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 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
412 * We don't protect the slave list iteration with a lock because:
413 * a. This operation is performed in IOCTL context,
414 * b. The operation is protected by the RTNL semaphore in the 8021q code,
415 * c. Holding a lock with BH disabled while directly calling a base driver
416 * entry point is generally a BAD idea.
418 * The design of synchronization/protection for this operation in the 8021q
419 * module is good for one or more VLAN devices over a single physical device
420 * and cannot be extended for a teaming solution like bonding, so there is a
421 * potential race condition here where a net device from the vlan group might
422 * be referenced (either by a base driver or the 8021q code) while it is being
423 * removed from the system. However, it turns out we're not making matters
424 * worse, and if it works for regular VLAN usage it will work here too.
428 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
429 * @bond_dev: bonding net device that got called
430 * @vid: vlan id being added
432 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
434 struct bonding
*bond
= netdev_priv(bond_dev
);
438 bond_for_each_slave(bond
, slave
, i
) {
439 struct net_device
*slave_dev
= slave
->dev
;
440 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
442 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
443 slave_ops
->ndo_vlan_rx_add_vid
) {
444 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vid
);
448 res
= bond_add_vlan(bond
, vid
);
450 pr_err("%s: Error: Failed to add vlan id %d\n",
451 bond_dev
->name
, vid
);
456 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
457 * @bond_dev: bonding net device that got called
458 * @vid: vlan id being removed
460 static void bond_vlan_rx_kill_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_kill_vid
) {
472 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vid
);
476 res
= bond_del_vlan(bond
, vid
);
478 pr_err("%s: Error: Failed to remove vlan id %d\n",
479 bond_dev
->name
, vid
);
483 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
485 struct vlan_entry
*vlan
;
486 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
488 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
489 !(slave_ops
->ndo_vlan_rx_add_vid
))
492 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
)
493 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
496 static void bond_del_vlans_from_slave(struct bonding
*bond
,
497 struct net_device
*slave_dev
)
499 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
500 struct vlan_entry
*vlan
;
502 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
503 !(slave_ops
->ndo_vlan_rx_kill_vid
))
506 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
509 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
513 /*------------------------------- Link status -------------------------------*/
516 * Set the carrier state for the master according to the state of its
517 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
518 * do special 802.3ad magic.
520 * Returns zero if carrier state does not change, nonzero if it does.
522 static int bond_set_carrier(struct bonding
*bond
)
527 if (bond
->slave_cnt
== 0)
530 if (bond
->params
.mode
== BOND_MODE_8023AD
)
531 return bond_3ad_set_carrier(bond
);
533 bond_for_each_slave(bond
, slave
, i
) {
534 if (slave
->link
== BOND_LINK_UP
) {
535 if (!netif_carrier_ok(bond
->dev
)) {
536 netif_carrier_on(bond
->dev
);
544 if (netif_carrier_ok(bond
->dev
)) {
545 netif_carrier_off(bond
->dev
);
552 * Get link speed and duplex from the slave's base driver
553 * using ethtool. If for some reason the call fails or the
554 * values are invalid, fake speed and duplex to 100/Full
557 static int bond_update_speed_duplex(struct slave
*slave
)
559 struct net_device
*slave_dev
= slave
->dev
;
560 struct ethtool_cmd etool
= { .cmd
= ETHTOOL_GSET
};
564 /* Fake speed and duplex */
565 slave
->speed
= SPEED_100
;
566 slave
->duplex
= DUPLEX_FULL
;
568 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
571 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
575 slave_speed
= ethtool_cmd_speed(&etool
);
576 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
579 switch (etool
.duplex
) {
587 slave
->speed
= slave_speed
;
588 slave
->duplex
= etool
.duplex
;
594 * if <dev> supports MII link status reporting, check its link status.
596 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
597 * depending upon the setting of the use_carrier parameter.
599 * Return either BMSR_LSTATUS, meaning that the link is up (or we
600 * can't tell and just pretend it is), or 0, meaning that the link is
603 * If reporting is non-zero, instead of faking link up, return -1 if
604 * both ETHTOOL and MII ioctls fail (meaning the device does not
605 * support them). If use_carrier is set, return whatever it says.
606 * It'd be nice if there was a good way to tell if a driver supports
607 * netif_carrier, but there really isn't.
609 static int bond_check_dev_link(struct bonding
*bond
,
610 struct net_device
*slave_dev
, int reporting
)
612 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
613 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
615 struct mii_ioctl_data
*mii
;
617 if (!reporting
&& !netif_running(slave_dev
))
620 if (bond
->params
.use_carrier
)
621 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
623 /* Try to get link status using Ethtool first. */
624 if (slave_dev
->ethtool_ops
) {
625 if (slave_dev
->ethtool_ops
->get_link
) {
628 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
630 return link
? BMSR_LSTATUS
: 0;
634 /* Ethtool can't be used, fallback to MII ioctls. */
635 ioctl
= slave_ops
->ndo_do_ioctl
;
637 /* TODO: set pointer to correct ioctl on a per team member */
638 /* bases to make this more efficient. that is, once */
639 /* we determine the correct ioctl, we will always */
640 /* call it and not the others for that team */
644 * We cannot assume that SIOCGMIIPHY will also read a
645 * register; not all network drivers (e.g., e100)
649 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
650 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
652 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
653 mii
->reg_num
= MII_BMSR
;
654 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
655 return mii
->val_out
& BMSR_LSTATUS
;
660 * If reporting, report that either there's no dev->do_ioctl,
661 * or both SIOCGMIIREG and get_link failed (meaning that we
662 * cannot report link status). If not reporting, pretend
665 return reporting
? -1 : BMSR_LSTATUS
;
668 /*----------------------------- Multicast list ------------------------------*/
671 * Push the promiscuity flag down to appropriate slaves
673 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
676 if (USES_PRIMARY(bond
->params
.mode
)) {
677 /* write lock already acquired */
678 if (bond
->curr_active_slave
) {
679 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
685 bond_for_each_slave(bond
, slave
, i
) {
686 err
= dev_set_promiscuity(slave
->dev
, inc
);
695 * Push the allmulti flag down to all slaves
697 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
700 if (USES_PRIMARY(bond
->params
.mode
)) {
701 /* write lock already acquired */
702 if (bond
->curr_active_slave
) {
703 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
709 bond_for_each_slave(bond
, slave
, i
) {
710 err
= dev_set_allmulti(slave
->dev
, inc
);
719 * Add a Multicast address to slaves
722 static void bond_mc_add(struct bonding
*bond
, void *addr
)
724 if (USES_PRIMARY(bond
->params
.mode
)) {
725 /* write lock already acquired */
726 if (bond
->curr_active_slave
)
727 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
732 bond_for_each_slave(bond
, slave
, i
)
733 dev_mc_add(slave
->dev
, addr
);
738 * Remove a multicast address from slave
741 static void bond_mc_del(struct bonding
*bond
, void *addr
)
743 if (USES_PRIMARY(bond
->params
.mode
)) {
744 /* write lock already acquired */
745 if (bond
->curr_active_slave
)
746 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
750 bond_for_each_slave(bond
, slave
, i
) {
751 dev_mc_del(slave
->dev
, addr
);
757 static void __bond_resend_igmp_join_requests(struct net_device
*dev
)
759 struct in_device
*in_dev
;
762 in_dev
= __in_dev_get_rcu(dev
);
764 ip_mc_rejoin_groups(in_dev
);
769 * Retrieve the list of registered multicast addresses for the bonding
770 * device and retransmit an IGMP JOIN request to the current active
773 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
775 struct net_device
*vlan_dev
;
776 struct vlan_entry
*vlan
;
778 read_lock(&bond
->lock
);
780 /* rejoin all groups on bond device */
781 __bond_resend_igmp_join_requests(bond
->dev
);
783 /* rejoin all groups on vlan devices */
784 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
786 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
790 __bond_resend_igmp_join_requests(vlan_dev
);
793 if (--bond
->igmp_retrans
> 0)
794 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
796 read_unlock(&bond
->lock
);
799 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
801 struct bonding
*bond
= container_of(work
, struct bonding
,
803 bond_resend_igmp_join_requests(bond
);
807 * flush all members of flush->mc_list from device dev->mc_list
809 static void bond_mc_list_flush(struct net_device
*bond_dev
,
810 struct net_device
*slave_dev
)
812 struct bonding
*bond
= netdev_priv(bond_dev
);
813 struct netdev_hw_addr
*ha
;
815 netdev_for_each_mc_addr(ha
, bond_dev
)
816 dev_mc_del(slave_dev
, ha
->addr
);
818 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
819 /* del lacpdu mc addr from mc list */
820 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
822 dev_mc_del(slave_dev
, lacpdu_multicast
);
826 /*--------------------------- Active slave change ---------------------------*/
829 * Update the mc list and multicast-related flags for the new and
830 * old active slaves (if any) according to the multicast mode, and
831 * promiscuous flags unconditionally.
833 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
834 struct slave
*old_active
)
836 struct netdev_hw_addr
*ha
;
838 if (!USES_PRIMARY(bond
->params
.mode
))
839 /* nothing to do - mc list is already up-to-date on
845 if (bond
->dev
->flags
& IFF_PROMISC
)
846 dev_set_promiscuity(old_active
->dev
, -1);
848 if (bond
->dev
->flags
& IFF_ALLMULTI
)
849 dev_set_allmulti(old_active
->dev
, -1);
851 netdev_for_each_mc_addr(ha
, bond
->dev
)
852 dev_mc_del(old_active
->dev
, ha
->addr
);
856 /* FIXME: Signal errors upstream. */
857 if (bond
->dev
->flags
& IFF_PROMISC
)
858 dev_set_promiscuity(new_active
->dev
, 1);
860 if (bond
->dev
->flags
& IFF_ALLMULTI
)
861 dev_set_allmulti(new_active
->dev
, 1);
863 netdev_for_each_mc_addr(ha
, bond
->dev
)
864 dev_mc_add(new_active
->dev
, ha
->addr
);
869 * bond_do_fail_over_mac
871 * Perform special MAC address swapping for fail_over_mac settings
873 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
875 static void bond_do_fail_over_mac(struct bonding
*bond
,
876 struct slave
*new_active
,
877 struct slave
*old_active
)
878 __releases(&bond
->curr_slave_lock
)
879 __releases(&bond
->lock
)
880 __acquires(&bond
->lock
)
881 __acquires(&bond
->curr_slave_lock
)
883 u8 tmp_mac
[ETH_ALEN
];
884 struct sockaddr saddr
;
887 switch (bond
->params
.fail_over_mac
) {
888 case BOND_FOM_ACTIVE
:
890 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
891 new_active
->dev
->addr_len
);
893 case BOND_FOM_FOLLOW
:
895 * if new_active && old_active, swap them
896 * if just old_active, do nothing (going to no active slave)
897 * if just new_active, set new_active to bond's MAC
902 write_unlock_bh(&bond
->curr_slave_lock
);
903 read_unlock(&bond
->lock
);
906 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
907 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
909 saddr
.sa_family
= new_active
->dev
->type
;
911 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
912 saddr
.sa_family
= bond
->dev
->type
;
915 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
917 pr_err("%s: Error %d setting MAC of slave %s\n",
918 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
925 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
926 saddr
.sa_family
= old_active
->dev
->type
;
928 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
930 pr_err("%s: Error %d setting MAC of slave %s\n",
931 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
933 read_lock(&bond
->lock
);
934 write_lock_bh(&bond
->curr_slave_lock
);
937 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
938 bond
->dev
->name
, bond
->params
.fail_over_mac
);
944 static bool bond_should_change_active(struct bonding
*bond
)
946 struct slave
*prim
= bond
->primary_slave
;
947 struct slave
*curr
= bond
->curr_active_slave
;
949 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
951 if (bond
->force_primary
) {
952 bond
->force_primary
= false;
955 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
956 (prim
->speed
< curr
->speed
||
957 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
959 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
965 * find_best_interface - select the best available slave to be the active one
966 * @bond: our bonding struct
968 * Warning: Caller must hold curr_slave_lock for writing.
970 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
972 struct slave
*new_active
, *old_active
;
973 struct slave
*bestslave
= NULL
;
974 int mintime
= bond
->params
.updelay
;
977 new_active
= bond
->curr_active_slave
;
979 if (!new_active
) { /* there were no active slaves left */
980 if (bond
->slave_cnt
> 0) /* found one slave */
981 new_active
= bond
->first_slave
;
983 return NULL
; /* still no slave, return NULL */
986 if ((bond
->primary_slave
) &&
987 bond
->primary_slave
->link
== BOND_LINK_UP
&&
988 bond_should_change_active(bond
)) {
989 new_active
= bond
->primary_slave
;
992 /* remember where to stop iterating over the slaves */
993 old_active
= new_active
;
995 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
996 if (new_active
->link
== BOND_LINK_UP
) {
998 } else if (new_active
->link
== BOND_LINK_BACK
&&
999 IS_UP(new_active
->dev
)) {
1000 /* link up, but waiting for stabilization */
1001 if (new_active
->delay
< mintime
) {
1002 mintime
= new_active
->delay
;
1003 bestslave
= new_active
;
1011 static bool bond_should_notify_peers(struct bonding
*bond
)
1013 struct slave
*slave
= bond
->curr_active_slave
;
1015 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1016 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
1018 if (!slave
|| !bond
->send_peer_notif
||
1019 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
1022 bond
->send_peer_notif
--;
1027 * change_active_interface - change the active slave into the specified one
1028 * @bond: our bonding struct
1029 * @new: the new slave to make the active one
1031 * Set the new slave to the bond's settings and unset them on the old
1032 * curr_active_slave.
