358 * was found, 0 otherwise.
360 * Assumes bond->lock is held.
362 static int bond_has_challenged_slaves(struct bonding
*bond
)
367 bond_for_each_slave(bond
, slave
, i
) {
368 if (slave
->dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
369 pr_debug("found VLAN challenged slave - %s\n",
375 pr_debug("no VLAN challenged slaves found\n");
380 * bond_next_vlan - safely skip to the next item in the vlans list.
381 * @bond: the bond we're working on
382 * @curr: item we're advancing from
384 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
385 * or @curr->next otherwise (even if it is @curr itself again).
387 * Caller must hold bond->lock
389 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
391 struct vlan_entry
*next
, *last
;
393 if (list_empty(&bond
->vlan_list
))
397 next
= list_entry(bond
->vlan_list
.next
,
398 struct vlan_entry
, vlan_list
);
400 last
= list_entry(bond
->vlan_list
.prev
,
401 struct vlan_entry
, vlan_list
);
403 next
= list_entry(bond
->vlan_list
.next
,
404 struct vlan_entry
, vlan_list
);
406 next
= list_entry(curr
->vlan_list
.next
,
407 struct vlan_entry
, vlan_list
);
415 * bond_dev_queue_xmit - Prepare skb for xmit.
417 * @bond: bond device that got this skb for tx.
418 * @skb: hw accel VLAN tagged skb to transmit
419 * @slave_dev: slave that is supposed to xmit this skbuff
421 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
422 struct net_device
*slave_dev
)
424 skb
->dev
= slave_dev
;
426 if (unlikely(netpoll_tx_running(slave_dev
)))
427 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
435 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
436 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
438 * a. This operation is performed in IOCTL context,
439 * b. The operation is protected by the RTNL semaphore in the 8021q code,
440 * c. Holding a lock with BH disabled while directly calling a base driver
441 * entry point is generally a BAD idea.
443 * The design of synchronization/protection for this operation in the 8021q
444 * module is good for one or more VLAN devices over a single physical device
445 * and cannot be extended for a teaming solution like bonding, so there is a
446 * potential race condition here where a net device from the vlan group might
447 * be referenced (either by a base driver or the 8021q code) while it is being
448 * removed from the system. However, it turns out we're not making matters
449 * worse, and if it works for regular VLAN usage it will work here too.
453 * bond_vlan_rx_register - Propagates registration to slaves
454 * @bond_dev: bonding net device that got called
455 * @grp: vlan group being registered
457 static void bond_vlan_rx_register(struct net_device
*bond_dev
,
458 struct vlan_group
*grp
)
460 struct bonding
*bond
= netdev_priv(bond_dev
);
464 write_lock_bh(&bond
->lock
);
466 write_unlock_bh(&bond
->lock
);
468 bond_for_each_slave(bond
, slave
, i
) {
469 struct net_device
*slave_dev
= slave
->dev
;
470 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
472 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
473 slave_ops
->ndo_vlan_rx_register
) {
474 slave_ops
->ndo_vlan_rx_register(slave_dev
, grp
);
480 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
481 * @bond_dev: bonding net device that got called
482 * @vid: vlan id being added
484 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
486 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_FILTER
) &&
495 slave_ops
->ndo_vlan_rx_add_vid
) {
496 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vid
);
500 res
= bond_add_vlan(bond
, vid
);
502 pr_err("%s: Error: Failed to add vlan id %d\n",
503 bond_dev
->name
, vid
);
508 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
509 * @bond_dev: bonding net device that got called
510 * @vid: vlan id being removed
512 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
514 struct bonding
*bond
= netdev_priv(bond_dev
);
516 struct net_device
*vlan_dev
;
519 bond_for_each_slave(bond
, slave
, i
) {
520 struct net_device
*slave_dev
= slave
->dev
;
521 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
523 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
524 slave_ops
->ndo_vlan_rx_kill_vid
) {
525 /* Save and then restore vlan_dev in the grp array,
526 * since the slave's driver might clear it.
528 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vid
);
529 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vid
);
530 vlan_group_set_device(bond
->vlgrp
, vid
, vlan_dev
);
534 res
= bond_del_vlan(bond
, vid
);
536 pr_err("%s: Error: Failed to remove vlan id %d\n",
537 bond_dev
->name
, vid
);
541 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
543 struct vlan_entry
*vlan
;
544 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
549 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
550 slave_ops
->ndo_vlan_rx_register
)
551 slave_ops
->ndo_vlan_rx_register(slave_dev
, bond
->vlgrp
);
553 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
554 !(slave_ops
->ndo_vlan_rx_add_vid
))
557 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
)
558 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
561 static void bond_del_vlans_from_slave(struct bonding
*bond
,
562 struct net_device
*slave_dev
)
564 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
565 struct vlan_entry
*vlan
;
566 struct net_device
*vlan_dev
;
571 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
572 !(slave_ops
->ndo_vlan_rx_kill_vid
))
575 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
578 /* Save and then restore vlan_dev in the grp array,
579 * since the slave's driver might clear it.
581 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
582 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
583 vlan_group_set_device(bond
->vlgrp
, vlan
->vlan_id
, vlan_dev
);
587 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
588 slave_ops
->ndo_vlan_rx_register
)
589 slave_ops
->ndo_vlan_rx_register(slave_dev
, NULL
);
592 /*------------------------------- Link status -------------------------------*/
595 * Set the carrier state for the master according to the state of its
596 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
597 * do special 802.3ad magic.
599 * Returns zero if carrier state does not change, nonzero if it does.
601 static int bond_set_carrier(struct bonding
*bond
)
606 if (bond
->slave_cnt
== 0)
609 if (bond
->params
.mode
== BOND_MODE_8023AD
)
610 return bond_3ad_set_carrier(bond
);
612 bond_for_each_slave(bond
, slave
, i
) {
613 if (slave
->link
== BOND_LINK_UP
) {
614 if (!netif_carrier_ok(bond
->dev
)) {
615 netif_carrier_on(bond
->dev
);
623 if (netif_carrier_ok(bond
->dev
)) {
624 netif_carrier_off(bond
->dev
);
631 * Get link speed and duplex from the slave's base driver
632 * using ethtool. If for some reason the call fails or the
633 * values are invalid, fake speed and duplex to 100/Full
636 static int bond_update_speed_duplex(struct slave
*slave
)
638 struct net_device
*slave_dev
= slave
->dev
;
639 struct ethtool_cmd etool
;
642 /* Fake speed and duplex */
643 slave
->speed
= SPEED_100
;
644 slave
->duplex
= DUPLEX_FULL
;
646 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
649 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
653 switch (etool
.speed
) {
663 switch (etool
.duplex
) {
671 slave
->speed
= etool
.speed
;
672 slave
->duplex
= etool
.duplex
;
678 * if <dev> supports MII link status reporting, check its link status.
680 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
681 * depending upon the setting of the use_carrier parameter.
683 * Return either BMSR_LSTATUS, meaning that the link is up (or we
684 * can't tell and just pretend it is), or 0, meaning that the link is
687 * If reporting is non-zero, instead of faking link up, return -1 if
688 * both ETHTOOL and MII ioctls fail (meaning the device does not
689 * support them). If use_carrier is set, return whatever it says.
690 * It'd be nice if there was a good way to tell if a driver supports
691 * netif_carrier, but there really isn't.
693 static int bond_check_dev_link(struct bonding
*bond
,
694 struct net_device
*slave_dev
, int reporting
)
696 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
697 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
699 struct mii_ioctl_data
*mii
;
701 if (!reporting
&& !netif_running(slave_dev
))
704 if (bond
->params
.use_carrier
)
705 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
707 /* Try to get link status using Ethtool first. */
708 if (slave_dev
->ethtool_ops
) {
709 if (slave_dev
->ethtool_ops
->get_link
) {
712 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
714 return link
? BMSR_LSTATUS
: 0;
718 /* Ethtool can't be used, fallback to MII ioctls. */
719 ioctl
= slave_ops
->ndo_do_ioctl
;
721 /* TODO: set pointer to correct ioctl on a per team member */
722 /* bases to make this more efficient. that is, once */
723 /* we determine the correct ioctl, we will always */
724 /* call it and not the others for that team */
728 * We cannot assume that SIOCGMIIPHY will also read a
729 * register; not all network drivers (e.g., e100)
733 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
734 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
736 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
737 mii
->reg_num
= MII_BMSR
;
738 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
739 return mii
->val_out
& BMSR_LSTATUS
;
744 * If reporting, report that either there's no dev->do_ioctl,
745 * or both SIOCGMIIREG and get_link failed (meaning that we
746 * cannot report link status). If not reporting, pretend
749 return reporting
? -1 : BMSR_LSTATUS
;
752 /*----------------------------- Multicast list ------------------------------*/
755 * Push the promiscuity flag down to appropriate slaves
757 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
760 if (USES_PRIMARY(bond
->params
.mode
)) {
761 /* write lock already acquired */
762 if (bond
->curr_active_slave
) {
763 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
769 bond_for_each_slave(bond
, slave
, i
) {
770 err
= dev_set_promiscuity(slave
->dev
, inc
);
779 * Push the allmulti flag down to all slaves
781 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
784 if (USES_PRIMARY(bond
->params
.mode
)) {
785 /* write lock already acquired */
786 if (bond
->curr_active_slave
) {
787 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
793 bond_for_each_slave(bond
, slave
, i
) {
794 err
= dev_set_allmulti(slave
->dev
, inc
);
803 * Add a Multicast address to slaves
806 static void bond_mc_add(struct bonding
*bond
, void *addr
)
808 if (USES_PRIMARY(bond
->params
.mode
)) {
809 /* write lock already acquired */
810 if (bond
->curr_active_slave
)
811 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
816 bond_for_each_slave(bond
, slave
, i
)
817 dev_mc_add(slave
->dev
, addr
);
822 * Remove a multicast address from slave
825 static void bond_mc_del(struct bonding
*bond
, void *addr
)
827 if (USES_PRIMARY(bond
->params
.mode
)) {
828 /* write lock already acquired */
829 if (bond
->curr_active_slave
)
830 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
834 bond_for_each_slave(bond
, slave
, i
) {
835 dev_mc_del(slave
->dev
, addr
);
841 static void __bond_resend_igmp_join_requests(struct net_device
*dev
)
843 struct in_device
*in_dev
;
846 in_dev
= __in_dev_get_rcu(dev
);
848 ip_mc_rejoin_groups(in_dev
);
853 * Retrieve the list of registered multicast addresses for the bonding
854 * device and retransmit an IGMP JOIN request to the current active
857 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
859 struct net_device
*vlan_dev
;
860 struct vlan_entry
*vlan
;
862 read_lock(&bond
->lock
);
864 /* rejoin all groups on bond device */
865 __bond_resend_igmp_join_requests(bond
->dev
);
867 /* rejoin all groups on vlan devices */
869 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
870 vlan_dev
= vlan_group_get_device(bond
->vlgrp
,
873 __bond_resend_igmp_join_requests(vlan_dev
);
877 if (--bond
->igmp_retrans
> 0)
878 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
880 read_unlock(&bond
->lock
);
883 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
885 struct bonding
*bond
= container_of(work
, struct bonding
,
887 bond_resend_igmp_join_requests(bond
);
891 * flush all members of flush->mc_list from device dev->mc_list
893 static void bond_mc_list_flush(struct net_device
*bond_dev
,
894 struct net_device
*slave_dev
)
896 struct bonding
*bond
= netdev_priv(bond_dev
);
897 struct netdev_hw_addr
*ha
;
899 netdev_for_each_mc_addr(ha
, bond_dev
)
900 dev_mc_del(slave_dev
, ha
->addr
);
902 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
903 /* del lacpdu mc addr from mc list */
904 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
906 dev_mc_del(slave_dev
, lacpdu_multicast
);
910 /*--------------------------- Active slave change ---------------------------*/
913 * Update the mc list and multicast-related flags for the new and
914 * old active slaves (if any) according to the multicast mode, and
915 * promiscuous flags unconditionally.
917 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
918 struct slave
*old_active
)
920 struct netdev_hw_addr
*ha
;
922 if (!USES_PRIMARY(bond
->params
.mode
))
923 /* nothing to do - mc list is already up-to-date on
929 if (bond
->dev
->flags
& IFF_PROMISC
)
930 dev_set_promiscuity(old_active
->dev
, -1);
932 if (bond
->dev
->flags
& IFF_ALLMULTI
)
933 dev_set_allmulti(old_active
->dev
, -1);
935 netdev_for_each_mc_addr(ha
, bond
->dev
)
936 dev_mc_del(old_active
->dev
, ha
->addr
);
940 /* FIXME: Signal errors upstream. */
941 if (bond
->dev
->flags
& IFF_PROMISC
)
942 dev_set_promiscuity(new_active
->dev
, 1);
944 if (bond
->dev
->flags
& IFF_ALLMULTI
)
945 dev_set_allmulti(new_active
->dev
, 1);
947 netdev_for_each_mc_addr(ha
, bond
->dev
)
948 dev_mc_add(new_active
->dev
, ha
->addr
);
953 * bond_do_fail_over_mac
955 * Perform special MAC address swapping for fail_over_mac settings
957 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
959 static void bond_do_fail_over_mac(struct bonding
*bond
,
960 struct slave
*new_active
,
961 struct slave
*old_active
)
962 __releases(&bond
->curr_slave_lock
)
963 __releases(&bond
->lock
)
964 __acquires(&bond
->lock
)
965 __acquires(&bond
->curr_slave_lock
)
967 u8 tmp_mac
[ETH_ALEN
];
968 struct sockaddr saddr
;
971 switch (bond
->params
.fail_over_mac
) {
972 case BOND_FOM_ACTIVE
:
974 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
975 new_active
->dev
->addr_len
);
977 case BOND_FOM_FOLLOW
:
979 * if new_active && old_active, swap them
980 * if just old_active, do nothing (going to no active slave)
981 * if just new_active, set new_active to bond's MAC
986 write_unlock_bh(&bond
->curr_slave_lock
);
987 read_unlock(&bond
->lock
);
990 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
991 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
993 saddr
.sa_family
= new_active
->dev
->type
;
995 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
996 saddr
.sa_family
= bond
->dev
->type
;
999 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
1001 pr_err("%s: Error %d setting MAC of slave %s\n",
1002 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1009 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1010 saddr
.sa_family
= old_active
->dev
->type
;
1012 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
1014 pr_err("%s: Error %d setting MAC of slave %s\n",
1015 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1017 read_lock(&bond
->lock
);
1018 write_lock_bh(&bond
->curr_slave_lock
);
1021 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1022 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1028 static bool bond_should_change_active(struct bonding
*bond
)
1030 struct slave
*prim
= bond
->primary_slave
;
1031 struct slave
*curr
= bond
->curr_active_slave
;
1033 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
1035 if (bond
->force_primary
) {
1036 bond
->force_primary
= false;
1039 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
1040 (prim
->speed
< curr
->speed
||
1041 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
1043 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
1049 * find_best_interface - select the best available slave to be the active one
1050 * @bond: our bonding struct
1052 * Warning: Caller must hold curr_slave_lock for writing.
