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/smp.h>
69 #include <linux/if_ether.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <linux/preempt.h>
77 #include <net/route.h>
78 #include <net/net_namespace.h>
79 #include <net/netns/generic.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV 0
88 #define BOND_LINK_ARP_INTERV 0
90 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
91 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
92 static int num_peer_notif
= 1;
93 static int miimon
= BOND_LINK_MON_INTERV
;
96 static int use_carrier
= 1;
99 static char *primary_reselect
;
100 static char *lacp_rate
;
101 static int min_links
;
102 static char *ad_select
;
103 static char *xmit_hash_policy
;
104 static int arp_interval
= BOND_LINK_ARP_INTERV
;
105 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
106 static char *arp_validate
;
107 static char *fail_over_mac
;
108 static int all_slaves_active
= 0;
109 static struct bond_params bonding_defaults
;
110 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
112 module_param(max_bonds
, int, 0);
113 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
114 module_param(tx_queues
, int, 0);
115 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
116 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
118 "failover event (alias of num_unsol_na)");
119 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
121 "failover event (alias of num_grat_arp)");
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; "
149 "0 for slow, 1 for fast");
150 module_param(ad_select
, charp
, 0);
151 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
152 "0 for stable (default), 1 for bandwidth, "
154 module_param(min_links
, int, 0);
155 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
157 module_param(xmit_hash_policy
, charp
, 0);
158 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
159 "0 for layer 2 (default), 1 for layer 3+4, "
161 module_param(arp_interval
, int, 0);
162 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
163 module_param_array(arp_ip_target
, charp
, NULL
, 0);
164 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
165 module_param(arp_validate
, charp
, 0);
166 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
167 "0 for none (default), 1 for active, "
168 "2 for backup, 3 for all");
169 module_param(fail_over_mac
, charp
, 0);
170 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
171 "the same MAC; 0 for none (default), "
172 "1 for active, 2 for follow");
173 module_param(all_slaves_active
, int, 0);
174 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
175 "by setting active flag for all slaves; "
176 "0 for never (default), 1 for always.");
177 module_param(resend_igmp
, int, 0);
178 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
181 /*----------------------------- Global variables ----------------------------*/
183 #ifdef CONFIG_NET_POLL_CONTROLLER
184 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
187 int bond_net_id __read_mostly
;
189 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
190 static int arp_ip_count
;
191 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
192 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
193 static int lacp_fast
;
195 const struct bond_parm_tbl bond_lacp_tbl
[] = {
196 { "slow", AD_LACP_SLOW
},
197 { "fast", AD_LACP_FAST
},
201 const struct bond_parm_tbl bond_mode_tbl
[] = {
202 { "balance-rr", BOND_MODE_ROUNDROBIN
},
203 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
204 { "balance-xor", BOND_MODE_XOR
},
205 { "broadcast", BOND_MODE_BROADCAST
},
206 { "802.3ad", BOND_MODE_8023AD
},
207 { "balance-tlb", BOND_MODE_TLB
},
208 { "balance-alb", BOND_MODE_ALB
},
212 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
213 { "layer2", BOND_XMIT_POLICY_LAYER2
},
214 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
215 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
219 const struct bond_parm_tbl arp_validate_tbl
[] = {
220 { "none", BOND_ARP_VALIDATE_NONE
},
221 { "active", BOND_ARP_VALIDATE_ACTIVE
},
222 { "backup", BOND_ARP_VALIDATE_BACKUP
},
223 { "all", BOND_ARP_VALIDATE_ALL
},
227 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
228 { "none", BOND_FOM_NONE
},
229 { "active", BOND_FOM_ACTIVE
},
230 { "follow", BOND_FOM_FOLLOW
},
234 const struct bond_parm_tbl pri_reselect_tbl
[] = {
235 { "always", BOND_PRI_RESELECT_ALWAYS
},
236 { "better", BOND_PRI_RESELECT_BETTER
},
237 { "failure", BOND_PRI_RESELECT_FAILURE
},
241 struct bond_parm_tbl ad_select_tbl
[] = {
242 { "stable", BOND_AD_STABLE
},
243 { "bandwidth", BOND_AD_BANDWIDTH
},
244 { "count", BOND_AD_COUNT
},
248 /*-------------------------- Forward declarations ---------------------------*/
250 static int bond_init(struct net_device
*bond_dev
);
251 static void bond_uninit(struct net_device
*bond_dev
);
253 /*---------------------------- General routines -----------------------------*/
255 const char *bond_mode_name(int mode
)
257 static const char *names
[] = {
258 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
259 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
260 [BOND_MODE_XOR
] = "load balancing (xor)",
261 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
262 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
263 [BOND_MODE_TLB
] = "transmit load balancing",
264 [BOND_MODE_ALB
] = "adaptive load balancing",
267 if (mode
< 0 || mode
> BOND_MODE_ALB
)
273 /*---------------------------------- VLAN -----------------------------------*/
276 * bond_add_vlan - add a new vlan id on bond
277 * @bond: bond that got the notification
278 * @vlan_id: the vlan id to add
280 * Returns -ENOMEM if allocation failed.
282 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
284 struct vlan_entry
*vlan
;
286 pr_debug("bond: %s, vlan id %d\n",
287 (bond
? bond
->dev
->name
: "None"), vlan_id
);
289 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
293 INIT_LIST_HEAD(&vlan
->vlan_list
);
294 vlan
->vlan_id
= vlan_id
;
296 write_lock_bh(&bond
->lock
);
298 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
300 write_unlock_bh(&bond
->lock
);
302 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
308 * bond_del_vlan - delete a vlan id from bond
309 * @bond: bond that got the notification
310 * @vlan_id: the vlan id to delete
312 * returns -ENODEV if @vlan_id was not found in @bond.
314 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
316 struct vlan_entry
*vlan
;
319 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
322 write_lock_bh(&bond
->lock
);
324 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
325 if (vlan
->vlan_id
== vlan_id
) {
326 list_del(&vlan
->vlan_list
);
328 if (bond_is_lb(bond
))
329 bond_alb_clear_vlan(bond
, vlan_id
);
331 pr_debug("removed VLAN ID %d from bond %s\n",
332 vlan_id
, bond
->dev
->name
);
341 pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 vlan_id
, bond
->dev
->name
);
345 write_unlock_bh(&bond
->lock
);
346 unblock_netpoll_tx();
351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
358 * Caller must hold bond->lock
360 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
362 struct vlan_entry
*next
, *last
;
364 if (list_empty(&bond
->vlan_list
))
368 next
= list_entry(bond
->vlan_list
.next
,
369 struct vlan_entry
, vlan_list
);
371 last
= list_entry(bond
->vlan_list
.prev
,
372 struct vlan_entry
, vlan_list
);
374 next
= list_entry(bond
->vlan_list
.next
,
375 struct vlan_entry
, vlan_list
);
377 next
= list_entry(curr
->vlan_list
.next
,
378 struct vlan_entry
, vlan_list
);
385 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
388 * bond_dev_queue_xmit - Prepare skb for xmit.
390 * @bond: bond device that got this skb for tx.
391 * @skb: hw accel VLAN tagged skb to transmit
392 * @slave_dev: slave that is supposed to xmit this skbuff
394 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
395 struct net_device
*slave_dev
)
397 skb
->dev
= slave_dev
;
400 skb
->queue_mapping
= bond_queue_mapping(skb
);
402 if (unlikely(netpoll_tx_running(slave_dev
)))
403 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
411 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
412 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
414 * a. This operation is performed in IOCTL context,
415 * b. The operation is protected by the RTNL semaphore in the 8021q code,
416 * c. Holding a lock with BH disabled while directly calling a base driver
417 * entry point is generally a BAD idea.
419 * The design of synchronization/protection for this operation in the 8021q
420 * module is good for one or more VLAN devices over a single physical device
421 * and cannot be extended for a teaming solution like bonding, so there is a
422 * potential race condition here where a net device from the vlan group might
423 * be referenced (either by a base driver or the 8021q code) while it is being
424 * removed from the system. However, it turns out we're not making matters
425 * worse, and if it works for regular VLAN usage it will work here too.
429 * bond_vlan_rx_register - Propagates registration to slaves
430 * @bond_dev: bonding net device that got called
431 * @grp: vlan group being registered
433 static void bond_vlan_rx_register(struct net_device
*bond_dev
,
434 struct vlan_group
*grp
)
436 struct bonding
*bond
= netdev_priv(bond_dev
);
440 write_lock_bh(&bond
->lock
);
442 write_unlock_bh(&bond
->lock
);
444 bond_for_each_slave(bond
, slave
, i
) {
445 struct net_device
*slave_dev
= slave
->dev
;
446 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
448 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
449 slave_ops
->ndo_vlan_rx_register
) {
450 slave_ops
->ndo_vlan_rx_register(slave_dev
, grp
);
456 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
457 * @bond_dev: bonding net device that got called
458 * @vid: vlan id being added
460 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
462 struct bonding
*bond
= netdev_priv(bond_dev
);
466 bond_for_each_slave(bond
, slave
, i
) {
467 struct net_device
*slave_dev
= slave
->dev
;
468 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
470 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
471 slave_ops
->ndo_vlan_rx_add_vid
) {
472 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vid
);
476 res
= bond_add_vlan(bond
, vid
);
478 pr_err("%s: Error: Failed to add vlan id %d\n",
479 bond_dev
->name
, vid
);
484 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
485 * @bond_dev: bonding net device that got called
486 * @vid: vlan id being removed
488 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
490 struct bonding
*bond
= netdev_priv(bond_dev
);
492 struct net_device
*vlan_dev
;
495 bond_for_each_slave(bond
, slave
, i
) {
496 struct net_device
*slave_dev
= slave
->dev
;
497 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
499 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
500 slave_ops
->ndo_vlan_rx_kill_vid
) {
501 /* Save and then restore vlan_dev in the grp array,
502 * since the slave's driver might clear it.
504 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vid
);
505 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vid
);
506 vlan_group_set_device(bond
->vlgrp
, vid
, vlan_dev
);
510 res
= bond_del_vlan(bond
, vid
);
512 pr_err("%s: Error: Failed to remove vlan id %d\n",
513 bond_dev
->name
, vid
);
517 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
519 struct vlan_entry
*vlan
;
520 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
525 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
526 slave_ops
->ndo_vlan_rx_register
)
527 slave_ops
->ndo_vlan_rx_register(slave_dev
, bond
->vlgrp
);
529 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
530 !(slave_ops
->ndo_vlan_rx_add_vid
))
533 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
)
534 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
537 static void bond_del_vlans_from_slave(struct bonding
*bond
,
538 struct net_device
*slave_dev
)
540 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
541 struct vlan_entry
*vlan
;
542 struct net_device
*vlan_dev
;
547 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
548 !(slave_ops
->ndo_vlan_rx_kill_vid
))
551 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
554 /* Save and then restore vlan_dev in the grp array,
555 * since the slave's driver might clear it.
557 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
558 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
559 vlan_group_set_device(bond
->vlgrp
, vlan
->vlan_id
, vlan_dev
);
563 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
564 slave_ops
->ndo_vlan_rx_register
)
565 slave_ops
->ndo_vlan_rx_register(slave_dev
, NULL
);
568 /*------------------------------- Link status -------------------------------*/
571 * Set the carrier state for the master according to the state of its
572 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
573 * do special 802.3ad magic.
575 * Returns zero if carrier state does not change, nonzero if it does.
577 static int bond_set_carrier(struct bonding
*bond
)
582 if (bond
->slave_cnt
== 0)
585 if (bond
->params
.mode
== BOND_MODE_8023AD
)
586 return bond_3ad_set_carrier(bond
);
588 bond_for_each_slave(bond
, slave
, i
) {
589 if (slave
->link
== BOND_LINK_UP
) {
590 if (!netif_carrier_ok(bond
->dev
)) {
591 netif_carrier_on(bond
->dev
);
599 if (netif_carrier_ok(bond
->dev
)) {
600 netif_carrier_off(bond
->dev
);
607 * Get link speed and duplex from the slave's base driver
608 * using ethtool. If for some reason the call fails or the
609 * values are invalid, fake speed and duplex to 100/Full
612 static int bond_update_speed_duplex(struct slave
*slave
)
614 struct net_device
*slave_dev
= slave
->dev
;
615 struct ethtool_cmd etool
= { .cmd
= ETHTOOL_GSET
};
619 /* Fake speed and duplex */
620 slave
->speed
= SPEED_100
;
621 slave
->duplex
= DUPLEX_FULL
;
623 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
626 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
630 slave_speed
= ethtool_cmd_speed(&etool
);
631 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
634 switch (etool
.duplex
) {
642 slave
->speed
= slave_speed
;
643 slave
->duplex
= etool
.duplex
;
649 * if <dev> supports MII link status reporting, check its link status.
651 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
652 * depending upon the setting of the use_carrier parameter.
654 * Return either BMSR_LSTATUS, meaning that the link is up (or we
655 * can't tell and just pretend it is), or 0, meaning that the link is
658 * If reporting is non-zero, instead of faking link up, return -1 if
659 * both ETHTOOL and MII ioctls fail (meaning the device does not
660 * support them). If use_carrier is set, return whatever it says.
661 * It'd be nice if there was a good way to tell if a driver supports
662 * netif_carrier, but there really isn't.