1033 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1035 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1036 * because it is apparently the best available slave we have, even though its
1037 * updelay hasn't timed out yet.
1039 * If new_active is not NULL, caller must hold bond->lock for read and
1040 * curr_slave_lock for write_bh.
1042 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1044 struct slave
*old_active
= bond
->curr_active_slave
;
1046 if (old_active
== new_active
)
1050 new_active
->jiffies
= jiffies
;
1052 if (new_active
->link
== BOND_LINK_BACK
) {
1053 if (USES_PRIMARY(bond
->params
.mode
)) {
1054 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1055 bond
->dev
->name
, new_active
->dev
->name
,
1056 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1059 new_active
->delay
= 0;
1060 new_active
->link
= BOND_LINK_UP
;
1062 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1063 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1065 if (bond_is_lb(bond
))
1066 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1068 if (USES_PRIMARY(bond
->params
.mode
)) {
1069 pr_info("%s: making interface %s the new active one.\n",
1070 bond
->dev
->name
, new_active
->dev
->name
);
1075 if (USES_PRIMARY(bond
->params
.mode
))
1076 bond_mc_swap(bond
, new_active
, old_active
);
1078 if (bond_is_lb(bond
)) {
1079 bond_alb_handle_active_change(bond
, new_active
);
1081 bond_set_slave_inactive_flags(old_active
);
1083 bond_set_slave_active_flags(new_active
);
1085 bond
->curr_active_slave
= new_active
;
1088 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1090 bond_set_slave_inactive_flags(old_active
);
1093 bool should_notify_peers
= false;
1095 bond_set_slave_active_flags(new_active
);
1097 if (bond
->params
.fail_over_mac
)
1098 bond_do_fail_over_mac(bond
, new_active
,
1101 if (netif_running(bond
->dev
)) {
1102 bond
->send_peer_notif
=
1103 bond
->params
.num_peer_notif
;
1104 should_notify_peers
=
1105 bond_should_notify_peers(bond
);
1108 write_unlock_bh(&bond
->curr_slave_lock
);
1109 read_unlock(&bond
->lock
);
1111 netdev_bonding_change(bond
->dev
, NETDEV_BONDING_FAILOVER
);
1112 if (should_notify_peers
)
1113 netdev_bonding_change(bond
->dev
,
1114 NETDEV_NOTIFY_PEERS
);
1116 read_lock(&bond
->lock
);
1117 write_lock_bh(&bond
->curr_slave_lock
);
1121 /* resend IGMP joins since active slave has changed or
1122 * all were sent on curr_active_slave.
1123 * resend only if bond is brought up with the affected
1124 * bonding modes and the retransmission is enabled */
1125 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
1126 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
1127 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
1128 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
1129 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 0);
1134 * bond_select_active_slave - select a new active slave, if needed
1135 * @bond: our bonding struct
1137 * This functions should be called when one of the following occurs:
1138 * - The old curr_active_slave has been released or lost its link.
1139 * - The primary_slave has got its link back.
1140 * - A slave has got its link back and there's no old curr_active_slave.
1142 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1144 void bond_select_active_slave(struct bonding
*bond
)
1146 struct slave
*best_slave
;
1149 best_slave
= bond_find_best_slave(bond
);
1150 if (best_slave
!= bond
->curr_active_slave
) {
1151 bond_change_active_slave(bond
, best_slave
);
1152 rv
= bond_set_carrier(bond
);
1156 if (netif_carrier_ok(bond
->dev
)) {
1157 pr_info("%s: first active interface up!\n",
1160 pr_info("%s: now running without any active interface !\n",
1166 /*--------------------------- slave list handling ---------------------------*/
1169 * This function attaches the slave to the end of list.
1171 * bond->lock held for writing by caller.
1173 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1175 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1176 new_slave
->next
= new_slave
;
1177 new_slave
->prev
= new_slave
;
1178 bond
->first_slave
= new_slave
;
1180 new_slave
->next
= bond
->first_slave
;
1181 new_slave
->prev
= bond
->first_slave
->prev
;
1182 new_slave
->next
->prev
= new_slave
;
1183 new_slave
->prev
->next
= new_slave
;
1190 * This function detaches the slave from the list.
1191 * WARNING: no check is made to verify if the slave effectively
1192 * belongs to <bond>.
1193 * Nothing is freed on return, structures are just unchained.
1194 * If any slave pointer in bond was pointing to <slave>,
1195 * it should be changed by the calling function.
1197 * bond->lock held for writing by caller.
1199 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1202 slave
->next
->prev
= slave
->prev
;
1205 slave
->prev
->next
= slave
->next
;
1207 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1208 if (bond
->slave_cnt
> 1) { /* there are more slave */
1209 bond
->first_slave
= slave
->next
;
1211 bond
->first_slave
= NULL
; /* slave was the last one */
1220 #ifdef CONFIG_NET_POLL_CONTROLLER
1221 static inline int slave_enable_netpoll(struct slave
*slave
)
1226 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
1231 np
->dev
= slave
->dev
;
1232 strlcpy(np
->dev_name
, slave
->dev
->name
, IFNAMSIZ
);
1233 err
= __netpoll_setup(np
);
1242 static inline void slave_disable_netpoll(struct slave
*slave
)
1244 struct netpoll
*np
= slave
->np
;
1250 synchronize_rcu_bh();
1251 __netpoll_cleanup(np
);
1254 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1256 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1258 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1263 static void bond_poll_controller(struct net_device
*bond_dev
)
1267 static void __bond_netpoll_cleanup(struct bonding
*bond
)
1269 struct slave
*slave
;
1272 bond_for_each_slave(bond
, slave
, i
)
1273 if (IS_UP(slave
->dev
))
1274 slave_disable_netpoll(slave
);
1276 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1278 struct bonding
*bond
= netdev_priv(bond_dev
);
1280 read_lock(&bond
->lock
);
1281 __bond_netpoll_cleanup(bond
);
1282 read_unlock(&bond
->lock
);
1285 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
1287 struct bonding
*bond
= netdev_priv(dev
);
1288 struct slave
*slave
;
1291 read_lock(&bond
->lock
);
1292 bond_for_each_slave(bond
, slave
, i
) {
1293 err
= slave_enable_netpoll(slave
);
1295 __bond_netpoll_cleanup(bond
);
1299 read_unlock(&bond
->lock
);
1303 static struct netpoll_info
*bond_netpoll_info(struct bonding
*bond
)
1305 return bond
->dev
->npinfo
;
1309 static inline int slave_enable_netpoll(struct slave
*slave
)
1313 static inline void slave_disable_netpoll(struct slave
*slave
)
1316 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1321 /*---------------------------------- IOCTL ----------------------------------*/
1323 static int bond_sethwaddr(struct net_device
*bond_dev
,
1324 struct net_device
*slave_dev
)
1326 pr_debug("bond_dev=%p\n", bond_dev
);
1327 pr_debug("slave_dev=%p\n", slave_dev
);
1328 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1329 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1333 static u32
bond_fix_features(struct net_device
*dev
, u32 features
)
1335 struct slave
*slave
;
1336 struct bonding
*bond
= netdev_priv(dev
);
1340 read_lock(&bond
->lock
);
1342 if (!bond
->first_slave
) {
1343 /* Disable adding VLANs to empty bond. But why? --mq */
1344 features
|= NETIF_F_VLAN_CHALLENGED
;
1349 features
&= ~NETIF_F_ONE_FOR_ALL
;
1350 features
|= NETIF_F_ALL_FOR_ALL
;
1352 bond_for_each_slave(bond
, slave
, i
) {
1353 features
= netdev_increment_features(features
,
1354 slave
->dev
->features
,
1359 read_unlock(&bond
->lock
);
1363 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1364 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1365 NETIF_F_HIGHDMA | NETIF_F_LRO)
1367 static void bond_compute_features(struct bonding
*bond
)
1369 struct slave
*slave
;
1370 struct net_device
*bond_dev
= bond
->dev
;
1371 u32 vlan_features
= BOND_VLAN_FEATURES
;
1372 unsigned short max_hard_header_len
= ETH_HLEN
;
1375 read_lock(&bond
->lock
);
1377 if (!bond
->first_slave
)
1380 bond_for_each_slave(bond
, slave
, i
) {
1381 vlan_features
= netdev_increment_features(vlan_features
,
1382 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1384 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1385 max_hard_header_len
= slave
->dev
->hard_header_len
;
1389 bond_dev
->vlan_features
= vlan_features
;
1390 bond_dev
->hard_header_len
= max_hard_header_len
;
1392 read_unlock(&bond
->lock
);
1394 netdev_change_features(bond_dev
);
1397 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1398 struct net_device
*slave_dev
)
1400 struct bonding
*bond
= netdev_priv(bond_dev
);
1402 bond_dev
->header_ops
= slave_dev
->header_ops
;
1404 bond_dev
->type
= slave_dev
->type
;
1405 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1406 bond_dev
->addr_len
= slave_dev
->addr_len
;
1408 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1409 slave_dev
->addr_len
);
1410 bond
->setup_by_slave
= 1;
1413 /* On bonding slaves other than the currently active slave, suppress
1414 * duplicates except for alb non-mcast/bcast.
1416 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1417 struct slave
*slave
,
1418 struct bonding
*bond
)
1420 if (bond_is_slave_inactive(slave
)) {
1421 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1422 skb
->pkt_type
!= PACKET_BROADCAST
&&
1423 skb
->pkt_type
!= PACKET_MULTICAST
)
1430 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1432 struct sk_buff
*skb
= *pskb
;
1433 struct slave
*slave
;
1434 struct bonding
*bond
;
1436 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1438 return RX_HANDLER_CONSUMED
;
1442 slave
= bond_slave_get_rcu(skb
->dev
);
1445 if (bond
->params
.arp_interval
)
1446 slave
->dev
->last_rx
= jiffies
;
1448 if (bond
->recv_probe
) {
1449 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
1452 bond
->recv_probe(nskb
, bond
, slave
);
1453 dev_kfree_skb(nskb
);
1457 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1458 return RX_HANDLER_EXACT
;
1461 skb
->dev
= bond
->dev
;
1463 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1464 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1465 skb
->pkt_type
== PACKET_HOST
) {
1467 if (unlikely(skb_cow_head(skb
,
1468 skb
->data
- skb_mac_header(skb
)))) {
1470 return RX_HANDLER_CONSUMED
;
1472 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1475 return RX_HANDLER_ANOTHER
;
1478 /* enslave device <slave> to bond device <master> */
1479 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1481 struct bonding
*bond
= netdev_priv(bond_dev
);
1482 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1483 struct slave
*new_slave
= NULL
;
1484 struct netdev_hw_addr
*ha
;
1485 struct sockaddr addr
;
1489 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1490 slave_ops
->ndo_do_ioctl
== NULL
) {
1491 pr_warning("%s: Warning: no link monitoring support for %s\n",
1492 bond_dev
->name
, slave_dev
->name
);
1495 /* already enslaved */
1496 if (slave_dev
->flags
& IFF_SLAVE
) {
1497 pr_debug("Error, Device was already enslaved\n");
1501 /* vlan challenged mutual exclusion */
1502 /* no need to lock since we're protected by rtnl_lock */
1503 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1504 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1505 if (bond_vlan_used(bond
)) {
1506 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1507 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1510 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1511 bond_dev
->name
, slave_dev
->name
,
1512 slave_dev
->name
, bond_dev
->name
);
1515 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1519 * Old ifenslave binaries are no longer supported. These can
1520 * be identified with moderate accuracy by the state of the slave:
1521 * the current ifenslave will set the interface down prior to
1522 * enslaving it; the old ifenslave will not.
1524 if ((slave_dev
->flags
& IFF_UP
)) {
1525 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1528 goto err_undo_flags
;
1531 /* set bonding device ether type by slave - bonding netdevices are
1532 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1533 * there is a need to override some of the type dependent attribs/funcs.
1535 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1536 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1538 if (bond
->slave_cnt
== 0) {
1539 if (bond_dev
->type
!= slave_dev
->type
) {
1540 pr_debug("%s: change device type from %d to %d\n",
1542 bond_dev
->type
, slave_dev
->type
);
1544 res
= netdev_bonding_change(bond_dev
,
1545 NETDEV_PRE_TYPE_CHANGE
);
1546 res
= notifier_to_errno(res
);
1548 pr_err("%s: refused to change device type\n",
1551 goto err_undo_flags
;
1554 /* Flush unicast and multicast addresses */
1555 dev_uc_flush(bond_dev
);
1556 dev_mc_flush(bond_dev
);
1558 if (slave_dev
->type
!= ARPHRD_ETHER
)
1559 bond_setup_by_slave(bond_dev
, slave_dev
);
1561 ether_setup(bond_dev
);
1563 netdev_bonding_change(bond_dev
,
1564 NETDEV_POST_TYPE_CHANGE
);
1566 } else if (bond_dev
->type
!= slave_dev
->type
) {
1567 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1569 slave_dev
->type
, bond_dev
->type
);
1571 goto err_undo_flags
;
1574 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1575 if (bond
->slave_cnt
== 0) {
1576 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1578 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1579 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1580 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",
1583 goto err_undo_flags
;
1587 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1589 /* If this is the first slave, then we need to set the master's hardware
1590 * address to be the same as the slave's. */
1591 if (is_zero_ether_addr(bond
->dev
->dev_addr
))
1592 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1593 slave_dev
->addr_len
);
1596 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1599 goto err_undo_flags
;
1603 * Set the new_slave's queue_id to be zero. Queue ID mapping
1604 * is set via sysfs or module option if desired.