1054 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1056 struct slave
*new_active
, *old_active
;
1057 struct slave
*bestslave
= NULL
;
1058 int mintime
= bond
->params
.updelay
;
1061 new_active
= bond
->curr_active_slave
;
1063 if (!new_active
) { /* there were no active slaves left */
1064 if (bond
->slave_cnt
> 0) /* found one slave */
1065 new_active
= bond
->first_slave
;
1067 return NULL
; /* still no slave, return NULL */
1070 if ((bond
->primary_slave
) &&
1071 bond
->primary_slave
->link
== BOND_LINK_UP
&&
1072 bond_should_change_active(bond
)) {
1073 new_active
= bond
->primary_slave
;
1076 /* remember where to stop iterating over the slaves */
1077 old_active
= new_active
;
1079 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1080 if (new_active
->link
== BOND_LINK_UP
) {
1082 } else if (new_active
->link
== BOND_LINK_BACK
&&
1083 IS_UP(new_active
->dev
)) {
1084 /* link up, but waiting for stabilization */
1085 if (new_active
->delay
< mintime
) {
1086 mintime
= new_active
->delay
;
1087 bestslave
= new_active
;
1096 * change_active_interface - change the active slave into the specified one
1097 * @bond: our bonding struct
1098 * @new: the new slave to make the active one
1100 * Set the new slave to the bond's settings and unset them on the old
1101 * curr_active_slave.
1102 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1104 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1105 * because it is apparently the best available slave we have, even though its
1106 * updelay hasn't timed out yet.
1108 * If new_active is not NULL, caller must hold bond->lock for read and
1109 * curr_slave_lock for write_bh.
1111 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1113 struct slave
*old_active
= bond
->curr_active_slave
;
1115 if (old_active
== new_active
)
1119 new_active
->jiffies
= jiffies
;
1121 if (new_active
->link
== BOND_LINK_BACK
) {
1122 if (USES_PRIMARY(bond
->params
.mode
)) {
1123 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1124 bond
->dev
->name
, new_active
->dev
->name
,
1125 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1128 new_active
->delay
= 0;
1129 new_active
->link
= BOND_LINK_UP
;
1131 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1132 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1134 if (bond_is_lb(bond
))
1135 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1137 if (USES_PRIMARY(bond
->params
.mode
)) {
1138 pr_info("%s: making interface %s the new active one.\n",
1139 bond
->dev
->name
, new_active
->dev
->name
);
1144 if (USES_PRIMARY(bond
->params
.mode
))
1145 bond_mc_swap(bond
, new_active
, old_active
);
1147 if (bond_is_lb(bond
)) {
1148 bond_alb_handle_active_change(bond
, new_active
);
1150 bond_set_slave_inactive_flags(old_active
);
1152 bond_set_slave_active_flags(new_active
);
1154 bond
->curr_active_slave
= new_active
;
1157 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1159 bond_set_slave_inactive_flags(old_active
);
1162 bond_set_slave_active_flags(new_active
);
1164 if (bond
->params
.fail_over_mac
)
1165 bond_do_fail_over_mac(bond
, new_active
,
1168 if (netif_running(bond
->dev
)) {
1169 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1170 bond_send_gratuitous_arp(bond
);
1172 bond
->send_unsol_na
= bond
->params
.num_unsol_na
;
1173 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 active slave has changed or
1187 * all were sent on curr_active_slave */
1188 if (((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
1189 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
) &&
1190 netif_running(bond
->dev
)) {
1191 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
1192 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 0);
1197 * bond_select_active_slave - select a new active slave, if needed
1198 * @bond: our bonding struct
1200 * This functions should be called when one of the following occurs:
1201 * - The old curr_active_slave has been released or lost its link.
1202 * - The primary_slave has got its link back.
1203 * - A slave has got its link back and there's no old curr_active_slave.
1205 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1207 void bond_select_active_slave(struct bonding
*bond
)
1209 struct slave
*best_slave
;
1212 best_slave
= bond_find_best_slave(bond
);
1213 if (best_slave
!= bond
->curr_active_slave
) {
1214 bond_change_active_slave(bond
, best_slave
);
1215 rv
= bond_set_carrier(bond
);
1219 if (netif_carrier_ok(bond
->dev
)) {
1220 pr_info("%s: first active interface up!\n",
1223 pr_info("%s: now running without any active interface !\n",
1229 /*--------------------------- slave list handling ---------------------------*/
1232 * This function attaches the slave to the end of list.
1234 * bond->lock held for writing by caller.
1236 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1238 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1239 new_slave
->next
= new_slave
;
1240 new_slave
->prev
= new_slave
;
1241 bond
->first_slave
= new_slave
;
1243 new_slave
->next
= bond
->first_slave
;
1244 new_slave
->prev
= bond
->first_slave
->prev
;
1245 new_slave
->next
->prev
= new_slave
;
1246 new_slave
->prev
->next
= new_slave
;
1253 * This function detaches the slave from the list.
1254 * WARNING: no check is made to verify if the slave effectively
1255 * belongs to <bond>.
1256 * Nothing is freed on return, structures are just unchained.
1257 * If any slave pointer in bond was pointing to <slave>,
1258 * it should be changed by the calling function.
1260 * bond->lock held for writing by caller.
1262 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1265 slave
->next
->prev
= slave
->prev
;
1268 slave
->prev
->next
= slave
->next
;
1270 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1271 if (bond
->slave_cnt
> 1) { /* there are more slave */
1272 bond
->first_slave
= slave
->next
;
1274 bond
->first_slave
= NULL
; /* slave was the last one */
1283 #ifdef CONFIG_NET_POLL_CONTROLLER
1284 static inline int slave_enable_netpoll(struct slave
*slave
)
1289 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
1294 np
->dev
= slave
->dev
;
1295 err
= __netpoll_setup(np
);
1304 static inline void slave_disable_netpoll(struct slave
*slave
)
1306 struct netpoll
*np
= slave
->np
;
1312 synchronize_rcu_bh();
1313 __netpoll_cleanup(np
);
1316 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1318 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1320 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1325 static void bond_poll_controller(struct net_device
*bond_dev
)
1329 static void __bond_netpoll_cleanup(struct bonding
*bond
)
1331 struct slave
*slave
;
1334 bond_for_each_slave(bond
, slave
, i
)
1335 if (IS_UP(slave
->dev
))
1336 slave_disable_netpoll(slave
);
1338 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1340 struct bonding
*bond
= netdev_priv(bond_dev
);
1342 read_lock(&bond
->lock
);
1343 __bond_netpoll_cleanup(bond
);
1344 read_unlock(&bond
->lock
);
1347 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
1349 struct bonding
*bond
= netdev_priv(dev
);
1350 struct slave
*slave
;
1353 read_lock(&bond
->lock
);
1354 bond_for_each_slave(bond
, slave
, i
) {
1355 if (!IS_UP(slave
->dev
))
1357 err
= slave_enable_netpoll(slave
);
1359 __bond_netpoll_cleanup(bond
);
1363 read_unlock(&bond
->lock
);
1367 static struct netpoll_info
*bond_netpoll_info(struct bonding
*bond
)
1369 return bond
->dev
->npinfo
;
1373 static inline int slave_enable_netpoll(struct slave
*slave
)
1377 static inline void slave_disable_netpoll(struct slave
*slave
)
1380 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1385 /*---------------------------------- IOCTL ----------------------------------*/
1387 static int bond_sethwaddr(struct net_device
*bond_dev
,
1388 struct net_device
*slave_dev
)
1390 pr_debug("bond_dev=%p\n", bond_dev
);
1391 pr_debug("slave_dev=%p\n", slave_dev
);
1392 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1393 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1397 #define BOND_VLAN_FEATURES \
1398 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1399 NETIF_F_HW_VLAN_FILTER)
1402 * Compute the common dev->feature set available to all slaves. Some
1403 * feature bits are managed elsewhere, so preserve those feature bits
1404 * on the master device.
1406 static int bond_compute_features(struct bonding
*bond
)
1408 struct slave
*slave
;
1409 struct net_device
*bond_dev
= bond
->dev
;
1410 u32 features
= bond_dev
->features
;
1411 u32 vlan_features
= 0;
1412 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1413 bond_dev
->hard_header_len
);
1416 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1417 features
|= NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1419 if (!bond
->first_slave
)
1422 features
&= ~NETIF_F_ONE_FOR_ALL
;
1424 vlan_features
= bond
->first_slave
->dev
->vlan_features
;
1425 bond_for_each_slave(bond
, slave
, i
) {
1426 features
= netdev_increment_features(features
,
1427 slave
->dev
->features
,
1428 NETIF_F_ONE_FOR_ALL
);
1429 vlan_features
= netdev_increment_features(vlan_features
,
1430 slave
->dev
->vlan_features
,
1431 NETIF_F_ONE_FOR_ALL
);
1432 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1433 max_hard_header_len
= slave
->dev
->hard_header_len
;
1437 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1438 bond_dev
->features
= netdev_fix_features(bond_dev
, features
);
1439 bond_dev
->vlan_features
= netdev_fix_features(bond_dev
, vlan_features
);
1440 bond_dev
->hard_header_len
= max_hard_header_len
;
1445 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1446 struct net_device
*slave_dev
)
1448 struct bonding
*bond
= netdev_priv(bond_dev
);
1450 bond_dev
->header_ops
= slave_dev
->header_ops
;
1452 bond_dev
->type
= slave_dev
->type
;
1453 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1454 bond_dev
->addr_len
= slave_dev
->addr_len
;
1456 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1457 slave_dev
->addr_len
);
1458 bond
->setup_by_slave
= 1;
1461 /* On bonding slaves other than the currently active slave, suppress
1462 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
1463 * ARP on active-backup slaves with arp_validate enabled.
1465 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1466 struct net_device
*slave_dev
,
1467 struct net_device
*bond_dev
)
1469 if (slave_dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
1470 if (slave_dev
->priv_flags
& IFF_SLAVE_NEEDARP
&&
1471 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
1474 if (bond_dev
->priv_flags
& IFF_MASTER_ALB
&&
1475 skb
->pkt_type
!= PACKET_BROADCAST
&&
1476 skb
->pkt_type
!= PACKET_MULTICAST
)
1479 if (bond_dev
->priv_flags
& IFF_MASTER_8023AD
&&
1480 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
1488 static struct sk_buff
*bond_handle_frame(struct sk_buff
*skb
)
1490 struct net_device
*slave_dev
;
1491 struct net_device
*bond_dev
;
1493 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1496 slave_dev
= skb
->dev
;
1497 bond_dev
= ACCESS_ONCE(slave_dev
->master
);
1498 if (unlikely(!bond_dev
))
1501 if (bond_dev
->priv_flags
& IFF_MASTER_ARPMON
)
1502 slave_dev
->last_rx
= jiffies
;
1504 if (bond_should_deliver_exact_match(skb
, slave_dev
, bond_dev
)) {
1505 skb
->deliver_no_wcard
= 1;
1509 skb
->dev
= bond_dev
;
1511 if (bond_dev
->priv_flags
& IFF_MASTER_ALB
&&
1512 bond_dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1513 skb
->pkt_type
== PACKET_HOST
) {
1515 if (unlikely(skb_cow_head(skb
,
1516 skb
->data
- skb_mac_header(skb
)))) {
1520 memcpy(eth_hdr(skb
)->h_dest
, bond_dev
->dev_addr
, ETH_ALEN
);
1526 /* enslave device <slave> to bond device <master> */
1527 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1529 struct bonding
*bond
= netdev_priv(bond_dev
);
1530 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1531 struct slave
*new_slave
= NULL
;
1532 struct netdev_hw_addr
*ha
;
1533 struct sockaddr addr
;
1535 int old_features
= bond_dev
->features
;
1538 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1539 slave_ops
->ndo_do_ioctl
== NULL
) {
1540 pr_warning("%s: Warning: no link monitoring support for %s\n",
1541 bond_dev
->name
, slave_dev
->name
);
1544 /* bond must be initialized by bond_open() before enslaving */
1545 if (!(bond_dev
->flags
& IFF_UP
)) {
1546 pr_warning("%s: master_dev is not up in bond_enslave\n",
1550 /* already enslaved */
1551 if (slave_dev
->flags
& IFF_SLAVE
) {
1552 pr_debug("Error, Device was already enslaved\n");
1556 /* vlan challenged mutual exclusion */
1557 /* no need to lock since we're protected by rtnl_lock */
1558 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1559 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1561 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1562 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1565 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1566 bond_dev
->name
, slave_dev
->name
,
1567 slave_dev
->name
, bond_dev
->name
);
1568 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1571 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1572 if (bond
->slave_cnt
== 0) {
1573 /* First slave, and it is not VLAN challenged,
1574 * so remove the block of adding VLANs over the bond.
1576 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1581 * Old ifenslave binaries are no longer supported. These can
1582 * be identified with moderate accuracy by the state of the slave:
1583 * the current ifenslave will set the interface down prior to
1584 * enslaving it; the old ifenslave will not.
1586 if ((slave_dev
->flags
& IFF_UP
)) {
1587 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1590 goto err_undo_flags
;
1593 /* set bonding device ether type by slave - bonding netdevices are
1594 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1595 * there is a need to override some of the type dependent attribs/funcs.
1597 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1598 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1600 if (bond
->slave_cnt
== 0) {
1601 if (bond_dev
->type
!= slave_dev
->type
) {
1602 pr_debug("%s: change device type from %d to %d\n",
1604 bond_dev
->type
, slave_dev
->type
);
1606 res
= netdev_bonding_change(bond_dev
,
1607 NETDEV_PRE_TYPE_CHANGE
);
1608 res
= notifier_to_errno(res
);
1610 pr_err("%s: refused to change device type\n",
1613 goto err_undo_flags
;
1616 /* Flush unicast and multicast addresses */
1617 dev_uc_flush(bond_dev
);
1618 dev_mc_flush(bond_dev
);
1620 if (slave_dev
->type
!= ARPHRD_ETHER
)
1621 bond_setup_by_slave(bond_dev
, slave_dev
);
1623 ether_setup(bond_dev
);
1625 netdev_bonding_change(bond_dev
,
1626 NETDEV_POST_TYPE_CHANGE
);
1628 } else if (bond_dev
->type
!= slave_dev
->type
) {
1629 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1631 slave_dev
->type
, bond_dev
->type
);
1633 goto err_undo_flags
;
1636 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1637 if (bond
->slave_cnt
== 0) {
1638 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1640 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1641 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1642 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",
1645 goto err_undo_flags
;
1649 /* If this is the first slave, then we need to set the master's hardware
1650 * address to be the same as the slave's. */
1651 if (is_zero_ether_addr(bond
->dev
->dev_addr
))
1652 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1653 slave_dev
->addr_len
);
1656 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1659 goto err_undo_flags
;
1663 * Set the new_slave's queue_id to be zero. Queue ID mapping
1664 * is set via sysfs or module option if desired.