664 static int bond_check_dev_link(struct bonding
*bond
,
665 struct net_device
*slave_dev
, int reporting
)
667 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
668 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
670 struct mii_ioctl_data
*mii
;
672 if (!reporting
&& !netif_running(slave_dev
))
675 if (bond
->params
.use_carrier
)
676 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
678 /* Try to get link status using Ethtool first. */
679 if (slave_dev
->ethtool_ops
) {
680 if (slave_dev
->ethtool_ops
->get_link
) {
683 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
685 return link
? BMSR_LSTATUS
: 0;
689 /* Ethtool can't be used, fallback to MII ioctls. */
690 ioctl
= slave_ops
->ndo_do_ioctl
;
692 /* TODO: set pointer to correct ioctl on a per team member */
693 /* bases to make this more efficient. that is, once */
694 /* we determine the correct ioctl, we will always */
695 /* call it and not the others for that team */
699 * We cannot assume that SIOCGMIIPHY will also read a
700 * register; not all network drivers (e.g., e100)
704 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
705 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
707 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
708 mii
->reg_num
= MII_BMSR
;
709 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
710 return mii
->val_out
& BMSR_LSTATUS
;
715 * If reporting, report that either there's no dev->do_ioctl,
716 * or both SIOCGMIIREG and get_link failed (meaning that we
717 * cannot report link status). If not reporting, pretend
720 return reporting
? -1 : BMSR_LSTATUS
;
723 /*----------------------------- Multicast list ------------------------------*/
726 * Push the promiscuity flag down to appropriate slaves
728 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
731 if (USES_PRIMARY(bond
->params
.mode
)) {
732 /* write lock already acquired */
733 if (bond
->curr_active_slave
) {
734 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
740 bond_for_each_slave(bond
, slave
, i
) {
741 err
= dev_set_promiscuity(slave
->dev
, inc
);
750 * Push the allmulti flag down to all slaves
752 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
755 if (USES_PRIMARY(bond
->params
.mode
)) {
756 /* write lock already acquired */
757 if (bond
->curr_active_slave
) {
758 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
764 bond_for_each_slave(bond
, slave
, i
) {
765 err
= dev_set_allmulti(slave
->dev
, inc
);
774 * Add a Multicast address to slaves
777 static void bond_mc_add(struct bonding
*bond
, void *addr
)
779 if (USES_PRIMARY(bond
->params
.mode
)) {
780 /* write lock already acquired */
781 if (bond
->curr_active_slave
)
782 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
787 bond_for_each_slave(bond
, slave
, i
)
788 dev_mc_add(slave
->dev
, addr
);
793 * Remove a multicast address from slave
796 static void bond_mc_del(struct bonding
*bond
, void *addr
)
798 if (USES_PRIMARY(bond
->params
.mode
)) {
799 /* write lock already acquired */
800 if (bond
->curr_active_slave
)
801 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
805 bond_for_each_slave(bond
, slave
, i
) {
806 dev_mc_del(slave
->dev
, addr
);
812 static void __bond_resend_igmp_join_requests(struct net_device
*dev
)
814 struct in_device
*in_dev
;
817 in_dev
= __in_dev_get_rcu(dev
);
819 ip_mc_rejoin_groups(in_dev
);
824 * Retrieve the list of registered multicast addresses for the bonding
825 * device and retransmit an IGMP JOIN request to the current active
828 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
830 struct net_device
*vlan_dev
;
831 struct vlan_entry
*vlan
;
833 read_lock(&bond
->lock
);
835 /* rejoin all groups on bond device */
836 __bond_resend_igmp_join_requests(bond
->dev
);
838 /* rejoin all groups on vlan devices */
840 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
841 vlan_dev
= vlan_group_get_device(bond
->vlgrp
,
844 __bond_resend_igmp_join_requests(vlan_dev
);
848 if (--bond
->igmp_retrans
> 0)
849 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
851 read_unlock(&bond
->lock
);
854 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
856 struct bonding
*bond
= container_of(work
, struct bonding
,
858 bond_resend_igmp_join_requests(bond
);
862 * flush all members of flush->mc_list from device dev->mc_list
864 static void bond_mc_list_flush(struct net_device
*bond_dev
,
865 struct net_device
*slave_dev
)
867 struct bonding
*bond
= netdev_priv(bond_dev
);
868 struct netdev_hw_addr
*ha
;
870 netdev_for_each_mc_addr(ha
, bond_dev
)
871 dev_mc_del(slave_dev
, ha
->addr
);
873 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
874 /* del lacpdu mc addr from mc list */
875 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
877 dev_mc_del(slave_dev
, lacpdu_multicast
);
881 /*--------------------------- Active slave change ---------------------------*/
884 * Update the mc list and multicast-related flags for the new and
885 * old active slaves (if any) according to the multicast mode, and
886 * promiscuous flags unconditionally.
888 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
889 struct slave
*old_active
)
891 struct netdev_hw_addr
*ha
;
893 if (!USES_PRIMARY(bond
->params
.mode
))
894 /* nothing to do - mc list is already up-to-date on
900 if (bond
->dev
->flags
& IFF_PROMISC
)
901 dev_set_promiscuity(old_active
->dev
, -1);
903 if (bond
->dev
->flags
& IFF_ALLMULTI
)
904 dev_set_allmulti(old_active
->dev
, -1);
906 netdev_for_each_mc_addr(ha
, bond
->dev
)
907 dev_mc_del(old_active
->dev
, ha
->addr
);
911 /* FIXME: Signal errors upstream. */
912 if (bond
->dev
->flags
& IFF_PROMISC
)
913 dev_set_promiscuity(new_active
->dev
, 1);
915 if (bond
->dev
->flags
& IFF_ALLMULTI
)
916 dev_set_allmulti(new_active
->dev
, 1);
918 netdev_for_each_mc_addr(ha
, bond
->dev
)
919 dev_mc_add(new_active
->dev
, ha
->addr
);
924 * bond_do_fail_over_mac
926 * Perform special MAC address swapping for fail_over_mac settings
928 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
930 static void bond_do_fail_over_mac(struct bonding
*bond
,
931 struct slave
*new_active
,
932 struct slave
*old_active
)
933 __releases(&bond
->curr_slave_lock
)
934 __releases(&bond
->lock
)
935 __acquires(&bond
->lock
)
936 __acquires(&bond
->curr_slave_lock
)
938 u8 tmp_mac
[ETH_ALEN
];
939 struct sockaddr saddr
;
942 switch (bond
->params
.fail_over_mac
) {
943 case BOND_FOM_ACTIVE
:
945 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
946 new_active
->dev
->addr_len
);
948 case BOND_FOM_FOLLOW
:
950 * if new_active && old_active, swap them
951 * if just old_active, do nothing (going to no active slave)
952 * if just new_active, set new_active to bond's MAC
957 write_unlock_bh(&bond
->curr_slave_lock
);
958 read_unlock(&bond
->lock
);
961 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
962 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
964 saddr
.sa_family
= new_active
->dev
->type
;
966 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
967 saddr
.sa_family
= bond
->dev
->type
;
970 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
972 pr_err("%s: Error %d setting MAC of slave %s\n",
973 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
980 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
981 saddr
.sa_family
= old_active
->dev
->type
;
983 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
985 pr_err("%s: Error %d setting MAC of slave %s\n",
986 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
988 read_lock(&bond
->lock
);
989 write_lock_bh(&bond
->curr_slave_lock
);
992 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
993 bond
->dev
->name
, bond
->params
.fail_over_mac
);
999 static bool bond_should_change_active(struct bonding
*bond
)
1001 struct slave
*prim
= bond
->primary_slave
;
1002 struct slave
*curr
= bond
->curr_active_slave
;
1004 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
1006 if (bond
->force_primary
) {
1007 bond
->force_primary
= false;
1010 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
1011 (prim
->speed
< curr
->speed
||
1012 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
1014 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
1020 * find_best_interface - select the best available slave to be the active one
1021 * @bond: our bonding struct
1023 * Warning: Caller must hold curr_slave_lock for writing.
1025 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1027 struct slave
*new_active
, *old_active
;
1028 struct slave
*bestslave
= NULL
;
1029 int mintime
= bond
->params
.updelay
;
1032 new_active
= bond
->curr_active_slave
;
1034 if (!new_active
) { /* there were no active slaves left */
1035 if (bond
->slave_cnt
> 0) /* found one slave */
1036 new_active
= bond
->first_slave
;
1038 return NULL
; /* still no slave, return NULL */
1041 if ((bond
->primary_slave
) &&
1042 bond
->primary_slave
->link
== BOND_LINK_UP
&&
1043 bond_should_change_active(bond
)) {
1044 new_active
= bond
->primary_slave
;
1047 /* remember where to stop iterating over the slaves */
1048 old_active
= new_active
;
1050 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1051 if (new_active
->link
== BOND_LINK_UP
) {
1053 } else if (new_active
->link
== BOND_LINK_BACK
&&
1054 IS_UP(new_active
->dev
)) {
1055 /* link up, but waiting for stabilization */
1056 if (new_active
->delay
< mintime
) {
1057 mintime
= new_active
->delay
;
1058 bestslave
= new_active
;
1066 static bool bond_should_notify_peers(struct bonding
*bond
)
1068 struct slave
*slave
= bond
->curr_active_slave
;
1070 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1071 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
1073 if (!slave
|| !bond
->send_peer_notif
||
1074 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
1077 bond
->send_peer_notif
--;
1082 * change_active_interface - change the active slave into the specified one
1083 * @bond: our bonding struct
1084 * @new: the new slave to make the active one
1086 * Set the new slave to the bond's settings and unset them on the old
1087 * curr_active_slave.
1088 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1090 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1091 * because it is apparently the best available slave we have, even though its
1092 * updelay hasn't timed out yet.
1094 * If new_active is not NULL, caller must hold bond->lock for read and
1095 * curr_slave_lock for write_bh.
1097 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1099 struct slave
*old_active
= bond
->curr_active_slave
;
1101 if (old_active
== new_active
)
1105 new_active
->jiffies
= jiffies
;
1107 if (new_active
->link
== BOND_LINK_BACK
) {
1108 if (USES_PRIMARY(bond
->params
.mode
)) {
1109 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1110 bond
->dev
->name
, new_active
->dev
->name
,
1111 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1114 new_active
->delay
= 0;
1115 new_active
->link
= BOND_LINK_UP
;
1117 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1118 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1120 if (bond_is_lb(bond
))
1121 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1123 if (USES_PRIMARY(bond
->params
.mode
)) {
1124 pr_info("%s: making interface %s the new active one.\n",
1125 bond
->dev
->name
, new_active
->dev
->name
);
1130 if (USES_PRIMARY(bond
->params
.mode
))
1131 bond_mc_swap(bond
, new_active
, old_active
);
1133 if (bond_is_lb(bond
)) {
1134 bond_alb_handle_active_change(bond
, new_active
);
1136 bond_set_slave_inactive_flags(old_active
);
1138 bond_set_slave_active_flags(new_active
);
1140 bond
->curr_active_slave
= new_active
;
1143 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1145 bond_set_slave_inactive_flags(old_active
);
1148 bool should_notify_peers
= false;
1150 bond_set_slave_active_flags(new_active
);
1152 if (bond
->params
.fail_over_mac
)
1153 bond_do_fail_over_mac(bond
, new_active
,
1156 if (netif_running(bond
->dev
)) {
1157 bond
->send_peer_notif
=
1158 bond
->params
.num_peer_notif
;
1159 should_notify_peers
=
1160 bond_should_notify_peers(bond
);
1163 write_unlock_bh(&bond
->curr_slave_lock
);
1164 read_unlock(&bond
->lock
);
1166 netdev_bonding_change(bond
->dev
, NETDEV_BONDING_FAILOVER
);
1167 if (should_notify_peers
)
1168 netdev_bonding_change(bond
->dev
,
1169 NETDEV_NOTIFY_PEERS
);
1171 read_lock(&bond
->lock
);
1172 write_lock_bh(&bond
->curr_slave_lock
);
1176 /* resend IGMP joins since active slave has changed or
1177 * all were sent on curr_active_slave.
1178 * resend only if bond is brought up with the affected
1179 * bonding modes and the retransmission is enabled */
1180 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
1181 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
1182 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
1183 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
1184 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 0);
1189 * bond_select_active_slave - select a new active slave, if needed
1190 * @bond: our bonding struct
1192 * This functions should be called when one of the following occurs:
1193 * - The old curr_active_slave has been released or lost its link.
1194 * - The primary_slave has got its link back.
1195 * - A slave has got its link back and there's no old curr_active_slave.
1197 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1199 void bond_select_active_slave(struct bonding
*bond
)
1201 struct slave
*best_slave
;
1204 best_slave
= bond_find_best_slave(bond
);
1205 if (best_slave
!= bond
->curr_active_slave
) {
1206 bond_change_active_slave(bond
, best_slave
);
1207 rv
= bond_set_carrier(bond
);
1211 if (netif_carrier_ok(bond
->dev
)) {
1212 pr_info("%s: first active interface up!\n",
1215 pr_info("%s: now running without any active interface !\n",
1221 /*--------------------------- slave list handling ---------------------------*/
1224 * This function attaches the slave to the end of list.
1226 * bond->lock held for writing by caller.
1228 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1230 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1231 new_slave
->next
= new_slave
;
1232 new_slave
->prev
= new_slave
;
1233 bond
->first_slave
= new_slave
;
1235 new_slave
->next
= bond
->first_slave
;
1236 new_slave
->prev
= bond
->first_slave
->prev
;
1237 new_slave
->next
->prev
= new_slave
;
1238 new_slave
->prev
->next
= new_slave
;
1245 * This function detaches the slave from the list.
1246 * WARNING: no check is made to verify if the slave effectively
1247 * belongs to <bond>.
1248 * Nothing is freed on return, structures are just unchained.
1249 * If any slave pointer in bond was pointing to <slave>,
1250 * it should be changed by the calling function.
1252 * bond->lock held for writing by caller.
