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
);
1562 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1565 netdev_bonding_change(bond_dev
,
1566 NETDEV_POST_TYPE_CHANGE
);
1568 } else if (bond_dev
->type
!= slave_dev
->type
) {
1569 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1571 slave_dev
->type
, bond_dev
->type
);
1573 goto err_undo_flags
;
1576 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1577 if (bond
->slave_cnt
== 0) {
1578 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1580 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1581 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1582 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",
1585 goto err_undo_flags
;
1589 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1591 /* If this is the first slave, then we need to set the master's hardware
1592 * address to be the same as the slave's. */
1593 if (is_zero_ether_addr(bond
->dev
->dev_addr
))
1594 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1595 slave_dev
->addr_len
);
1598 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1601 goto err_undo_flags
;
1605 * Set the new_slave's queue_id to be zero. Queue ID mapping
1606 * is set via sysfs or module option if desired.
1608 new_slave
->queue_id
= 0;
1610 /* Save slave's original mtu and then set it to match the bond */
1611 new_slave
->original_mtu
= slave_dev
->mtu
;
1612 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1614 pr_debug("Error %d calling dev_set_mtu\n", res
);
1619 * Save slave's original ("permanent") mac address for modes
1620 * that need it, and for restoring it upon release, and then
1621 * set it to the master's address
1623 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1625 if (!bond
->params
.fail_over_mac
) {
1627 * Set slave to master's mac address. The application already
1628 * set the master's mac address to that of the first slave
1630 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1631 addr
.sa_family
= slave_dev
->type
;
1632 res
= dev_set_mac_address(slave_dev
, &addr
);
1634 pr_debug("Error %d calling set_mac_address\n", res
);
1635 goto err_restore_mtu
;
1639 res
= netdev_set_bond_master(slave_dev
, bond_dev
);
1641 pr_debug("Error %d calling netdev_set_bond_master\n", res
);
1642 goto err_restore_mac
;
1645 /* open the slave since the application closed it */
1646 res
= dev_open(slave_dev
);
1648 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1649 goto err_unset_master
;
1652 new_slave
->bond
= bond
;
1653 new_slave
->dev
= slave_dev
;
1654 slave_dev
->priv_flags
|= IFF_BONDING
;
1656 if (bond_is_lb(bond
)) {
1657 /* bond_alb_init_slave() must be called before all other stages since
1658 * it might fail and we do not want to have to undo everything
1660 res
= bond_alb_init_slave(bond
, new_slave
);
1665 /* If the mode USES_PRIMARY, then the new slave gets the
1666 * master's promisc (and mc) settings only if it becomes the
1667 * curr_active_slave, and that is taken care of later when calling
1668 * bond_change_active()
1670 if (!USES_PRIMARY(bond
->params
.mode
)) {
1671 /* set promiscuity level to new slave */
1672 if (bond_dev
->flags
& IFF_PROMISC
) {
1673 res
= dev_set_promiscuity(slave_dev
, 1);
1678 /* set allmulti level to new slave */
1679 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1680 res
= dev_set_allmulti(slave_dev
, 1);
1685 netif_addr_lock_bh(bond_dev
);
1686 /* upload master's mc_list to new slave */
1687 netdev_for_each_mc_addr(ha
, bond_dev
)
1688 dev_mc_add(slave_dev
, ha
->addr
);
1689 netif_addr_unlock_bh(bond_dev
);
1692 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1693 /* add lacpdu mc addr to mc list */
1694 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1696 dev_mc_add(slave_dev
, lacpdu_multicast
);
1699 bond_add_vlans_on_slave(bond
, slave_dev
);
1701 write_lock_bh(&bond
->lock
);
1703 bond_attach_slave(bond
, new_slave
);
1705 new_slave
->delay
= 0;
1706 new_slave
->link_failure_count
= 0;
1708 write_unlock_bh(&bond
->lock
);
1710 bond_compute_features(bond
);
1712 read_lock(&bond
->lock
);
1714 new_slave
->last_arp_rx
= jiffies
;
1716 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1717 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1719 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1721 * miimon is set but a bonded network driver
1722 * does not support ETHTOOL/MII and
1723 * arp_interval is not set. Note: if
1724 * use_carrier is enabled, we will never go
1725 * here (because netif_carrier is always
1726 * supported); thus, we don't need to change
1727 * the messages for netif_carrier.