1666 new_slave
->queue_id
= 0;
1668 /* Save slave's original mtu and then set it to match the bond */
1669 new_slave
->original_mtu
= slave_dev
->mtu
;
1670 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1672 pr_debug("Error %d calling dev_set_mtu\n", res
);
1677 * Save slave's original ("permanent") mac address for modes
1678 * that need it, and for restoring it upon release, and then
1679 * set it to the master's address
1681 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1683 if (!bond
->params
.fail_over_mac
) {
1685 * Set slave to master's mac address. The application already
1686 * set the master's mac address to that of the first slave
1688 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1689 addr
.sa_family
= slave_dev
->type
;
1690 res
= dev_set_mac_address(slave_dev
, &addr
);
1692 pr_debug("Error %d calling set_mac_address\n", res
);
1693 goto err_restore_mtu
;
1697 res
= netdev_set_bond_master(slave_dev
, bond_dev
);
1699 pr_debug("Error %d calling netdev_set_bond_master\n", res
);
1700 goto err_restore_mac
;
1702 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
, NULL
);
1704 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1705 goto err_unset_master
;
1708 /* open the slave since the application closed it */
1709 res
= dev_open(slave_dev
);
1711 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1712 goto err_unreg_rxhandler
;
1715 new_slave
->dev
= slave_dev
;
1716 slave_dev
->priv_flags
|= IFF_BONDING
;
1718 if (bond_is_lb(bond
)) {
1719 /* bond_alb_init_slave() must be called before all other stages since
1720 * it might fail and we do not want to have to undo everything
1722 res
= bond_alb_init_slave(bond
, new_slave
);
1727 /* If the mode USES_PRIMARY, then the new slave gets the
1728 * master's promisc (and mc) settings only if it becomes the
1729 * curr_active_slave, and that is taken care of later when calling
1730 * bond_change_active()
1732 if (!USES_PRIMARY(bond
->params
.mode
)) {
1733 /* set promiscuity level to new slave */
1734 if (bond_dev
->flags
& IFF_PROMISC
) {
1735 res
= dev_set_promiscuity(slave_dev
, 1);
1740 /* set allmulti level to new slave */
1741 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1742 res
= dev_set_allmulti(slave_dev
, 1);
1747 netif_addr_lock_bh(bond_dev
);
1748 /* upload master's mc_list to new slave */
1749 netdev_for_each_mc_addr(ha
, bond_dev
)
1750 dev_mc_add(slave_dev
, ha
->addr
);
1751 netif_addr_unlock_bh(bond_dev
);
1754 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1755 /* add lacpdu mc addr to mc list */
1756 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1758 dev_mc_add(slave_dev
, lacpdu_multicast
);
1761 bond_add_vlans_on_slave(bond
, slave_dev
);
1763 write_lock_bh(&bond
->lock
);
1765 bond_attach_slave(bond
, new_slave
);
1767 new_slave
->delay
= 0;
1768 new_slave
->link_failure_count
= 0;
1770 bond_compute_features(bond
);
1772 write_unlock_bh(&bond
->lock
);
1774 read_lock(&bond
->lock
);
1776 new_slave
->last_arp_rx
= jiffies
;
1778 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1779 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1781 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1783 * miimon is set but a bonded network driver
1784 * does not support ETHTOOL/MII and
1785 * arp_interval is not set. Note: if
1786 * use_carrier is enabled, we will never go
1787 * here (because netif_carrier is always
1788 * supported); thus, we don't need to change
1789 * the messages for netif_carrier.
1791 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",
1792 bond_dev
->name
, slave_dev
->name
);
1793 } else if (link_reporting
== -1) {
1794 /* unable get link status using mii/ethtool */
1795 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",
1796 bond_dev
->name
, slave_dev
->name
);
1800 /* check for initial state */
1801 if (!bond
->params
.miimon
||
1802 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1803 if (bond
->params
.updelay
) {
1804 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1805 new_slave
->link
= BOND_LINK_BACK
;
1806 new_slave
->delay
= bond
->params
.updelay
;
1808 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1809 new_slave
->link
= BOND_LINK_UP
;
1811 new_slave
->jiffies
= jiffies
;
1813 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1814 new_slave
->link
= BOND_LINK_DOWN
;
1817 if (bond_update_speed_duplex(new_slave
) &&
1818 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1819 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1820 bond_dev
->name
, new_slave
->dev
->name
);
1822 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1823 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1828 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1829 /* if there is a primary slave, remember it */
1830 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1831 bond
->primary_slave
= new_slave
;
1832 bond
->force_primary
= true;
1836 write_lock_bh(&bond
->curr_slave_lock
);
1838 switch (bond
->params
.mode
) {
1839 case BOND_MODE_ACTIVEBACKUP
:
1840 bond_set_slave_inactive_flags(new_slave
);
1841 bond_select_active_slave(bond
);
1843 case BOND_MODE_8023AD
:
1844 /* in 802.3ad mode, the internal mechanism
1845 * will activate the slaves in the selected
1848 bond_set_slave_inactive_flags(new_slave
);
1849 /* if this is the first slave */
1850 if (bond
->slave_cnt
== 1) {
1851 SLAVE_AD_INFO(new_slave
).id
= 1;
1852 /* Initialize AD with the number of times that the AD timer is called in 1 second
1853 * can be called only after the mac address of the bond is set
1855 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1856 bond
->params
.lacp_fast
);
1858 SLAVE_AD_INFO(new_slave
).id
=
1859 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1862 bond_3ad_bind_slave(new_slave
);
1866 new_slave
->state
= BOND_STATE_ACTIVE
;
1867 bond_set_slave_inactive_flags(new_slave
);
1868 bond_select_active_slave(bond
);
1871 pr_debug("This slave is always active in trunk mode\n");
1873 /* always active in trunk mode */
1874 new_slave
->state
= BOND_STATE_ACTIVE
;
1876 /* In trunking mode there is little meaning to curr_active_slave
1877 * anyway (it holds no special properties of the bond device),
1878 * so we can change it without calling change_active_interface()
1880 if (!bond
->curr_active_slave
)
1881 bond
->curr_active_slave
= new_slave
;
1884 } /* switch(bond_mode) */
1886 write_unlock_bh(&bond
->curr_slave_lock
);
1888 bond_set_carrier(bond
);
1890 #ifdef CONFIG_NET_POLL_CONTROLLER
1891 slave_dev
->npinfo
= bond_netpoll_info(bond
);
1892 if (slave_dev
->npinfo
) {
1893 if (slave_enable_netpoll(new_slave
)) {
1894 read_unlock(&bond
->lock
);
1895 pr_info("Error, %s: master_dev is using netpoll, "
1896 "but new slave device does not support netpoll.\n",
1904 read_unlock(&bond
->lock
);
1906 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1910 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1911 bond_dev
->name
, slave_dev
->name
,
1912 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1913 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1915 /* enslave is successful */
1918 /* Undo stages on error */
1920 dev_close(slave_dev
);
1922 err_unreg_rxhandler
:
1923 netdev_rx_handler_unregister(slave_dev
);
1926 netdev_set_bond_master(slave_dev
, NULL
);
1929 if (!bond
->params
.fail_over_mac
) {
1930 /* XXX TODO - fom follow mode needs to change master's
1931 * MAC if this slave's MAC is in use by the bond, or at
1932 * least print a warning.
1934 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1935 addr
.sa_family
= slave_dev
->type
;
1936 dev_set_mac_address(slave_dev
, &addr
);
1940 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1946 bond_dev
->features
= old_features
;
1952 * Try to release the slave device <slave> from the bond device <master>
1953 * It is legal to access curr_active_slave without a lock because all the function
1956 * The rules for slave state should be:
1957 * for Active/Backup:
1958 * Active stays on all backups go down
1959 * for Bonded connections:
1960 * The first up interface should be left on and all others downed.
1962 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1964 struct bonding
*bond
= netdev_priv(bond_dev
);
1965 struct slave
*slave
, *oldcurrent
;
1966 struct sockaddr addr
;
1968 /* slave is not a slave or master is not master of this slave */
1969 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1970 (slave_dev
->master
!= bond_dev
)) {
1971 pr_err("%s: Error: cannot release %s.\n",
1972 bond_dev
->name
, slave_dev
->name
);
1977 netdev_bonding_change(bond_dev
, NETDEV_BONDING_DESLAVE
);
1978 write_lock_bh(&bond
->lock
);
1980 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1982 /* not a slave of this bond */
1983 pr_info("%s: %s not enslaved\n",
1984 bond_dev
->name
, slave_dev
->name
);
1985 write_unlock_bh(&bond
->lock
);
1986 unblock_netpoll_tx();
1990 if (!bond
->params
.fail_over_mac
) {
1991 if (!compare_ether_addr(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1992 bond
->slave_cnt
> 1)
1993 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",
1994 bond_dev
->name
, slave_dev
->name
,
1996 bond_dev
->name
, slave_dev
->name
);
1999 /* Inform AD package of unbinding of slave. */
2000 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2001 /* must be called before the slave is
2002 * detached from the list
2004 bond_3ad_unbind_slave(slave
);
2007 pr_info("%s: releasing %s interface %s\n",
2009 (slave
->state
== BOND_STATE_ACTIVE
) ? "active" : "backup",
2012 oldcurrent
= bond
->curr_active_slave
;
2014 bond
->current_arp_slave
= NULL
;
2016 /* release the slave from its bond */
2017 bond_detach_slave(bond
, slave
);
2019 bond_compute_features(bond
);
2021 if (bond
->primary_slave
== slave
)
2022 bond
->primary_slave
= NULL
;
2024 if (oldcurrent
== slave
)
2025 bond_change_active_slave(bond
, NULL
);
2027 if (bond_is_lb(bond
)) {
2028 /* Must be called only after the slave has been
2029 * detached from the list and the curr_active_slave
2030 * has been cleared (if our_slave == old_current),
2031 * but before a new active slave is selected.
2033 write_unlock_bh(&bond
->lock
);
2034 bond_alb_deinit_slave(bond
, slave
);
2035 write_lock_bh(&bond
->lock
);
2038 if (oldcurrent
== slave
) {
2040 * Note that we hold RTNL over this sequence, so there
2041 * is no concern that another slave add/remove event
2044 write_unlock_bh(&bond
->lock
);
2045 read_lock(&bond
->lock
);
2046 write_lock_bh(&bond
->curr_slave_lock
);
2048 bond_select_active_slave(bond
);
2050 write_unlock_bh(&bond
->curr_slave_lock
);
2051 read_unlock(&bond
->lock
);
2052 write_lock_bh(&bond
->lock
);
2055 if (bond
->slave_cnt
== 0) {
2056 bond_set_carrier(bond
);
2058 /* if the last slave was removed, zero the mac address
2059 * of the master so it will be set by the application
2060 * to the mac address of the first slave
2062 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2065 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2067 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2068 bond_dev
->name
, bond_dev
->name
);
2069 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2072 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
2073 !bond_has_challenged_slaves(bond
)) {
2074 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2075 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
2076 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
2079 write_unlock_bh(&bond
->lock
);
2080 unblock_netpoll_tx();
2082 /* must do this from outside any spinlocks */
2083 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2085 bond_del_vlans_from_slave(bond
, slave_dev
);
2087 /* If the mode USES_PRIMARY, then we should only remove its
2088 * promisc and mc settings if it was the curr_active_slave, but that was
2089 * already taken care of above when we detached the slave
2091 if (!USES_PRIMARY(bond
->params
.mode
)) {
2092 /* unset promiscuity level from slave */
2093 if (bond_dev
->flags
& IFF_PROMISC
)
2094 dev_set_promiscuity(slave_dev
, -1);
2096 /* unset allmulti level from slave */
2097 if (bond_dev
->flags
& IFF_ALLMULTI
)
2098 dev_set_allmulti(slave_dev
, -1);
2100 /* flush master's mc_list from slave */
2101 netif_addr_lock_bh(bond_dev
);
2102 bond_mc_list_flush(bond_dev
, slave_dev
);
2103 netif_addr_unlock_bh(bond_dev
);
2106 netdev_rx_handler_unregister(slave_dev
);
2107 netdev_set_bond_master(slave_dev
, NULL
);
2109 slave_disable_netpoll(slave
);
2111 /* close slave before restoring its mac address */
2112 dev_close(slave_dev
);
2114 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2115 /* restore original ("permanent") mac address */
2116 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2117 addr
.sa_family
= slave_dev
->type
;
2118 dev_set_mac_address(slave_dev
, &addr
);
2121 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2123 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2124 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
2129 return 0; /* deletion OK */
2133 * First release a slave and than destroy the bond if no more slaves are left.
2134 * Must be under rtnl_lock when this function is called.
2136 static int bond_release_and_destroy(struct net_device
*bond_dev
,
2137 struct net_device
*slave_dev
)
2139 struct bonding
*bond
= netdev_priv(bond_dev
);
2142 ret
= bond_release(bond_dev
, slave_dev
);
2143 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2144 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
2145 pr_info("%s: destroying bond %s.\n",
2146 bond_dev
->name
, bond_dev
->name
);
2147 unregister_netdevice(bond_dev
);
2153 * This function releases all slaves.
2155 static int bond_release_all(struct net_device
*bond_dev
)
2157 struct bonding
*bond
= netdev_priv(bond_dev
);
2158 struct slave
*slave
;
2159 struct net_device
*slave_dev
;
2160 struct sockaddr addr
;
2162 write_lock_bh(&bond
->lock
);
2164 netif_carrier_off(bond_dev
);
2166 if (bond
->slave_cnt
== 0)
2169 bond
->current_arp_slave
= NULL
;
2170 bond
->primary_slave
= NULL
;
2171 bond_change_active_slave(bond
, NULL
);
2173 while ((slave
= bond
->first_slave
) != NULL
) {
2174 /* Inform AD package of unbinding of slave
2175 * before slave is detached from the list.
2177 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2178 bond_3ad_unbind_slave(slave
);
2180 slave_dev
= slave
->dev
;
2181 bond_detach_slave(bond
, slave
);
2183 /* now that the slave is detached, unlock and perform
2184 * all the undo steps that should not be called from
2187 write_unlock_bh(&bond
->lock
);
2189 if (bond_is_lb(bond
)) {
2190 /* must be called only after the slave
2191 * has been detached from the list
2193 bond_alb_deinit_slave(bond
, slave
);
2196 bond_compute_features(bond
);
2198 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2199 bond_del_vlans_from_slave(bond
, slave_dev
);
2201 /* If the mode USES_PRIMARY, then we should only remove its
2202 * promisc and mc settings if it was the curr_active_slave, but that was
2203 * already taken care of above when we detached the slave
2205 if (!USES_PRIMARY(bond
->params
.mode
)) {
2206 /* unset promiscuity level from slave */
2207 if (bond_dev
->flags
& IFF_PROMISC
)
2208 dev_set_promiscuity(slave_dev
, -1);
2210 /* unset allmulti level from slave */
2211 if (bond_dev
->flags
& IFF_ALLMULTI
)
2212 dev_set_allmulti(slave_dev
, -1);
2214 /* flush master's mc_list from slave */
2215 netif_addr_lock_bh(bond_dev
);
2216 bond_mc_list_flush(bond_dev
, slave_dev
);
2217 netif_addr_unlock_bh(bond_dev
);
2220 netdev_rx_handler_unregister(slave_dev
);
2221 netdev_set_bond_master(slave_dev
, NULL
);
2223 slave_disable_netpoll(slave
);
2225 /* close slave before restoring its mac address */
2226 dev_close(slave_dev
);
2228 if (!bond
->params
.fail_over_mac
) {
2229 /* restore original ("permanent") mac address*/
2230 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2231 addr
.sa_family
= slave_dev
->type
;
2232 dev_set_mac_address(slave_dev
, &addr
);
2235 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2236 IFF_SLAVE_INACTIVE
);
2240 /* re-acquire the lock before getting the next slave */
2241 write_lock_bh(&bond
->lock
);
2244 /* zero the mac address of the master so it will be
2245 * set by the application to the mac address of the
2248 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2251 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2253 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2254 bond_dev
->name
, bond_dev
->name
);
2255 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2259 pr_info("%s: released all slaves\n", bond_dev
->name
);
2262 write_unlock_bh(&bond
->lock
);
2267 * This function changes the active slave to slave <slave_dev>.