1254 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1257 slave
->next
->prev
= slave
->prev
;
1260 slave
->prev
->next
= slave
->next
;
1262 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1263 if (bond
->slave_cnt
> 1) { /* there are more slave */
1264 bond
->first_slave
= slave
->next
;
1266 bond
->first_slave
= NULL
; /* slave was the last one */
1275 #ifdef CONFIG_NET_POLL_CONTROLLER
1276 static inline int slave_enable_netpoll(struct slave
*slave
)
1281 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
1286 np
->dev
= slave
->dev
;
1287 strlcpy(np
->dev_name
, slave
->dev
->name
, IFNAMSIZ
);
1288 err
= __netpoll_setup(np
);
1297 static inline void slave_disable_netpoll(struct slave
*slave
)
1299 struct netpoll
*np
= slave
->np
;
1305 synchronize_rcu_bh();
1306 __netpoll_cleanup(np
);
1309 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1311 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1313 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1318 static void bond_poll_controller(struct net_device
*bond_dev
)
1322 static void __bond_netpoll_cleanup(struct bonding
*bond
)
1324 struct slave
*slave
;
1327 bond_for_each_slave(bond
, slave
, i
)
1328 if (IS_UP(slave
->dev
))
1329 slave_disable_netpoll(slave
);
1331 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1333 struct bonding
*bond
= netdev_priv(bond_dev
);
1335 read_lock(&bond
->lock
);
1336 __bond_netpoll_cleanup(bond
);
1337 read_unlock(&bond
->lock
);
1340 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
1342 struct bonding
*bond
= netdev_priv(dev
);
1343 struct slave
*slave
;
1346 read_lock(&bond
->lock
);
1347 bond_for_each_slave(bond
, slave
, i
) {
1348 err
= slave_enable_netpoll(slave
);
1350 __bond_netpoll_cleanup(bond
);
1354 read_unlock(&bond
->lock
);
1358 static struct netpoll_info
*bond_netpoll_info(struct bonding
*bond
)
1360 return bond
->dev
->npinfo
;
1364 static inline int slave_enable_netpoll(struct slave
*slave
)
1368 static inline void slave_disable_netpoll(struct slave
*slave
)
1371 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1376 /*---------------------------------- IOCTL ----------------------------------*/
1378 static int bond_sethwaddr(struct net_device
*bond_dev
,
1379 struct net_device
*slave_dev
)
1381 pr_debug("bond_dev=%p\n", bond_dev
);
1382 pr_debug("slave_dev=%p\n", slave_dev
);
1383 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1384 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1388 static u32
bond_fix_features(struct net_device
*dev
, u32 features
)
1390 struct slave
*slave
;
1391 struct bonding
*bond
= netdev_priv(dev
);
1395 read_lock(&bond
->lock
);
1397 if (!bond
->first_slave
) {
1398 /* Disable adding VLANs to empty bond. But why? --mq */
1399 features
|= NETIF_F_VLAN_CHALLENGED
;
1404 features
&= ~NETIF_F_ONE_FOR_ALL
;
1405 features
|= NETIF_F_ALL_FOR_ALL
;
1407 bond_for_each_slave(bond
, slave
, i
) {
1408 features
= netdev_increment_features(features
,
1409 slave
->dev
->features
,
1414 read_unlock(&bond
->lock
);
1418 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1419 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1420 NETIF_F_HIGHDMA | NETIF_F_LRO)
1422 static void bond_compute_features(struct bonding
*bond
)
1424 struct slave
*slave
;
1425 struct net_device
*bond_dev
= bond
->dev
;
1426 u32 vlan_features
= BOND_VLAN_FEATURES
;
1427 unsigned short max_hard_header_len
= ETH_HLEN
;
1430 read_lock(&bond
->lock
);
1432 if (!bond
->first_slave
)
1435 bond_for_each_slave(bond
, slave
, i
) {
1436 vlan_features
= netdev_increment_features(vlan_features
,
1437 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1439 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1440 max_hard_header_len
= slave
->dev
->hard_header_len
;
1444 bond_dev
->vlan_features
= vlan_features
;
1445 bond_dev
->hard_header_len
= max_hard_header_len
;
1447 read_unlock(&bond
->lock
);
1449 netdev_change_features(bond_dev
);
1452 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1453 struct net_device
*slave_dev
)
1455 struct bonding
*bond
= netdev_priv(bond_dev
);
1457 bond_dev
->header_ops
= slave_dev
->header_ops
;
1459 bond_dev
->type
= slave_dev
->type
;
1460 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1461 bond_dev
->addr_len
= slave_dev
->addr_len
;
1463 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1464 slave_dev
->addr_len
);
1465 bond
->setup_by_slave
= 1;
1468 /* On bonding slaves other than the currently active slave, suppress
1469 * duplicates except for alb non-mcast/bcast.
1471 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1472 struct slave
*slave
,
1473 struct bonding
*bond
)
1475 if (bond_is_slave_inactive(slave
)) {
1476 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1477 skb
->pkt_type
!= PACKET_BROADCAST
&&
1478 skb
->pkt_type
!= PACKET_MULTICAST
)
1485 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1487 struct sk_buff
*skb
= *pskb
;
1488 struct slave
*slave
;
1489 struct bonding
*bond
;
1491 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1493 return RX_HANDLER_CONSUMED
;
1497 slave
= bond_slave_get_rcu(skb
->dev
);
1500 if (bond
->params
.arp_interval
)
1501 slave
->dev
->last_rx
= jiffies
;
1503 if (bond
->recv_probe
) {
1504 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
1507 bond
->recv_probe(nskb
, bond
, slave
);
1508 dev_kfree_skb(nskb
);
1512 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1513 return RX_HANDLER_EXACT
;
1516 skb
->dev
= bond
->dev
;
1518 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1519 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1520 skb
->pkt_type
== PACKET_HOST
) {
1522 if (unlikely(skb_cow_head(skb
,
1523 skb
->data
- skb_mac_header(skb
)))) {
1525 return RX_HANDLER_CONSUMED
;
1527 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1530 return RX_HANDLER_ANOTHER
;
1533 /* enslave device <slave> to bond device <master> */
1534 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1536 struct bonding
*bond
= netdev_priv(bond_dev
);
1537 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1538 struct slave
*new_slave
= NULL
;
1539 struct netdev_hw_addr
*ha
;
1540 struct sockaddr addr
;
1544 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1545 slave_ops
->ndo_do_ioctl
== NULL
) {
1546 pr_warning("%s: Warning: no link monitoring support for %s\n",
1547 bond_dev
->name
, slave_dev
->name
);
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
);
1570 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1574 * Old ifenslave binaries are no longer supported. These can
1575 * be identified with moderate accuracy by the state of the slave:
1576 * the current ifenslave will set the interface down prior to
1577 * enslaving it; the old ifenslave will not.
1579 if ((slave_dev
->flags
& IFF_UP
)) {
1580 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1583 goto err_undo_flags
;
1586 /* set bonding device ether type by slave - bonding netdevices are
1587 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1588 * there is a need to override some of the type dependent attribs/funcs.
1590 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1591 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1593 if (bond
->slave_cnt
== 0) {
1594 if (bond_dev
->type
!= slave_dev
->type
) {
1595 pr_debug("%s: change device type from %d to %d\n",
1597 bond_dev
->type
, slave_dev
->type
);
1599 res
= netdev_bonding_change(bond_dev
,
1600 NETDEV_PRE_TYPE_CHANGE
);
1601 res
= notifier_to_errno(res
);
1603 pr_err("%s: refused to change device type\n",
1606 goto err_undo_flags
;
1609 /* Flush unicast and multicast addresses */
1610 dev_uc_flush(bond_dev
);
1611 dev_mc_flush(bond_dev
);
1613 if (slave_dev
->type
!= ARPHRD_ETHER
)
1614 bond_setup_by_slave(bond_dev
, slave_dev
);
1616 ether_setup(bond_dev
);
1618 netdev_bonding_change(bond_dev
,
1619 NETDEV_POST_TYPE_CHANGE
);
1621 } else if (bond_dev
->type
!= slave_dev
->type
) {
1622 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1624 slave_dev
->type
, bond_dev
->type
);
1626 goto err_undo_flags
;
1629 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1630 if (bond
->slave_cnt
== 0) {
1631 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1633 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1634 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1635 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",
1638 goto err_undo_flags
;
1642 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1644 /* If this is the first slave, then we need to set the master's hardware
1645 * address to be the same as the slave's. */
1646 if (is_zero_ether_addr(bond
->dev
->dev_addr
))
1647 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1648 slave_dev
->addr_len
);
1651 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1654 goto err_undo_flags
;
1658 * Set the new_slave's queue_id to be zero. Queue ID mapping
1659 * is set via sysfs or module option if desired.
1661 new_slave
->queue_id
= 0;
1663 /* Save slave's original mtu and then set it to match the bond */
1664 new_slave
->original_mtu
= slave_dev
->mtu
;
1665 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1667 pr_debug("Error %d calling dev_set_mtu\n", res
);
1672 * Save slave's original ("permanent") mac address for modes
1673 * that need it, and for restoring it upon release, and then
1674 * set it to the master's address
1676 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1678 if (!bond
->params
.fail_over_mac
) {
1680 * Set slave to master's mac address. The application already
1681 * set the master's mac address to that of the first slave
1683 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1684 addr
.sa_family
= slave_dev
->type
;
1685 res
= dev_set_mac_address(slave_dev
, &addr
);
1687 pr_debug("Error %d calling set_mac_address\n", res
);
1688 goto err_restore_mtu
;
1692 res
= netdev_set_bond_master(slave_dev
, bond_dev
);
1694 pr_debug("Error %d calling netdev_set_bond_master\n", res
);
1695 goto err_restore_mac
;
1698 /* open the slave since the application closed it */
1699 res
= dev_open(slave_dev
);
1701 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1702 goto err_unset_master
;
1705 new_slave
->bond
= bond
;
1706 new_slave
->dev
= slave_dev
;
1707 slave_dev
->priv_flags
|= IFF_BONDING
;
1709 if (bond_is_lb(bond
)) {
1710 /* bond_alb_init_slave() must be called before all other stages since
1711 * it might fail and we do not want to have to undo everything
1713 res
= bond_alb_init_slave(bond
, new_slave
);
1718 /* If the mode USES_PRIMARY, then the new slave gets the
1719 * master's promisc (and mc) settings only if it becomes the
1720 * curr_active_slave, and that is taken care of later when calling
1721 * bond_change_active()
1723 if (!USES_PRIMARY(bond
->params
.mode
)) {
1724 /* set promiscuity level to new slave */
1725 if (bond_dev
->flags
& IFF_PROMISC
) {
1726 res
= dev_set_promiscuity(slave_dev
, 1);
1731 /* set allmulti level to new slave */
1732 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1733 res
= dev_set_allmulti(slave_dev
, 1);
1738 netif_addr_lock_bh(bond_dev
);
1739 /* upload master's mc_list to new slave */
1740 netdev_for_each_mc_addr(ha
, bond_dev
)
1741 dev_mc_add(slave_dev
, ha
->addr
);
1742 netif_addr_unlock_bh(bond_dev
);
1745 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1746 /* add lacpdu mc addr to mc list */
1747 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1749 dev_mc_add(slave_dev
, lacpdu_multicast
);
1752 bond_add_vlans_on_slave(bond
, slave_dev
);
1754 write_lock_bh(&bond
->lock
);
1756 bond_attach_slave(bond
, new_slave
);
1758 new_slave
->delay
= 0;
1759 new_slave
->link_failure_count
= 0;
1761 write_unlock_bh(&bond
->lock
);
1763 bond_compute_features(bond
);
1765 read_lock(&bond
->lock
);
1767 new_slave
->last_arp_rx
= jiffies
;
1769 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1770 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1772 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1774 * miimon is set but a bonded network driver
1775 * does not support ETHTOOL/MII and
1776 * arp_interval is not set. Note: if
1777 * use_carrier is enabled, we will never go
1778 * here (because netif_carrier is always
1779 * supported); thus, we don't need to change
1780 * the messages for netif_carrier.
1782 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",
1783 bond_dev
->name
, slave_dev
->name
);
1784 } else if (link_reporting
== -1) {
1785 /* unable get link status using mii/ethtool */
1786 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",
1787 bond_dev
->name
, slave_dev
->name
);
1791 /* check for initial state */
1792 if (!bond
->params
.miimon
||
1793 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1794 if (bond
->params
.updelay
) {
1795 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1796 new_slave
->link
= BOND_LINK_BACK
;
1797 new_slave
->delay
= bond
->params
.updelay
;
1799 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1800 new_slave
->link
= BOND_LINK_UP
;
1802 new_slave
->jiffies
= jiffies
;
1804 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1805 new_slave
->link
= BOND_LINK_DOWN
;
1808 if (bond_update_speed_duplex(new_slave
) &&
1809 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1810 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1811 bond_dev
->name
, new_slave
->dev
->name
);
1813 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1814 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1819 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1820 /* if there is a primary slave, remember it */
1821 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1822 bond
->primary_slave
= new_slave
;
1823 bond
->force_primary
= true;
1827 write_lock_bh(&bond
->curr_slave_lock
);
1829 switch (bond
->params
.mode
) {
1830 case BOND_MODE_ACTIVEBACKUP
:
1831 bond_set_slave_inactive_flags(new_slave
);
1832 bond_select_active_slave(bond
);
1834 case BOND_MODE_8023AD
:
1835 /* in 802.3ad mode, the internal mechanism
1836 * will activate the slaves in the selected
1839 bond_set_slave_inactive_flags(new_slave
);
1840 /* if this is the first slave */
1841 if (bond
->slave_cnt
== 1) {
1842 SLAVE_AD_INFO(new_slave
).id
= 1;
1843 /* Initialize AD with the number of times that the AD timer is called in 1 second
1844 * can be called only after the mac address of the bond is set
1846 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1848 SLAVE_AD_INFO(new_slave
).id
=
1849 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1852 bond_3ad_bind_slave(new_slave
);
1856 bond_set_active_slave(new_slave
);
1857 bond_set_slave_inactive_flags(new_slave
);
1858 bond_select_active_slave(bond
);
1861 pr_debug("This slave is always active in trunk mode\n");
1863 /* always active in trunk mode */
1864 bond_set_active_slave(new_slave
);
1866 /* In trunking mode there is little meaning to curr_active_slave
1867 * anyway (it holds no special properties of the bond device),
1868 * so we can change it without calling change_active_interface()
1870 if (!bond
->curr_active_slave
)
1871 bond
->curr_active_slave
= new_slave
;
1874 } /* switch(bond_mode) */
1876 write_unlock_bh(&bond
->curr_slave_lock
);
1878 bond_set_carrier(bond
);
1880 #ifdef CONFIG_NET_POLL_CONTROLLER
1881 slave_dev
->npinfo
= bond_netpoll_info(bond
);
1882 if (slave_dev
->npinfo
) {
1883 if (slave_enable_netpoll(new_slave
)) {
1884 read_unlock(&bond
->lock
);
1885 pr_info("Error, %s: master_dev is using netpoll, "
1886 "but new slave device does not support netpoll.\n",
1894 read_unlock(&bond
->lock
);
1896 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1900 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1903 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1904 goto err_dest_symlinks
;
1907 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1908 bond_dev
->name
, slave_dev
->name
,
1909 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1910 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1912 /* enslave is successful */
1915 /* Undo stages on error */
1917 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1920 dev_close(slave_dev
);
1923 netdev_set_bond_master(slave_dev
, NULL
);
1926 if (!bond
->params
.fail_over_mac
) {
1927 /* XXX TODO - fom follow mode needs to change master's
1928 * MAC if this slave's MAC is in use by the bond, or at
1929 * least print a warning.