1606 new_slave
->queue_id
= 0;
1608 /* Save slave's original mtu and then set it to match the bond */
1609 new_slave
->original_mtu
= slave_dev
->mtu
;
1610 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1612 pr_debug("Error %d calling dev_set_mtu\n", res
);
1617 * Save slave's original ("permanent") mac address for modes
1618 * that need it, and for restoring it upon release, and then
1619 * set it to the master's address
1621 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1623 if (!bond
->params
.fail_over_mac
) {
1625 * Set slave to master's mac address. The application already
1626 * set the master's mac address to that of the first slave
1628 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1629 addr
.sa_family
= slave_dev
->type
;
1630 res
= dev_set_mac_address(slave_dev
, &addr
);
1632 pr_debug("Error %d calling set_mac_address\n", res
);
1633 goto err_restore_mtu
;
1637 res
= netdev_set_bond_master(slave_dev
, bond_dev
);
1639 pr_debug("Error %d calling netdev_set_bond_master\n", res
);
1640 goto err_restore_mac
;
1643 /* open the slave since the application closed it */
1644 res
= dev_open(slave_dev
);
1646 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1647 goto err_unset_master
;
1650 new_slave
->bond
= bond
;
1651 new_slave
->dev
= slave_dev
;
1652 slave_dev
->priv_flags
|= IFF_BONDING
;
1654 if (bond_is_lb(bond
)) {
1655 /* bond_alb_init_slave() must be called before all other stages since
1656 * it might fail and we do not want to have to undo everything
1658 res
= bond_alb_init_slave(bond
, new_slave
);
1663 /* If the mode USES_PRIMARY, then the new slave gets the
1664 * master's promisc (and mc) settings only if it becomes the
1665 * curr_active_slave, and that is taken care of later when calling
1666 * bond_change_active()
1668 if (!USES_PRIMARY(bond
->params
.mode
)) {
1669 /* set promiscuity level to new slave */
1670 if (bond_dev
->flags
& IFF_PROMISC
) {
1671 res
= dev_set_promiscuity(slave_dev
, 1);
1676 /* set allmulti level to new slave */
1677 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1678 res
= dev_set_allmulti(slave_dev
, 1);
1683 netif_addr_lock_bh(bond_dev
);
1684 /* upload master's mc_list to new slave */
1685 netdev_for_each_mc_addr(ha
, bond_dev
)
1686 dev_mc_add(slave_dev
, ha
->addr
);
1687 netif_addr_unlock_bh(bond_dev
);
1690 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1691 /* add lacpdu mc addr to mc list */
1692 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1694 dev_mc_add(slave_dev
, lacpdu_multicast
);
1697 bond_add_vlans_on_slave(bond
, slave_dev
);
1699 write_lock_bh(&bond
->lock
);
1701 bond_attach_slave(bond
, new_slave
);
1703 new_slave
->delay
= 0;
1704 new_slave
->link_failure_count
= 0;
1706 write_unlock_bh(&bond
->lock
);
1708 bond_compute_features(bond
);
1710 read_lock(&bond
->lock
);
1712 new_slave
->last_arp_rx
= jiffies
;
1714 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1715 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1717 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1719 * miimon is set but a bonded network driver
1720 * does not support ETHTOOL/MII and
1721 * arp_interval is not set. Note: if
1722 * use_carrier is enabled, we will never go
1723 * here (because netif_carrier is always
1724 * supported); thus, we don't need to change
1725 * the messages for netif_carrier.
1727 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",
1728 bond_dev
->name
, slave_dev
->name
);
1729 } else if (link_reporting
== -1) {
1730 /* unable get link status using mii/ethtool */
1731 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",
1732 bond_dev
->name
, slave_dev
->name
);
1736 /* check for initial state */
1737 if (!bond
->params
.miimon
||
1738 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1739 if (bond
->params
.updelay
) {
1740 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1741 new_slave
->link
= BOND_LINK_BACK
;
1742 new_slave
->delay
= bond
->params
.updelay
;
1744 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1745 new_slave
->link
= BOND_LINK_UP
;
1747 new_slave
->jiffies
= jiffies
;
1749 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1750 new_slave
->link
= BOND_LINK_DOWN
;
1753 if (bond_update_speed_duplex(new_slave
) &&
1754 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1755 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1756 bond_dev
->name
, new_slave
->dev
->name
);
1758 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1759 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1764 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1765 /* if there is a primary slave, remember it */
1766 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1767 bond
->primary_slave
= new_slave
;
1768 bond
->force_primary
= true;
1772 write_lock_bh(&bond
->curr_slave_lock
);
1774 switch (bond
->params
.mode
) {
1775 case BOND_MODE_ACTIVEBACKUP
:
1776 bond_set_slave_inactive_flags(new_slave
);
1777 bond_select_active_slave(bond
);
1779 case BOND_MODE_8023AD
:
1780 /* in 802.3ad mode, the internal mechanism
1781 * will activate the slaves in the selected
1784 bond_set_slave_inactive_flags(new_slave
);
1785 /* if this is the first slave */
1786 if (bond
->slave_cnt
== 1) {
1787 SLAVE_AD_INFO(new_slave
).id
= 1;
1788 /* Initialize AD with the number of times that the AD timer is called in 1 second
1789 * can be called only after the mac address of the bond is set
1791 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1793 SLAVE_AD_INFO(new_slave
).id
=
1794 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1797 bond_3ad_bind_slave(new_slave
);
1801 bond_set_active_slave(new_slave
);
1802 bond_set_slave_inactive_flags(new_slave
);
1803 bond_select_active_slave(bond
);
1806 pr_debug("This slave is always active in trunk mode\n");
1808 /* always active in trunk mode */
1809 bond_set_active_slave(new_slave
);
1811 /* In trunking mode there is little meaning to curr_active_slave
1812 * anyway (it holds no special properties of the bond device),
1813 * so we can change it without calling change_active_interface()
1815 if (!bond
->curr_active_slave
)
1816 bond
->curr_active_slave
= new_slave
;
1819 } /* switch(bond_mode) */
1821 write_unlock_bh(&bond
->curr_slave_lock
);
1823 bond_set_carrier(bond
);
1825 #ifdef CONFIG_NET_POLL_CONTROLLER
1826 slave_dev
->npinfo
= bond_netpoll_info(bond
);
1827 if (slave_dev
->npinfo
) {
1828 if (slave_enable_netpoll(new_slave
)) {
1829 read_unlock(&bond
->lock
);
1830 pr_info("Error, %s: master_dev is using netpoll, "
1831 "but new slave device does not support netpoll.\n",
1839 read_unlock(&bond
->lock
);
1841 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1845 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1848 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1849 goto err_dest_symlinks
;
1852 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1853 bond_dev
->name
, slave_dev
->name
,
1854 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1855 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1857 /* enslave is successful */
1860 /* Undo stages on error */
1862 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1865 dev_close(slave_dev
);
1868 netdev_set_bond_master(slave_dev
, NULL
);
1871 if (!bond
->params
.fail_over_mac
) {
1872 /* XXX TODO - fom follow mode needs to change master's
1873 * MAC if this slave's MAC is in use by the bond, or at
1874 * least print a warning.
1876 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1877 addr
.sa_family
= slave_dev
->type
;
1878 dev_set_mac_address(slave_dev
, &addr
);
1882 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1888 bond_compute_features(bond
);
1894 * Try to release the slave device <slave> from the bond device <master>
1895 * It is legal to access curr_active_slave without a lock because all the function
1898 * The rules for slave state should be:
1899 * for Active/Backup:
1900 * Active stays on all backups go down
1901 * for Bonded connections:
1902 * The first up interface should be left on and all others downed.
1904 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1906 struct bonding
*bond
= netdev_priv(bond_dev
);
1907 struct slave
*slave
, *oldcurrent
;
1908 struct sockaddr addr
;
1909 u32 old_features
= bond_dev
->features
;
1911 /* slave is not a slave or master is not master of this slave */
1912 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1913 (slave_dev
->master
!= bond_dev
)) {
1914 pr_err("%s: Error: cannot release %s.\n",
1915 bond_dev
->name
, slave_dev
->name
);
1920 netdev_bonding_change(bond_dev
, NETDEV_RELEASE
);
1921 write_lock_bh(&bond
->lock
);
1923 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1925 /* not a slave of this bond */
1926 pr_info("%s: %s not enslaved\n",
1927 bond_dev
->name
, slave_dev
->name
);
1928 write_unlock_bh(&bond
->lock
);
1929 unblock_netpoll_tx();
1933 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1934 * for this slave anymore.
1936 netdev_rx_handler_unregister(slave_dev
);
1937 write_unlock_bh(&bond
->lock
);
1939 write_lock_bh(&bond
->lock
);
1941 if (!bond
->params
.fail_over_mac
) {
1942 if (!compare_ether_addr(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1943 bond
->slave_cnt
> 1)
1944 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",
1945 bond_dev
->name
, slave_dev
->name
,
1947 bond_dev
->name
, slave_dev
->name
);
1950 /* Inform AD package of unbinding of slave. */
1951 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1952 /* must be called before the slave is
1953 * detached from the list
1955 bond_3ad_unbind_slave(slave
);
1958 pr_info("%s: releasing %s interface %s\n",
1960 bond_is_active_slave(slave
) ? "active" : "backup",
1963 oldcurrent
= bond
->curr_active_slave
;
1965 bond
->current_arp_slave
= NULL
;
1967 /* release the slave from its bond */
1968 bond_detach_slave(bond
, slave
);
1970 if (bond
->primary_slave
== slave
)
1971 bond
->primary_slave
= NULL
;
1973 if (oldcurrent
== slave
)
1974 bond_change_active_slave(bond
, NULL
);
1976 if (bond_is_lb(bond
)) {
1977 /* Must be called only after the slave has been
1978 * detached from the list and the curr_active_slave
1979 * has been cleared (if our_slave == old_current),
1980 * but before a new active slave is selected.
1982 write_unlock_bh(&bond
->lock
);
1983 bond_alb_deinit_slave(bond
, slave
);
1984 write_lock_bh(&bond
->lock
);
1987 if (oldcurrent
== slave
) {
1989 * Note that we hold RTNL over this sequence, so there
1990 * is no concern that another slave add/remove event
1993 write_unlock_bh(&bond
->lock
);
1994 read_lock(&bond
->lock
);
1995 write_lock_bh(&bond
->curr_slave_lock
);
1997 bond_select_active_slave(bond
);
1999 write_unlock_bh(&bond
->curr_slave_lock
);
2000 read_unlock(&bond
->lock
);
2001 write_lock_bh(&bond
->lock
);
2004 if (bond
->slave_cnt
== 0) {
2005 bond_set_carrier(bond
);
2007 /* if the last slave was removed, zero the mac address
2008 * of the master so it will be set by the application
2009 * to the mac address of the first slave
2011 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2013 if (bond_vlan_used(bond
)) {
2014 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2015 bond_dev
->name
, bond_dev
->name
);
2016 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2021 write_unlock_bh(&bond
->lock
);
2022 unblock_netpoll_tx();
2024 bond_compute_features(bond
);
2025 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
2026 (old_features
& NETIF_F_VLAN_CHALLENGED
))
2027 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2028 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
2030 /* must do this from outside any spinlocks */
2031 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2033 bond_del_vlans_from_slave(bond
, slave_dev
);
2035 /* If the mode USES_PRIMARY, then we should only remove its
2036 * promisc and mc settings if it was the curr_active_slave, but that was
2037 * already taken care of above when we detached the slave
2039 if (!USES_PRIMARY(bond
->params
.mode
)) {
2040 /* unset promiscuity level from slave */
2041 if (bond_dev
->flags
& IFF_PROMISC
)
2042 dev_set_promiscuity(slave_dev
, -1);
2044 /* unset allmulti level from slave */
2045 if (bond_dev
->flags
& IFF_ALLMULTI
)
2046 dev_set_allmulti(slave_dev
, -1);
2048 /* flush master's mc_list from slave */
2049 netif_addr_lock_bh(bond_dev
);
2050 bond_mc_list_flush(bond_dev
, slave_dev
);
2051 netif_addr_unlock_bh(bond_dev
);
2054 netdev_set_bond_master(slave_dev
, NULL
);
2056 slave_disable_netpoll(slave
);
2058 /* close slave before restoring its mac address */
2059 dev_close(slave_dev
);
2061 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2062 /* restore original ("permanent") mac address */
2063 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2064 addr
.sa_family
= slave_dev
->type
;
2065 dev_set_mac_address(slave_dev
, &addr
);
2068 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2070 slave_dev
->priv_flags
&= ~IFF_BONDING
;
2074 return 0; /* deletion OK */
2078 * First release a slave and then destroy the bond if no more slaves are left.