2268 * It returns -EINVAL in the following cases.
2269 * - <slave_dev> is not found in the list.
2270 * - There is not active slave now.
2271 * - <slave_dev> is already active.
2272 * - The link state of <slave_dev> is not BOND_LINK_UP.
2273 * - <slave_dev> is not running.
2274 * In these cases, this function does nothing.
2275 * In the other cases, current_slave pointer is changed and 0 is returned.
2277 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2279 struct bonding
*bond
= netdev_priv(bond_dev
);
2280 struct slave
*old_active
= NULL
;
2281 struct slave
*new_active
= NULL
;
2284 if (!USES_PRIMARY(bond
->params
.mode
))
2287 /* Verify that master_dev is indeed the master of slave_dev */
2288 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2291 read_lock(&bond
->lock
);
2293 read_lock(&bond
->curr_slave_lock
);
2294 old_active
= bond
->curr_active_slave
;
2295 read_unlock(&bond
->curr_slave_lock
);
2297 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2300 * Changing to the current active: do nothing; return success.
2302 if (new_active
&& (new_active
== old_active
)) {
2303 read_unlock(&bond
->lock
);
2309 (new_active
->link
== BOND_LINK_UP
) &&
2310 IS_UP(new_active
->dev
)) {
2312 write_lock_bh(&bond
->curr_slave_lock
);
2313 bond_change_active_slave(bond
, new_active
);
2314 write_unlock_bh(&bond
->curr_slave_lock
);
2315 unblock_netpoll_tx();
2319 read_unlock(&bond
->lock
);
2324 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2326 struct bonding
*bond
= netdev_priv(bond_dev
);
2328 info
->bond_mode
= bond
->params
.mode
;
2329 info
->miimon
= bond
->params
.miimon
;
2331 read_lock(&bond
->lock
);
2332 info
->num_slaves
= bond
->slave_cnt
;
2333 read_unlock(&bond
->lock
);
2338 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2340 struct bonding
*bond
= netdev_priv(bond_dev
);
2341 struct slave
*slave
;
2342 int i
, res
= -ENODEV
;
2344 read_lock(&bond
->lock
);
2346 bond_for_each_slave(bond
, slave
, i
) {
2347 if (i
== (int)info
->slave_id
) {
2349 strcpy(info
->slave_name
, slave
->dev
->name
);
2350 info
->link
= slave
->link
;
2351 info
->state
= slave
->state
;
2352 info
->link_failure_count
= slave
->link_failure_count
;
2357 read_unlock(&bond
->lock
);
2362 /*-------------------------------- Monitoring -------------------------------*/
2365 static int bond_miimon_inspect(struct bonding
*bond
)
2367 struct slave
*slave
;
2368 int i
, link_state
, commit
= 0;
2369 bool ignore_updelay
;
2371 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2373 bond_for_each_slave(bond
, slave
, i
) {
2374 slave
->new_link
= BOND_LINK_NOCHANGE
;
2376 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2378 switch (slave
->link
) {
2383 slave
->link
= BOND_LINK_FAIL
;
2384 slave
->delay
= bond
->params
.downdelay
;
2386 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2388 (bond
->params
.mode
==
2389 BOND_MODE_ACTIVEBACKUP
) ?
2390 ((slave
->state
== BOND_STATE_ACTIVE
) ?
2391 "active " : "backup ") : "",
2393 bond
->params
.downdelay
* bond
->params
.miimon
);
2396 case BOND_LINK_FAIL
:
2399 * recovered before downdelay expired
2401 slave
->link
= BOND_LINK_UP
;
2402 slave
->jiffies
= jiffies
;
2403 pr_info("%s: link status up again after %d ms for interface %s.\n",
2405 (bond
->params
.downdelay
- slave
->delay
) *
2406 bond
->params
.miimon
,
2411 if (slave
->delay
<= 0) {
2412 slave
->new_link
= BOND_LINK_DOWN
;
2420 case BOND_LINK_DOWN
:
2424 slave
->link
= BOND_LINK_BACK
;
2425 slave
->delay
= bond
->params
.updelay
;
2428 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2429 bond
->dev
->name
, slave
->dev
->name
,
2430 ignore_updelay
? 0 :
2431 bond
->params
.updelay
*
2432 bond
->params
.miimon
);
2435 case BOND_LINK_BACK
:
2437 slave
->link
= BOND_LINK_DOWN
;
2438 pr_info("%s: link status down again after %d ms for interface %s.\n",
2440 (bond
->params
.updelay
- slave
->delay
) *
2441 bond
->params
.miimon
,
2450 if (slave
->delay
<= 0) {
2451 slave
->new_link
= BOND_LINK_UP
;
2453 ignore_updelay
= false;
2465 static void bond_miimon_commit(struct bonding
*bond
)
2467 struct slave
*slave
;
2470 bond_for_each_slave(bond
, slave
, i
) {
2471 switch (slave
->new_link
) {
2472 case BOND_LINK_NOCHANGE
:
2476 slave
->link
= BOND_LINK_UP
;
2477 slave
->jiffies
= jiffies
;
2479 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2480 /* prevent it from being the active one */
2481 slave
->state
= BOND_STATE_BACKUP
;
2482 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2483 /* make it immediately active */
2484 slave
->state
= BOND_STATE_ACTIVE
;
2485 } else if (slave
!= bond
->primary_slave
) {
2486 /* prevent it from being the active one */
2487 slave
->state
= BOND_STATE_BACKUP
;
2490 bond_update_speed_duplex(slave
);
2492 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2493 bond
->dev
->name
, slave
->dev
->name
,
2494 slave
->speed
, slave
->duplex
? "full" : "half");
2496 /* notify ad that the link status has changed */
2497 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2498 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2500 if (bond_is_lb(bond
))
2501 bond_alb_handle_link_change(bond
, slave
,
2504 if (!bond
->curr_active_slave
||
2505 (slave
== bond
->primary_slave
))
2510 case BOND_LINK_DOWN
:
2511 if (slave
->link_failure_count
< UINT_MAX
)
2512 slave
->link_failure_count
++;
2514 slave
->link
= BOND_LINK_DOWN
;
2516 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2517 bond
->params
.mode
== BOND_MODE_8023AD
)
2518 bond_set_slave_inactive_flags(slave
);
2520 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2521 bond
->dev
->name
, slave
->dev
->name
);
2523 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2524 bond_3ad_handle_link_change(slave
,
2527 if (bond_is_lb(bond
))
2528 bond_alb_handle_link_change(bond
, slave
,
2531 if (slave
== bond
->curr_active_slave
)
2537 pr_err("%s: invalid new link %d on slave %s\n",
2538 bond
->dev
->name
, slave
->new_link
,
2540 slave
->new_link
= BOND_LINK_NOCHANGE
;
2548 write_lock_bh(&bond
->curr_slave_lock
);
2549 bond_select_active_slave(bond
);
2550 write_unlock_bh(&bond
->curr_slave_lock
);
2551 unblock_netpoll_tx();
2554 bond_set_carrier(bond
);
2560 * Really a wrapper that splits the mii monitor into two phases: an
2561 * inspection, then (if inspection indicates something needs to be done)
2562 * an acquisition of appropriate locks followed by a commit phase to
2563 * implement whatever link state changes are indicated.
2565 void bond_mii_monitor(struct work_struct
*work
)
2567 struct bonding
*bond
= container_of(work
, struct bonding
,
2570 read_lock(&bond
->lock
);
2571 if (bond
->kill_timers
)
2574 if (bond
->slave_cnt
== 0)
2577 if (bond
->send_grat_arp
) {
2578 read_lock(&bond
->curr_slave_lock
);
2579 bond_send_gratuitous_arp(bond
);
2580 read_unlock(&bond
->curr_slave_lock
);
2583 if (bond
->send_unsol_na
) {
2584 read_lock(&bond
->curr_slave_lock
);
2585 bond_send_unsolicited_na(bond
);
2586 read_unlock(&bond
->curr_slave_lock
);
2589 if (bond_miimon_inspect(bond
)) {
2590 read_unlock(&bond
->lock
);
2592 read_lock(&bond
->lock
);
2594 bond_miimon_commit(bond
);
2596 read_unlock(&bond
->lock
);
2597 rtnl_unlock(); /* might sleep, hold no other locks */
2598 read_lock(&bond
->lock
);
2602 if (bond
->params
.miimon
)
2603 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2604 msecs_to_jiffies(bond
->params
.miimon
));
2606 read_unlock(&bond
->lock
);
2609 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2611 struct in_device
*idev
;
2612 struct in_ifaddr
*ifa
;
2619 idev
= __in_dev_get_rcu(dev
);
2623 ifa
= idev
->ifa_list
;
2627 addr
= ifa
->ifa_local
;
2633 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2635 struct vlan_entry
*vlan
;
2637 if (ip
== bond
->master_ip
)
2640 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2641 if (ip
== vlan
->vlan_ip
)
2649 * We go to the (large) trouble of VLAN tagging ARP frames because
2650 * switches in VLAN mode (especially if ports are configured as
2651 * "native" to a VLAN) might not pass non-tagged frames.
2653 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2655 struct sk_buff
*skb
;
2657 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2658 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2660 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2661 NULL
, slave_dev
->dev_addr
, NULL
);
2664 pr_err("ARP packet allocation failed\n");
2668 skb
= vlan_put_tag(skb
, vlan_id
);
2670 pr_err("failed to insert VLAN tag\n");
2678 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2681 __be32
*targets
= bond
->params
.arp_targets
;
2682 struct vlan_entry
*vlan
;
2683 struct net_device
*vlan_dev
;
2687 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2690 pr_debug("basa: target %x\n", targets
[i
]);
2692 pr_debug("basa: empty vlan: arp_send\n");
2693 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2694 bond
->master_ip
, 0);
2699 * If VLANs are configured, we do a route lookup to
2700 * determine which VLAN interface would be used, so we
2701 * can tag the ARP with the proper VLAN tag.
2703 memset(&fl
, 0, sizeof(fl
));
2704 fl
.fl4_dst
= targets
[i
];
2705 fl
.fl4_tos
= RTO_ONLINK
;
2707 rt
= ip_route_output_key(dev_net(bond
->dev
), &fl
);
2709 if (net_ratelimit()) {
2710 pr_warning("%s: no route to arp_ip_target %pI4\n",
2711 bond
->dev
->name
, &fl
.fl4_dst
);
2717 * This target is not on a VLAN
2719 if (rt
->dst
.dev
== bond
->dev
) {
2721 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2722 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2723 bond
->master_ip
, 0);
2728 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2729 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2730 if (vlan_dev
== rt
->dst
.dev
) {
2731 vlan_id
= vlan
->vlan_id
;
2732 pr_debug("basa: vlan match on %s %d\n",
2733 vlan_dev
->name
, vlan_id
);
2740 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2741 vlan
->vlan_ip
, vlan_id
);
2745 if (net_ratelimit()) {
2746 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2747 bond
->dev
->name
, &fl
.fl4_dst
,
2748 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2755 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2756 * for each VLAN above us.
2758 * Caller must hold curr_slave_lock for read or better
2760 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2762 struct slave
*slave
= bond
->curr_active_slave
;
2763 struct vlan_entry
*vlan
;
2764 struct net_device
*vlan_dev
;
2766 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2767 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
2769 if (!slave
|| !bond
->send_grat_arp
||
2770 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2773 bond
->send_grat_arp
--;
2775 if (bond
->master_ip
) {
2776 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2777 bond
->master_ip
, 0);
2783 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2784 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2785 if (vlan
->vlan_ip
) {
2786 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2787 vlan
->vlan_ip
, vlan
->vlan_id
);
2792 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2795 __be32
*targets
= bond
->params
.arp_targets
;
2797 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2798 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2799 &sip
, &tip
, i
, &targets
[i
],
2800 bond_has_this_ip(bond
, tip
));
2801 if (sip
== targets
[i
]) {
2802 if (bond_has_this_ip(bond
, tip
))
2803 slave
->last_arp_rx
= jiffies
;
2809 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2812 struct slave
*slave
;
2813 struct bonding
*bond
;
2814 unsigned char *arp_ptr
;
2817 if (dev
->priv_flags
& IFF_802_1Q_VLAN
) {
2819 * When using VLANS and bonding, dev and oriv_dev may be
2820 * incorrect if the physical interface supports VLAN
2821 * acceleration. With this change ARP validation now
2822 * works for hosts only reachable on the VLAN interface.
2824 dev
= vlan_dev_real_dev(dev
);
2825 orig_dev
= dev_get_by_index_rcu(dev_net(skb
->dev
),skb
->skb_iif
);
2828 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2831 bond
= netdev_priv(dev
);
2832 read_lock(&bond
->lock
);
2834 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2835 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2836 orig_dev
? orig_dev
->name
: "NULL");
2838 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2839 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2842 skb
= skb_share_check(skb
, GFP_ATOMIC
);
2846 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2850 if (arp
->ar_hln
!= dev
->addr_len
||
2851 skb
->pkt_type
== PACKET_OTHERHOST
||
2852 skb
->pkt_type
== PACKET_LOOPBACK
||
2853 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2854 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2858 arp_ptr
= (unsigned char *)(arp
+ 1);
2859 arp_ptr
+= dev
->addr_len
;
2860 memcpy(&sip
, arp_ptr
, 4);
2861 arp_ptr
+= 4 + dev
->addr_len
;
2862 memcpy(&tip
, arp_ptr
, 4);
2864 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2865 bond
->dev
->name
, slave
->dev
->name
, slave
->state
,
2866 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2870 * Backup slaves won't see the ARP reply, but do come through
2871 * here for each ARP probe (so we swap the sip/tip to validate
2872 * the probe). In a "redundant switch, common router" type of
2873 * configuration, the ARP probe will (hopefully) travel from
2874 * the active, through one switch, the router, then the other
2875 * switch before reaching the backup.
2877 if (slave
->state
== BOND_STATE_ACTIVE
)
2878 bond_validate_arp(bond
, slave
, sip
, tip
);
2880 bond_validate_arp(bond
, slave
, tip
, sip
);
2883 read_unlock(&bond
->lock
);
2886 return NET_RX_SUCCESS
;
2890 * this function is called regularly to monitor each slave's link
2891 * ensuring that traffic is being sent and received when arp monitoring
2892 * is used in load-balancing mode. if the adapter has been dormant, then an
2893 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2894 * arp monitoring in active backup mode.
2896 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2898 struct bonding
*bond
= container_of(work
, struct bonding
,
2900 struct slave
*slave
, *oldcurrent
;
2901 int do_failover
= 0;
2905 read_lock(&bond
->lock
);
2907 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2909 if (bond
->kill_timers
)
2912 if (bond
->slave_cnt
== 0)
2915 read_lock(&bond
->curr_slave_lock
);
2916 oldcurrent
= bond
->curr_active_slave
;
2917 read_unlock(&bond
->curr_slave_lock
);
2919 /* see if any of the previous devices are up now (i.e. they have
2920 * xmt and rcv traffic). the curr_active_slave does not come into
2921 * the picture unless it is null. also, slave->jiffies is not needed
2922 * here because we send an arp on each slave and give a slave as
2923 * long as it needs to get the tx/rx within the delta.