1931 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1932 addr
.sa_family
= slave_dev
->type
;
1933 dev_set_mac_address(slave_dev
, &addr
);
1937 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1943 bond_compute_features(bond
);
1949 * Try to release the slave device <slave> from the bond device <master>
1950 * It is legal to access curr_active_slave without a lock because all the function
1953 * The rules for slave state should be:
1954 * for Active/Backup:
1955 * Active stays on all backups go down
1956 * for Bonded connections:
1957 * The first up interface should be left on and all others downed.
1959 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1961 struct bonding
*bond
= netdev_priv(bond_dev
);
1962 struct slave
*slave
, *oldcurrent
;
1963 struct sockaddr addr
;
1964 u32 old_features
= bond_dev
->features
;
1966 /* slave is not a slave or master is not master of this slave */
1967 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1968 (slave_dev
->master
!= bond_dev
)) {
1969 pr_err("%s: Error: cannot release %s.\n",
1970 bond_dev
->name
, slave_dev
->name
);
1975 netdev_bonding_change(bond_dev
, NETDEV_RELEASE
);
1976 write_lock_bh(&bond
->lock
);
1978 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1980 /* not a slave of this bond */
1981 pr_info("%s: %s not enslaved\n",
1982 bond_dev
->name
, slave_dev
->name
);
1983 write_unlock_bh(&bond
->lock
);
1984 unblock_netpoll_tx();
1988 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1989 * for this slave anymore.
1991 netdev_rx_handler_unregister(slave_dev
);
1992 write_unlock_bh(&bond
->lock
);
1994 write_lock_bh(&bond
->lock
);
1996 if (!bond
->params
.fail_over_mac
) {
1997 if (!compare_ether_addr(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1998 bond
->slave_cnt
> 1)
1999 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",
2000 bond_dev
->name
, slave_dev
->name
,
2002 bond_dev
->name
, slave_dev
->name
);
2005 /* Inform AD package of unbinding of slave. */
2006 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2007 /* must be called before the slave is
2008 * detached from the list
2010 bond_3ad_unbind_slave(slave
);
2013 pr_info("%s: releasing %s interface %s\n",
2015 bond_is_active_slave(slave
) ? "active" : "backup",
2018 oldcurrent
= bond
->curr_active_slave
;
2020 bond
->current_arp_slave
= NULL
;
2022 /* release the slave from its bond */
2023 bond_detach_slave(bond
, slave
);
2025 if (bond
->primary_slave
== slave
)
2026 bond
->primary_slave
= NULL
;
2028 if (oldcurrent
== slave
)
2029 bond_change_active_slave(bond
, NULL
);
2031 if (bond_is_lb(bond
)) {
2032 /* Must be called only after the slave has been
2033 * detached from the list and the curr_active_slave
2034 * has been cleared (if our_slave == old_current),
2035 * but before a new active slave is selected.
2037 write_unlock_bh(&bond
->lock
);
2038 bond_alb_deinit_slave(bond
, slave
);
2039 write_lock_bh(&bond
->lock
);
2042 if (oldcurrent
== slave
) {
2044 * Note that we hold RTNL over this sequence, so there
2045 * is no concern that another slave add/remove event
2048 write_unlock_bh(&bond
->lock
);
2049 read_lock(&bond
->lock
);
2050 write_lock_bh(&bond
->curr_slave_lock
);
2052 bond_select_active_slave(bond
);
2054 write_unlock_bh(&bond
->curr_slave_lock
);
2055 read_unlock(&bond
->lock
);
2056 write_lock_bh(&bond
->lock
);
2059 if (bond
->slave_cnt
== 0) {
2060 bond_set_carrier(bond
);
2062 /* if the last slave was removed, zero the mac address
2063 * of the master so it will be set by the application
2064 * to the mac address of the first slave
2066 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2069 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2070 bond_dev
->name
, bond_dev
->name
);
2071 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2076 write_unlock_bh(&bond
->lock
);
2077 unblock_netpoll_tx();
2079 bond_compute_features(bond
);
2080 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
2081 (old_features
& NETIF_F_VLAN_CHALLENGED
))
2082 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2083 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
2085 /* must do this from outside any spinlocks */
2086 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2088 bond_del_vlans_from_slave(bond
, slave_dev
);
2090 /* If the mode USES_PRIMARY, then we should only remove its
2091 * promisc and mc settings if it was the curr_active_slave, but that was
2092 * already taken care of above when we detached the slave
2094 if (!USES_PRIMARY(bond
->params
.mode
)) {
2095 /* unset promiscuity level from slave */
2096 if (bond_dev
->flags
& IFF_PROMISC
)
2097 dev_set_promiscuity(slave_dev
, -1);
2099 /* unset allmulti level from slave */
2100 if (bond_dev
->flags
& IFF_ALLMULTI
)
2101 dev_set_allmulti(slave_dev
, -1);
2103 /* flush master's mc_list from slave */
2104 netif_addr_lock_bh(bond_dev
);
2105 bond_mc_list_flush(bond_dev
, slave_dev
);
2106 netif_addr_unlock_bh(bond_dev
);
2109 netdev_set_bond_master(slave_dev
, NULL
);
2111 slave_disable_netpoll(slave
);
2113 /* close slave before restoring its mac address */
2114 dev_close(slave_dev
);
2116 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2117 /* restore original ("permanent") mac address */
2118 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2119 addr
.sa_family
= slave_dev
->type
;
2120 dev_set_mac_address(slave_dev
, &addr
);
2123 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2125 slave_dev
->priv_flags
&= ~IFF_BONDING
;
2129 return 0; /* deletion OK */
2133 * First release a slave and then 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 /* unregister rx_handler early so bond_handle_frame wouldn't
2190 * be called for this slave anymore.
2192 netdev_rx_handler_unregister(slave_dev
);
2195 if (bond_is_lb(bond
)) {
2196 /* must be called only after the slave
2197 * has been detached from the list
2199 bond_alb_deinit_slave(bond
, slave
);
2202 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2203 bond_del_vlans_from_slave(bond
, slave_dev
);
2205 /* If the mode USES_PRIMARY, then we should only remove its
2206 * promisc and mc settings if it was the curr_active_slave, but that was
2207 * already taken care of above when we detached the slave
2209 if (!USES_PRIMARY(bond
->params
.mode
)) {
2210 /* unset promiscuity level from slave */
2211 if (bond_dev
->flags
& IFF_PROMISC
)
2212 dev_set_promiscuity(slave_dev
, -1);
2214 /* unset allmulti level from slave */
2215 if (bond_dev
->flags
& IFF_ALLMULTI
)
2216 dev_set_allmulti(slave_dev
, -1);
2218 /* flush master's mc_list from slave */
2219 netif_addr_lock_bh(bond_dev
);
2220 bond_mc_list_flush(bond_dev
, slave_dev
);
2221 netif_addr_unlock_bh(bond_dev
);
2224 netdev_set_bond_master(slave_dev
, NULL
);
2226 slave_disable_netpoll(slave
);
2228 /* close slave before restoring its mac address */
2229 dev_close(slave_dev
);
2231 if (!bond
->params
.fail_over_mac
) {
2232 /* restore original ("permanent") mac address*/
2233 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2234 addr
.sa_family
= slave_dev
->type
;
2235 dev_set_mac_address(slave_dev
, &addr
);
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 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2252 bond_dev
->name
, bond_dev
->name
);
2253 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2257 pr_info("%s: released all slaves\n", bond_dev
->name
);
2260 write_unlock_bh(&bond
->lock
);
2262 bond_compute_features(bond
);
2268 * This function changes the active slave to slave <slave_dev>.
2269 * It returns -EINVAL in the following cases.
2270 * - <slave_dev> is not found in the list.
2271 * - There is not active slave now.
2272 * - <slave_dev> is already active.
2273 * - The link state of <slave_dev> is not BOND_LINK_UP.
2274 * - <slave_dev> is not running.
2275 * In these cases, this function does nothing.
2276 * In the other cases, current_slave pointer is changed and 0 is returned.
2278 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2280 struct bonding
*bond
= netdev_priv(bond_dev
);
2281 struct slave
*old_active
= NULL
;
2282 struct slave
*new_active
= NULL
;
2285 if (!USES_PRIMARY(bond
->params
.mode
))
2288 /* Verify that master_dev is indeed the master of slave_dev */
2289 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2292 read_lock(&bond
->lock
);
2294 read_lock(&bond
->curr_slave_lock
);
2295 old_active
= bond
->curr_active_slave
;
2296 read_unlock(&bond
->curr_slave_lock
);
2298 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2301 * Changing to the current active: do nothing; return success.
2303 if (new_active
&& (new_active
== old_active
)) {
2304 read_unlock(&bond
->lock
);
2310 (new_active
->link
== BOND_LINK_UP
) &&
2311 IS_UP(new_active
->dev
)) {
2313 write_lock_bh(&bond
->curr_slave_lock
);
2314 bond_change_active_slave(bond
, new_active
);
2315 write_unlock_bh(&bond
->curr_slave_lock
);
2316 unblock_netpoll_tx();
2320 read_unlock(&bond
->lock
);
2325 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2327 struct bonding
*bond
= netdev_priv(bond_dev
);
2329 info
->bond_mode
= bond
->params
.mode
;
2330 info
->miimon
= bond
->params
.miimon
;
2332 read_lock(&bond
->lock
);
2333 info
->num_slaves
= bond
->slave_cnt
;
2334 read_unlock(&bond
->lock
);
2339 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2341 struct bonding
*bond
= netdev_priv(bond_dev
);
2342 struct slave
*slave
;
2343 int i
, res
= -ENODEV
;
2345 read_lock(&bond
->lock
);
2347 bond_for_each_slave(bond
, slave
, i
) {
2348 if (i
== (int)info
->slave_id
) {
2350 strcpy(info
->slave_name
, slave
->dev
->name
);
2351 info
->link
= slave
->link
;
2352 info
->state
= bond_slave_state(slave
);
2353 info
->link_failure_count
= slave
->link_failure_count
;
2358 read_unlock(&bond
->lock
);
2363 /*-------------------------------- Monitoring -------------------------------*/
2366 static int bond_miimon_inspect(struct bonding
*bond
)
2368 struct slave
*slave
;
2369 int i
, link_state
, commit
= 0;
2370 bool ignore_updelay
;
2372 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2374 bond_for_each_slave(bond
, slave
, i
) {
2375 slave
->new_link
= BOND_LINK_NOCHANGE
;
2377 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2379 switch (slave
->link
) {
2384 slave
->link
= BOND_LINK_FAIL
;
2385 slave
->delay
= bond
->params
.downdelay
;
2387 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2389 (bond
->params
.mode
==
2390 BOND_MODE_ACTIVEBACKUP
) ?
2391 (bond_is_active_slave(slave
) ?
2392 "active " : "backup ") : "",
2394 bond
->params
.downdelay
* bond
->params
.miimon
);
2397 case BOND_LINK_FAIL
:
2400 * recovered before downdelay expired
2402 slave
->link
= BOND_LINK_UP
;
2403 slave
->jiffies
= jiffies
;
2404 pr_info("%s: link status up again after %d ms for interface %s.\n",
2406 (bond
->params
.downdelay
- slave
->delay
) *
2407 bond
->params
.miimon
,
2412 if (slave
->delay
<= 0) {
2413 slave
->new_link
= BOND_LINK_DOWN
;
2421 case BOND_LINK_DOWN
:
2425 slave
->link
= BOND_LINK_BACK
;
2426 slave
->delay
= bond
->params
.updelay
;
2429 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2430 bond
->dev
->name
, slave
->dev
->name
,
2431 ignore_updelay
? 0 :
2432 bond
->params
.updelay
*
2433 bond
->params
.miimon
);
2436 case BOND_LINK_BACK
:
2438 slave
->link
= BOND_LINK_DOWN
;
2439 pr_info("%s: link status down again after %d ms for interface %s.\n",
2441 (bond
->params
.updelay
- slave
->delay
) *
2442 bond
->params
.miimon
,
2451 if (slave
->delay
<= 0) {
2452 slave
->new_link
= BOND_LINK_UP
;
2454 ignore_updelay
= false;
2466 static void bond_miimon_commit(struct bonding
*bond
)
2468 struct slave
*slave
;
2471 bond_for_each_slave(bond
, slave
, i
) {
2472 switch (slave
->new_link
) {
2473 case BOND_LINK_NOCHANGE
:
2477 slave
->link
= BOND_LINK_UP
;
2478 slave
->jiffies
= jiffies
;
2480 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2481 /* prevent it from being the active one */
2482 bond_set_backup_slave(slave
);
2483 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2484 /* make it immediately active */
2485 bond_set_active_slave(slave
);
2486 } else if (slave
!= bond
->primary_slave
) {
2487 /* prevent it from being the active one */
2488 bond_set_backup_slave(slave
);
2491 bond_update_speed_duplex(slave
);
2493 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2494 bond
->dev
->name
, slave
->dev
->name
,
2495 slave
->speed
, slave
->duplex
? "full" : "half");
2497 /* notify ad that the link status has changed */
2498 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2499 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2501 if (bond_is_lb(bond
))
2502 bond_alb_handle_link_change(bond
, slave
,
2505 if (!bond
->curr_active_slave
||
2506 (slave
== bond
->primary_slave
))
2511 case BOND_LINK_DOWN
:
2512 if (slave
->link_failure_count
< UINT_MAX
)
2513 slave
->link_failure_count
++;
2515 slave
->link
= BOND_LINK_DOWN
;
2517 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2518 bond
->params
.mode
== BOND_MODE_8023AD
)
2519 bond_set_slave_inactive_flags(slave
);
2521 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2522 bond
->dev
->name
, slave
->dev
->name
);
2524 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2525 bond_3ad_handle_link_change(slave
,
2528 if (bond_is_lb(bond
))
2529 bond_alb_handle_link_change(bond
, slave
,
2532 if (slave
== bond
->curr_active_slave
)
2538 pr_err("%s: invalid new link %d on slave %s\n",
2539 bond
->dev
->name
, slave
->new_link
,
2541 slave
->new_link
= BOND_LINK_NOCHANGE
;
2549 write_lock_bh(&bond
->curr_slave_lock
);
2550 bond_select_active_slave(bond
);
2551 write_unlock_bh(&bond
->curr_slave_lock
);
2552 unblock_netpoll_tx();
2555 bond_set_carrier(bond
);
2561 * Really a wrapper that splits the mii monitor into two phases: an
2562 * inspection, then (if inspection indicates something needs to be done)
2563 * an acquisition of appropriate locks followed by a commit phase to
2564 * implement whatever link state changes are indicated.