2079 * Must be under rtnl_lock when this function is called.
2081 static int bond_release_and_destroy(struct net_device
*bond_dev
,
2082 struct net_device
*slave_dev
)
2084 struct bonding
*bond
= netdev_priv(bond_dev
);
2087 ret
= bond_release(bond_dev
, slave_dev
);
2088 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2089 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
2090 pr_info("%s: destroying bond %s.\n",
2091 bond_dev
->name
, bond_dev
->name
);
2092 unregister_netdevice(bond_dev
);
2098 * This function releases all slaves.
2100 static int bond_release_all(struct net_device
*bond_dev
)
2102 struct bonding
*bond
= netdev_priv(bond_dev
);
2103 struct slave
*slave
;
2104 struct net_device
*slave_dev
;
2105 struct sockaddr addr
;
2107 write_lock_bh(&bond
->lock
);
2109 netif_carrier_off(bond_dev
);
2111 if (bond
->slave_cnt
== 0)
2114 bond
->current_arp_slave
= NULL
;
2115 bond
->primary_slave
= NULL
;
2116 bond_change_active_slave(bond
, NULL
);
2118 while ((slave
= bond
->first_slave
) != NULL
) {
2119 /* Inform AD package of unbinding of slave
2120 * before slave is detached from the list.
2122 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2123 bond_3ad_unbind_slave(slave
);
2125 slave_dev
= slave
->dev
;
2126 bond_detach_slave(bond
, slave
);
2128 /* now that the slave is detached, unlock and perform
2129 * all the undo steps that should not be called from
2132 write_unlock_bh(&bond
->lock
);
2134 /* unregister rx_handler early so bond_handle_frame wouldn't
2135 * be called for this slave anymore.
2137 netdev_rx_handler_unregister(slave_dev
);
2140 if (bond_is_lb(bond
)) {
2141 /* must be called only after the slave
2142 * has been detached from the list
2144 bond_alb_deinit_slave(bond
, slave
);
2147 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2148 bond_del_vlans_from_slave(bond
, slave_dev
);
2150 /* If the mode USES_PRIMARY, then we should only remove its
2151 * promisc and mc settings if it was the curr_active_slave, but that was
2152 * already taken care of above when we detached the slave
2154 if (!USES_PRIMARY(bond
->params
.mode
)) {
2155 /* unset promiscuity level from slave */
2156 if (bond_dev
->flags
& IFF_PROMISC
)
2157 dev_set_promiscuity(slave_dev
, -1);
2159 /* unset allmulti level from slave */
2160 if (bond_dev
->flags
& IFF_ALLMULTI
)
2161 dev_set_allmulti(slave_dev
, -1);
2163 /* flush master's mc_list from slave */
2164 netif_addr_lock_bh(bond_dev
);
2165 bond_mc_list_flush(bond_dev
, slave_dev
);
2166 netif_addr_unlock_bh(bond_dev
);
2169 netdev_set_bond_master(slave_dev
, NULL
);
2171 slave_disable_netpoll(slave
);
2173 /* close slave before restoring its mac address */
2174 dev_close(slave_dev
);
2176 if (!bond
->params
.fail_over_mac
) {
2177 /* restore original ("permanent") mac address*/
2178 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2179 addr
.sa_family
= slave_dev
->type
;
2180 dev_set_mac_address(slave_dev
, &addr
);
2185 /* re-acquire the lock before getting the next slave */
2186 write_lock_bh(&bond
->lock
);
2189 /* zero the mac address of the master so it will be
2190 * set by the application to the mac address of the
2193 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2195 if (bond_vlan_used(bond
)) {
2196 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2197 bond_dev
->name
, bond_dev
->name
);
2198 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2202 pr_info("%s: released all slaves\n", bond_dev
->name
);
2205 write_unlock_bh(&bond
->lock
);
2207 bond_compute_features(bond
);
2213 * This function changes the active slave to slave <slave_dev>.
2214 * It returns -EINVAL in the following cases.
2215 * - <slave_dev> is not found in the list.
2216 * - There is not active slave now.
2217 * - <slave_dev> is already active.
2218 * - The link state of <slave_dev> is not BOND_LINK_UP.
2219 * - <slave_dev> is not running.
2220 * In these cases, this function does nothing.
2221 * In the other cases, current_slave pointer is changed and 0 is returned.
2223 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2225 struct bonding
*bond
= netdev_priv(bond_dev
);
2226 struct slave
*old_active
= NULL
;
2227 struct slave
*new_active
= NULL
;
2230 if (!USES_PRIMARY(bond
->params
.mode
))
2233 /* Verify that master_dev is indeed the master of slave_dev */
2234 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2237 read_lock(&bond
->lock
);
2239 read_lock(&bond
->curr_slave_lock
);
2240 old_active
= bond
->curr_active_slave
;
2241 read_unlock(&bond
->curr_slave_lock
);
2243 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2246 * Changing to the current active: do nothing; return success.
2248 if (new_active
&& (new_active
== old_active
)) {
2249 read_unlock(&bond
->lock
);
2255 (new_active
->link
== BOND_LINK_UP
) &&
2256 IS_UP(new_active
->dev
)) {
2258 write_lock_bh(&bond
->curr_slave_lock
);
2259 bond_change_active_slave(bond
, new_active
);
2260 write_unlock_bh(&bond
->curr_slave_lock
);
2261 unblock_netpoll_tx();
2265 read_unlock(&bond
->lock
);
2270 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2272 struct bonding
*bond
= netdev_priv(bond_dev
);
2274 info
->bond_mode
= bond
->params
.mode
;
2275 info
->miimon
= bond
->params
.miimon
;
2277 read_lock(&bond
->lock
);
2278 info
->num_slaves
= bond
->slave_cnt
;
2279 read_unlock(&bond
->lock
);
2284 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2286 struct bonding
*bond
= netdev_priv(bond_dev
);
2287 struct slave
*slave
;
2288 int i
, res
= -ENODEV
;
2290 read_lock(&bond
->lock
);
2292 bond_for_each_slave(bond
, slave
, i
) {
2293 if (i
== (int)info
->slave_id
) {
2295 strcpy(info
->slave_name
, slave
->dev
->name
);
2296 info
->link
= slave
->link
;
2297 info
->state
= bond_slave_state(slave
);
2298 info
->link_failure_count
= slave
->link_failure_count
;
2303 read_unlock(&bond
->lock
);
2308 /*-------------------------------- Monitoring -------------------------------*/
2311 static int bond_miimon_inspect(struct bonding
*bond
)
2313 struct slave
*slave
;
2314 int i
, link_state
, commit
= 0;
2315 bool ignore_updelay
;
2317 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2319 bond_for_each_slave(bond
, slave
, i
) {
2320 slave
->new_link
= BOND_LINK_NOCHANGE
;
2322 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2324 switch (slave
->link
) {
2329 slave
->link
= BOND_LINK_FAIL
;
2330 slave
->delay
= bond
->params
.downdelay
;
2332 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2334 (bond
->params
.mode
==
2335 BOND_MODE_ACTIVEBACKUP
) ?
2336 (bond_is_active_slave(slave
) ?
2337 "active " : "backup ") : "",
2339 bond
->params
.downdelay
* bond
->params
.miimon
);
2342 case BOND_LINK_FAIL
:
2345 * recovered before downdelay expired
2347 slave
->link
= BOND_LINK_UP
;
2348 slave
->jiffies
= jiffies
;
2349 pr_info("%s: link status up again after %d ms for interface %s.\n",
2351 (bond
->params
.downdelay
- slave
->delay
) *
2352 bond
->params
.miimon
,
2357 if (slave
->delay
<= 0) {
2358 slave
->new_link
= BOND_LINK_DOWN
;
2366 case BOND_LINK_DOWN
:
2370 slave
->link
= BOND_LINK_BACK
;
2371 slave
->delay
= bond
->params
.updelay
;
2374 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2375 bond
->dev
->name
, slave
->dev
->name
,
2376 ignore_updelay
? 0 :
2377 bond
->params
.updelay
*
2378 bond
->params
.miimon
);
2381 case BOND_LINK_BACK
:
2383 slave
->link
= BOND_LINK_DOWN
;
2384 pr_info("%s: link status down again after %d ms for interface %s.\n",
2386 (bond
->params
.updelay
- slave
->delay
) *
2387 bond
->params
.miimon
,
2396 if (slave
->delay
<= 0) {
2397 slave
->new_link
= BOND_LINK_UP
;
2399 ignore_updelay
= false;
2411 static void bond_miimon_commit(struct bonding
*bond
)
2413 struct slave
*slave
;
2416 bond_for_each_slave(bond
, slave
, i
) {
2417 switch (slave
->new_link
) {
2418 case BOND_LINK_NOCHANGE
:
2422 slave
->link
= BOND_LINK_UP
;
2423 slave
->jiffies
= jiffies
;
2425 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2426 /* prevent it from being the active one */
2427 bond_set_backup_slave(slave
);
2428 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2429 /* make it immediately active */
2430 bond_set_active_slave(slave
);
2431 } else if (slave
!= bond
->primary_slave
) {
2432 /* prevent it from being the active one */
2433 bond_set_backup_slave(slave
);
2436 bond_update_speed_duplex(slave
);
2438 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2439 bond
->dev
->name
, slave
->dev
->name
,
2440 slave
->speed
, slave
->duplex
? "full" : "half");
2442 /* notify ad that the link status has changed */
2443 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2444 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2446 if (bond_is_lb(bond
))
2447 bond_alb_handle_link_change(bond
, slave
,
2450 if (!bond
->curr_active_slave
||
2451 (slave
== bond
->primary_slave
))
2456 case BOND_LINK_DOWN
:
2457 if (slave
->link_failure_count
< UINT_MAX
)
2458 slave
->link_failure_count
++;
2460 slave
->link
= BOND_LINK_DOWN
;
2462 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2463 bond
->params
.mode
== BOND_MODE_8023AD
)
2464 bond_set_slave_inactive_flags(slave
);
2466 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2467 bond
->dev
->name
, slave
->dev
->name
);
2469 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2470 bond_3ad_handle_link_change(slave
,
2473 if (bond_is_lb(bond
))
2474 bond_alb_handle_link_change(bond
, slave
,
2477 if (slave
== bond
->curr_active_slave
)
2483 pr_err("%s: invalid new link %d on slave %s\n",
2484 bond
->dev
->name
, slave
->new_link
,
2486 slave
->new_link
= BOND_LINK_NOCHANGE
;
2494 write_lock_bh(&bond
->curr_slave_lock
);
2495 bond_select_active_slave(bond
);
2496 write_unlock_bh(&bond
->curr_slave_lock
);
2497 unblock_netpoll_tx();
2500 bond_set_carrier(bond
);
2506 * Really a wrapper that splits the mii monitor into two phases: an
2507 * inspection, then (if inspection indicates something needs to be done)
2508 * an acquisition of appropriate locks followed by a commit phase to
2509 * implement whatever link state changes are indicated.
2511 void bond_mii_monitor(struct work_struct
*work
)
2513 struct bonding
*bond
= container_of(work
, struct bonding
,
2515 bool should_notify_peers
= false;
2517 read_lock(&bond
->lock
);
2518 if (bond
->kill_timers
)
2521 if (bond
->slave_cnt
== 0)
2524 should_notify_peers
= bond_should_notify_peers(bond
);
2526 if (bond_miimon_inspect(bond
)) {
2527 read_unlock(&bond
->lock
);
2529 read_lock(&bond
->lock
);
2531 bond_miimon_commit(bond
);
2533 read_unlock(&bond
->lock
);
2534 rtnl_unlock(); /* might sleep, hold no other locks */
2535 read_lock(&bond
->lock
);
2539 if (bond
->params
.miimon
)
2540 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2541 msecs_to_jiffies(bond
->params
.miimon
));
2543 read_unlock(&bond
->lock
);
2545 if (should_notify_peers
) {
2547 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
2552 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2554 struct in_device
*idev
;
2555 struct in_ifaddr
*ifa
;
2562 idev
= __in_dev_get_rcu(dev
);
2566 ifa
= idev
->ifa_list
;
2570 addr
= ifa
->ifa_local
;
2576 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2578 struct vlan_entry
*vlan
;
2580 if (ip
== bond
->master_ip
)
2583 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2584 if (ip
== vlan
->vlan_ip
)
2592 * We go to the (large) trouble of VLAN tagging ARP frames because
2593 * switches in VLAN mode (especially if ports are configured as
2594 * "native" to a VLAN) might not pass non-tagged frames.
2596 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2598 struct sk_buff
*skb
;
2600 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2601 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2603 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2604 NULL
, slave_dev
->dev_addr
, NULL
);
2607 pr_err("ARP packet allocation failed\n");
2611 skb
= vlan_put_tag(skb
, vlan_id
);
2613 pr_err("failed to insert VLAN tag\n");
2621 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2624 __be32
*targets
= bond
->params
.arp_targets
;
2625 struct vlan_entry
*vlan
;
2626 struct net_device
*vlan_dev
;
2629 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2632 pr_debug("basa: target %x\n", targets
[i
]);
2633 if (!bond_vlan_used(bond
)) {
2634 pr_debug("basa: empty vlan: arp_send\n");
2635 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2636 bond
->master_ip
, 0);
2641 * If VLANs are configured, we do a route lookup to
2642 * determine which VLAN interface would be used, so we
2643 * can tag the ARP with the proper VLAN tag.