2924 * TODO: what about up/down delay in arp mode? it wasn't here before
2927 bond_for_each_slave(bond
, slave
, i
) {
2928 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2930 if (slave
->link
!= BOND_LINK_UP
) {
2931 if (time_in_range(jiffies
,
2932 trans_start
- delta_in_ticks
,
2933 trans_start
+ delta_in_ticks
) &&
2934 time_in_range(jiffies
,
2935 slave
->dev
->last_rx
- delta_in_ticks
,
2936 slave
->dev
->last_rx
+ delta_in_ticks
)) {
2938 slave
->link
= BOND_LINK_UP
;
2939 slave
->state
= BOND_STATE_ACTIVE
;
2941 /* primary_slave has no meaning in round-robin
2942 * mode. the window of a slave being up and
2943 * curr_active_slave being null after enslaving
2947 pr_info("%s: link status definitely up for interface %s, ",
2952 pr_info("%s: interface %s is now up\n",
2958 /* slave->link == BOND_LINK_UP */
2960 /* not all switches will respond to an arp request
2961 * when the source ip is 0, so don't take the link down
2962 * if we don't know our ip yet
2964 if (!time_in_range(jiffies
,
2965 trans_start
- delta_in_ticks
,
2966 trans_start
+ 2 * delta_in_ticks
) ||
2967 !time_in_range(jiffies
,
2968 slave
->dev
->last_rx
- delta_in_ticks
,
2969 slave
->dev
->last_rx
+ 2 * delta_in_ticks
)) {
2971 slave
->link
= BOND_LINK_DOWN
;
2972 slave
->state
= BOND_STATE_BACKUP
;
2974 if (slave
->link_failure_count
< UINT_MAX
)
2975 slave
->link_failure_count
++;
2977 pr_info("%s: interface %s is now down.\n",
2981 if (slave
== oldcurrent
)
2986 /* note: if switch is in round-robin mode, all links
2987 * must tx arp to ensure all links rx an arp - otherwise
2988 * links may oscillate or not come up at all; if switch is
2989 * in something like xor mode, there is nothing we can
2990 * do - all replies will be rx'ed on same link causing slaves
2991 * to be unstable during low/no traffic periods
2993 if (IS_UP(slave
->dev
))
2994 bond_arp_send_all(bond
, slave
);
2999 write_lock_bh(&bond
->curr_slave_lock
);
3001 bond_select_active_slave(bond
);
3003 write_unlock_bh(&bond
->curr_slave_lock
);
3004 unblock_netpoll_tx();
3008 if (bond
->params
.arp_interval
)
3009 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3011 read_unlock(&bond
->lock
);
3015 * Called to inspect slaves for active-backup mode ARP monitor link state
3016 * changes. Sets new_link in slaves to specify what action should take
3017 * place for the slave. Returns 0 if no changes are found, >0 if changes
3018 * to link states must be committed.
3020 * Called with bond->lock held for read.
3022 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
3024 struct slave
*slave
;
3026 unsigned long trans_start
;
3028 bond_for_each_slave(bond
, slave
, i
) {
3029 slave
->new_link
= BOND_LINK_NOCHANGE
;
3031 if (slave
->link
!= BOND_LINK_UP
) {
3032 if (time_in_range(jiffies
,
3033 slave_last_rx(bond
, slave
) - delta_in_ticks
,
3034 slave_last_rx(bond
, slave
) + delta_in_ticks
)) {
3036 slave
->new_link
= BOND_LINK_UP
;
3044 * Give slaves 2*delta after being enslaved or made
3045 * active. This avoids bouncing, as the last receive
3046 * times need a full ARP monitor cycle to be updated.
3048 if (time_in_range(jiffies
,
3049 slave
->jiffies
- delta_in_ticks
,
3050 slave
->jiffies
+ 2 * delta_in_ticks
))
3054 * Backup slave is down if:
3055 * - No current_arp_slave AND
3056 * - more than 3*delta since last receive AND
3057 * - the bond has an IP address
3059 * Note: a non-null current_arp_slave indicates
3060 * the curr_active_slave went down and we are
3061 * searching for a new one; under this condition
3062 * we only take the curr_active_slave down - this
3063 * gives each slave a chance to tx/rx traffic
3064 * before being taken out
3066 if (slave
->state
== BOND_STATE_BACKUP
&&
3067 !bond
->current_arp_slave
&&
3068 !time_in_range(jiffies
,
3069 slave_last_rx(bond
, slave
) - delta_in_ticks
,
3070 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
)) {
3072 slave
->new_link
= BOND_LINK_DOWN
;
3077 * Active slave is down if:
3078 * - more than 2*delta since transmitting OR
3079 * - (more than 2*delta since receive AND
3080 * the bond has an IP address)
3082 trans_start
= dev_trans_start(slave
->dev
);
3083 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
3084 (!time_in_range(jiffies
,
3085 trans_start
- delta_in_ticks
,
3086 trans_start
+ 2 * delta_in_ticks
) ||
3087 !time_in_range(jiffies
,
3088 slave_last_rx(bond
, slave
) - delta_in_ticks
,
3089 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
))) {
3091 slave
->new_link
= BOND_LINK_DOWN
;
3100 * Called to commit link state changes noted by inspection step of
3101 * active-backup mode ARP monitor.
3103 * Called with RTNL and bond->lock for read.
3105 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
3107 struct slave
*slave
;
3109 unsigned long trans_start
;
3111 bond_for_each_slave(bond
, slave
, i
) {
3112 switch (slave
->new_link
) {
3113 case BOND_LINK_NOCHANGE
:
3117 trans_start
= dev_trans_start(slave
->dev
);
3118 if ((!bond
->curr_active_slave
&&
3119 time_in_range(jiffies
,
3120 trans_start
- delta_in_ticks
,
3121 trans_start
+ delta_in_ticks
)) ||
3122 bond
->curr_active_slave
!= slave
) {
3123 slave
->link
= BOND_LINK_UP
;
3124 bond
->current_arp_slave
= NULL
;
3126 pr_info("%s: link status definitely up for interface %s.\n",
3127 bond
->dev
->name
, slave
->dev
->name
);
3129 if (!bond
->curr_active_slave
||
3130 (slave
== bond
->primary_slave
))
3137 case BOND_LINK_DOWN
:
3138 if (slave
->link_failure_count
< UINT_MAX
)
3139 slave
->link_failure_count
++;
3141 slave
->link
= BOND_LINK_DOWN
;
3142 bond_set_slave_inactive_flags(slave
);
3144 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3145 bond
->dev
->name
, slave
->dev
->name
);
3147 if (slave
== bond
->curr_active_slave
) {
3148 bond
->current_arp_slave
= NULL
;
3155 pr_err("%s: impossible: new_link %d on slave %s\n",
3156 bond
->dev
->name
, slave
->new_link
,
3164 write_lock_bh(&bond
->curr_slave_lock
);
3165 bond_select_active_slave(bond
);
3166 write_unlock_bh(&bond
->curr_slave_lock
);
3167 unblock_netpoll_tx();
3170 bond_set_carrier(bond
);
3174 * Send ARP probes for active-backup mode ARP monitor.
3176 * Called with bond->lock held for read.
3178 static void bond_ab_arp_probe(struct bonding
*bond
)
3180 struct slave
*slave
;
3183 read_lock(&bond
->curr_slave_lock
);
3185 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3186 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3187 bond
->current_arp_slave
->dev
->name
,
3188 bond
->curr_active_slave
->dev
->name
);
3190 if (bond
->curr_active_slave
) {
3191 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3192 read_unlock(&bond
->curr_slave_lock
);
3196 read_unlock(&bond
->curr_slave_lock
);
3198 /* if we don't have a curr_active_slave, search for the next available
3199 * backup slave from the current_arp_slave and make it the candidate
3200 * for becoming the curr_active_slave
3203 if (!bond
->current_arp_slave
) {
3204 bond
->current_arp_slave
= bond
->first_slave
;
3205 if (!bond
->current_arp_slave
)
3209 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3211 /* search for next candidate */
3212 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3213 if (IS_UP(slave
->dev
)) {
3214 slave
->link
= BOND_LINK_BACK
;
3215 bond_set_slave_active_flags(slave
);
3216 bond_arp_send_all(bond
, slave
);
3217 slave
->jiffies
= jiffies
;
3218 bond
->current_arp_slave
= slave
;
3222 /* if the link state is up at this point, we
3223 * mark it down - this can happen if we have
3224 * simultaneous link failures and
3225 * reselect_active_interface doesn't make this
3226 * one the current slave so it is still marked
3227 * up when it is actually down
3229 if (slave
->link
== BOND_LINK_UP
) {
3230 slave
->link
= BOND_LINK_DOWN
;
3231 if (slave
->link_failure_count
< UINT_MAX
)
3232 slave
->link_failure_count
++;
3234 bond_set_slave_inactive_flags(slave
);
3236 pr_info("%s: backup interface %s is now down.\n",
3237 bond
->dev
->name
, slave
->dev
->name
);
3242 void bond_activebackup_arp_mon(struct work_struct
*work
)
3244 struct bonding
*bond
= container_of(work
, struct bonding
,
3248 read_lock(&bond
->lock
);
3250 if (bond
->kill_timers
)
3253 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3255 if (bond
->slave_cnt
== 0)
3258 if (bond
->send_grat_arp
) {
3259 read_lock(&bond
->curr_slave_lock
);
3260 bond_send_gratuitous_arp(bond
);
3261 read_unlock(&bond
->curr_slave_lock
);
3264 if (bond
->send_unsol_na
) {
3265 read_lock(&bond
->curr_slave_lock
);
3266 bond_send_unsolicited_na(bond
);
3267 read_unlock(&bond
->curr_slave_lock
);
3270 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3271 read_unlock(&bond
->lock
);
3273 read_lock(&bond
->lock
);
3275 bond_ab_arp_commit(bond
, delta_in_ticks
);
3277 read_unlock(&bond
->lock
);
3279 read_lock(&bond
->lock
);
3282 bond_ab_arp_probe(bond
);
3285 if (bond
->params
.arp_interval
)
3286 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3288 read_unlock(&bond
->lock
);
3291 /*------------------------------ proc/seq_file-------------------------------*/
3293 #ifdef CONFIG_PROC_FS
3295 static void *bond_info_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3297 __acquires(&bond
->lock
)
3299 struct bonding
*bond
= seq
->private;
3301 struct slave
*slave
;
3304 /* make sure the bond won't be taken away */
3306 read_lock(&bond
->lock
);
3309 return SEQ_START_TOKEN
;
3311 bond_for_each_slave(bond
, slave
, i
) {
3319 static void *bond_info_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3321 struct bonding
*bond
= seq
->private;
3322 struct slave
*slave
= v
;
3325 if (v
== SEQ_START_TOKEN
)
3326 return bond
->first_slave
;
3328 slave
= slave
->next
;
3330 return (slave
== bond
->first_slave
) ? NULL
: slave
;
3333 static void bond_info_seq_stop(struct seq_file
*seq
, void *v
)
3334 __releases(&bond
->lock
)
3337 struct bonding
*bond
= seq
->private;
3339 read_unlock(&bond
->lock
);
3343 static void bond_info_show_master(struct seq_file
*seq
)
3345 struct bonding
*bond
= seq
->private;
3349 read_lock(&bond
->curr_slave_lock
);
3350 curr
= bond
->curr_active_slave
;
3351 read_unlock(&bond
->curr_slave_lock
);
3353 seq_printf(seq
, "Bonding Mode: %s",
3354 bond_mode_name(bond
->params
.mode
));
3356 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
&&
3357 bond
->params
.fail_over_mac
)
3358 seq_printf(seq
, " (fail_over_mac %s)",
3359 fail_over_mac_tbl
[bond
->params
.fail_over_mac
].modename
);
3361 seq_printf(seq
, "\n");
3363 if (bond
->params
.mode
== BOND_MODE_XOR
||
3364 bond
->params
.mode
== BOND_MODE_8023AD
) {
3365 seq_printf(seq
, "Transmit Hash Policy: %s (%d)\n",
3366 xmit_hashtype_tbl
[bond
->params
.xmit_policy
].modename
,
3367 bond
->params
.xmit_policy
);
3370 if (USES_PRIMARY(bond
->params
.mode
)) {
3371 seq_printf(seq
, "Primary Slave: %s",
3372 (bond
->primary_slave
) ?
3373 bond
->primary_slave
->dev
->name
: "None");
3374 if (bond
->primary_slave
)
3375 seq_printf(seq
, " (primary_reselect %s)",
3376 pri_reselect_tbl
[bond
->params
.primary_reselect
].modename
);
3378 seq_printf(seq
, "\nCurrently Active Slave: %s\n",
3379 (curr
) ? curr
->dev
->name
: "None");
3382 seq_printf(seq
, "MII Status: %s\n", netif_carrier_ok(bond
->dev
) ?