2566 void bond_mii_monitor(struct work_struct
*work
)
2568 struct bonding
*bond
= container_of(work
, struct bonding
,
2570 bool should_notify_peers
= false;
2572 read_lock(&bond
->lock
);
2573 if (bond
->kill_timers
)
2576 if (bond
->slave_cnt
== 0)
2579 should_notify_peers
= bond_should_notify_peers(bond
);
2581 if (bond_miimon_inspect(bond
)) {
2582 read_unlock(&bond
->lock
);
2584 read_lock(&bond
->lock
);
2586 bond_miimon_commit(bond
);
2588 read_unlock(&bond
->lock
);
2589 rtnl_unlock(); /* might sleep, hold no other locks */
2590 read_lock(&bond
->lock
);
2594 if (bond
->params
.miimon
)
2595 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2596 msecs_to_jiffies(bond
->params
.miimon
));
2598 read_unlock(&bond
->lock
);
2600 if (should_notify_peers
) {
2602 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
2607 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2609 struct in_device
*idev
;
2610 struct in_ifaddr
*ifa
;
2617 idev
= __in_dev_get_rcu(dev
);
2621 ifa
= idev
->ifa_list
;
2625 addr
= ifa
->ifa_local
;
2631 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2633 struct vlan_entry
*vlan
;
2635 if (ip
== bond
->master_ip
)
2638 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2639 if (ip
== vlan
->vlan_ip
)
2647 * We go to the (large) trouble of VLAN tagging ARP frames because
2648 * switches in VLAN mode (especially if ports are configured as
2649 * "native" to a VLAN) might not pass non-tagged frames.
2651 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2653 struct sk_buff
*skb
;
2655 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2656 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2658 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2659 NULL
, slave_dev
->dev_addr
, NULL
);
2662 pr_err("ARP packet allocation failed\n");
2666 skb
= vlan_put_tag(skb
, vlan_id
);
2668 pr_err("failed to insert VLAN tag\n");
2676 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2679 __be32
*targets
= bond
->params
.arp_targets
;
2680 struct vlan_entry
*vlan
;
2681 struct net_device
*vlan_dev
;
2684 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2687 pr_debug("basa: target %x\n", targets
[i
]);
2689 pr_debug("basa: empty vlan: arp_send\n");
2690 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2691 bond
->master_ip
, 0);
2696 * If VLANs are configured, we do a route lookup to
2697 * determine which VLAN interface would be used, so we
2698 * can tag the ARP with the proper VLAN tag.
2700 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2703 if (net_ratelimit()) {
2704 pr_warning("%s: no route to arp_ip_target %pI4\n",
2705 bond
->dev
->name
, &targets
[i
]);
2711 * This target is not on a VLAN
2713 if (rt
->dst
.dev
== bond
->dev
) {
2715 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2716 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2717 bond
->master_ip
, 0);
2722 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2723 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2724 if (vlan_dev
== rt
->dst
.dev
) {
2725 vlan_id
= vlan
->vlan_id
;
2726 pr_debug("basa: vlan match on %s %d\n",
2727 vlan_dev
->name
, vlan_id
);
2734 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2735 vlan
->vlan_ip
, vlan_id
);
2739 if (net_ratelimit()) {
2740 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2741 bond
->dev
->name
, &targets
[i
],
2742 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2748 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2751 __be32
*targets
= bond
->params
.arp_targets
;
2753 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2754 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2755 &sip
, &tip
, i
, &targets
[i
],
2756 bond_has_this_ip(bond
, tip
));
2757 if (sip
== targets
[i
]) {
2758 if (bond_has_this_ip(bond
, tip
))
2759 slave
->last_arp_rx
= jiffies
;
2765 static void bond_arp_rcv(struct sk_buff
*skb
, struct bonding
*bond
,
2766 struct slave
*slave
)
2769 unsigned char *arp_ptr
;
2772 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2775 read_lock(&bond
->lock
);
2777 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2778 bond
->dev
->name
, skb
->dev
->name
);
2780 if (!pskb_may_pull(skb
, arp_hdr_len(bond
->dev
)))
2784 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2785 skb
->pkt_type
== PACKET_OTHERHOST
||
2786 skb
->pkt_type
== PACKET_LOOPBACK
||
2787 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2788 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2792 arp_ptr
= (unsigned char *)(arp
+ 1);
2793 arp_ptr
+= bond
->dev
->addr_len
;
2794 memcpy(&sip
, arp_ptr
, 4);
2795 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2796 memcpy(&tip
, arp_ptr
, 4);
2798 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2799 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2800 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2804 * Backup slaves won't see the ARP reply, but do come through
2805 * here for each ARP probe (so we swap the sip/tip to validate
2806 * the probe). In a "redundant switch, common router" type of
2807 * configuration, the ARP probe will (hopefully) travel from
2808 * the active, through one switch, the router, then the other
2809 * switch before reaching the backup.
2811 if (bond_is_active_slave(slave
))
2812 bond_validate_arp(bond
, slave
, sip
, tip
);
2814 bond_validate_arp(bond
, slave
, tip
, sip
);
2817 read_unlock(&bond
->lock
);
2821 * this function is called regularly to monitor each slave's link
2822 * ensuring that traffic is being sent and received when arp monitoring
2823 * is used in load-balancing mode. if the adapter has been dormant, then an
2824 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2825 * arp monitoring in active backup mode.
2827 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2829 struct bonding
*bond
= container_of(work
, struct bonding
,
2831 struct slave
*slave
, *oldcurrent
;
2832 int do_failover
= 0;
2836 read_lock(&bond
->lock
);
2838 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2840 if (bond
->kill_timers
)
2843 if (bond
->slave_cnt
== 0)
2846 read_lock(&bond
->curr_slave_lock
);
2847 oldcurrent
= bond
->curr_active_slave
;
2848 read_unlock(&bond
->curr_slave_lock
);
2850 /* see if any of the previous devices are up now (i.e. they have
2851 * xmt and rcv traffic). the curr_active_slave does not come into
2852 * the picture unless it is null. also, slave->jiffies is not needed
2853 * here because we send an arp on each slave and give a slave as
2854 * long as it needs to get the tx/rx within the delta.
2855 * TODO: what about up/down delay in arp mode? it wasn't here before
2858 bond_for_each_slave(bond
, slave
, i
) {
2859 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2861 if (slave
->link
!= BOND_LINK_UP
) {
2862 if (time_in_range(jiffies
,
2863 trans_start
- delta_in_ticks
,
2864 trans_start
+ delta_in_ticks
) &&
2865 time_in_range(jiffies
,
2866 slave
->dev
->last_rx
- delta_in_ticks
,
2867 slave
->dev
->last_rx
+ delta_in_ticks
)) {
2869 slave
->link
= BOND_LINK_UP
;
2870 bond_set_active_slave(slave
);
2872 /* primary_slave has no meaning in round-robin
2873 * mode. the window of a slave being up and
2874 * curr_active_slave being null after enslaving
2878 pr_info("%s: link status definitely up for interface %s, ",
2883 pr_info("%s: interface %s is now up\n",
2889 /* slave->link == BOND_LINK_UP */
2891 /* not all switches will respond to an arp request
2892 * when the source ip is 0, so don't take the link down
2893 * if we don't know our ip yet
2895 if (!time_in_range(jiffies
,
2896 trans_start
- delta_in_ticks
,
2897 trans_start
+ 2 * delta_in_ticks
) ||
2898 !time_in_range(jiffies
,
2899 slave
->dev
->last_rx
- delta_in_ticks
,
2900 slave
->dev
->last_rx
+ 2 * delta_in_ticks
)) {
2902 slave
->link
= BOND_LINK_DOWN
;
2903 bond_set_backup_slave(slave
);
2905 if (slave
->link_failure_count
< UINT_MAX
)
2906 slave
->link_failure_count
++;
2908 pr_info("%s: interface %s is now down.\n",
2912 if (slave
== oldcurrent
)
2917 /* note: if switch is in round-robin mode, all links
2918 * must tx arp to ensure all links rx an arp - otherwise
2919 * links may oscillate or not come up at all; if switch is
2920 * in something like xor mode, there is nothing we can
2921 * do - all replies will be rx'ed on same link causing slaves
2922 * to be unstable during low/no traffic periods
2924 if (IS_UP(slave
->dev
))
2925 bond_arp_send_all(bond
, slave
);
2930 write_lock_bh(&bond
->curr_slave_lock
);
2932 bond_select_active_slave(bond
);
2934 write_unlock_bh(&bond
->curr_slave_lock
);
2935 unblock_netpoll_tx();
2939 if (bond
->params
.arp_interval
)
2940 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2942 read_unlock(&bond
->lock
);
2946 * Called to inspect slaves for active-backup mode ARP monitor link state
2947 * changes. Sets new_link in slaves to specify what action should take
2948 * place for the slave. Returns 0 if no changes are found, >0 if changes
2949 * to link states must be committed.
2951 * Called with bond->lock held for read.
2953 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2955 struct slave
*slave
;
2957 unsigned long trans_start
;
2959 bond_for_each_slave(bond
, slave
, i
) {
2960 slave
->new_link
= BOND_LINK_NOCHANGE
;
2962 if (slave
->link
!= BOND_LINK_UP
) {
2963 if (time_in_range(jiffies
,
2964 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2965 slave_last_rx(bond
, slave
) + delta_in_ticks
)) {
2967 slave
->new_link
= BOND_LINK_UP
;
2975 * Give slaves 2*delta after being enslaved or made
2976 * active. This avoids bouncing, as the last receive
2977 * times need a full ARP monitor cycle to be updated.
2979 if (time_in_range(jiffies
,
2980 slave
->jiffies
- delta_in_ticks
,
2981 slave
->jiffies
+ 2 * delta_in_ticks
))
2985 * Backup slave is down if:
2986 * - No current_arp_slave AND
2987 * - more than 3*delta since last receive AND
2988 * - the bond has an IP address
2990 * Note: a non-null current_arp_slave indicates
2991 * the curr_active_slave went down and we are
2992 * searching for a new one; under this condition
2993 * we only take the curr_active_slave down - this
2994 * gives each slave a chance to tx/rx traffic
2995 * before being taken out
2997 if (!bond_is_active_slave(slave
) &&
2998 !bond
->current_arp_slave
&&
2999 !time_in_range(jiffies
,
3000 slave_last_rx(bond
, slave
) - delta_in_ticks
,
3001 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
)) {
3003 slave
->new_link
= BOND_LINK_DOWN
;
3008 * Active slave is down if:
3009 * - more than 2*delta since transmitting OR
3010 * - (more than 2*delta since receive AND
3011 * the bond has an IP address)
3013 trans_start
= dev_trans_start(slave
->dev
);
3014 if (bond_is_active_slave(slave
) &&
3015 (!time_in_range(jiffies
,
3016 trans_start
- delta_in_ticks
,
3017 trans_start
+ 2 * delta_in_ticks
) ||
3018 !time_in_range(jiffies
,
3019 slave_last_rx(bond
, slave
) - delta_in_ticks
,
3020 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
))) {
3022 slave
->new_link
= BOND_LINK_DOWN
;
3031 * Called to commit link state changes noted by inspection step of
3032 * active-backup mode ARP monitor.
3034 * Called with RTNL and bond->lock for read.
3036 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
3038 struct slave
*slave
;
3040 unsigned long trans_start
;
3042 bond_for_each_slave(bond
, slave
, i
) {
3043 switch (slave
->new_link
) {
3044 case BOND_LINK_NOCHANGE
:
3048 trans_start
= dev_trans_start(slave
->dev
);
3049 if ((!bond
->curr_active_slave
&&
3050 time_in_range(jiffies
,
3051 trans_start
- delta_in_ticks
,
3052 trans_start
+ delta_in_ticks
)) ||
3053 bond
->curr_active_slave
!= slave
) {
3054 slave
->link
= BOND_LINK_UP
;
3055 bond
->current_arp_slave
= NULL
;
3057 pr_info("%s: link status definitely up for interface %s.\n",
3058 bond
->dev
->name
, slave
->dev
->name
);
3060 if (!bond
->curr_active_slave
||
3061 (slave
== bond
->primary_slave
))
3068 case BOND_LINK_DOWN
:
3069 if (slave
->link_failure_count
< UINT_MAX
)
3070 slave
->link_failure_count
++;
3072 slave
->link
= BOND_LINK_DOWN
;
3073 bond_set_slave_inactive_flags(slave
);
3075 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3076 bond
->dev
->name
, slave
->dev
->name
);
3078 if (slave
== bond
->curr_active_slave
) {
3079 bond
->current_arp_slave
= NULL
;
3086 pr_err("%s: impossible: new_link %d on slave %s\n",
3087 bond
->dev
->name
, slave
->new_link
,
3095 write_lock_bh(&bond
->curr_slave_lock
);
3096 bond_select_active_slave(bond
);
3097 write_unlock_bh(&bond
->curr_slave_lock
);
3098 unblock_netpoll_tx();
3101 bond_set_carrier(bond
);
3105 * Send ARP probes for active-backup mode ARP monitor.