2645 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2648 if (net_ratelimit()) {
2649 pr_warning("%s: no route to arp_ip_target %pI4\n",
2650 bond
->dev
->name
, &targets
[i
]);
2656 * This target is not on a VLAN
2658 if (rt
->dst
.dev
== bond
->dev
) {
2660 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2661 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2662 bond
->master_ip
, 0);
2667 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2669 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
2672 if (vlan_dev
== rt
->dst
.dev
) {
2673 vlan_id
= vlan
->vlan_id
;
2674 pr_debug("basa: vlan match on %s %d\n",
2675 vlan_dev
->name
, vlan_id
);
2682 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2683 vlan
->vlan_ip
, vlan_id
);
2687 if (net_ratelimit()) {
2688 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2689 bond
->dev
->name
, &targets
[i
],
2690 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2696 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2699 __be32
*targets
= bond
->params
.arp_targets
;
2701 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2702 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2703 &sip
, &tip
, i
, &targets
[i
],
2704 bond_has_this_ip(bond
, tip
));
2705 if (sip
== targets
[i
]) {
2706 if (bond_has_this_ip(bond
, tip
))
2707 slave
->last_arp_rx
= jiffies
;
2713 static void bond_arp_rcv(struct sk_buff
*skb
, struct bonding
*bond
,
2714 struct slave
*slave
)
2717 unsigned char *arp_ptr
;
2720 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2723 read_lock(&bond
->lock
);
2725 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2726 bond
->dev
->name
, skb
->dev
->name
);
2728 if (!pskb_may_pull(skb
, arp_hdr_len(bond
->dev
)))
2732 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2733 skb
->pkt_type
== PACKET_OTHERHOST
||
2734 skb
->pkt_type
== PACKET_LOOPBACK
||
2735 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2736 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2740 arp_ptr
= (unsigned char *)(arp
+ 1);
2741 arp_ptr
+= bond
->dev
->addr_len
;
2742 memcpy(&sip
, arp_ptr
, 4);
2743 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2744 memcpy(&tip
, arp_ptr
, 4);
2746 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2747 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2748 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2752 * Backup slaves won't see the ARP reply, but do come through
2753 * here for each ARP probe (so we swap the sip/tip to validate
2754 * the probe). In a "redundant switch, common router" type of
2755 * configuration, the ARP probe will (hopefully) travel from
2756 * the active, through one switch, the router, then the other
2757 * switch before reaching the backup.
2759 if (bond_is_active_slave(slave
))
2760 bond_validate_arp(bond
, slave
, sip
, tip
);
2762 bond_validate_arp(bond
, slave
, tip
, sip
);
2765 read_unlock(&bond
->lock
);
2769 * this function is called regularly to monitor each slave's link
2770 * ensuring that traffic is being sent and received when arp monitoring
2771 * is used in load-balancing mode. if the adapter has been dormant, then an
2772 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2773 * arp monitoring in active backup mode.
2775 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2777 struct bonding
*bond
= container_of(work
, struct bonding
,
2779 struct slave
*slave
, *oldcurrent
;
2780 int do_failover
= 0;
2784 read_lock(&bond
->lock
);
2786 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2788 if (bond
->kill_timers
)
2791 if (bond
->slave_cnt
== 0)
2794 read_lock(&bond
->curr_slave_lock
);
2795 oldcurrent
= bond
->curr_active_slave
;
2796 read_unlock(&bond
->curr_slave_lock
);
2798 /* see if any of the previous devices are up now (i.e. they have
2799 * xmt and rcv traffic). the curr_active_slave does not come into
2800 * the picture unless it is null. also, slave->jiffies is not needed
2801 * here because we send an arp on each slave and give a slave as
2802 * long as it needs to get the tx/rx within the delta.
2803 * TODO: what about up/down delay in arp mode? it wasn't here before
2806 bond_for_each_slave(bond
, slave
, i
) {
2807 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2809 if (slave
->link
!= BOND_LINK_UP
) {
2810 if (time_in_range(jiffies
,
2811 trans_start
- delta_in_ticks
,
2812 trans_start
+ delta_in_ticks
) &&
2813 time_in_range(jiffies
,
2814 slave
->dev
->last_rx
- delta_in_ticks
,
2815 slave
->dev
->last_rx
+ delta_in_ticks
)) {
2817 slave
->link
= BOND_LINK_UP
;
2818 bond_set_active_slave(slave
);
2820 /* primary_slave has no meaning in round-robin
2821 * mode. the window of a slave being up and
2822 * curr_active_slave being null after enslaving
2826 pr_info("%s: link status definitely up for interface %s, ",
2831 pr_info("%s: interface %s is now up\n",
2837 /* slave->link == BOND_LINK_UP */
2839 /* not all switches will respond to an arp request
2840 * when the source ip is 0, so don't take the link down
2841 * if we don't know our ip yet
2843 if (!time_in_range(jiffies
,
2844 trans_start
- delta_in_ticks
,
2845 trans_start
+ 2 * delta_in_ticks
) ||
2846 !time_in_range(jiffies
,
2847 slave
->dev
->last_rx
- delta_in_ticks
,
2848 slave
->dev
->last_rx
+ 2 * delta_in_ticks
)) {
2850 slave
->link
= BOND_LINK_DOWN
;
2851 bond_set_backup_slave(slave
);
2853 if (slave
->link_failure_count
< UINT_MAX
)
2854 slave
->link_failure_count
++;
2856 pr_info("%s: interface %s is now down.\n",
2860 if (slave
== oldcurrent
)
2865 /* note: if switch is in round-robin mode, all links
2866 * must tx arp to ensure all links rx an arp - otherwise
2867 * links may oscillate or not come up at all; if switch is
2868 * in something like xor mode, there is nothing we can
2869 * do - all replies will be rx'ed on same link causing slaves
2870 * to be unstable during low/no traffic periods
2872 if (IS_UP(slave
->dev
))
2873 bond_arp_send_all(bond
, slave
);
2878 write_lock_bh(&bond
->curr_slave_lock
);
2880 bond_select_active_slave(bond
);
2882 write_unlock_bh(&bond
->curr_slave_lock
);
2883 unblock_netpoll_tx();
2887 if (bond
->params
.arp_interval
)
2888 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2890 read_unlock(&bond
->lock
);
2894 * Called to inspect slaves for active-backup mode ARP monitor link state
2895 * changes. Sets new_link in slaves to specify what action should take
2896 * place for the slave. Returns 0 if no changes are found, >0 if changes
2897 * to link states must be committed.
2899 * Called with bond->lock held for read.
2901 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2903 struct slave
*slave
;
2905 unsigned long trans_start
;
2907 bond_for_each_slave(bond
, slave
, i
) {
2908 slave
->new_link
= BOND_LINK_NOCHANGE
;
2910 if (slave
->link
!= BOND_LINK_UP
) {
2911 if (time_in_range(jiffies
,
2912 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2913 slave_last_rx(bond
, slave
) + delta_in_ticks
)) {
2915 slave
->new_link
= BOND_LINK_UP
;
2923 * Give slaves 2*delta after being enslaved or made
2924 * active. This avoids bouncing, as the last receive
2925 * times need a full ARP monitor cycle to be updated.
2927 if (time_in_range(jiffies
,
2928 slave
->jiffies
- delta_in_ticks
,
2929 slave
->jiffies
+ 2 * delta_in_ticks
))
2933 * Backup slave is down if:
2934 * - No current_arp_slave AND
2935 * - more than 3*delta since last receive AND
2936 * - the bond has an IP address
2938 * Note: a non-null current_arp_slave indicates
2939 * the curr_active_slave went down and we are
2940 * searching for a new one; under this condition
2941 * we only take the curr_active_slave down - this
2942 * gives each slave a chance to tx/rx traffic
2943 * before being taken out
2945 if (!bond_is_active_slave(slave
) &&
2946 !bond
->current_arp_slave
&&
2947 !time_in_range(jiffies
,
2948 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2949 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
)) {
2951 slave
->new_link
= BOND_LINK_DOWN
;
2956 * Active slave is down if:
2957 * - more than 2*delta since transmitting OR
2958 * - (more than 2*delta since receive AND
2959 * the bond has an IP address)
2961 trans_start
= dev_trans_start(slave
->dev
);
2962 if (bond_is_active_slave(slave
) &&
2963 (!time_in_range(jiffies
,
2964 trans_start
- delta_in_ticks
,
2965 trans_start
+ 2 * delta_in_ticks
) ||
2966 !time_in_range(jiffies
,
2967 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2968 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
))) {
2970 slave
->new_link
= BOND_LINK_DOWN
;
2979 * Called to commit link state changes noted by inspection step of
2980 * active-backup mode ARP monitor.
2982 * Called with RTNL and bond->lock for read.
2984 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2986 struct slave
*slave
;
2988 unsigned long trans_start
;
2990 bond_for_each_slave(bond
, slave
, i
) {
2991 switch (slave
->new_link
) {
2992 case BOND_LINK_NOCHANGE
:
2996 trans_start
= dev_trans_start(slave
->dev
);
2997 if ((!bond
->curr_active_slave
&&
2998 time_in_range(jiffies
,
2999 trans_start
- delta_in_ticks
,
3000 trans_start
+ delta_in_ticks
)) ||
3001 bond
->curr_active_slave
!= slave
) {
3002 slave
->link
= BOND_LINK_UP
;
3003 bond
->current_arp_slave
= NULL
;
3005 pr_info("%s: link status definitely up for interface %s.\n",
3006 bond
->dev
->name
, slave
->dev
->name
);
3008 if (!bond
->curr_active_slave
||
3009 (slave
== bond
->primary_slave
))
3016 case BOND_LINK_DOWN
:
3017 if (slave
->link_failure_count
< UINT_MAX
)
3018 slave
->link_failure_count
++;
3020 slave
->link
= BOND_LINK_DOWN
;
3021 bond_set_slave_inactive_flags(slave
);
3023 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3024 bond
->dev
->name
, slave
->dev
->name
);
3026 if (slave
== bond
->curr_active_slave
) {
3027 bond
->current_arp_slave
= NULL
;
3034 pr_err("%s: impossible: new_link %d on slave %s\n",
3035 bond
->dev
->name
, slave
->new_link
,
3043 write_lock_bh(&bond
->curr_slave_lock
);
3044 bond_select_active_slave(bond
);
3045 write_unlock_bh(&bond
->curr_slave_lock
);
3046 unblock_netpoll_tx();
3049 bond_set_carrier(bond
);
3053 * Send ARP probes for active-backup mode ARP monitor.
3055 * Called with bond->lock held for read.
3057 static void bond_ab_arp_probe(struct bonding
*bond
)
3059 struct slave
*slave
;
3062 read_lock(&bond
->curr_slave_lock
);
3064 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3065 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3066 bond
->current_arp_slave
->dev
->name
,
3067 bond
->curr_active_slave
->dev
->name
);
3069 if (bond
->curr_active_slave
) {
3070 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3071 read_unlock(&bond
->curr_slave_lock
);
3075 read_unlock(&bond
->curr_slave_lock
);
3077 /* if we don't have a curr_active_slave, search for the next available
3078 * backup slave from the current_arp_slave and make it the candidate
3079 * for becoming the curr_active_slave
3082 if (!bond
->current_arp_slave
) {
3083 bond
->current_arp_slave
= bond
->first_slave
;
3084 if (!bond
->current_arp_slave
)
3088 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3090 /* search for next candidate */
3091 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3092 if (IS_UP(slave
->dev
)) {
3093 slave
->link
= BOND_LINK_BACK
;
3094 bond_set_slave_active_flags(slave
);
3095 bond_arp_send_all(bond
, slave
);
3096 slave
->jiffies
= jiffies
;
3097 bond
->current_arp_slave
= slave
;
3101 /* if the link state is up at this point, we
3102 * mark it down - this can happen if we have
3103 * simultaneous link failures and
3104 * reselect_active_interface doesn't make this
3105 * one the current slave so it is still marked
3106 * up when it is actually down
3108 if (slave
->link
== BOND_LINK_UP
) {
3109 slave
->link
= BOND_LINK_DOWN
;
3110 if (slave
->link_failure_count
< UINT_MAX
)
3111 slave
->link_failure_count
++;
3113 bond_set_slave_inactive_flags(slave
);
3115 pr_info("%s: backup interface %s is now down.\n",
3116 bond
->dev
->name
, slave
->dev
->name
);
3121 void bond_activebackup_arp_mon(struct work_struct
*work
)
3123 struct bonding
*bond
= container_of(work
, struct bonding
,
3125 bool should_notify_peers
= false;
3128 read_lock(&bond
->lock
);
3130 if (bond
->kill_timers
)
3133 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3135 if (bond
->slave_cnt
== 0)
3138 should_notify_peers
= bond_should_notify_peers(bond
);
3140 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3141 read_unlock(&bond
->lock
);
3143 read_lock(&bond
->lock
);
3145 bond_ab_arp_commit(bond
, delta_in_ticks
);
3147 read_unlock(&bond
->lock
);
3149 read_lock(&bond
->lock
);
3152 bond_ab_arp_probe(bond
);
3155 if (bond
->params
.arp_interval
)
3156 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3158 read_unlock(&bond
->lock
);
3160 if (should_notify_peers
) {
3162 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
3167 /*-------------------------- netdev event handling --------------------------*/
3170 * Change device name
3172 static int bond_event_changename(struct bonding
*bond
)
3174 bond_remove_proc_entry(bond
);
3175 bond_create_proc_entry(bond
);
3177 bond_debug_reregister(bond
);
3182 static int bond_master_netdev_event(unsigned long event
,
3183 struct net_device
*bond_dev
)
3185 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3188 case NETDEV_CHANGENAME
:
3189 return bond_event_changename(event_bond
);
3197 static int bond_slave_netdev_event(unsigned long event
,
3198 struct net_device
*slave_dev
)
3200 struct net_device
*bond_dev
= slave_dev
->master
;
3201 struct bonding
*bond
= netdev_priv(bond_dev
);
3204 case NETDEV_UNREGISTER
:
3206 if (bond
->setup_by_slave
)
3207 bond_release_and_destroy(bond_dev
, slave_dev
);
3209 bond_release(bond_dev
, slave_dev
);
3213 if (bond
->params
.mode
== BOND_MODE_8023AD
|| bond_is_lb(bond
)) {
3214 struct slave
*slave
;
3216 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3218 u32 old_speed
= slave
->speed
;
3219 u8 old_duplex
= slave
->duplex
;
3221 bond_update_speed_duplex(slave
);
3223 if (bond_is_lb(bond
))
3226 if (old_speed
!= slave
->speed
)
3227 bond_3ad_adapter_speed_changed(slave
);
3228 if (old_duplex
!= slave
->duplex
)
3229 bond_3ad_adapter_duplex_changed(slave
);
3236 * ... Or is it this?