1729 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",
1730 bond_dev
->name
, slave_dev
->name
);
1731 } else if (link_reporting
== -1) {
1732 /* unable get link status using mii/ethtool */
1733 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",
1734 bond_dev
->name
, slave_dev
->name
);
1738 /* check for initial state */
1739 if (!bond
->params
.miimon
||
1740 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1741 if (bond
->params
.updelay
) {
1742 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1743 new_slave
->link
= BOND_LINK_BACK
;
1744 new_slave
->delay
= bond
->params
.updelay
;
1746 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1747 new_slave
->link
= BOND_LINK_UP
;
1749 new_slave
->jiffies
= jiffies
;
1751 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1752 new_slave
->link
= BOND_LINK_DOWN
;
1755 if (bond_update_speed_duplex(new_slave
) &&
1756 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1757 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1758 bond_dev
->name
, new_slave
->dev
->name
);
1760 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1761 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1766 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1767 /* if there is a primary slave, remember it */
1768 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1769 bond
->primary_slave
= new_slave
;
1770 bond
->force_primary
= true;
1774 write_lock_bh(&bond
->curr_slave_lock
);
1776 switch (bond
->params
.mode
) {
1777 case BOND_MODE_ACTIVEBACKUP
:
1778 bond_set_slave_inactive_flags(new_slave
);
1779 bond_select_active_slave(bond
);
1781 case BOND_MODE_8023AD
:
1782 /* in 802.3ad mode, the internal mechanism
1783 * will activate the slaves in the selected
1786 bond_set_slave_inactive_flags(new_slave
);
1787 /* if this is the first slave */
1788 if (bond
->slave_cnt
== 1) {
1789 SLAVE_AD_INFO(new_slave
).id
= 1;
1790 /* Initialize AD with the number of times that the AD timer is called in 1 second
1791 * can be called only after the mac address of the bond is set
1793 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1795 SLAVE_AD_INFO(new_slave
).id
=
1796 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1799 bond_3ad_bind_slave(new_slave
);
1803 bond_set_active_slave(new_slave
);
1804 bond_set_slave_inactive_flags(new_slave
);
1805 bond_select_active_slave(bond
);
1808 pr_debug("This slave is always active in trunk mode\n");
1810 /* always active in trunk mode */
1811 bond_set_active_slave(new_slave
);
1813 /* In trunking mode there is little meaning to curr_active_slave
1814 * anyway (it holds no special properties of the bond device),
1815 * so we can change it without calling change_active_interface()
1817 if (!bond
->curr_active_slave
)
1818 bond
->curr_active_slave
= new_slave
;
1821 } /* switch(bond_mode) */
1823 write_unlock_bh(&bond
->curr_slave_lock
);
1825 bond_set_carrier(bond
);
1827 #ifdef CONFIG_NET_POLL_CONTROLLER
1828 slave_dev
->npinfo
= bond_netpoll_info(bond
);
1829 if (slave_dev
->npinfo
) {
1830 if (slave_enable_netpoll(new_slave
)) {
1831 read_unlock(&bond
->lock
);
1832 pr_info("Error, %s: master_dev is using netpoll, "
1833 "but new slave device does not support netpoll.\n",
1841 read_unlock(&bond
->lock
);
1843 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1847 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1850 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1851 goto err_dest_symlinks
;
1854 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1855 bond_dev
->name
, slave_dev
->name
,
1856 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1857 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1859 /* enslave is successful */
1862 /* Undo stages on error */
1864 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1867 dev_close(slave_dev
);
1870 netdev_set_bond_master(slave_dev
, NULL
);
1873 if (!bond
->params
.fail_over_mac
) {
1874 /* XXX TODO - fom follow mode needs to change master's
1875 * MAC if this slave's MAC is in use by the bond, or at
1876 * least print a warning.
1878 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1879 addr
.sa_family
= slave_dev
->type
;
1880 dev_set_mac_address(slave_dev
, &addr
);
1884 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1890 bond_compute_features(bond
);
1896 * Try to release the slave device <slave> from the bond device <master>
1897 * It is legal to access curr_active_slave without a lock because all the function
1900 * The rules for slave state should be:
1901 * for Active/Backup:
1902 * Active stays on all backups go down
1903 * for Bonded connections:
1904 * The first up interface should be left on and all others downed.
1906 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1908 struct bonding
*bond
= netdev_priv(bond_dev
);
1909 struct slave
*slave
, *oldcurrent
;
1910 struct sockaddr addr
;
1911 u32 old_features
= bond_dev
->features
;
1913 /* slave is not a slave or master is not master of this slave */
1914 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1915 (slave_dev
->master
!= bond_dev
)) {
1916 pr_err("%s: Error: cannot release %s.\n",
1917 bond_dev
->name
, slave_dev
->name
);
1922 netdev_bonding_change(bond_dev
, NETDEV_RELEASE
);
1923 write_lock_bh(&bond
->lock
);
1925 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1927 /* not a slave of this bond */
1928 pr_info("%s: %s not enslaved\n",
1929 bond_dev
->name
, slave_dev
->name
);
1930 write_unlock_bh(&bond
->lock
);
1931 unblock_netpoll_tx();
1935 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1936 * for this slave anymore.