3107 * Called with bond->lock held for read.
3109 static void bond_ab_arp_probe(struct bonding
*bond
)
3111 struct slave
*slave
;
3114 read_lock(&bond
->curr_slave_lock
);
3116 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3117 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3118 bond
->current_arp_slave
->dev
->name
,
3119 bond
->curr_active_slave
->dev
->name
);
3121 if (bond
->curr_active_slave
) {
3122 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3123 read_unlock(&bond
->curr_slave_lock
);
3127 read_unlock(&bond
->curr_slave_lock
);
3129 /* if we don't have a curr_active_slave, search for the next available
3130 * backup slave from the current_arp_slave and make it the candidate
3131 * for becoming the curr_active_slave
3134 if (!bond
->current_arp_slave
) {
3135 bond
->current_arp_slave
= bond
->first_slave
;
3136 if (!bond
->current_arp_slave
)
3140 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3142 /* search for next candidate */
3143 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3144 if (IS_UP(slave
->dev
)) {
3145 slave
->link
= BOND_LINK_BACK
;
3146 bond_set_slave_active_flags(slave
);
3147 bond_arp_send_all(bond
, slave
);
3148 slave
->jiffies
= jiffies
;
3149 bond
->current_arp_slave
= slave
;
3153 /* if the link state is up at this point, we
3154 * mark it down - this can happen if we have
3155 * simultaneous link failures and
3156 * reselect_active_interface doesn't make this
3157 * one the current slave so it is still marked
3158 * up when it is actually down
3160 if (slave
->link
== BOND_LINK_UP
) {
3161 slave
->link
= BOND_LINK_DOWN
;
3162 if (slave
->link_failure_count
< UINT_MAX
)
3163 slave
->link_failure_count
++;
3165 bond_set_slave_inactive_flags(slave
);
3167 pr_info("%s: backup interface %s is now down.\n",
3168 bond
->dev
->name
, slave
->dev
->name
);
3173 void bond_activebackup_arp_mon(struct work_struct
*work
)
3175 struct bonding
*bond
= container_of(work
, struct bonding
,
3177 bool should_notify_peers
= false;
3180 read_lock(&bond
->lock
);
3182 if (bond
->kill_timers
)
3185 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3187 if (bond
->slave_cnt
== 0)
3190 should_notify_peers
= bond_should_notify_peers(bond
);
3192 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3193 read_unlock(&bond
->lock
);
3195 read_lock(&bond
->lock
);
3197 bond_ab_arp_commit(bond
, delta_in_ticks
);
3199 read_unlock(&bond
->lock
);
3201 read_lock(&bond
->lock
);
3204 bond_ab_arp_probe(bond
);
3207 if (bond
->params
.arp_interval
)
3208 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3210 read_unlock(&bond
->lock
);
3212 if (should_notify_peers
) {
3214 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
3219 /*-------------------------- netdev event handling --------------------------*/
3222 * Change device name
3224 static int bond_event_changename(struct bonding
*bond
)
3226 bond_remove_proc_entry(bond
);
3227 bond_create_proc_entry(bond
);
3229 bond_debug_reregister(bond
);
3234 static int bond_master_netdev_event(unsigned long event
,
3235 struct net_device
*bond_dev
)
3237 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3240 case NETDEV_CHANGENAME
:
3241 return bond_event_changename(event_bond
);
3249 static int bond_slave_netdev_event(unsigned long event
,
3250 struct net_device
*slave_dev
)
3252 struct net_device
*bond_dev
= slave_dev
->master
;
3253 struct bonding
*bond
= netdev_priv(bond_dev
);
3256 case NETDEV_UNREGISTER
:
3258 if (bond
->setup_by_slave
)
3259 bond_release_and_destroy(bond_dev
, slave_dev
);
3261 bond_release(bond_dev
, slave_dev
);
3265 if (bond
->params
.mode
== BOND_MODE_8023AD
|| bond_is_lb(bond
)) {
3266 struct slave
*slave
;
3268 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3270 u32 old_speed
= slave
->speed
;
3271 u8 old_duplex
= slave
->duplex
;
3273 bond_update_speed_duplex(slave
);
3275 if (bond_is_lb(bond
))
3278 if (old_speed
!= slave
->speed
)
3279 bond_3ad_adapter_speed_changed(slave
);
3280 if (old_duplex
!= slave
->duplex
)
3281 bond_3ad_adapter_duplex_changed(slave
);
3288 * ... Or is it this?
3291 case NETDEV_CHANGEMTU
:
3293 * TODO: Should slaves be allowed to
3294 * independently alter their MTU? For
3295 * an active-backup bond, slaves need
3296 * not be the same type of device, so
3297 * MTUs may vary. For other modes,
3298 * slaves arguably should have the
3299 * same MTUs. To do this, we'd need to
3300 * take over the slave's change_mtu
3301 * function for the duration of their
3305 case NETDEV_CHANGENAME
:
3307 * TODO: handle changing the primary's name
3310 case NETDEV_FEAT_CHANGE
:
3311 bond_compute_features(bond
);
3321 * bond_netdev_event: handle netdev notifier chain events.
3323 * This function receives events for the netdev chain. The caller (an
3324 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3325 * locks for us to safely manipulate the slave devices (RTNL lock,
3328 static int bond_netdev_event(struct notifier_block
*this,
3329 unsigned long event
, void *ptr
)
3331 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3333 pr_debug("event_dev: %s, event: %lx\n",
3334 event_dev
? event_dev
->name
: "None",
3337 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3340 if (event_dev
->flags
& IFF_MASTER
) {
3341 pr_debug("IFF_MASTER\n");
3342 return bond_master_netdev_event(event
, event_dev
);
3345 if (event_dev
->flags
& IFF_SLAVE
) {
3346 pr_debug("IFF_SLAVE\n");
3347 return bond_slave_netdev_event(event
, event_dev
);
3354 * bond_inetaddr_event: handle inetaddr notifier chain events.
3356 * We keep track of device IPs primarily to use as source addresses in
3357 * ARP monitor probes (rather than spewing out broadcasts all the time).
3359 * We track one IP for the main device (if it has one), plus one per VLAN.
3361 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3363 struct in_ifaddr
*ifa
= ptr
;
3364 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3365 struct bond_net
*bn
= net_generic(dev_net(event_dev
), bond_net_id
);
3366 struct bonding
*bond
;
3367 struct vlan_entry
*vlan
;
3369 list_for_each_entry(bond
, &bn
->dev_list
, bond_list
) {
3370 if (bond
->dev
== event_dev
) {
3373 bond
->master_ip
= ifa
->ifa_local
;
3376 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3383 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3386 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3387 if (vlan_dev
== event_dev
) {
3390 vlan
->vlan_ip
= ifa
->ifa_local
;
3394 bond_glean_dev_ip(vlan_dev
);
3405 static struct notifier_block bond_netdev_notifier
= {
3406 .notifier_call
= bond_netdev_event
,
3409 static struct notifier_block bond_inetaddr_notifier
= {
3410 .notifier_call
= bond_inetaddr_event
,
3413 /*---------------------------- Hashing Policies -----------------------------*/
3416 * Hash for the output device based upon layer 2 and layer 3 data. If
3417 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3419 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3421 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3422 struct iphdr
*iph
= ip_hdr(skb
);
3424 if (skb
->protocol
== htons(ETH_P_IP
)) {
3425 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3426 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3429 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3433 * Hash for the output device based upon layer 3 and layer 4 data. If
3434 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3435 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3437 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3439 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3440 struct iphdr
*iph
= ip_hdr(skb
);
3441 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3444 if (skb
->protocol
== htons(ETH_P_IP
)) {
3445 if (!ip_is_fragment(iph
) &&
3446 (iph
->protocol
== IPPROTO_TCP
||
3447 iph
->protocol
== IPPROTO_UDP
)) {
3448 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3450 return (layer4_xor
^
3451 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3455 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3459 * Hash for the output device based upon layer 2 data
3461 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3463 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3465 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3468 /*-------------------------- Device entry points ----------------------------*/
3470 static int bond_open(struct net_device
*bond_dev
)
3472 struct bonding
*bond
= netdev_priv(bond_dev
);
3474 bond
->kill_timers
= 0;
3476 INIT_DELAYED_WORK(&bond
->mcast_work
, bond_resend_igmp_join_requests_delayed
);
3478 if (bond_is_lb(bond
)) {
3479 /* bond_alb_initialize must be called before the timer
3482 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3483 /* something went wrong - fail the open operation */
3487 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3488 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3491 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3492 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3493 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3496 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3497 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3498 INIT_DELAYED_WORK(&bond
->arp_work
,
3499 bond_activebackup_arp_mon
);
3501 INIT_DELAYED_WORK(&bond
->arp_work
,
3502 bond_loadbalance_arp_mon
);
3504 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3505 if (bond
->params
.arp_validate
)
3506 bond
->recv_probe
= bond_arp_rcv
;
3509 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3510 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3511 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3512 /* register to receive LACPDUs */
3513 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3514 bond_3ad_initiate_agg_selection(bond
, 1);
3520 static int bond_close(struct net_device
*bond_dev
)
3522 struct bonding
*bond
= netdev_priv(bond_dev
);
3524 write_lock_bh(&bond
->lock
);
3526 bond
->send_peer_notif
= 0;
3528 /* signal timers not to re-arm */
3529 bond
->kill_timers
= 1;
3531 write_unlock_bh(&bond
->lock
);
3533 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3534 cancel_delayed_work(&bond
->mii_work
);
3537 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3538 cancel_delayed_work(&bond
->arp_work
);
3541 switch (bond
->params
.mode
) {
3542 case BOND_MODE_8023AD
:
3543 cancel_delayed_work(&bond
->ad_work
);
3547 cancel_delayed_work(&bond
->alb_work
);
3553 if (delayed_work_pending(&bond
->mcast_work
))
3554 cancel_delayed_work(&bond
->mcast_work
);
3556 if (bond_is_lb(bond
)) {
3557 /* Must be called only after all
3558 * slaves have been released
3560 bond_alb_deinitialize(bond
);
3562 bond
->recv_probe
= NULL
;
3567 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3568 struct rtnl_link_stats64
*stats
)
3570 struct bonding
*bond
= netdev_priv(bond_dev
);
3571 struct rtnl_link_stats64 temp
;
3572 struct slave
*slave
;
3575 memset(stats
, 0, sizeof(*stats
));
3577 read_lock_bh(&bond
->lock
);
3579 bond_for_each_slave(bond
, slave
, i
) {
3580 const struct rtnl_link_stats64
*sstats
=
3581 dev_get_stats(slave
->dev
, &temp
);
3583 stats
->rx_packets
+= sstats
->rx_packets
;
3584 stats
->rx_bytes
+= sstats
->rx_bytes
;
3585 stats
->rx_errors
+= sstats
->rx_errors
;
3586 stats
->rx_dropped
+= sstats
->rx_dropped
;
3588 stats
->tx_packets
+= sstats
->tx_packets
;
3589 stats
->tx_bytes
+= sstats
->tx_bytes
;
3590 stats
->tx_errors
+= sstats
->tx_errors
;
3591 stats
->tx_dropped
+= sstats
->tx_dropped
;
3593 stats
->multicast
+= sstats
->multicast
;
3594 stats
->collisions
+= sstats
->collisions
;
3596 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3597 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3598 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3599 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3600 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3601 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3603 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3604 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3605 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3606 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3607 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3610 read_unlock_bh(&bond
->lock
);
3615 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3617 struct net_device
*slave_dev
= NULL
;
3618 struct ifbond k_binfo
;
3619 struct ifbond __user
*u_binfo
= NULL
;
3620 struct ifslave k_sinfo
;
3621 struct ifslave __user
*u_sinfo
= NULL
;
3622 struct mii_ioctl_data
*mii
= NULL
;
3625 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3637 * We do this again just in case we were called by SIOCGMIIREG
3638 * instead of SIOCGMIIPHY.
3645 if (mii
->reg_num
== 1) {
3646 struct bonding
*bond
= netdev_priv(bond_dev
);
3648 read_lock(&bond
->lock
);
3649 read_lock(&bond
->curr_slave_lock
);
3650 if (netif_carrier_ok(bond
->dev
))
3651 mii
->val_out
= BMSR_LSTATUS
;
3653 read_unlock(&bond
->curr_slave_lock
);
3654 read_unlock(&bond
->lock
);
3658 case BOND_INFO_QUERY_OLD
:
3659 case SIOCBONDINFOQUERY
:
3660 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3662 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3665 res
= bond_info_query(bond_dev
, &k_binfo
);
3667 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3671 case BOND_SLAVE_INFO_QUERY_OLD
:
3672 case SIOCBONDSLAVEINFOQUERY
:
3673 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3675 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3678 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3680 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3689 if (!capable(CAP_NET_ADMIN
))
3692 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
3694 pr_debug("slave_dev=%p:\n", slave_dev
);
3699 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3701 case BOND_ENSLAVE_OLD
:
3702 case SIOCBONDENSLAVE
:
3703 res
= bond_enslave(bond_dev
, slave_dev
);
3705 case BOND_RELEASE_OLD
:
3706 case SIOCBONDRELEASE
:
3707 res
= bond_release(bond_dev
, slave_dev
);
3709 case BOND_SETHWADDR_OLD
:
3710 case SIOCBONDSETHWADDR
:
3711 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3713 case BOND_CHANGE_ACTIVE_OLD
:
3714 case SIOCBONDCHANGEACTIVE
:
3715 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3727 static bool bond_addr_in_mc_list(unsigned char *addr
,
3728 struct netdev_hw_addr_list
*list
,
3731 struct netdev_hw_addr
*ha
;
3733 netdev_hw_addr_list_for_each(ha
, list
)
3734 if (!memcmp(ha
->addr
, addr
, addrlen
))
3740 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3742 struct bonding
*bond
= netdev_priv(bond_dev
);
3743 struct netdev_hw_addr
*ha
;
3747 * Do promisc before checking multicast_mode
3749 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
))
3751 * FIXME: Need to handle the error when one of the multi-slaves
3754 bond_set_promiscuity(bond
, 1);
3757 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
))
3758 bond_set_promiscuity(bond
, -1);
3761 /* set allmulti flag to slaves */
3762 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
))
3764 * FIXME: Need to handle the error when one of the multi-slaves
3767 bond_set_allmulti(bond
, 1);
3770 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
))
3771 bond_set_allmulti(bond
, -1);
3774 read_lock(&bond
->lock
);
3776 bond
->flags
= bond_dev
->flags
;
3778 /* looking for addresses to add to slaves' mc list */
3779 netdev_for_each_mc_addr(ha
, bond_dev
) {
3780 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
3781 bond_dev
->addr_len
);
3783 bond_mc_add(bond
, ha
->addr
);
3786 /* looking for addresses to delete from slaves' list */
3787 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
3788 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
3789 bond_dev
->addr_len
);
3791 bond_mc_del(bond
, ha
->addr
);
3794 /* save master's multicast list */
3795 __hw_addr_flush(&bond
->mc_list
);
3796 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
3797 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
3799 read_unlock(&bond
->lock
);
3802 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
3804 struct bonding
*bond
= netdev_priv(dev
);
3805 struct slave
*slave
= bond
->first_slave
;
3808 const struct net_device_ops
*slave_ops
3809 = slave
->dev
->netdev_ops
;
3810 if (slave_ops
->ndo_neigh_setup
)
3811 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
3817 * Change the MTU of all of a master's slaves to match the master
3819 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3821 struct bonding
*bond
= netdev_priv(bond_dev
);
3822 struct slave
*slave
, *stop_at
;
3826 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3827 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3829 /* Can't hold bond->lock with bh disabled here since
3830 * some base drivers panic. On the other hand we can't
3831 * hold bond->lock without bh disabled because we'll
3832 * deadlock. The only solution is to rely on the fact
3833 * that we're under rtnl_lock here, and the slaves
3834 * list won't change. This doesn't solve the problem
3835 * of setting the slave's MTU while it is
3836 * transmitting, but the assumption is that the base
3837 * driver can handle that.