3239 case NETDEV_CHANGEMTU
:
3241 * TODO: Should slaves be allowed to
3242 * independently alter their MTU? For
3243 * an active-backup bond, slaves need
3244 * not be the same type of device, so
3245 * MTUs may vary. For other modes,
3246 * slaves arguably should have the
3247 * same MTUs. To do this, we'd need to
3248 * take over the slave's change_mtu
3249 * function for the duration of their
3253 case NETDEV_CHANGENAME
:
3255 * TODO: handle changing the primary's name
3258 case NETDEV_FEAT_CHANGE
:
3259 bond_compute_features(bond
);
3269 * bond_netdev_event: handle netdev notifier chain events.
3271 * This function receives events for the netdev chain. The caller (an
3272 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3273 * locks for us to safely manipulate the slave devices (RTNL lock,
3276 static int bond_netdev_event(struct notifier_block
*this,
3277 unsigned long event
, void *ptr
)
3279 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3281 pr_debug("event_dev: %s, event: %lx\n",
3282 event_dev
? event_dev
->name
: "None",
3285 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3288 if (event_dev
->flags
& IFF_MASTER
) {
3289 pr_debug("IFF_MASTER\n");
3290 return bond_master_netdev_event(event
, event_dev
);
3293 if (event_dev
->flags
& IFF_SLAVE
) {
3294 pr_debug("IFF_SLAVE\n");
3295 return bond_slave_netdev_event(event
, event_dev
);
3302 * bond_inetaddr_event: handle inetaddr notifier chain events.
3304 * We keep track of device IPs primarily to use as source addresses in
3305 * ARP monitor probes (rather than spewing out broadcasts all the time).
3307 * We track one IP for the main device (if it has one), plus one per VLAN.
3309 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3311 struct in_ifaddr
*ifa
= ptr
;
3312 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3313 struct bond_net
*bn
= net_generic(dev_net(event_dev
), bond_net_id
);
3314 struct bonding
*bond
;
3315 struct vlan_entry
*vlan
;
3317 list_for_each_entry(bond
, &bn
->dev_list
, bond_list
) {
3318 if (bond
->dev
== event_dev
) {
3321 bond
->master_ip
= ifa
->ifa_local
;
3324 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3331 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3332 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
3334 if (vlan_dev
== event_dev
) {
3337 vlan
->vlan_ip
= ifa
->ifa_local
;
3341 bond_glean_dev_ip(vlan_dev
);
3352 static struct notifier_block bond_netdev_notifier
= {
3353 .notifier_call
= bond_netdev_event
,
3356 static struct notifier_block bond_inetaddr_notifier
= {
3357 .notifier_call
= bond_inetaddr_event
,
3360 /*---------------------------- Hashing Policies -----------------------------*/
3363 * Hash for the output device based upon layer 2 and layer 3 data. If
3364 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3366 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3368 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3369 struct iphdr
*iph
= ip_hdr(skb
);
3371 if (skb
->protocol
== htons(ETH_P_IP
)) {
3372 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3373 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3376 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3380 * Hash for the output device based upon layer 3 and layer 4 data. If
3381 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3382 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3384 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3386 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3387 struct iphdr
*iph
= ip_hdr(skb
);
3388 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3391 if (skb
->protocol
== htons(ETH_P_IP
)) {
3392 if (!ip_is_fragment(iph
) &&
3393 (iph
->protocol
== IPPROTO_TCP
||
3394 iph
->protocol
== IPPROTO_UDP
)) {
3395 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3397 return (layer4_xor
^
3398 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3402 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3406 * Hash for the output device based upon layer 2 data
3408 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3410 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3412 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3415 /*-------------------------- Device entry points ----------------------------*/
3417 static int bond_open(struct net_device
*bond_dev
)
3419 struct bonding
*bond
= netdev_priv(bond_dev
);
3421 bond
->kill_timers
= 0;
3423 INIT_DELAYED_WORK(&bond
->mcast_work
, bond_resend_igmp_join_requests_delayed
);
3425 if (bond_is_lb(bond
)) {
3426 /* bond_alb_initialize must be called before the timer
3429 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3430 /* something went wrong - fail the open operation */
3434 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3435 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3438 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3439 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3440 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3443 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3444 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3445 INIT_DELAYED_WORK(&bond
->arp_work
,
3446 bond_activebackup_arp_mon
);
3448 INIT_DELAYED_WORK(&bond
->arp_work
,
3449 bond_loadbalance_arp_mon
);
3451 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3452 if (bond
->params
.arp_validate
)
3453 bond
->recv_probe
= bond_arp_rcv
;
3456 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3457 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3458 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3459 /* register to receive LACPDUs */
3460 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3461 bond_3ad_initiate_agg_selection(bond
, 1);
3467 static int bond_close(struct net_device
*bond_dev
)
3469 struct bonding
*bond
= netdev_priv(bond_dev
);
3471 write_lock_bh(&bond
->lock
);
3473 bond
->send_peer_notif
= 0;
3475 /* signal timers not to re-arm */
3476 bond
->kill_timers
= 1;
3478 write_unlock_bh(&bond
->lock
);
3480 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3481 cancel_delayed_work(&bond
->mii_work
);
3484 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3485 cancel_delayed_work(&bond
->arp_work
);
3488 switch (bond
->params
.mode
) {
3489 case BOND_MODE_8023AD
:
3490 cancel_delayed_work(&bond
->ad_work
);
3494 cancel_delayed_work(&bond
->alb_work
);
3500 if (delayed_work_pending(&bond
->mcast_work
))
3501 cancel_delayed_work(&bond
->mcast_work
);
3503 if (bond_is_lb(bond
)) {
3504 /* Must be called only after all
3505 * slaves have been released
3507 bond_alb_deinitialize(bond
);
3509 bond
->recv_probe
= NULL
;
3514 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3515 struct rtnl_link_stats64
*stats
)
3517 struct bonding
*bond
= netdev_priv(bond_dev
);
3518 struct rtnl_link_stats64 temp
;
3519 struct slave
*slave
;
3522 memset(stats
, 0, sizeof(*stats
));
3524 read_lock_bh(&bond
->lock
);
3526 bond_for_each_slave(bond
, slave
, i
) {
3527 const struct rtnl_link_stats64
*sstats
=
3528 dev_get_stats(slave
->dev
, &temp
);
3530 stats
->rx_packets
+= sstats
->rx_packets
;
3531 stats
->rx_bytes
+= sstats
->rx_bytes
;
3532 stats
->rx_errors
+= sstats
->rx_errors
;
3533 stats
->rx_dropped
+= sstats
->rx_dropped
;
3535 stats
->tx_packets
+= sstats
->tx_packets
;
3536 stats
->tx_bytes
+= sstats
->tx_bytes
;
3537 stats
->tx_errors
+= sstats
->tx_errors
;
3538 stats
->tx_dropped
+= sstats
->tx_dropped
;
3540 stats
->multicast
+= sstats
->multicast
;
3541 stats
->collisions
+= sstats
->collisions
;
3543 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3544 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3545 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3546 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3547 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3548 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3550 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3551 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3552 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3553 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3554 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3557 read_unlock_bh(&bond
->lock
);
3562 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3564 struct net_device
*slave_dev
= NULL
;
3565 struct ifbond k_binfo
;
3566 struct ifbond __user
*u_binfo
= NULL
;
3567 struct ifslave k_sinfo
;
3568 struct ifslave __user
*u_sinfo
= NULL
;
3569 struct mii_ioctl_data
*mii
= NULL
;
3572 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3584 * We do this again just in case we were called by SIOCGMIIREG
3585 * instead of SIOCGMIIPHY.
3592 if (mii
->reg_num
== 1) {
3593 struct bonding
*bond
= netdev_priv(bond_dev
);
3595 read_lock(&bond
->lock
);
3596 read_lock(&bond
->curr_slave_lock
);
3597 if (netif_carrier_ok(bond
->dev
))
3598 mii
->val_out
= BMSR_LSTATUS
;
3600 read_unlock(&bond
->curr_slave_lock
);
3601 read_unlock(&bond
->lock
);
3605 case BOND_INFO_QUERY_OLD
:
3606 case SIOCBONDINFOQUERY
:
3607 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3609 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3612 res
= bond_info_query(bond_dev
, &k_binfo
);
3614 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3618 case BOND_SLAVE_INFO_QUERY_OLD
:
3619 case SIOCBONDSLAVEINFOQUERY
:
3620 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3622 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3625 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3627 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3636 if (!capable(CAP_NET_ADMIN
))
3639 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
3641 pr_debug("slave_dev=%p:\n", slave_dev
);
3646 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3648 case BOND_ENSLAVE_OLD
:
3649 case SIOCBONDENSLAVE
:
3650 res
= bond_enslave(bond_dev
, slave_dev
);
3652 case BOND_RELEASE_OLD
:
3653 case SIOCBONDRELEASE
:
3654 res
= bond_release(bond_dev
, slave_dev
);
3656 case BOND_SETHWADDR_OLD
:
3657 case SIOCBONDSETHWADDR
:
3658 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3660 case BOND_CHANGE_ACTIVE_OLD
:
3661 case SIOCBONDCHANGEACTIVE
:
3662 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3674 static bool bond_addr_in_mc_list(unsigned char *addr
,
3675 struct netdev_hw_addr_list
*list
,
3678 struct netdev_hw_addr
*ha
;
3680 netdev_hw_addr_list_for_each(ha
, list
)
3681 if (!memcmp(ha
->addr
, addr
, addrlen
))
3687 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3689 struct bonding
*bond
= netdev_priv(bond_dev
);
3690 struct netdev_hw_addr
*ha
;
3694 * Do promisc before checking multicast_mode
3696 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
))
3698 * FIXME: Need to handle the error when one of the multi-slaves
3701 bond_set_promiscuity(bond
, 1);
3704 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
))
3705 bond_set_promiscuity(bond
, -1);
3708 /* set allmulti flag to slaves */
3709 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
))
3711 * FIXME: Need to handle the error when one of the multi-slaves
3714 bond_set_allmulti(bond
, 1);
3717 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
))
3718 bond_set_allmulti(bond
, -1);
3721 read_lock(&bond
->lock
);
3723 bond
->flags
= bond_dev
->flags
;
3725 /* looking for addresses to add to slaves' mc list */
3726 netdev_for_each_mc_addr(ha
, bond_dev
) {
3727 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
3728 bond_dev
->addr_len
);
3730 bond_mc_add(bond
, ha
->addr
);
3733 /* looking for addresses to delete from slaves' list */
3734 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
3735 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
3736 bond_dev
->addr_len
);
3738 bond_mc_del(bond
, ha
->addr
);
3741 /* save master's multicast list */
3742 __hw_addr_flush(&bond
->mc_list
);
3743 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
3744 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
3746 read_unlock(&bond
->lock
);
3749 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
3751 struct bonding
*bond
= netdev_priv(dev
);
3752 struct slave
*slave
= bond
->first_slave
;
3755 const struct net_device_ops
*slave_ops
3756 = slave
->dev
->netdev_ops
;
3757 if (slave_ops
->ndo_neigh_setup
)
3758 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
3764 * Change the MTU of all of a master's slaves to match the master
3766 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3768 struct bonding
*bond
= netdev_priv(bond_dev
);
3769 struct slave
*slave
, *stop_at
;
3773 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3774 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3776 /* Can't hold bond->lock with bh disabled here since
3777 * some base drivers panic. On the other hand we can't
3778 * hold bond->lock without bh disabled because we'll
3779 * deadlock. The only solution is to rely on the fact
3780 * that we're under rtnl_lock here, and the slaves
3781 * list won't change. This doesn't solve the problem
3782 * of setting the slave's MTU while it is
3783 * transmitting, but the assumption is that the base
3784 * driver can handle that.