3384 seq_printf(seq
, "MII Polling Interval (ms): %d\n", bond
->params
.miimon
);
3385 seq_printf(seq
, "Up Delay (ms): %d\n",
3386 bond
->params
.updelay
* bond
->params
.miimon
);
3387 seq_printf(seq
, "Down Delay (ms): %d\n",
3388 bond
->params
.downdelay
* bond
->params
.miimon
);
3391 /* ARP information */
3392 if (bond
->params
.arp_interval
> 0) {
3394 seq_printf(seq
, "ARP Polling Interval (ms): %d\n",
3395 bond
->params
.arp_interval
);
3397 seq_printf(seq
, "ARP IP target/s (n.n.n.n form):");
3399 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
3400 if (!bond
->params
.arp_targets
[i
])
3403 seq_printf(seq
, ",");
3404 seq_printf(seq
, " %pI4", &bond
->params
.arp_targets
[i
]);
3407 seq_printf(seq
, "\n");
3410 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3411 struct ad_info ad_info
;
3413 seq_puts(seq
, "\n802.3ad info\n");
3414 seq_printf(seq
, "LACP rate: %s\n",
3415 (bond
->params
.lacp_fast
) ? "fast" : "slow");
3416 seq_printf(seq
, "Aggregator selection policy (ad_select): %s\n",
3417 ad_select_tbl
[bond
->params
.ad_select
].modename
);
3419 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3420 seq_printf(seq
, "bond %s has no active aggregator\n",
3423 seq_printf(seq
, "Active Aggregator Info:\n");
3425 seq_printf(seq
, "\tAggregator ID: %d\n",
3426 ad_info
.aggregator_id
);
3427 seq_printf(seq
, "\tNumber of ports: %d\n",
3429 seq_printf(seq
, "\tActor Key: %d\n",
3431 seq_printf(seq
, "\tPartner Key: %d\n",
3432 ad_info
.partner_key
);
3433 seq_printf(seq
, "\tPartner Mac Address: %pM\n",
3434 ad_info
.partner_system
);
3439 static void bond_info_show_slave(struct seq_file
*seq
,
3440 const struct slave
*slave
)
3442 struct bonding
*bond
= seq
->private;
3444 seq_printf(seq
, "\nSlave Interface: %s\n", slave
->dev
->name
);
3445 seq_printf(seq
, "MII Status: %s\n",
3446 (slave
->link
== BOND_LINK_UP
) ? "up" : "down");
3447 seq_printf(seq
, "Speed: %d Mbps\n", slave
->speed
);
3448 seq_printf(seq
, "Duplex: %s\n", slave
->duplex
? "full" : "half");
3449 seq_printf(seq
, "Link Failure Count: %u\n",
3450 slave
->link_failure_count
);
3452 seq_printf(seq
, "Permanent HW addr: %pM\n", slave
->perm_hwaddr
);
3454 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3455 const struct aggregator
*agg
3456 = SLAVE_AD_INFO(slave
).port
.aggregator
;
3459 seq_printf(seq
, "Aggregator ID: %d\n",
3460 agg
->aggregator_identifier
);
3462 seq_puts(seq
, "Aggregator ID: N/A\n");
3464 seq_printf(seq
, "Slave queue ID: %d\n", slave
->queue_id
);
3467 static int bond_info_seq_show(struct seq_file
*seq
, void *v
)
3469 if (v
== SEQ_START_TOKEN
) {
3470 seq_printf(seq
, "%s\n", version
);
3471 bond_info_show_master(seq
);
3473 bond_info_show_slave(seq
, v
);
3478 static const struct seq_operations bond_info_seq_ops
= {
3479 .start
= bond_info_seq_start
,
3480 .next
= bond_info_seq_next
,
3481 .stop
= bond_info_seq_stop
,
3482 .show
= bond_info_seq_show
,
3485 static int bond_info_open(struct inode
*inode
, struct file
*file
)
3487 struct seq_file
*seq
;
3488 struct proc_dir_entry
*proc
;
3491 res
= seq_open(file
, &bond_info_seq_ops
);
3493 /* recover the pointer buried in proc_dir_entry data */
3494 seq
= file
->private_data
;
3496 seq
->private = proc
->data
;
3502 static const struct file_operations bond_info_fops
= {
3503 .owner
= THIS_MODULE
,
3504 .open
= bond_info_open
,
3506 .llseek
= seq_lseek
,
3507 .release
= seq_release
,
3510 static void bond_create_proc_entry(struct bonding
*bond
)
3512 struct net_device
*bond_dev
= bond
->dev
;
3513 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
3516 bond
->proc_entry
= proc_create_data(bond_dev
->name
,
3517 S_IRUGO
, bn
->proc_dir
,
3518 &bond_info_fops
, bond
);
3519 if (bond
->proc_entry
== NULL
)
3520 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3521 DRV_NAME
, bond_dev
->name
);
3523 memcpy(bond
->proc_file_name
, bond_dev
->name
, IFNAMSIZ
);
3527 static void bond_remove_proc_entry(struct bonding
*bond
)
3529 struct net_device
*bond_dev
= bond
->dev
;
3530 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
3532 if (bn
->proc_dir
&& bond
->proc_entry
) {
3533 remove_proc_entry(bond
->proc_file_name
, bn
->proc_dir
);
3534 memset(bond
->proc_file_name
, 0, IFNAMSIZ
);
3535 bond
->proc_entry
= NULL
;
3539 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3540 * Caller must hold rtnl_lock.
3542 static void __net_init
bond_create_proc_dir(struct bond_net
*bn
)
3544 if (!bn
->proc_dir
) {
3545 bn
->proc_dir
= proc_mkdir(DRV_NAME
, bn
->net
->proc_net
);
3547 pr_warning("Warning: cannot create /proc/net/%s\n",
3552 /* Destroy the bonding directory under /proc/net, if empty.
3553 * Caller must hold rtnl_lock.
3555 static void __net_exit
bond_destroy_proc_dir(struct bond_net
*bn
)
3558 remove_proc_entry(DRV_NAME
, bn
->net
->proc_net
);
3559 bn
->proc_dir
= NULL
;
3563 #else /* !CONFIG_PROC_FS */
3565 static void bond_create_proc_entry(struct bonding
*bond
)
3569 static void bond_remove_proc_entry(struct bonding
*bond
)
3573 static inline void bond_create_proc_dir(struct bond_net
*bn
)
3577 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
3581 #endif /* CONFIG_PROC_FS */
3584 /*-------------------------- netdev event handling --------------------------*/
3587 * Change device name
3589 static int bond_event_changename(struct bonding
*bond
)
3591 bond_remove_proc_entry(bond
);
3592 bond_create_proc_entry(bond
);
3594 bond_debug_reregister(bond
);
3599 static int bond_master_netdev_event(unsigned long event
,
3600 struct net_device
*bond_dev
)
3602 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3605 case NETDEV_CHANGENAME
:
3606 return bond_event_changename(event_bond
);
3614 static int bond_slave_netdev_event(unsigned long event
,
3615 struct net_device
*slave_dev
)
3617 struct net_device
*bond_dev
= slave_dev
->master
;
3618 struct bonding
*bond
= netdev_priv(bond_dev
);
3621 case NETDEV_UNREGISTER
:
3623 if (bond
->setup_by_slave
)
3624 bond_release_and_destroy(bond_dev
, slave_dev
);
3626 bond_release(bond_dev
, slave_dev
);
3630 if (bond
->params
.mode
== BOND_MODE_8023AD
|| bond_is_lb(bond
)) {
3631 struct slave
*slave
;
3633 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3635 u16 old_speed
= slave
->speed
;
3636 u16 old_duplex
= slave
->duplex
;
3638 bond_update_speed_duplex(slave
);
3640 if (bond_is_lb(bond
))
3643 if (old_speed
!= slave
->speed
)
3644 bond_3ad_adapter_speed_changed(slave
);
3645 if (old_duplex
!= slave
->duplex
)
3646 bond_3ad_adapter_duplex_changed(slave
);
3653 * ... Or is it this?
3656 case NETDEV_CHANGEMTU
:
3658 * TODO: Should slaves be allowed to
3659 * independently alter their MTU? For
3660 * an active-backup bond, slaves need
3661 * not be the same type of device, so
3662 * MTUs may vary. For other modes,
3663 * slaves arguably should have the
3664 * same MTUs. To do this, we'd need to
3665 * take over the slave's change_mtu
3666 * function for the duration of their
3670 case NETDEV_CHANGENAME
:
3672 * TODO: handle changing the primary's name
3675 case NETDEV_FEAT_CHANGE
:
3676 bond_compute_features(bond
);
3686 * bond_netdev_event: handle netdev notifier chain events.
3688 * This function receives events for the netdev chain. The caller (an
3689 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3690 * locks for us to safely manipulate the slave devices (RTNL lock,
3693 static int bond_netdev_event(struct notifier_block
*this,
3694 unsigned long event
, void *ptr
)
3696 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3698 pr_debug("event_dev: %s, event: %lx\n",
3699 event_dev
? event_dev
->name
: "None",
3702 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3705 if (event_dev
->flags
& IFF_MASTER
) {
3706 pr_debug("IFF_MASTER\n");
3707 return bond_master_netdev_event(event
, event_dev
);
3710 if (event_dev
->flags
& IFF_SLAVE
) {
3711 pr_debug("IFF_SLAVE\n");
3712 return bond_slave_netdev_event(event
, event_dev
);
3719 * bond_inetaddr_event: handle inetaddr notifier chain events.
3721 * We keep track of device IPs primarily to use as source addresses in
3722 * ARP monitor probes (rather than spewing out broadcasts all the time).
3724 * We track one IP for the main device (if it has one), plus one per VLAN.
3726 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3728 struct in_ifaddr
*ifa
= ptr
;
3729 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3730 struct bond_net
*bn
= net_generic(dev_net(event_dev
), bond_net_id
);
3731 struct bonding
*bond
;
3732 struct vlan_entry
*vlan
;
3734 list_for_each_entry(bond
, &bn
->dev_list
, bond_list
) {
3735 if (bond
->dev
== event_dev
) {
3738 bond
->master_ip
= ifa
->ifa_local
;
3741 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3748 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3751 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3752 if (vlan_dev
== event_dev
) {
3755 vlan
->vlan_ip
= ifa
->ifa_local
;
3759 bond_glean_dev_ip(vlan_dev
);
3770 static struct notifier_block bond_netdev_notifier
= {
3771 .notifier_call
= bond_netdev_event
,
3774 static struct notifier_block bond_inetaddr_notifier
= {
3775 .notifier_call
= bond_inetaddr_event
,
3778 /*-------------------------- Packet type handling ---------------------------*/
3780 /* register to receive lacpdus on a bond */
3781 static void bond_register_lacpdu(struct bonding
*bond
)
3783 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3785 /* initialize packet type */
3786 pk_type
->type
= PKT_TYPE_LACPDU
;
3787 pk_type
->dev
= bond
->dev
;
3788 pk_type
->func
= bond_3ad_lacpdu_recv
;
3790 dev_add_pack(pk_type
);
3793 /* unregister to receive lacpdus on a bond */
3794 static void bond_unregister_lacpdu(struct bonding
*bond
)
3796 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3799 void bond_register_arp(struct bonding
*bond
)
3801 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3806 pt
->type
= htons(ETH_P_ARP
);
3807 pt
->dev
= bond
->dev
;
3808 pt
->func
= bond_arp_rcv
;
3812 void bond_unregister_arp(struct bonding
*bond
)
3814 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3816 dev_remove_pack(pt
);
3820 /*---------------------------- Hashing Policies -----------------------------*/
3823 * Hash for the output device based upon layer 2 and layer 3 data. If
3824 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3826 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3828 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3829 struct iphdr
*iph
= ip_hdr(skb
);
3831 if (skb
->protocol
== htons(ETH_P_IP
)) {
3832 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3833 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3836 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3840 * Hash for the output device based upon layer 3 and layer 4 data. If
3841 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3842 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3844 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3846 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3847 struct iphdr
*iph
= ip_hdr(skb
);
3848 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3851 if (skb
->protocol
== htons(ETH_P_IP
)) {
3852 if (!(iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) &&
3853 (iph
->protocol
== IPPROTO_TCP
||
3854 iph
->protocol
== IPPROTO_UDP
)) {
3855 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3857 return (layer4_xor
^
3858 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3862 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3866 * Hash for the output device based upon layer 2 data
3868 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3870 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3872 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3875 /*-------------------------- Device entry points ----------------------------*/
3877 static int bond_open(struct net_device
*bond_dev
)
3879 struct bonding
*bond
= netdev_priv(bond_dev
);
3881 bond
->kill_timers
= 0;
3883 INIT_DELAYED_WORK(&bond
->mcast_work
, bond_resend_igmp_join_requests_delayed
);
3885 if (bond_is_lb(bond
)) {
3886 /* bond_alb_initialize must be called before the timer
3889 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3890 /* something went wrong - fail the open operation */
3894 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3895 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3898 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3899 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3900 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3903 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3904 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3905 INIT_DELAYED_WORK(&bond
->arp_work
,
3906 bond_activebackup_arp_mon
);
3908 INIT_DELAYED_WORK(&bond
->arp_work
,
3909 bond_loadbalance_arp_mon
);
3911 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3912 if (bond
->params
.arp_validate
)
3913 bond_register_arp(bond
);
3916 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3917 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3918 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3919 /* register to receive LACPDUs */
3920 bond_register_lacpdu(bond
);
3921 bond_3ad_initiate_agg_selection(bond
, 1);
3927 static int bond_close(struct net_device
*bond_dev
)
3929 struct bonding
*bond
= netdev_priv(bond_dev
);
3931 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3932 /* Unregister the receive of LACPDUs */
3933 bond_unregister_lacpdu(bond
);
3936 if (bond
->params
.arp_validate
)
3937 bond_unregister_arp(bond
);
3939 write_lock_bh(&bond
->lock
);
3941 bond
->send_grat_arp
= 0;
3942 bond
->send_unsol_na
= 0;
3944 /* signal timers not to re-arm */
3945 bond
->kill_timers
= 1;
3947 write_unlock_bh(&bond
->lock
);
3949 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3950 cancel_delayed_work(&bond
->mii_work
);
3953 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3954 cancel_delayed_work(&bond
->arp_work
);
3957 switch (bond
->params
.mode
) {
3958 case BOND_MODE_8023AD
:
3959 cancel_delayed_work(&bond
->ad_work
);
3963 cancel_delayed_work(&bond
->alb_work
);
3969 if (delayed_work_pending(&bond
->mcast_work
))
3970 cancel_delayed_work(&bond
->mcast_work
);
3972 if (bond_is_lb(bond
)) {
3973 /* Must be called only after all
3974 * slaves have been released
3976 bond_alb_deinitialize(bond
);
3982 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3983 struct rtnl_link_stats64
*stats
)
3985 struct bonding
*bond
= netdev_priv(bond_dev
);
3986 struct rtnl_link_stats64 temp
;
3987 struct slave
*slave
;
3990 memset(stats
, 0, sizeof(*stats
));
3992 read_lock_bh(&bond
->lock
);
3994 bond_for_each_slave(bond
, slave
, i
) {
3995 const struct rtnl_link_stats64
*sstats
=
3996 dev_get_stats(slave
->dev
, &temp
);
3998 stats
->rx_packets
+= sstats
->rx_packets
;
3999 stats
->rx_bytes
+= sstats
->rx_bytes
;
4000 stats
->rx_errors
+= sstats
->rx_errors
;
4001 stats
->rx_dropped
+= sstats
->rx_dropped
;
4003 stats
->tx_packets
+= sstats
->tx_packets
;
4004 stats
->tx_bytes
+= sstats
->tx_bytes
;
4005 stats
->tx_errors
+= sstats
->tx_errors
;
4006 stats
->tx_dropped
+= sstats
->tx_dropped
;
4008 stats
->multicast
+= sstats
->multicast
;
4009 stats
->collisions
+= sstats
->collisions
;
4011 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
4012 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
4013 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
4014 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
4015 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
4016 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
4018 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
4019 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
4020 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
4021 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
4022 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
4025 read_unlock_bh(&bond
->lock
);
4030 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
4032 struct net_device
*slave_dev
= NULL
;
4033 struct ifbond k_binfo
;
4034 struct ifbond __user
*u_binfo
= NULL
;
4035 struct ifslave k_sinfo
;
4036 struct ifslave __user
*u_sinfo
= NULL
;
4037 struct mii_ioctl_data
*mii
= NULL
;
4040 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
4052 * We do this again just in case we were called by SIOCGMIIREG
4053 * instead of SIOCGMIIPHY.