3839 * TODO: figure out a way to safely iterate the slaves
3840 * list, but without holding a lock around the actual
3841 * call to the base driver.
3844 bond_for_each_slave(bond
, slave
, i
) {
3845 pr_debug("s %p s->p %p c_m %p\n",
3848 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3850 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3853 /* If we failed to set the slave's mtu to the new value
3854 * we must abort the operation even in ACTIVE_BACKUP
3855 * mode, because if we allow the backup slaves to have
3856 * different mtu values than the active slave we'll
3857 * need to change their mtu when doing a failover. That
3858 * means changing their mtu from timer context, which
3859 * is probably not a good idea.
3861 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3866 bond_dev
->mtu
= new_mtu
;
3871 /* unwind from head to the slave that failed */
3873 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3876 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3878 pr_debug("unwind err %d dev %s\n",
3879 tmp_res
, slave
->dev
->name
);
3889 * Note that many devices must be down to change the HW address, and
3890 * downing the master releases all slaves. We can make bonds full of
3891 * bonding devices to test this, however.
3893 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3895 struct bonding
*bond
= netdev_priv(bond_dev
);
3896 struct sockaddr
*sa
= addr
, tmp_sa
;
3897 struct slave
*slave
, *stop_at
;
3901 if (bond
->params
.mode
== BOND_MODE_ALB
)
3902 return bond_alb_set_mac_address(bond_dev
, addr
);
3905 pr_debug("bond=%p, name=%s\n",
3906 bond
, bond_dev
? bond_dev
->name
: "None");
3909 * If fail_over_mac is set to active, do nothing and return
3910 * success. Returning an error causes ifenslave to fail.
3912 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
3915 if (!is_valid_ether_addr(sa
->sa_data
))
3916 return -EADDRNOTAVAIL
;
3918 /* Can't hold bond->lock with bh disabled here since
3919 * some base drivers panic. On the other hand we can't
3920 * hold bond->lock without bh disabled because we'll
3921 * deadlock. The only solution is to rely on the fact
3922 * that we're under rtnl_lock here, and the slaves
3923 * list won't change. This doesn't solve the problem
3924 * of setting the slave's hw address while it is
3925 * transmitting, but the assumption is that the base
3926 * driver can handle that.
3928 * TODO: figure out a way to safely iterate the slaves
3929 * list, but without holding a lock around the actual
3930 * call to the base driver.
3933 bond_for_each_slave(bond
, slave
, i
) {
3934 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3935 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3937 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3939 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3943 res
= dev_set_mac_address(slave
->dev
, addr
);
3945 /* TODO: consider downing the slave
3947 * User should expect communications
3948 * breakage anyway until ARP finish
3951 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3957 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3961 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3962 tmp_sa
.sa_family
= bond_dev
->type
;
3964 /* unwind from head to the slave that failed */
3966 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3969 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3971 pr_debug("unwind err %d dev %s\n",
3972 tmp_res
, slave
->dev
->name
);
3979 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3981 struct bonding
*bond
= netdev_priv(bond_dev
);
3982 struct slave
*slave
, *start_at
;
3983 int i
, slave_no
, res
= 1;
3984 struct iphdr
*iph
= ip_hdr(skb
);
3987 * Start with the curr_active_slave that joined the bond as the
3988 * default for sending IGMP traffic. For failover purposes one
3989 * needs to maintain some consistency for the interface that will
3990 * send the join/membership reports. The curr_active_slave found
3991 * will send all of this type of traffic.
3993 if ((iph
->protocol
== IPPROTO_IGMP
) &&
3994 (skb
->protocol
== htons(ETH_P_IP
))) {
3996 read_lock(&bond
->curr_slave_lock
);
3997 slave
= bond
->curr_active_slave
;
3998 read_unlock(&bond
->curr_slave_lock
);
4004 * Concurrent TX may collide on rr_tx_counter; we accept
4005 * that as being rare enough not to justify using an
4008 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4010 bond_for_each_slave(bond
, slave
, i
) {
4018 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4019 if (IS_UP(slave
->dev
) &&
4020 (slave
->link
== BOND_LINK_UP
) &&
4021 bond_is_active_slave(slave
)) {
4022 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4029 /* no suitable interface, frame not sent */
4033 return NETDEV_TX_OK
;
4038 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4039 * the bond has a usable interface.
4041 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4043 struct bonding
*bond
= netdev_priv(bond_dev
);
4046 read_lock(&bond
->curr_slave_lock
);
4048 if (bond
->curr_active_slave
)
4049 res
= bond_dev_queue_xmit(bond
, skb
,
4050 bond
->curr_active_slave
->dev
);
4053 /* no suitable interface, frame not sent */
4056 read_unlock(&bond
->curr_slave_lock
);
4058 return NETDEV_TX_OK
;
4062 * In bond_xmit_xor() , we determine the output device by using a pre-
4063 * determined xmit_hash_policy(), If the selected device is not enabled,
4064 * find the next active slave.
4066 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4068 struct bonding
*bond
= netdev_priv(bond_dev
);
4069 struct slave
*slave
, *start_at
;
4074 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4076 bond_for_each_slave(bond
, slave
, i
) {
4084 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4085 if (IS_UP(slave
->dev
) &&
4086 (slave
->link
== BOND_LINK_UP
) &&
4087 bond_is_active_slave(slave
)) {
4088 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4094 /* no suitable interface, frame not sent */
4098 return NETDEV_TX_OK
;
4102 * in broadcast mode, we send everything to all usable interfaces.
4104 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4106 struct bonding
*bond
= netdev_priv(bond_dev
);
4107 struct slave
*slave
, *start_at
;
4108 struct net_device
*tx_dev
= NULL
;
4112 read_lock(&bond
->curr_slave_lock
);
4113 start_at
= bond
->curr_active_slave
;
4114 read_unlock(&bond
->curr_slave_lock
);
4119 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4120 if (IS_UP(slave
->dev
) &&
4121 (slave
->link
== BOND_LINK_UP
) &&
4122 bond_is_active_slave(slave
)) {
4124 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4126 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4131 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4133 dev_kfree_skb(skb2
);
4137 tx_dev
= slave
->dev
;
4142 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4146 /* no suitable interface, frame not sent */
4149 /* frame sent to all suitable interfaces */
4150 return NETDEV_TX_OK
;
4153 /*------------------------- Device initialization ---------------------------*/
4155 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4157 switch (bond
->params
.xmit_policy
) {
4158 case BOND_XMIT_POLICY_LAYER23
:
4159 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4161 case BOND_XMIT_POLICY_LAYER34
:
4162 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4164 case BOND_XMIT_POLICY_LAYER2
:
4166 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4172 * Lookup the slave that corresponds to a qid
4174 static inline int bond_slave_override(struct bonding
*bond
,
4175 struct sk_buff
*skb
)
4178 struct slave
*slave
= NULL
;
4179 struct slave
*check_slave
;
4181 if (!skb
->queue_mapping
)
4184 /* Find out if any slaves have the same mapping as this skb. */
4185 bond_for_each_slave(bond
, check_slave
, i
) {
4186 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4187 slave
= check_slave
;
4192 /* If the slave isn't UP, use default transmit policy. */
4193 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4194 (slave
->link
== BOND_LINK_UP
)) {
4195 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4202 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4205 * This helper function exists to help dev_pick_tx get the correct
4206 * destination queue. Using a helper function skips a call to
4207 * skb_tx_hash and will put the skbs in the queue we expect on their
4208 * way down to the bonding driver.
4210 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
4213 * Save the original txq to restore before passing to the driver
4215 bond_queue_mapping(skb
) = skb
->queue_mapping
;
4217 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
4219 txq
-= dev
->real_num_tx_queues
;
4220 } while (txq
>= dev
->real_num_tx_queues
);
4225 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4227 struct bonding
*bond
= netdev_priv(dev
);
4229 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4230 if (!bond_slave_override(bond
, skb
))
4231 return NETDEV_TX_OK
;
4234 switch (bond
->params
.mode
) {
4235 case BOND_MODE_ROUNDROBIN
:
4236 return bond_xmit_roundrobin(skb
, dev
);
4237 case BOND_MODE_ACTIVEBACKUP
:
4238 return bond_xmit_activebackup(skb
, dev
);
4240 return bond_xmit_xor(skb
, dev
);
4241 case BOND_MODE_BROADCAST
:
4242 return bond_xmit_broadcast(skb
, dev
);
4243 case BOND_MODE_8023AD
:
4244 return bond_3ad_xmit_xor(skb
, dev
);
4247 return bond_alb_xmit(skb
, dev
);
4249 /* Should never happen, mode already checked */
4250 pr_err("%s: Error: Unknown bonding mode %d\n",
4251 dev
->name
, bond
->params
.mode
);
4254 return NETDEV_TX_OK
;
4258 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4260 struct bonding
*bond
= netdev_priv(dev
);
4261 netdev_tx_t ret
= NETDEV_TX_OK
;
4264 * If we risk deadlock from transmitting this in the
4265 * netpoll path, tell netpoll to queue the frame for later tx
4267 if (is_netpoll_tx_blocked(dev
))
4268 return NETDEV_TX_BUSY
;
4270 read_lock(&bond
->lock
);
4272 if (bond
->slave_cnt
)
4273 ret
= __bond_start_xmit(skb
, dev
);
4277 read_unlock(&bond
->lock
);
4283 * set bond mode specific net device operations
4285 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4287 struct net_device
*bond_dev
= bond
->dev
;
4290 case BOND_MODE_ROUNDROBIN
:
4292 case BOND_MODE_ACTIVEBACKUP
:
4295 bond_set_xmit_hash_policy(bond
);
4297 case BOND_MODE_BROADCAST
:
4299 case BOND_MODE_8023AD
:
4300 bond_set_xmit_hash_policy(bond
);
4307 /* Should never happen, mode already checked */
4308 pr_err("%s: Error: Unknown bonding mode %d\n",
4309 bond_dev
->name
, mode
);
4314 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4315 struct ethtool_drvinfo
*drvinfo
)
4317 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4318 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4319 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4322 static const struct ethtool_ops bond_ethtool_ops
= {
4323 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4324 .get_link
= ethtool_op_get_link
,
4327 static const struct net_device_ops bond_netdev_ops
= {
4328 .ndo_init
= bond_init
,
4329 .ndo_uninit
= bond_uninit
,
4330 .ndo_open
= bond_open
,
4331 .ndo_stop
= bond_close
,
4332 .ndo_start_xmit
= bond_start_xmit
,
4333 .ndo_select_queue
= bond_select_queue
,
4334 .ndo_get_stats64
= bond_get_stats
,
4335 .ndo_do_ioctl
= bond_do_ioctl
,
4336 .ndo_set_multicast_list
= bond_set_multicast_list
,
4337 .ndo_change_mtu
= bond_change_mtu
,
4338 .ndo_set_mac_address
= bond_set_mac_address
,
4339 .ndo_neigh_setup
= bond_neigh_setup
,
4340 .ndo_vlan_rx_register
= bond_vlan_rx_register
,
4341 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4342 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4343 #ifdef CONFIG_NET_POLL_CONTROLLER
4344 .ndo_netpoll_setup
= bond_netpoll_setup
,
4345 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4346 .ndo_poll_controller
= bond_poll_controller
,
4348 .ndo_add_slave
= bond_enslave
,
4349 .ndo_del_slave
= bond_release
,
4350 .ndo_fix_features
= bond_fix_features
,
4353 static void bond_destructor(struct net_device
*bond_dev
)
4355 struct bonding
*bond
= netdev_priv(bond_dev
);
4357 destroy_workqueue(bond
->wq
);
4358 free_netdev(bond_dev
);
4361 static void bond_setup(struct net_device
*bond_dev
)
4363 struct bonding
*bond
= netdev_priv(bond_dev
);
4365 /* initialize rwlocks */
4366 rwlock_init(&bond
->lock
);
4367 rwlock_init(&bond
->curr_slave_lock
);
4369 bond
->params
= bonding_defaults
;
4371 /* Initialize pointers */
4372 bond
->dev
= bond_dev
;
4373 INIT_LIST_HEAD(&bond
->vlan_list
);
4375 /* Initialize the device entry points */
4376 ether_setup(bond_dev
);
4377 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4378 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4379 bond_set_mode_ops(bond
, bond
->params
.mode
);
4381 bond_dev
->destructor
= bond_destructor
;
4383 /* Initialize the device options */
4384 bond_dev
->tx_queue_len
= 0;
4385 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4386 bond_dev
->priv_flags
|= IFF_BONDING
;
4387 bond_dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
4389 /* At first, we block adding VLANs. That's the only way to
4390 * prevent problems that occur when adding VLANs over an
4391 * empty bond. The block will be removed once non-challenged
4392 * slaves are enslaved.