1938 netdev_rx_handler_unregister(slave_dev
);
1939 write_unlock_bh(&bond
->lock
);
1941 write_lock_bh(&bond
->lock
);
1943 if (!bond
->params
.fail_over_mac
) {
1944 if (!compare_ether_addr(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1945 bond
->slave_cnt
> 1)
1946 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",
1947 bond_dev
->name
, slave_dev
->name
,
1949 bond_dev
->name
, slave_dev
->name
);
1952 /* Inform AD package of unbinding of slave. */
1953 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1954 /* must be called before the slave is
1955 * detached from the list
1957 bond_3ad_unbind_slave(slave
);
1960 pr_info("%s: releasing %s interface %s\n",
1962 bond_is_active_slave(slave
) ? "active" : "backup",
1965 oldcurrent
= bond
->curr_active_slave
;
1967 bond
->current_arp_slave
= NULL
;
1969 /* release the slave from its bond */
1970 bond_detach_slave(bond
, slave
);
1972 if (bond
->primary_slave
== slave
)
1973 bond
->primary_slave
= NULL
;
1975 if (oldcurrent
== slave
)
1976 bond_change_active_slave(bond
, NULL
);
1978 if (bond_is_lb(bond
)) {
1979 /* Must be called only after the slave has been
1980 * detached from the list and the curr_active_slave
1981 * has been cleared (if our_slave == old_current),
1982 * but before a new active slave is selected.
1984 write_unlock_bh(&bond
->lock
);
1985 bond_alb_deinit_slave(bond
, slave
);
1986 write_lock_bh(&bond
->lock
);
1989 if (oldcurrent
== slave
) {
1991 * Note that we hold RTNL over this sequence, so there
1992 * is no concern that another slave add/remove event
1995 write_unlock_bh(&bond
->lock
);
1996 read_lock(&bond
->lock
);
1997 write_lock_bh(&bond
->curr_slave_lock
);
1999 bond_select_active_slave(bond
);
2001 write_unlock_bh(&bond
->curr_slave_lock
);
2002 read_unlock(&bond
->lock
);
2003 write_lock_bh(&bond
->lock
);
2006 if (bond
->slave_cnt
== 0) {
2007 bond_set_carrier(bond
);
2009 /* if the last slave was removed, zero the mac address
2010 * of the master so it will be set by the application
2011 * to the mac address of the first slave
2013 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2015 if (bond_vlan_used(bond
)) {
2016 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2017 bond_dev
->name
, bond_dev
->name
);
2018 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2023 write_unlock_bh(&bond
->lock
);
2024 unblock_netpoll_tx();
2026 bond_compute_features(bond
);
2027 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
2028 (old_features
& NETIF_F_VLAN_CHALLENGED
))
2029 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2030 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
2032 /* must do this from outside any spinlocks */
2033 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2035 bond_del_vlans_from_slave(bond
, slave_dev
);
2037 /* If the mode USES_PRIMARY, then we should only remove its
2038 * promisc and mc settings if it was the curr_active_slave, but that was
2039 * already taken care of above when we detached the slave
2041 if (!USES_PRIMARY(bond
->params
.mode
)) {
2042 /* unset promiscuity level from slave */
2043 if (bond_dev
->flags
& IFF_PROMISC
)
2044 dev_set_promiscuity(slave_dev
, -1);
2046 /* unset allmulti level from slave */
2047 if (bond_dev
->flags
& IFF_ALLMULTI
)
2048 dev_set_allmulti(slave_dev
, -1);
2050 /* flush master's mc_list from slave */
2051 netif_addr_lock_bh(bond_dev
);
2052 bond_mc_list_flush(bond_dev
, slave_dev
);
2053 netif_addr_unlock_bh(bond_dev
);
2056 netdev_set_bond_master(slave_dev
, NULL
);
2058 slave_disable_netpoll(slave
);
2060 /* close slave before restoring its mac address */
2061 dev_close(slave_dev
);
2063 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2064 /* restore original ("permanent") mac address */
2065 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2066 addr
.sa_family
= slave_dev
->type
;
2067 dev_set_mac_address(slave_dev
, &addr
);
2070 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2072 slave_dev
->priv_flags
&= ~IFF_BONDING
;
2076 return 0; /* deletion OK */
2080 * First release a slave and then destroy the bond if no more slaves are left.