4060 if (mii
->reg_num
== 1) {
4061 struct bonding
*bond
= netdev_priv(bond_dev
);
4063 read_lock(&bond
->lock
);
4064 read_lock(&bond
->curr_slave_lock
);
4065 if (netif_carrier_ok(bond
->dev
))
4066 mii
->val_out
= BMSR_LSTATUS
;
4068 read_unlock(&bond
->curr_slave_lock
);
4069 read_unlock(&bond
->lock
);
4073 case BOND_INFO_QUERY_OLD
:
4074 case SIOCBONDINFOQUERY
:
4075 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
4077 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
4080 res
= bond_info_query(bond_dev
, &k_binfo
);
4082 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
4086 case BOND_SLAVE_INFO_QUERY_OLD
:
4087 case SIOCBONDSLAVEINFOQUERY
:
4088 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
4090 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
4093 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
4095 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
4104 if (!capable(CAP_NET_ADMIN
))
4107 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
4109 pr_debug("slave_dev=%p:\n", slave_dev
);
4114 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
4116 case BOND_ENSLAVE_OLD
:
4117 case SIOCBONDENSLAVE
:
4118 res
= bond_enslave(bond_dev
, slave_dev
);
4120 case BOND_RELEASE_OLD
:
4121 case SIOCBONDRELEASE
:
4122 res
= bond_release(bond_dev
, slave_dev
);
4124 case BOND_SETHWADDR_OLD
:
4125 case SIOCBONDSETHWADDR
:
4126 res
= bond_sethwaddr(bond_dev
, slave_dev
);
4128 case BOND_CHANGE_ACTIVE_OLD
:
4129 case SIOCBONDCHANGEACTIVE
:
4130 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
4142 static bool bond_addr_in_mc_list(unsigned char *addr
,
4143 struct netdev_hw_addr_list
*list
,
4146 struct netdev_hw_addr
*ha
;
4148 netdev_hw_addr_list_for_each(ha
, list
)
4149 if (!memcmp(ha
->addr
, addr
, addrlen
))
4155 static void bond_set_multicast_list(struct net_device
*bond_dev
)
4157 struct bonding
*bond
= netdev_priv(bond_dev
);
4158 struct netdev_hw_addr
*ha
;
4162 * Do promisc before checking multicast_mode
4164 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
))
4166 * FIXME: Need to handle the error when one of the multi-slaves
4169 bond_set_promiscuity(bond
, 1);
4172 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
))
4173 bond_set_promiscuity(bond
, -1);
4176 /* set allmulti flag to slaves */
4177 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
))
4179 * FIXME: Need to handle the error when one of the multi-slaves
4182 bond_set_allmulti(bond
, 1);
4185 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
))
4186 bond_set_allmulti(bond
, -1);
4189 read_lock(&bond
->lock
);
4191 bond
->flags
= bond_dev
->flags
;
4193 /* looking for addresses to add to slaves' mc list */
4194 netdev_for_each_mc_addr(ha
, bond_dev
) {
4195 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
4196 bond_dev
->addr_len
);
4198 bond_mc_add(bond
, ha
->addr
);
4201 /* looking for addresses to delete from slaves' list */
4202 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
4203 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
4204 bond_dev
->addr_len
);
4206 bond_mc_del(bond
, ha
->addr
);
4209 /* save master's multicast list */
4210 __hw_addr_flush(&bond
->mc_list
);
4211 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
4212 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
4214 read_unlock(&bond
->lock
);
4217 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
4219 struct bonding
*bond
= netdev_priv(dev
);
4220 struct slave
*slave
= bond
->first_slave
;
4223 const struct net_device_ops
*slave_ops
4224 = slave
->dev
->netdev_ops
;
4225 if (slave_ops
->ndo_neigh_setup
)
4226 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
4232 * Change the MTU of all of a master's slaves to match the master
4234 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
4236 struct bonding
*bond
= netdev_priv(bond_dev
);
4237 struct slave
*slave
, *stop_at
;
4241 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
4242 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
4244 /* Can't hold bond->lock with bh disabled here since
4245 * some base drivers panic. On the other hand we can't
4246 * hold bond->lock without bh disabled because we'll
4247 * deadlock. The only solution is to rely on the fact
4248 * that we're under rtnl_lock here, and the slaves
4249 * list won't change. This doesn't solve the problem
4250 * of setting the slave's MTU while it is
4251 * transmitting, but the assumption is that the base
4252 * driver can handle that.
4254 * TODO: figure out a way to safely iterate the slaves
4255 * list, but without holding a lock around the actual
4256 * call to the base driver.
4259 bond_for_each_slave(bond
, slave
, i
) {
4260 pr_debug("s %p s->p %p c_m %p\n",
4263 slave
->dev
->netdev_ops
->ndo_change_mtu
);
4265 res
= dev_set_mtu(slave
->dev
, new_mtu
);
4268 /* If we failed to set the slave's mtu to the new value
4269 * we must abort the operation even in ACTIVE_BACKUP
4270 * mode, because if we allow the backup slaves to have
4271 * different mtu values than the active slave we'll
4272 * need to change their mtu when doing a failover. That
4273 * means changing their mtu from timer context, which
4274 * is probably not a good idea.
4276 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4281 bond_dev
->mtu
= new_mtu
;
4286 /* unwind from head to the slave that failed */
4288 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4291 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
4293 pr_debug("unwind err %d dev %s\n",
4294 tmp_res
, slave
->dev
->name
);
4304 * Note that many devices must be down to change the HW address, and
4305 * downing the master releases all slaves. We can make bonds full of
4306 * bonding devices to test this, however.
4308 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4310 struct bonding
*bond
= netdev_priv(bond_dev
);
4311 struct sockaddr
*sa
= addr
, tmp_sa
;
4312 struct slave
*slave
, *stop_at
;
4316 if (bond
->params
.mode
== BOND_MODE_ALB
)
4317 return bond_alb_set_mac_address(bond_dev
, addr
);
4320 pr_debug("bond=%p, name=%s\n",
4321 bond
, bond_dev
? bond_dev
->name
: "None");
4324 * If fail_over_mac is set to active, do nothing and return
4325 * success. Returning an error causes ifenslave to fail.
4327 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4330 if (!is_valid_ether_addr(sa
->sa_data
))
4331 return -EADDRNOTAVAIL
;
4333 /* Can't hold bond->lock with bh disabled here since
4334 * some base drivers panic. On the other hand we can't
4335 * hold bond->lock without bh disabled because we'll
4336 * deadlock. The only solution is to rely on the fact
4337 * that we're under rtnl_lock here, and the slaves
4338 * list won't change. This doesn't solve the problem
4339 * of setting the slave's hw address while it is
4340 * transmitting, but the assumption is that the base
4341 * driver can handle that.
4343 * TODO: figure out a way to safely iterate the slaves
4344 * list, but without holding a lock around the actual
4345 * call to the base driver.
4348 bond_for_each_slave(bond
, slave
, i
) {
4349 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
4350 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
4352 if (slave_ops
->ndo_set_mac_address
== NULL
) {
4354 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
4358 res
= dev_set_mac_address(slave
->dev
, addr
);
4360 /* TODO: consider downing the slave
4362 * User should expect communications
4363 * breakage anyway until ARP finish
4366 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4372 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4376 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4377 tmp_sa
.sa_family
= bond_dev
->type
;
4379 /* unwind from head to the slave that failed */
4381 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4384 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4386 pr_debug("unwind err %d dev %s\n",
4387 tmp_res
, slave
->dev
->name
);
4394 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4396 struct bonding
*bond
= netdev_priv(bond_dev
);
4397 struct slave
*slave
, *start_at
;
4398 int i
, slave_no
, res
= 1;
4399 struct iphdr
*iph
= ip_hdr(skb
);
4401 read_lock(&bond
->lock
);
4403 if (!BOND_IS_OK(bond
))
4406 * Start with the curr_active_slave that joined the bond as the
4407 * default for sending IGMP traffic. For failover purposes one
4408 * needs to maintain some consistency for the interface that will
4409 * send the join/membership reports. The curr_active_slave found
4410 * will send all of this type of traffic.
4412 if ((iph
->protocol
== IPPROTO_IGMP
) &&
4413 (skb
->protocol
== htons(ETH_P_IP
))) {
4415 read_lock(&bond
->curr_slave_lock
);
4416 slave
= bond
->curr_active_slave
;
4417 read_unlock(&bond
->curr_slave_lock
);
4423 * Concurrent TX may collide on rr_tx_counter; we accept
4424 * that as being rare enough not to justify using an
4427 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4429 bond_for_each_slave(bond
, slave
, i
) {
4437 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4438 if (IS_UP(slave
->dev
) &&
4439 (slave
->link
== BOND_LINK_UP
) &&
4440 (slave
->state
== BOND_STATE_ACTIVE
)) {
4441 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4448 /* no suitable interface, frame not sent */
4451 read_unlock(&bond
->lock
);
4452 return NETDEV_TX_OK
;
4457 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4458 * the bond has a usable interface.
4460 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4462 struct bonding
*bond
= netdev_priv(bond_dev
);
4465 read_lock(&bond
->lock
);
4466 read_lock(&bond
->curr_slave_lock
);
4468 if (!BOND_IS_OK(bond
))
4471 if (!bond
->curr_active_slave
)
4474 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4478 /* no suitable interface, frame not sent */
4481 read_unlock(&bond
->curr_slave_lock
);
4482 read_unlock(&bond
->lock
);
4483 return NETDEV_TX_OK
;
4487 * In bond_xmit_xor() , we determine the output device by using a pre-
4488 * determined xmit_hash_policy(), If the selected device is not enabled,
4489 * find the next active slave.
4491 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4493 struct bonding
*bond
= netdev_priv(bond_dev
);
4494 struct slave
*slave
, *start_at
;
4499 read_lock(&bond
->lock
);
4501 if (!BOND_IS_OK(bond
))
4504 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4506 bond_for_each_slave(bond
, slave
, i
) {
4514 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4515 if (IS_UP(slave
->dev
) &&
4516 (slave
->link
== BOND_LINK_UP
) &&
4517 (slave
->state
== BOND_STATE_ACTIVE
)) {
4518 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4525 /* no suitable interface, frame not sent */
4528 read_unlock(&bond
->lock
);
4529 return NETDEV_TX_OK
;
4533 * in broadcast mode, we send everything to all usable interfaces.
4535 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4537 struct bonding
*bond
= netdev_priv(bond_dev
);
4538 struct slave
*slave
, *start_at
;
4539 struct net_device
*tx_dev
= NULL
;
4543 read_lock(&bond
->lock
);
4545 if (!BOND_IS_OK(bond
))
4548 read_lock(&bond
->curr_slave_lock
);
4549 start_at
= bond
->curr_active_slave
;
4550 read_unlock(&bond
->curr_slave_lock
);
4555 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4556 if (IS_UP(slave
->dev
) &&
4557 (slave
->link
== BOND_LINK_UP
) &&
4558 (slave
->state
== BOND_STATE_ACTIVE
)) {
4560 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4562 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4567 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4569 dev_kfree_skb(skb2
);
4573 tx_dev
= slave
->dev
;
4578 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4582 /* no suitable interface, frame not sent */
4585 /* frame sent to all suitable interfaces */
4586 read_unlock(&bond
->lock
);
4587 return NETDEV_TX_OK
;
4590 /*------------------------- Device initialization ---------------------------*/
4592 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4594 switch (bond
->params
.xmit_policy
) {
4595 case BOND_XMIT_POLICY_LAYER23
:
4596 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4598 case BOND_XMIT_POLICY_LAYER34
:
4599 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4601 case BOND_XMIT_POLICY_LAYER2
:
4603 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4609 * Lookup the slave that corresponds to a qid
4611 static inline int bond_slave_override(struct bonding
*bond
,
4612 struct sk_buff
*skb
)
4615 struct slave
*slave
= NULL
;
4616 struct slave
*check_slave
;
4618 read_lock(&bond
->lock
);
4620 if (!BOND_IS_OK(bond
) || !skb
->queue_mapping
)
4623 /* Find out if any slaves have the same mapping as this skb. */
4624 bond_for_each_slave(bond
, check_slave
, i
) {
4625 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4626 slave
= check_slave
;
4631 /* If the slave isn't UP, use default transmit policy. */
4632 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4633 (slave
->link
== BOND_LINK_UP
)) {
4634 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4638 read_unlock(&bond
->lock
);
4642 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4645 * This helper function exists to help dev_pick_tx get the correct
4646 * destination queue. Using a helper function skips the a call to
4647 * skb_tx_hash and will put the skbs in the queue we expect on their
4648 * way down to the bonding driver.
4650 return skb
->queue_mapping
;
4653 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4655 struct bonding
*bond
= netdev_priv(dev
);
4658 * If we risk deadlock from transmitting this in the
4659 * netpoll path, tell netpoll to queue the frame for later tx
4661 if (is_netpoll_tx_blocked(dev
))
4662 return NETDEV_TX_BUSY
;
4664 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4665 if (!bond_slave_override(bond
, skb
))
4666 return NETDEV_TX_OK
;
4669 switch (bond
->params
.mode
) {
4670 case BOND_MODE_ROUNDROBIN
:
4671 return bond_xmit_roundrobin(skb
, dev
);
4672 case BOND_MODE_ACTIVEBACKUP
:
4673 return bond_xmit_activebackup(skb
, dev
);
4675 return bond_xmit_xor(skb
, dev
);
4676 case BOND_MODE_BROADCAST
:
4677 return bond_xmit_broadcast(skb
, dev
);
4678 case BOND_MODE_8023AD
:
4679 return bond_3ad_xmit_xor(skb
, dev
);
4682 return bond_alb_xmit(skb
, dev
);
4684 /* Should never happen, mode already checked */
4685 pr_err("%s: Error: Unknown bonding mode %d\n",
4686 dev
->name
, bond
->params
.mode
);
4689 return NETDEV_TX_OK
;
4695 * set bond mode specific net device operations
4697 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4699 struct net_device
*bond_dev
= bond
->dev
;
4702 case BOND_MODE_ROUNDROBIN
:
4704 case BOND_MODE_ACTIVEBACKUP
:
4707 bond_set_xmit_hash_policy(bond
);
4709 case BOND_MODE_BROADCAST
:
4711 case BOND_MODE_8023AD
:
4712 bond_set_master_3ad_flags(bond
);
4713 bond_set_xmit_hash_policy(bond
);
4716 bond_set_master_alb_flags(bond
);
4721 /* Should never happen, mode already checked */
4722 pr_err("%s: Error: Unknown bonding mode %d\n",
4723 bond_dev
->name
, mode
);
4728 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4729 struct ethtool_drvinfo
*drvinfo
)
4731 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4732 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4733 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4736 static const struct ethtool_ops bond_ethtool_ops
= {
4737 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4738 .get_link
= ethtool_op_get_link
,
4739 .get_tx_csum
= ethtool_op_get_tx_csum
,
4740 .get_sg
= ethtool_op_get_sg
,
4741 .get_tso
= ethtool_op_get_tso
,
4742 .get_ufo
= ethtool_op_get_ufo
,
4743 .get_flags
= ethtool_op_get_flags
,
4746 static const struct net_device_ops bond_netdev_ops
= {
4747 .ndo_init
= bond_init
,
4748 .ndo_uninit
= bond_uninit
,
4749 .ndo_open
= bond_open
,
4750 .ndo_stop
= bond_close
,
4751 .ndo_start_xmit
= bond_start_xmit
,
4752 .ndo_select_queue
= bond_select_queue
,
4753 .ndo_get_stats64
= bond_get_stats
,
4754 .ndo_do_ioctl
= bond_do_ioctl
,
4755 .ndo_set_multicast_list
= bond_set_multicast_list
,
4756 .ndo_change_mtu
= bond_change_mtu
,
4757 .ndo_set_mac_address
= bond_set_mac_address
,
4758 .ndo_neigh_setup
= bond_neigh_setup
,
4759 .ndo_vlan_rx_register
= bond_vlan_rx_register
,
4760 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4761 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4762 #ifdef CONFIG_NET_POLL_CONTROLLER
4763 .ndo_netpoll_setup
= bond_netpoll_setup
,
4764 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4765 .ndo_poll_controller
= bond_poll_controller
,
4767 .ndo_add_slave
= bond_enslave
,
4768 .ndo_del_slave
= bond_release
,
4771 static void bond_destructor(struct net_device
*bond_dev
)
4773 struct bonding
*bond
= netdev_priv(bond_dev
);
4775 destroy_workqueue(bond
->wq
);
4776 free_netdev(bond_dev
);
4779 static void bond_setup(struct net_device
*bond_dev
)
4781 struct bonding
*bond
= netdev_priv(bond_dev
);
4783 /* initialize rwlocks */
4784 rwlock_init(&bond
->lock
);
4785 rwlock_init(&bond
->curr_slave_lock
);
4787 bond
->params
= bonding_defaults
;
4789 /* Initialize pointers */
4790 bond
->dev
= bond_dev
;
4791 INIT_LIST_HEAD(&bond
->vlan_list
);
4793 /* Initialize the device entry points */
4794 ether_setup(bond_dev
);
4795 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4796 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4797 bond_set_mode_ops(bond
, bond
->params
.mode
);
4799 bond_dev
->destructor
= bond_destructor
;
4801 /* Initialize the device options */
4802 bond_dev
->tx_queue_len
= 0;
4803 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4804 bond_dev
->priv_flags
|= IFF_BONDING
;
4805 bond_dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
4807 if (bond
->params
.arp_interval
)
4808 bond_dev
->priv_flags
|= IFF_MASTER_ARPMON
;
4810 /* At first, we block adding VLANs. That's the only way to
4811 * prevent problems that occur when adding VLANs over an
4812 * empty bond. The block will be removed once non-challenged
4813 * slaves are enslaved.