3786 * TODO: figure out a way to safely iterate the slaves
3787 * list, but without holding a lock around the actual
3788 * call to the base driver.
3791 bond_for_each_slave(bond
, slave
, i
) {
3792 pr_debug("s %p s->p %p c_m %p\n",
3795 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3797 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3800 /* If we failed to set the slave's mtu to the new value
3801 * we must abort the operation even in ACTIVE_BACKUP
3802 * mode, because if we allow the backup slaves to have
3803 * different mtu values than the active slave we'll
3804 * need to change their mtu when doing a failover. That
3805 * means changing their mtu from timer context, which
3806 * is probably not a good idea.
3808 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3813 bond_dev
->mtu
= new_mtu
;
3818 /* unwind from head to the slave that failed */
3820 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3823 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3825 pr_debug("unwind err %d dev %s\n",
3826 tmp_res
, slave
->dev
->name
);
3836 * Note that many devices must be down to change the HW address, and
3837 * downing the master releases all slaves. We can make bonds full of
3838 * bonding devices to test this, however.
3840 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3842 struct bonding
*bond
= netdev_priv(bond_dev
);
3843 struct sockaddr
*sa
= addr
, tmp_sa
;
3844 struct slave
*slave
, *stop_at
;
3848 if (bond
->params
.mode
== BOND_MODE_ALB
)
3849 return bond_alb_set_mac_address(bond_dev
, addr
);
3852 pr_debug("bond=%p, name=%s\n",
3853 bond
, bond_dev
? bond_dev
->name
: "None");
3856 * If fail_over_mac is set to active, do nothing and return
3857 * success. Returning an error causes ifenslave to fail.
3859 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
3862 if (!is_valid_ether_addr(sa
->sa_data
))
3863 return -EADDRNOTAVAIL
;
3865 /* Can't hold bond->lock with bh disabled here since
3866 * some base drivers panic. On the other hand we can't
3867 * hold bond->lock without bh disabled because we'll
3868 * deadlock. The only solution is to rely on the fact
3869 * that we're under rtnl_lock here, and the slaves
3870 * list won't change. This doesn't solve the problem
3871 * of setting the slave's hw address while it is
3872 * transmitting, but the assumption is that the base
3873 * driver can handle that.
3875 * TODO: figure out a way to safely iterate the slaves
3876 * list, but without holding a lock around the actual
3877 * call to the base driver.
3880 bond_for_each_slave(bond
, slave
, i
) {
3881 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3882 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3884 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3886 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3890 res
= dev_set_mac_address(slave
->dev
, addr
);
3892 /* TODO: consider downing the slave
3894 * User should expect communications
3895 * breakage anyway until ARP finish
3898 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3904 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3908 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3909 tmp_sa
.sa_family
= bond_dev
->type
;
3911 /* unwind from head to the slave that failed */
3913 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3916 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3918 pr_debug("unwind err %d dev %s\n",
3919 tmp_res
, slave
->dev
->name
);
3926 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3928 struct bonding
*bond
= netdev_priv(bond_dev
);
3929 struct slave
*slave
, *start_at
;
3930 int i
, slave_no
, res
= 1;
3931 struct iphdr
*iph
= ip_hdr(skb
);
3934 * Start with the curr_active_slave that joined the bond as the
3935 * default for sending IGMP traffic. For failover purposes one
3936 * needs to maintain some consistency for the interface that will
3937 * send the join/membership reports. The curr_active_slave found
3938 * will send all of this type of traffic.
3940 if ((iph
->protocol
== IPPROTO_IGMP
) &&
3941 (skb
->protocol
== htons(ETH_P_IP
))) {
3943 read_lock(&bond
->curr_slave_lock
);
3944 slave
= bond
->curr_active_slave
;
3945 read_unlock(&bond
->curr_slave_lock
);
3951 * Concurrent TX may collide on rr_tx_counter; we accept
3952 * that as being rare enough not to justify using an
3955 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
3957 bond_for_each_slave(bond
, slave
, i
) {
3965 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3966 if (IS_UP(slave
->dev
) &&
3967 (slave
->link
== BOND_LINK_UP
) &&
3968 bond_is_active_slave(slave
)) {
3969 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3976 /* no suitable interface, frame not sent */
3980 return NETDEV_TX_OK
;
3985 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3986 * the bond has a usable interface.
3988 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3990 struct bonding
*bond
= netdev_priv(bond_dev
);
3993 read_lock(&bond
->curr_slave_lock
);
3995 if (bond
->curr_active_slave
)
3996 res
= bond_dev_queue_xmit(bond
, skb
,
3997 bond
->curr_active_slave
->dev
);
4000 /* no suitable interface, frame not sent */
4003 read_unlock(&bond
->curr_slave_lock
);
4005 return NETDEV_TX_OK
;
4009 * In bond_xmit_xor() , we determine the output device by using a pre-
4010 * determined xmit_hash_policy(), If the selected device is not enabled,
4011 * find the next active slave.
4013 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4015 struct bonding
*bond
= netdev_priv(bond_dev
);
4016 struct slave
*slave
, *start_at
;
4021 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4023 bond_for_each_slave(bond
, slave
, i
) {
4031 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4032 if (IS_UP(slave
->dev
) &&
4033 (slave
->link
== BOND_LINK_UP
) &&
4034 bond_is_active_slave(slave
)) {
4035 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4041 /* no suitable interface, frame not sent */
4045 return NETDEV_TX_OK
;
4049 * in broadcast mode, we send everything to all usable interfaces.
4051 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4053 struct bonding
*bond
= netdev_priv(bond_dev
);
4054 struct slave
*slave
, *start_at
;
4055 struct net_device
*tx_dev
= NULL
;
4059 read_lock(&bond
->curr_slave_lock
);
4060 start_at
= bond
->curr_active_slave
;
4061 read_unlock(&bond
->curr_slave_lock
);
4066 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4067 if (IS_UP(slave
->dev
) &&
4068 (slave
->link
== BOND_LINK_UP
) &&
4069 bond_is_active_slave(slave
)) {
4071 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4073 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4078 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4080 dev_kfree_skb(skb2
);
4084 tx_dev
= slave
->dev
;
4089 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4093 /* no suitable interface, frame not sent */
4096 /* frame sent to all suitable interfaces */
4097 return NETDEV_TX_OK
;
4100 /*------------------------- Device initialization ---------------------------*/
4102 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4104 switch (bond
->params
.xmit_policy
) {
4105 case BOND_XMIT_POLICY_LAYER23
:
4106 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4108 case BOND_XMIT_POLICY_LAYER34
:
4109 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4111 case BOND_XMIT_POLICY_LAYER2
:
4113 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4119 * Lookup the slave that corresponds to a qid
4121 static inline int bond_slave_override(struct bonding
*bond
,
4122 struct sk_buff
*skb
)
4125 struct slave
*slave
= NULL
;
4126 struct slave
*check_slave
;
4128 if (!skb
->queue_mapping
)
4131 /* Find out if any slaves have the same mapping as this skb. */
4132 bond_for_each_slave(bond
, check_slave
, i
) {
4133 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4134 slave
= check_slave
;
4139 /* If the slave isn't UP, use default transmit policy. */
4140 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4141 (slave
->link
== BOND_LINK_UP
)) {
4142 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4149 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4152 * This helper function exists to help dev_pick_tx get the correct
4153 * destination queue. Using a helper function skips a call to
4154 * skb_tx_hash and will put the skbs in the queue we expect on their
4155 * way down to the bonding driver.
4157 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
4160 * Save the original txq to restore before passing to the driver
4162 bond_queue_mapping(skb
) = skb
->queue_mapping
;
4164 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
4166 txq
-= dev
->real_num_tx_queues
;
4167 } while (txq
>= dev
->real_num_tx_queues
);
4172 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4174 struct bonding
*bond
= netdev_priv(dev
);
4176 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4177 if (!bond_slave_override(bond
, skb
))
4178 return NETDEV_TX_OK
;
4181 switch (bond
->params
.mode
) {
4182 case BOND_MODE_ROUNDROBIN
:
4183 return bond_xmit_roundrobin(skb
, dev
);
4184 case BOND_MODE_ACTIVEBACKUP
:
4185 return bond_xmit_activebackup(skb
, dev
);
4187 return bond_xmit_xor(skb
, dev
);
4188 case BOND_MODE_BROADCAST
:
4189 return bond_xmit_broadcast(skb
, dev
);
4190 case BOND_MODE_8023AD
:
4191 return bond_3ad_xmit_xor(skb
, dev
);
4194 return bond_alb_xmit(skb
, dev
);
4196 /* Should never happen, mode already checked */
4197 pr_err("%s: Error: Unknown bonding mode %d\n",
4198 dev
->name
, bond
->params
.mode
);
4201 return NETDEV_TX_OK
;
4205 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4207 struct bonding
*bond
= netdev_priv(dev
);
4208 netdev_tx_t ret
= NETDEV_TX_OK
;
4211 * If we risk deadlock from transmitting this in the
4212 * netpoll path, tell netpoll to queue the frame for later tx
4214 if (is_netpoll_tx_blocked(dev
))
4215 return NETDEV_TX_BUSY
;
4217 read_lock(&bond
->lock
);
4219 if (bond
->slave_cnt
)
4220 ret
= __bond_start_xmit(skb
, dev
);
4224 read_unlock(&bond
->lock
);
4230 * set bond mode specific net device operations
4232 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4234 struct net_device
*bond_dev
= bond
->dev
;
4237 case BOND_MODE_ROUNDROBIN
:
4239 case BOND_MODE_ACTIVEBACKUP
:
4242 bond_set_xmit_hash_policy(bond
);
4244 case BOND_MODE_BROADCAST
:
4246 case BOND_MODE_8023AD
:
4247 bond_set_xmit_hash_policy(bond
);
4254 /* Should never happen, mode already checked */
4255 pr_err("%s: Error: Unknown bonding mode %d\n",
4256 bond_dev
->name
, mode
);
4261 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4262 struct ethtool_drvinfo
*drvinfo
)
4264 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4265 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4266 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4269 static const struct ethtool_ops bond_ethtool_ops
= {
4270 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4271 .get_link
= ethtool_op_get_link
,
4274 static const struct net_device_ops bond_netdev_ops
= {
4275 .ndo_init
= bond_init
,
4276 .ndo_uninit
= bond_uninit
,
4277 .ndo_open
= bond_open
,
4278 .ndo_stop
= bond_close
,
4279 .ndo_start_xmit
= bond_start_xmit
,
4280 .ndo_select_queue
= bond_select_queue
,
4281 .ndo_get_stats64
= bond_get_stats
,
4282 .ndo_do_ioctl
= bond_do_ioctl
,
4283 .ndo_set_multicast_list
= bond_set_multicast_list
,
4284 .ndo_change_mtu
= bond_change_mtu
,
4285 .ndo_set_mac_address
= bond_set_mac_address
,
4286 .ndo_neigh_setup
= bond_neigh_setup
,
4287 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4288 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4289 #ifdef CONFIG_NET_POLL_CONTROLLER
4290 .ndo_netpoll_setup
= bond_netpoll_setup
,
4291 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4292 .ndo_poll_controller
= bond_poll_controller
,
4294 .ndo_add_slave
= bond_enslave
,
4295 .ndo_del_slave
= bond_release
,
4296 .ndo_fix_features
= bond_fix_features
,
4299 static void bond_destructor(struct net_device
*bond_dev
)
4301 struct bonding
*bond
= netdev_priv(bond_dev
);
4303 destroy_workqueue(bond
->wq
);
4304 free_netdev(bond_dev
);
4307 static void bond_setup(struct net_device
*bond_dev
)
4309 struct bonding
*bond
= netdev_priv(bond_dev
);
4311 /* initialize rwlocks */
4312 rwlock_init(&bond
->lock
);
4313 rwlock_init(&bond
->curr_slave_lock
);
4315 bond
->params
= bonding_defaults
;
4317 /* Initialize pointers */
4318 bond
->dev
= bond_dev
;
4319 INIT_LIST_HEAD(&bond
->vlan_list
);
4321 /* Initialize the device entry points */
4322 ether_setup(bond_dev
);
4323 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4324 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4325 bond_set_mode_ops(bond
, bond
->params
.mode
);
4327 bond_dev
->destructor
= bond_destructor
;
4329 /* Initialize the device options */
4330 bond_dev
->tx_queue_len
= 0;
4331 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4332 bond_dev
->priv_flags
|= IFF_BONDING
;
4333 bond_dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
4335 /* At first, we block adding VLANs. That's the only way to
4336 * prevent problems that occur when adding VLANs over an
4337 * empty bond. The block will be removed once non-challenged
4338 * slaves are enslaved.