2081 * Must be under rtnl_lock when this function is called.
2083 static int bond_release_and_destroy(struct net_device
*bond_dev
,
2084 struct net_device
*slave_dev
)
2086 struct bonding
*bond
= netdev_priv(bond_dev
);
2089 ret
= bond_release(bond_dev
, slave_dev
);
2090 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2091 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
2092 pr_info("%s: destroying bond %s.\n",
2093 bond_dev
->name
, bond_dev
->name
);
2094 unregister_netdevice(bond_dev
);
2100 * This function releases all slaves.
2102 static int bond_release_all(struct net_device
*bond_dev
)
2104 struct bonding
*bond
= netdev_priv(bond_dev
);
2105 struct slave
*slave
;
2106 struct net_device
*slave_dev
;
2107 struct sockaddr addr
;
2109 write_lock_bh(&bond
->lock
);
2111 netif_carrier_off(bond_dev
);
2113 if (bond
->slave_cnt
== 0)
2116 bond
->current_arp_slave
= NULL
;
2117 bond
->primary_slave
= NULL
;
2118 bond_change_active_slave(bond
, NULL
);
2120 while ((slave
= bond
->first_slave
) != NULL
) {
2121 /* Inform AD package of unbinding of slave
2122 * before slave is detached from the list.
2124 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2125 bond_3ad_unbind_slave(slave
);
2127 slave_dev
= slave
->dev
;
2128 bond_detach_slave(bond
, slave
);
2130 /* now that the slave is detached, unlock and perform
2131 * all the undo steps that should not be called from
2134 write_unlock_bh(&bond
->lock
);
2136 /* unregister rx_handler early so bond_handle_frame wouldn't
2137 * be called for this slave anymore.
2139 netdev_rx_handler_unregister(slave_dev
);
2142 if (bond_is_lb(bond
)) {
2143 /* must be called only after the slave
2144 * has been detached from the list
2146 bond_alb_deinit_slave(bond
, slave
);
2149 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2150 bond_del_vlans_from_slave(bond
, slave_dev
);
2152 /* If the mode USES_PRIMARY, then we should only remove its
2153 * promisc and mc settings if it was the curr_active_slave, but that was
2154 * already taken care of above when we detached the slave
2156 if (!USES_PRIMARY(bond
->params
.mode
)) {
2157 /* unset promiscuity level from slave */
2158 if (bond_dev
->flags
& IFF_PROMISC
)
2159 dev_set_promiscuity(slave_dev
, -1);
2161 /* unset allmulti level from slave */
2162 if (bond_dev
->flags
& IFF_ALLMULTI
)
2163 dev_set_allmulti(slave_dev
, -1);
2165 /* flush master's mc_list from slave */
2166 netif_addr_lock_bh(bond_dev
);
2167 bond_mc_list_flush(bond_dev
, slave_dev
);
2168 netif_addr_unlock_bh(bond_dev
);
2171 netdev_set_bond_master(slave_dev
, NULL
);
2173 slave_disable_netpoll(slave
);
2175 /* close slave before restoring its mac address */
2176 dev_close(slave_dev
);
2178 if (!bond
->params
.fail_over_mac
) {
2179 /* restore original ("permanent") mac address*/
2180 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2181 addr
.sa_family
= slave_dev
->type
;
2182 dev_set_mac_address(slave_dev
, &addr
);
2187 /* re-acquire the lock before getting the next slave */
2188 write_lock_bh(&bond
->lock
);
2191 /* zero the mac address of the master so it will be
2192 * set by the application to the mac address of the
2195 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2197 if (bond_vlan_used(bond
)) {
2198 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2199 bond_dev
->name
, bond_dev
->name
);
2200 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2204 pr_info("%s: released all slaves\n", bond_dev
->name
);
2207 write_unlock_bh(&bond
->lock
);
2209 bond_compute_features(bond
);
2215 * This function changes the active slave to slave <slave_dev>.
2216 * It returns -EINVAL in the following cases.
2217 * - <slave_dev> is not found in the list.
2218 * - There is not active slave now.
2219 * - <slave_dev> is already active.
2220 * - The link state of <slave_dev> is not BOND_LINK_UP.
2221 * - <slave_dev> is not running.
2222 * In these cases, this function does nothing.
2223 * In the other cases, current_slave pointer is changed and 0 is returned.