4815 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4817 /* don't acquire bond device's netif_tx_lock when
4819 bond_dev
->features
|= NETIF_F_LLTX
;
4821 /* By default, we declare the bond to be fully
4822 * VLAN hardware accelerated capable. Special
4823 * care is taken in the various xmit functions
4824 * when there are slaves that are not hw accel
4827 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4828 NETIF_F_HW_VLAN_RX
|
4829 NETIF_F_HW_VLAN_FILTER
);
4831 /* By default, we enable GRO on bonding devices.
4832 * Actual support requires lowlevel drivers are GRO ready.
4834 bond_dev
->features
|= NETIF_F_GRO
;
4837 static void bond_work_cancel_all(struct bonding
*bond
)
4839 write_lock_bh(&bond
->lock
);
4840 bond
->kill_timers
= 1;
4841 write_unlock_bh(&bond
->lock
);
4843 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4844 cancel_delayed_work(&bond
->mii_work
);
4846 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4847 cancel_delayed_work(&bond
->arp_work
);
4849 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4850 delayed_work_pending(&bond
->alb_work
))
4851 cancel_delayed_work(&bond
->alb_work
);
4853 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4854 delayed_work_pending(&bond
->ad_work
))
4855 cancel_delayed_work(&bond
->ad_work
);
4857 if (delayed_work_pending(&bond
->mcast_work
))
4858 cancel_delayed_work(&bond
->mcast_work
);
4862 * Destroy a bonding device.
4863 * Must be under rtnl_lock when this function is called.
4865 static void bond_uninit(struct net_device
*bond_dev
)
4867 struct bonding
*bond
= netdev_priv(bond_dev
);
4868 struct vlan_entry
*vlan
, *tmp
;
4870 bond_netpoll_cleanup(bond_dev
);
4872 /* Release the bonded slaves */
4873 bond_release_all(bond_dev
);
4875 list_del(&bond
->bond_list
);
4877 bond_work_cancel_all(bond
);
4879 bond_remove_proc_entry(bond
);
4881 bond_debug_unregister(bond
);
4883 __hw_addr_flush(&bond
->mc_list
);
4885 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4886 list_del(&vlan
->vlan_list
);
4891 /*------------------------- Module initialization ---------------------------*/
4894 * Convert string input module parms. Accept either the
4895 * number of the mode or its string name. A bit complicated because
4896 * some mode names are substrings of other names, and calls from sysfs
4897 * may have whitespace in the name (trailing newlines, for example).
4899 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4901 int modeint
= -1, i
, rv
;
4902 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4904 for (p
= (char *)buf
; *p
; p
++)
4905 if (!(isdigit(*p
) || isspace(*p
)))
4909 rv
= sscanf(buf
, "%20s", modestr
);
4911 rv
= sscanf(buf
, "%d", &modeint
);
4916 for (i
= 0; tbl
[i
].modename
; i
++) {
4917 if (modeint
== tbl
[i
].mode
)
4919 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4926 static int bond_check_params(struct bond_params
*params
)
4928 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4931 * Convert string parameters.
4934 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4935 if (bond_mode
== -1) {
4936 pr_err("Error: Invalid bonding mode \"%s\"\n",
4937 mode
== NULL
? "NULL" : mode
);
4942 if (xmit_hash_policy
) {
4943 if ((bond_mode
!= BOND_MODE_XOR
) &&
4944 (bond_mode
!= BOND_MODE_8023AD
)) {
4945 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4946 bond_mode_name(bond_mode
));
4948 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4950 if (xmit_hashtype
== -1) {
4951 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4952 xmit_hash_policy
== NULL
? "NULL" :
4960 if (bond_mode
!= BOND_MODE_8023AD
) {
4961 pr_info("lacp_rate param is irrelevant in mode %s\n",
4962 bond_mode_name(bond_mode
));
4964 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4965 if (lacp_fast
== -1) {
4966 pr_err("Error: Invalid lacp rate \"%s\"\n",
4967 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4974 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4975 if (params
->ad_select
== -1) {
4976 pr_err("Error: Invalid ad_select \"%s\"\n",
4977 ad_select
== NULL
? "NULL" : ad_select
);
4981 if (bond_mode
!= BOND_MODE_8023AD
) {
4982 pr_warning("ad_select param only affects 802.3ad mode\n");
4985 params
->ad_select
= BOND_AD_STABLE
;
4988 if (max_bonds
< 0) {
4989 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4990 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4991 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4995 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4996 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4997 miimon
= BOND_LINK_MON_INTERV
;
5001 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5006 if (downdelay
< 0) {
5007 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5008 downdelay
, INT_MAX
);
5012 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
5013 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5018 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
5019 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
5024 if (num_unsol_na
< 0 || num_unsol_na
> 255) {
5025 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5030 /* reset values for 802.3ad */
5031 if (bond_mode
== BOND_MODE_8023AD
) {
5033 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");
5034 pr_warning("Forcing miimon to 100msec\n");
5039 if (tx_queues
< 1 || tx_queues
> 255) {
5040 pr_warning("Warning: tx_queues (%d) should be between "
5041 "1 and 255, resetting to %d\n",
5042 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
5043 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
5046 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
5047 pr_warning("Warning: all_slaves_active module parameter (%d), "
5048 "not of valid value (0/1), so it was set to "
5049 "0\n", all_slaves_active
);
5050 all_slaves_active
= 0;
5053 if (resend_igmp
< 0 || resend_igmp
> 255) {
5054 pr_warning("Warning: resend_igmp (%d) should be between "
5055 "0 and 255, resetting to %d\n",
5056 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
5057 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
5060 /* reset values for TLB/ALB */
5061 if ((bond_mode
== BOND_MODE_TLB
) ||
5062 (bond_mode
== BOND_MODE_ALB
)) {
5064 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");
5065 pr_warning("Forcing miimon to 100msec\n");
5070 if (bond_mode
== BOND_MODE_ALB
) {
5071 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",
5076 if (updelay
|| downdelay
) {
5077 /* just warn the user the up/down delay will have
5078 * no effect since miimon is zero...
5080 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",
5081 updelay
, downdelay
);
5084 /* don't allow arp monitoring */
5086 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5087 miimon
, arp_interval
);
5091 if ((updelay
% miimon
) != 0) {
5092 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5094 (updelay
/ miimon
) * miimon
);
5099 if ((downdelay
% miimon
) != 0) {
5100 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5102 (downdelay
/ miimon
) * miimon
);
5105 downdelay
/= miimon
;
5108 if (arp_interval
< 0) {
5109 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5110 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
5111 arp_interval
= BOND_LINK_ARP_INTERV
;
5114 for (arp_ip_count
= 0;
5115 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
5117 /* not complete check, but should be good enough to
5119 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
5120 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5121 arp_ip_target
[arp_ip_count
]);
5124 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
5125 arp_target
[arp_ip_count
] = ip
;
5129 if (arp_interval
&& !arp_ip_count
) {
5130 /* don't allow arping if no arp_ip_target given... */
5131 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5137 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
5138 pr_err("arp_validate only supported in active-backup mode\n");
5141 if (!arp_interval
) {
5142 pr_err("arp_validate requires arp_interval\n");
5146 arp_validate_value
= bond_parse_parm(arp_validate
,
5148 if (arp_validate_value
== -1) {
5149 pr_err("Error: invalid arp_validate \"%s\"\n",
5150 arp_validate
== NULL
? "NULL" : arp_validate
);
5154 arp_validate_value
= 0;
5157 pr_info("MII link monitoring set to %d ms\n", miimon
);
5158 } else if (arp_interval
) {
5161 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5163 arp_validate_tbl
[arp_validate_value
].modename
,
5166 for (i
= 0; i
< arp_ip_count
; i
++)
5167 pr_info(" %s", arp_ip_target
[i
]);
5171 } else if (max_bonds
) {
5172 /* miimon and arp_interval not set, we need one so things
5173 * work as expected, see bonding.txt for details
5175 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");
5178 if (primary
&& !USES_PRIMARY(bond_mode
)) {
5179 /* currently, using a primary only makes sense
5180 * in active backup, TLB or ALB modes
5182 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5183 primary
, bond_mode_name(bond_mode
));
5187 if (primary
&& primary_reselect
) {
5188 primary_reselect_value
= bond_parse_parm(primary_reselect
,
5190 if (primary_reselect_value
== -1) {
5191 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5193 NULL
? "NULL" : primary_reselect
);
5197 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
5200 if (fail_over_mac
) {
5201 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
5203 if (fail_over_mac_value
== -1) {
5204 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5205 arp_validate
== NULL
? "NULL" : arp_validate
);
5209 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
5210 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5212 fail_over_mac_value
= BOND_FOM_NONE
;
5215 /* fill params struct with the proper values */
5216 params
->mode
= bond_mode
;
5217 params
->xmit_policy
= xmit_hashtype
;
5218 params
->miimon
= miimon
;
5219 params
->num_grat_arp
= num_grat_arp
;
5220 params
->num_unsol_na
= num_unsol_na
;
5221 params
->arp_interval
= arp_interval
;
5222 params
->arp_validate
= arp_validate_value
;
5223 params
->updelay
= updelay
;
5224 params
->downdelay
= downdelay
;
5225 params
->use_carrier
= use_carrier
;
5226 params
->lacp_fast
= lacp_fast
;
5227 params
->primary
[0] = 0;
5228 params
->primary_reselect
= primary_reselect_value
;
5229 params
->fail_over_mac
= fail_over_mac_value
;
5230 params
->tx_queues
= tx_queues
;
5231 params
->all_slaves_active
= all_slaves_active
;
5232 params
->resend_igmp
= resend_igmp
;
5235 strncpy(params
->primary
, primary
, IFNAMSIZ
);
5236 params
->primary
[IFNAMSIZ
- 1] = 0;
5239 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
5244 static struct lock_class_key bonding_netdev_xmit_lock_key
;
5245 static struct lock_class_key bonding_netdev_addr_lock_key
;
5247 static void bond_set_lockdep_class_one(struct net_device
*dev
,
5248 struct netdev_queue
*txq
,
5251 lockdep_set_class(&txq
->_xmit_lock
,
5252 &bonding_netdev_xmit_lock_key
);
5255 static void bond_set_lockdep_class(struct net_device
*dev
)
5257 lockdep_set_class(&dev
->addr_list_lock
,
5258 &bonding_netdev_addr_lock_key
);
5259 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
5263 * Called from registration process
5265 static int bond_init(struct net_device
*bond_dev
)
5267 struct bonding
*bond
= netdev_priv(bond_dev
);
5268 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
5270 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
5272 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
5276 bond_set_lockdep_class(bond_dev
);
5278 netif_carrier_off(bond_dev
);
5280 bond_create_proc_entry(bond
);
5281 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
5283 bond_prepare_sysfs_group(bond
);
5285 bond_debug_register(bond
);
5287 __hw_addr_init(&bond
->mc_list
);
5291 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
5293 if (tb
[IFLA_ADDRESS
]) {
5294 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
5296 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
5297 return -EADDRNOTAVAIL
;
5302 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
5304 .priv_size
= sizeof(struct bonding
),
5305 .setup
= bond_setup
,
5306 .validate
= bond_validate
,
5309 /* Create a new bond based on the specified name and bonding parameters.
5310 * If name is NULL, obtain a suitable "bond%d" name for us.
5311 * Caller must NOT hold rtnl_lock; we need to release it here before we
5312 * set up our sysfs entries.
5314 int bond_create(struct net
*net
, const char *name
)
5316 struct net_device
*bond_dev
;
5321 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
), name
? name
: "",
5322 bond_setup
, tx_queues
);
5324 pr_err("%s: eek! can't alloc netdev!\n", name
);
5329 dev_net_set(bond_dev
, net
);
5330 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
5333 res
= dev_alloc_name(bond_dev
, "bond%d");
5338 * If we're given a name to register
5339 * we need to ensure that its not already
5343 if (__dev_get_by_name(net
, name
) != NULL
)
5347 res
= register_netdevice(bond_dev
);
5352 bond_destructor(bond_dev
);
5356 static int __net_init
bond_net_init(struct net
*net
)
5358 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5361 INIT_LIST_HEAD(&bn
->dev_list
);
5363 bond_create_proc_dir(bn
);
5368 static void __net_exit
bond_net_exit(struct net
*net
)
5370 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5372 bond_destroy_proc_dir(bn
);
5375 static struct pernet_operations bond_net_ops
= {
5376 .init
= bond_net_init
,
5377 .exit
= bond_net_exit
,
5379 .size
= sizeof(struct bond_net
),
5382 static int __init
bonding_init(void)
5387 pr_info("%s", version
);
5389 res
= bond_check_params(&bonding_defaults
);
5393 res
= register_pernet_subsys(&bond_net_ops
);
5397 res
= rtnl_link_register(&bond_link_ops
);
5401 bond_create_debugfs();
5403 for (i
= 0; i
< max_bonds
; i
++) {
5404 res
= bond_create(&init_net
, NULL
);
5409 res
= bond_create_sysfs();
5413 register_netdevice_notifier(&bond_netdev_notifier
);
5414 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5415 bond_register_ipv6_notifier();
5419 rtnl_link_unregister(&bond_link_ops
);
5421 unregister_pernet_subsys(&bond_net_ops
);
5426 static void __exit
bonding_exit(void)
5428 unregister_netdevice_notifier(&bond_netdev_notifier
);
5429 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5430 bond_unregister_ipv6_notifier();
5432 bond_destroy_sysfs();
5433 bond_destroy_debugfs();
5435 rtnl_link_unregister(&bond_link_ops
);
5436 unregister_pernet_subsys(&bond_net_ops
);
5438 #ifdef CONFIG_NET_POLL_CONTROLLER
5440 * Make sure we don't have an imbalance on our netpoll blocking
5442 WARN_ON(atomic_read(&netpoll_block_tx
));
5446 module_init(bonding_init
);
5447 module_exit(bonding_exit
);
5448 MODULE_LICENSE("GPL");
5449 MODULE_VERSION(DRV_VERSION
);
5450 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5451 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5452 MODULE_ALIAS_RTNL_LINK("bond");