4340 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4342 /* don't acquire bond device's netif_tx_lock when
4344 bond_dev
->features
|= NETIF_F_LLTX
;
4346 /* By default, we declare the bond to be fully
4347 * VLAN hardware accelerated capable. Special
4348 * care is taken in the various xmit functions
4349 * when there are slaves that are not hw accel
4353 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4354 NETIF_F_HW_VLAN_TX
|
4355 NETIF_F_HW_VLAN_RX
|
4356 NETIF_F_HW_VLAN_FILTER
;
4358 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_NO_CSUM
);
4359 bond_dev
->features
|= bond_dev
->hw_features
;
4362 static void bond_work_cancel_all(struct bonding
*bond
)
4364 write_lock_bh(&bond
->lock
);
4365 bond
->kill_timers
= 1;
4366 write_unlock_bh(&bond
->lock
);
4368 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4369 cancel_delayed_work(&bond
->mii_work
);
4371 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4372 cancel_delayed_work(&bond
->arp_work
);
4374 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4375 delayed_work_pending(&bond
->alb_work
))
4376 cancel_delayed_work(&bond
->alb_work
);
4378 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4379 delayed_work_pending(&bond
->ad_work
))
4380 cancel_delayed_work(&bond
->ad_work
);
4382 if (delayed_work_pending(&bond
->mcast_work
))
4383 cancel_delayed_work(&bond
->mcast_work
);
4387 * Destroy a bonding device.
4388 * Must be under rtnl_lock when this function is called.
4390 static void bond_uninit(struct net_device
*bond_dev
)
4392 struct bonding
*bond
= netdev_priv(bond_dev
);
4393 struct vlan_entry
*vlan
, *tmp
;
4395 bond_netpoll_cleanup(bond_dev
);
4397 /* Release the bonded slaves */
4398 bond_release_all(bond_dev
);
4400 list_del(&bond
->bond_list
);
4402 bond_work_cancel_all(bond
);
4404 bond_remove_proc_entry(bond
);
4406 bond_debug_unregister(bond
);
4408 __hw_addr_flush(&bond
->mc_list
);
4410 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4411 list_del(&vlan
->vlan_list
);
4416 /*------------------------- Module initialization ---------------------------*/
4419 * Convert string input module parms. Accept either the
4420 * number of the mode or its string name. A bit complicated because
4421 * some mode names are substrings of other names, and calls from sysfs
4422 * may have whitespace in the name (trailing newlines, for example).
4424 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4426 int modeint
= -1, i
, rv
;
4427 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4429 for (p
= (char *)buf
; *p
; p
++)
4430 if (!(isdigit(*p
) || isspace(*p
)))
4434 rv
= sscanf(buf
, "%20s", modestr
);
4436 rv
= sscanf(buf
, "%d", &modeint
);
4441 for (i
= 0; tbl
[i
].modename
; i
++) {
4442 if (modeint
== tbl
[i
].mode
)
4444 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4451 static int bond_check_params(struct bond_params
*params
)
4453 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4456 * Convert string parameters.
4459 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4460 if (bond_mode
== -1) {
4461 pr_err("Error: Invalid bonding mode \"%s\"\n",
4462 mode
== NULL
? "NULL" : mode
);
4467 if (xmit_hash_policy
) {
4468 if ((bond_mode
!= BOND_MODE_XOR
) &&
4469 (bond_mode
!= BOND_MODE_8023AD
)) {
4470 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4471 bond_mode_name(bond_mode
));
4473 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4475 if (xmit_hashtype
== -1) {
4476 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4477 xmit_hash_policy
== NULL
? "NULL" :
4485 if (bond_mode
!= BOND_MODE_8023AD
) {
4486 pr_info("lacp_rate param is irrelevant in mode %s\n",
4487 bond_mode_name(bond_mode
));
4489 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4490 if (lacp_fast
== -1) {
4491 pr_err("Error: Invalid lacp rate \"%s\"\n",
4492 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4499 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4500 if (params
->ad_select
== -1) {
4501 pr_err("Error: Invalid ad_select \"%s\"\n",
4502 ad_select
== NULL
? "NULL" : ad_select
);
4506 if (bond_mode
!= BOND_MODE_8023AD
) {
4507 pr_warning("ad_select param only affects 802.3ad mode\n");
4510 params
->ad_select
= BOND_AD_STABLE
;
4513 if (max_bonds
< 0) {
4514 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4515 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4516 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4520 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4521 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4522 miimon
= BOND_LINK_MON_INTERV
;
4526 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4531 if (downdelay
< 0) {
4532 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4533 downdelay
, INT_MAX
);
4537 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4538 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4543 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4544 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4549 /* reset values for 802.3ad */
4550 if (bond_mode
== BOND_MODE_8023AD
) {
4552 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");
4553 pr_warning("Forcing miimon to 100msec\n");
4558 if (tx_queues
< 1 || tx_queues
> 255) {
4559 pr_warning("Warning: tx_queues (%d) should be between "
4560 "1 and 255, resetting to %d\n",
4561 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4562 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4565 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4566 pr_warning("Warning: all_slaves_active module parameter (%d), "
4567 "not of valid value (0/1), so it was set to "
4568 "0\n", all_slaves_active
);
4569 all_slaves_active
= 0;
4572 if (resend_igmp
< 0 || resend_igmp
> 255) {
4573 pr_warning("Warning: resend_igmp (%d) should be between "
4574 "0 and 255, resetting to %d\n",
4575 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4576 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4579 /* reset values for TLB/ALB */
4580 if ((bond_mode
== BOND_MODE_TLB
) ||
4581 (bond_mode
== BOND_MODE_ALB
)) {
4583 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");
4584 pr_warning("Forcing miimon to 100msec\n");
4589 if (bond_mode
== BOND_MODE_ALB
) {
4590 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",
4595 if (updelay
|| downdelay
) {
4596 /* just warn the user the up/down delay will have
4597 * no effect since miimon is zero...
4599 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",
4600 updelay
, downdelay
);
4603 /* don't allow arp monitoring */
4605 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4606 miimon
, arp_interval
);
4610 if ((updelay
% miimon
) != 0) {
4611 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4613 (updelay
/ miimon
) * miimon
);
4618 if ((downdelay
% miimon
) != 0) {
4619 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4621 (downdelay
/ miimon
) * miimon
);
4624 downdelay
/= miimon
;
4627 if (arp_interval
< 0) {
4628 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4629 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4630 arp_interval
= BOND_LINK_ARP_INTERV
;
4633 for (arp_ip_count
= 0;
4634 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4636 /* not complete check, but should be good enough to
4638 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4639 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4640 arp_ip_target
[arp_ip_count
]);
4643 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4644 arp_target
[arp_ip_count
] = ip
;
4648 if (arp_interval
&& !arp_ip_count
) {
4649 /* don't allow arping if no arp_ip_target given... */
4650 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4656 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4657 pr_err("arp_validate only supported in active-backup mode\n");
4660 if (!arp_interval
) {
4661 pr_err("arp_validate requires arp_interval\n");
4665 arp_validate_value
= bond_parse_parm(arp_validate
,
4667 if (arp_validate_value
== -1) {
4668 pr_err("Error: invalid arp_validate \"%s\"\n",
4669 arp_validate
== NULL
? "NULL" : arp_validate
);
4673 arp_validate_value
= 0;
4676 pr_info("MII link monitoring set to %d ms\n", miimon
);
4677 } else if (arp_interval
) {
4680 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4682 arp_validate_tbl
[arp_validate_value
].modename
,
4685 for (i
= 0; i
< arp_ip_count
; i
++)
4686 pr_info(" %s", arp_ip_target
[i
]);
4690 } else if (max_bonds
) {
4691 /* miimon and arp_interval not set, we need one so things
4692 * work as expected, see bonding.txt for details
4694 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");
4697 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4698 /* currently, using a primary only makes sense
4699 * in active backup, TLB or ALB modes
4701 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4702 primary
, bond_mode_name(bond_mode
));
4706 if (primary
&& primary_reselect
) {
4707 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4709 if (primary_reselect_value
== -1) {
4710 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4712 NULL
? "NULL" : primary_reselect
);
4716 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4719 if (fail_over_mac
) {
4720 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4722 if (fail_over_mac_value
== -1) {
4723 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4724 arp_validate
== NULL
? "NULL" : arp_validate
);
4728 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4729 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4731 fail_over_mac_value
= BOND_FOM_NONE
;
4734 /* fill params struct with the proper values */
4735 params
->mode
= bond_mode
;
4736 params
->xmit_policy
= xmit_hashtype
;
4737 params
->miimon
= miimon
;
4738 params
->num_peer_notif
= num_peer_notif
;
4739 params
->arp_interval
= arp_interval
;
4740 params
->arp_validate
= arp_validate_value
;
4741 params
->updelay
= updelay
;
4742 params
->downdelay
= downdelay
;
4743 params
->use_carrier
= use_carrier
;
4744 params
->lacp_fast
= lacp_fast
;
4745 params
->primary
[0] = 0;
4746 params
->primary_reselect
= primary_reselect_value
;
4747 params
->fail_over_mac
= fail_over_mac_value
;
4748 params
->tx_queues
= tx_queues
;
4749 params
->all_slaves_active
= all_slaves_active
;
4750 params
->resend_igmp
= resend_igmp
;
4751 params
->min_links
= min_links
;
4754 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4755 params
->primary
[IFNAMSIZ
- 1] = 0;
4758 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4763 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4764 static struct lock_class_key bonding_netdev_addr_lock_key
;
4766 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4767 struct netdev_queue
*txq
,
4770 lockdep_set_class(&txq
->_xmit_lock
,
4771 &bonding_netdev_xmit_lock_key
);
4774 static void bond_set_lockdep_class(struct net_device
*dev
)
4776 lockdep_set_class(&dev
->addr_list_lock
,
4777 &bonding_netdev_addr_lock_key
);
4778 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4782 * Called from registration process
4784 static int bond_init(struct net_device
*bond_dev
)
4786 struct bonding
*bond
= netdev_priv(bond_dev
);
4787 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4788 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4790 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4793 * Initialize locks that may be required during
4794 * en/deslave operations. All of the bond_open work
4795 * (of which this is part) should really be moved to
4796 * a phase prior to dev_open
4798 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4799 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4801 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4805 bond_set_lockdep_class(bond_dev
);
4807 bond_create_proc_entry(bond
);
4808 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4810 bond_prepare_sysfs_group(bond
);
4812 bond_debug_register(bond
);
4814 __hw_addr_init(&bond
->mc_list
);
4818 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4820 if (tb
[IFLA_ADDRESS
]) {
4821 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4823 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4824 return -EADDRNOTAVAIL
;
4829 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
4831 .priv_size
= sizeof(struct bonding
),
4832 .setup
= bond_setup
,
4833 .validate
= bond_validate
,
4836 /* Create a new bond based on the specified name and bonding parameters.
4837 * If name is NULL, obtain a suitable "bond%d" name for us.
4838 * Caller must NOT hold rtnl_lock; we need to release it here before we
4839 * set up our sysfs entries.
4841 int bond_create(struct net
*net
, const char *name
)
4843 struct net_device
*bond_dev
;
4848 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4849 name
? name
: "bond%d",
4850 bond_setup
, tx_queues
);
4852 pr_err("%s: eek! can't alloc netdev!\n", name
);
4857 dev_net_set(bond_dev
, net
);
4858 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4860 res
= register_netdevice(bond_dev
);
4862 netif_carrier_off(bond_dev
);
4866 bond_destructor(bond_dev
);
4870 static int __net_init
bond_net_init(struct net
*net
)
4872 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4875 INIT_LIST_HEAD(&bn
->dev_list
);
4877 bond_create_proc_dir(bn
);
4882 static void __net_exit
bond_net_exit(struct net
*net
)
4884 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4886 bond_destroy_proc_dir(bn
);
4889 static struct pernet_operations bond_net_ops
= {
4890 .init
= bond_net_init
,
4891 .exit
= bond_net_exit
,
4893 .size
= sizeof(struct bond_net
),
4896 static int __init
bonding_init(void)
4901 pr_info("%s", bond_version
);
4903 res
= bond_check_params(&bonding_defaults
);
4907 res
= register_pernet_subsys(&bond_net_ops
);
4911 res
= rtnl_link_register(&bond_link_ops
);
4915 bond_create_debugfs();
4917 for (i
= 0; i
< max_bonds
; i
++) {
4918 res
= bond_create(&init_net
, NULL
);
4923 res
= bond_create_sysfs();
4927 register_netdevice_notifier(&bond_netdev_notifier
);
4928 register_inetaddr_notifier(&bond_inetaddr_notifier
);
4932 rtnl_link_unregister(&bond_link_ops
);
4934 unregister_pernet_subsys(&bond_net_ops
);
4939 static void __exit
bonding_exit(void)
4941 unregister_netdevice_notifier(&bond_netdev_notifier
);
4942 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
4944 bond_destroy_sysfs();
4945 bond_destroy_debugfs();
4947 rtnl_link_unregister(&bond_link_ops
);
4948 unregister_pernet_subsys(&bond_net_ops
);
4950 #ifdef CONFIG_NET_POLL_CONTROLLER
4952 * Make sure we don't have an imbalance on our netpoll blocking
4954 WARN_ON(atomic_read(&netpoll_block_tx
));
4958 module_init(bonding_init
);
4959 module_exit(bonding_exit
);
4960 MODULE_LICENSE("GPL");
4961 MODULE_VERSION(DRV_VERSION
);
4962 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4963 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4964 MODULE_ALIAS_RTNL_LINK("bond");