4394 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4396 /* don't acquire bond device's netif_tx_lock when
4398 bond_dev
->features
|= NETIF_F_LLTX
;
4400 /* By default, we declare the bond to be fully
4401 * VLAN hardware accelerated capable. Special
4402 * care is taken in the various xmit functions
4403 * when there are slaves that are not hw accel
4407 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4408 NETIF_F_HW_VLAN_TX
|
4409 NETIF_F_HW_VLAN_RX
|
4410 NETIF_F_HW_VLAN_FILTER
;
4412 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_NO_CSUM
);
4413 bond_dev
->features
|= bond_dev
->hw_features
;
4416 static void bond_work_cancel_all(struct bonding
*bond
)
4418 write_lock_bh(&bond
->lock
);
4419 bond
->kill_timers
= 1;
4420 write_unlock_bh(&bond
->lock
);
4422 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4423 cancel_delayed_work(&bond
->mii_work
);
4425 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4426 cancel_delayed_work(&bond
->arp_work
);
4428 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4429 delayed_work_pending(&bond
->alb_work
))
4430 cancel_delayed_work(&bond
->alb_work
);
4432 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4433 delayed_work_pending(&bond
->ad_work
))
4434 cancel_delayed_work(&bond
->ad_work
);
4436 if (delayed_work_pending(&bond
->mcast_work
))
4437 cancel_delayed_work(&bond
->mcast_work
);
4441 * Destroy a bonding device.
4442 * Must be under rtnl_lock when this function is called.
4444 static void bond_uninit(struct net_device
*bond_dev
)
4446 struct bonding
*bond
= netdev_priv(bond_dev
);
4447 struct vlan_entry
*vlan
, *tmp
;
4449 bond_netpoll_cleanup(bond_dev
);
4451 /* Release the bonded slaves */
4452 bond_release_all(bond_dev
);
4454 list_del(&bond
->bond_list
);
4456 bond_work_cancel_all(bond
);
4458 bond_remove_proc_entry(bond
);
4460 bond_debug_unregister(bond
);
4462 __hw_addr_flush(&bond
->mc_list
);
4464 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4465 list_del(&vlan
->vlan_list
);
4470 /*------------------------- Module initialization ---------------------------*/
4473 * Convert string input module parms. Accept either the
4474 * number of the mode or its string name. A bit complicated because
4475 * some mode names are substrings of other names, and calls from sysfs
4476 * may have whitespace in the name (trailing newlines, for example).
4478 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4480 int modeint
= -1, i
, rv
;
4481 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4483 for (p
= (char *)buf
; *p
; p
++)
4484 if (!(isdigit(*p
) || isspace(*p
)))
4488 rv
= sscanf(buf
, "%20s", modestr
);
4490 rv
= sscanf(buf
, "%d", &modeint
);
4495 for (i
= 0; tbl
[i
].modename
; i
++) {
4496 if (modeint
== tbl
[i
].mode
)
4498 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4505 static int bond_check_params(struct bond_params
*params
)
4507 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4510 * Convert string parameters.
4513 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4514 if (bond_mode
== -1) {
4515 pr_err("Error: Invalid bonding mode \"%s\"\n",
4516 mode
== NULL
? "NULL" : mode
);
4521 if (xmit_hash_policy
) {
4522 if ((bond_mode
!= BOND_MODE_XOR
) &&
4523 (bond_mode
!= BOND_MODE_8023AD
)) {
4524 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4525 bond_mode_name(bond_mode
));
4527 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4529 if (xmit_hashtype
== -1) {
4530 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4531 xmit_hash_policy
== NULL
? "NULL" :
4539 if (bond_mode
!= BOND_MODE_8023AD
) {
4540 pr_info("lacp_rate param is irrelevant in mode %s\n",
4541 bond_mode_name(bond_mode
));
4543 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4544 if (lacp_fast
== -1) {
4545 pr_err("Error: Invalid lacp rate \"%s\"\n",
4546 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4553 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4554 if (params
->ad_select
== -1) {
4555 pr_err("Error: Invalid ad_select \"%s\"\n",
4556 ad_select
== NULL
? "NULL" : ad_select
);
4560 if (bond_mode
!= BOND_MODE_8023AD
) {
4561 pr_warning("ad_select param only affects 802.3ad mode\n");
4564 params
->ad_select
= BOND_AD_STABLE
;
4567 if (max_bonds
< 0) {
4568 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4569 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4570 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4574 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4575 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4576 miimon
= BOND_LINK_MON_INTERV
;
4580 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4585 if (downdelay
< 0) {
4586 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4587 downdelay
, INT_MAX
);
4591 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4592 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4597 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4598 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4603 /* reset values for 802.3ad */
4604 if (bond_mode
== BOND_MODE_8023AD
) {
4606 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");
4607 pr_warning("Forcing miimon to 100msec\n");
4612 if (tx_queues
< 1 || tx_queues
> 255) {
4613 pr_warning("Warning: tx_queues (%d) should be between "
4614 "1 and 255, resetting to %d\n",
4615 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4616 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4619 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4620 pr_warning("Warning: all_slaves_active module parameter (%d), "
4621 "not of valid value (0/1), so it was set to "
4622 "0\n", all_slaves_active
);
4623 all_slaves_active
= 0;
4626 if (resend_igmp
< 0 || resend_igmp
> 255) {
4627 pr_warning("Warning: resend_igmp (%d) should be between "
4628 "0 and 255, resetting to %d\n",
4629 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4630 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4633 /* reset values for TLB/ALB */
4634 if ((bond_mode
== BOND_MODE_TLB
) ||
4635 (bond_mode
== BOND_MODE_ALB
)) {
4637 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");
4638 pr_warning("Forcing miimon to 100msec\n");
4643 if (bond_mode
== BOND_MODE_ALB
) {
4644 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",
4649 if (updelay
|| downdelay
) {
4650 /* just warn the user the up/down delay will have
4651 * no effect since miimon is zero...
4653 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",
4654 updelay
, downdelay
);
4657 /* don't allow arp monitoring */
4659 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4660 miimon
, arp_interval
);
4664 if ((updelay
% miimon
) != 0) {
4665 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4667 (updelay
/ miimon
) * miimon
);
4672 if ((downdelay
% miimon
) != 0) {
4673 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4675 (downdelay
/ miimon
) * miimon
);
4678 downdelay
/= miimon
;
4681 if (arp_interval
< 0) {
4682 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4683 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4684 arp_interval
= BOND_LINK_ARP_INTERV
;
4687 for (arp_ip_count
= 0;
4688 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4690 /* not complete check, but should be good enough to
4692 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4693 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4694 arp_ip_target
[arp_ip_count
]);
4697 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4698 arp_target
[arp_ip_count
] = ip
;
4702 if (arp_interval
&& !arp_ip_count
) {
4703 /* don't allow arping if no arp_ip_target given... */
4704 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4710 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4711 pr_err("arp_validate only supported in active-backup mode\n");
4714 if (!arp_interval
) {
4715 pr_err("arp_validate requires arp_interval\n");
4719 arp_validate_value
= bond_parse_parm(arp_validate
,
4721 if (arp_validate_value
== -1) {
4722 pr_err("Error: invalid arp_validate \"%s\"\n",
4723 arp_validate
== NULL
? "NULL" : arp_validate
);
4727 arp_validate_value
= 0;
4730 pr_info("MII link monitoring set to %d ms\n", miimon
);
4731 } else if (arp_interval
) {
4734 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4736 arp_validate_tbl
[arp_validate_value
].modename
,
4739 for (i
= 0; i
< arp_ip_count
; i
++)
4740 pr_info(" %s", arp_ip_target
[i
]);
4744 } else if (max_bonds
) {
4745 /* miimon and arp_interval not set, we need one so things
4746 * work as expected, see bonding.txt for details
4748 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");
4751 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4752 /* currently, using a primary only makes sense
4753 * in active backup, TLB or ALB modes
4755 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4756 primary
, bond_mode_name(bond_mode
));
4760 if (primary
&& primary_reselect
) {
4761 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4763 if (primary_reselect_value
== -1) {
4764 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4766 NULL
? "NULL" : primary_reselect
);
4770 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4773 if (fail_over_mac
) {
4774 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4776 if (fail_over_mac_value
== -1) {
4777 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4778 arp_validate
== NULL
? "NULL" : arp_validate
);
4782 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4783 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4785 fail_over_mac_value
= BOND_FOM_NONE
;
4788 /* fill params struct with the proper values */
4789 params
->mode
= bond_mode
;
4790 params
->xmit_policy
= xmit_hashtype
;
4791 params
->miimon
= miimon
;
4792 params
->num_peer_notif
= num_peer_notif
;
4793 params
->arp_interval
= arp_interval
;
4794 params
->arp_validate
= arp_validate_value
;
4795 params
->updelay
= updelay
;
4796 params
->downdelay
= downdelay
;
4797 params
->use_carrier
= use_carrier
;
4798 params
->lacp_fast
= lacp_fast
;
4799 params
->primary
[0] = 0;
4800 params
->primary_reselect
= primary_reselect_value
;
4801 params
->fail_over_mac
= fail_over_mac_value
;
4802 params
->tx_queues
= tx_queues
;
4803 params
->all_slaves_active
= all_slaves_active
;
4804 params
->resend_igmp
= resend_igmp
;
4805 params
->min_links
= min_links
;
4808 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4809 params
->primary
[IFNAMSIZ
- 1] = 0;
4812 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4817 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4818 static struct lock_class_key bonding_netdev_addr_lock_key
;
4820 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4821 struct netdev_queue
*txq
,
4824 lockdep_set_class(&txq
->_xmit_lock
,
4825 &bonding_netdev_xmit_lock_key
);
4828 static void bond_set_lockdep_class(struct net_device
*dev
)
4830 lockdep_set_class(&dev
->addr_list_lock
,
4831 &bonding_netdev_addr_lock_key
);
4832 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4836 * Called from registration process
4838 static int bond_init(struct net_device
*bond_dev
)
4840 struct bonding
*bond
= netdev_priv(bond_dev
);
4841 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4842 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4844 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4847 * Initialize locks that may be required during
4848 * en/deslave operations. All of the bond_open work
4849 * (of which this is part) should really be moved to
4850 * a phase prior to dev_open
4852 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4853 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4855 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4859 bond_set_lockdep_class(bond_dev
);
4861 bond_create_proc_entry(bond
);
4862 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4864 bond_prepare_sysfs_group(bond
);
4866 bond_debug_register(bond
);
4868 __hw_addr_init(&bond
->mc_list
);
4872 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4874 if (tb
[IFLA_ADDRESS
]) {
4875 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4877 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4878 return -EADDRNOTAVAIL
;
4883 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
4885 .priv_size
= sizeof(struct bonding
),
4886 .setup
= bond_setup
,
4887 .validate
= bond_validate
,
4890 /* Create a new bond based on the specified name and bonding parameters.
4891 * If name is NULL, obtain a suitable "bond%d" name for us.
4892 * Caller must NOT hold rtnl_lock; we need to release it here before we
4893 * set up our sysfs entries.
4895 int bond_create(struct net
*net
, const char *name
)
4897 struct net_device
*bond_dev
;
4902 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4903 name
? name
: "bond%d",
4904 bond_setup
, tx_queues
);
4906 pr_err("%s: eek! can't alloc netdev!\n", name
);
4911 dev_net_set(bond_dev
, net
);
4912 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4914 res
= register_netdevice(bond_dev
);
4916 netif_carrier_off(bond_dev
);
4920 bond_destructor(bond_dev
);
4924 static int __net_init
bond_net_init(struct net
*net
)
4926 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4929 INIT_LIST_HEAD(&bn
->dev_list
);
4931 bond_create_proc_dir(bn
);
4936 static void __net_exit
bond_net_exit(struct net
*net
)
4938 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4940 bond_destroy_proc_dir(bn
);
4943 static struct pernet_operations bond_net_ops
= {
4944 .init
= bond_net_init
,
4945 .exit
= bond_net_exit
,
4947 .size
= sizeof(struct bond_net
),
4950 static int __init
bonding_init(void)
4955 pr_info("%s", bond_version
);
4957 res
= bond_check_params(&bonding_defaults
);
4961 res
= register_pernet_subsys(&bond_net_ops
);
4965 res
= rtnl_link_register(&bond_link_ops
);
4969 bond_create_debugfs();
4971 for (i
= 0; i
< max_bonds
; i
++) {
4972 res
= bond_create(&init_net
, NULL
);
4977 res
= bond_create_sysfs();
4981 register_netdevice_notifier(&bond_netdev_notifier
);
4982 register_inetaddr_notifier(&bond_inetaddr_notifier
);
4986 rtnl_link_unregister(&bond_link_ops
);
4988 unregister_pernet_subsys(&bond_net_ops
);
4993 static void __exit
bonding_exit(void)
4995 unregister_netdevice_notifier(&bond_netdev_notifier
);
4996 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
4998 bond_destroy_sysfs();
4999 bond_destroy_debugfs();
5001 rtnl_link_unregister(&bond_link_ops
);
5002 unregister_pernet_subsys(&bond_net_ops
);
5004 #ifdef CONFIG_NET_POLL_CONTROLLER
5006 * Make sure we don't have an imbalance on our netpoll blocking
5008 WARN_ON(atomic_read(&netpoll_block_tx
));
5012 module_init(bonding_init
);
5013 module_exit(bonding_exit
);
5014 MODULE_LICENSE("GPL");
5015 MODULE_VERSION(DRV_VERSION
);
5016 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5017 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5018 MODULE_ALIAS_RTNL_LINK("bond");