2225 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2227 struct bonding
*bond
= netdev_priv(bond_dev
);
2228 struct slave
*old_active
= NULL
;
2229 struct slave
*new_active
= NULL
;
2232 if (!USES_PRIMARY(bond
->params
.mode
))
2235 /* Verify that master_dev is indeed the master of slave_dev */
2236 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2239 read_lock(&bond
->lock
);
2241 read_lock(&bond
->curr_slave_lock
);
2242 old_active
= bond
->curr_active_slave
;
2243 read_unlock(&bond
->curr_slave_lock
);
2245 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2248 * Changing to the current active: do nothing; return success.
2250 if (new_active
&& (new_active
== old_active
)) {
2251 read_unlock(&bond
->lock
);
2257 (new_active
->link
== BOND_LINK_UP
) &&
2258 IS_UP(new_active
->dev
)) {
2260 write_lock_bh(&bond
->curr_slave_lock
);
2261 bond_change_active_slave(bond
, new_active
);
2262 write_unlock_bh(&bond
->curr_slave_lock
);
2263 unblock_netpoll_tx();
2267 read_unlock(&bond
->lock
);
2272 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2274 struct bonding
*bond
= netdev_priv(bond_dev
);
2276 info
->bond_mode
= bond
->params
.mode
;
2277 info
->miimon
= bond
->params
.miimon
;
2279 read_lock(&bond
->lock
);
2280 info
->num_slaves
= bond
->slave_cnt
;
2281 read_unlock(&bond
->lock
);
2286 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2288 struct bonding
*bond
= netdev_priv(bond_dev
);
2289 struct slave
*slave
;
2290 int i
, res
= -ENODEV
;
2292 read_lock(&bond
->lock
);
2294 bond_for_each_slave(bond
, slave
, i
) {
2295 if (i
== (int)info
->slave_id
) {
2297 strcpy(info
->slave_name
, slave
->dev
->name
);
2298 info
->link
= slave
->link
;
2299 info
->state
= bond_slave_state(slave
);
2300 info
->link_failure_count
= slave
->link_failure_count
;
2305 read_unlock(&bond
->lock
);
2310 /*-------------------------------- Monitoring -------------------------------*/
2313 static int bond_miimon_inspect(struct bonding
*bond
)
2315 struct slave
*slave
;
2316 int i
, link_state
, commit
= 0;
2317 bool ignore_updelay
;
2319 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2321 bond_for_each_slave(bond
, slave
, i
) {
2322 slave
->new_link
= BOND_LINK_NOCHANGE
;
2324 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2326 switch (slave
->link
) {
2331 slave
->link
= BOND_LINK_FAIL
;
2332 slave
->delay
= bond
->params
.downdelay
;
2334 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2336 (bond
->params
.mode
==
2337 BOND_MODE_ACTIVEBACKUP
) ?
2338 (bond_is_active_slave(slave
) ?
2339 "active " : "backup ") : "",
2341 bond
->params
.downdelay
* bond
->params
.miimon
);
2344 case BOND_LINK_FAIL
:
2347 * recovered before downdelay expired
2349 slave
->link
= BOND_LINK_UP
;
2350 slave
->jiffies
= jiffies
;
2351 pr_info("%s: link status up again after %d ms for interface %s.\n",
2353 (bond
->params
.downdelay
- slave
->delay
) *
2354 bond
->params
.miimon
,
2359 if (slave
->delay
<= 0) {
2360 slave
->new_link
= BOND_LINK_DOWN
;
2368 case BOND_LINK_DOWN
:
2372 slave
->link
= BOND_LINK_BACK
;
2373 slave
->delay
= bond
->params
.updelay
;
2376 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2377 bond
->dev
->name
, slave
->dev
->name
,
2378 ignore_updelay
? 0 :
2379 bond
->params
.updelay
*
2380 bond
->params
.miimon
);
2383 case BOND_LINK_BACK
:
2385 slave
->link
= BOND_LINK_DOWN
;
2386 pr_info("%s: link status down again after %d ms for interface %s.\n",
2388 (bond
->params
.updelay
- slave
->delay
) *
2389 bond
->params
.miimon
,
2398 if (slave
->delay
<= 0) {
2399 slave
->new_link
= BOND_LINK_UP
;
2401 ignore_updelay
= false;
2413 static void bond_miimon_commit(struct bonding
*bond
)
2415 struct slave
*slave
;
2418 bond_for_each_slave(bond
, slave
, i
) {
2419 switch (slave
->new_link
) {
2420 case BOND_LINK_NOCHANGE
:
2424 slave
->link
= BOND_LINK_UP
;
2425 slave
->jiffies
= jiffies
;
2427 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2428 /* prevent it from being the active one */
2429 bond_set_backup_slave(slave
);
2430 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2431 /* make it immediately active */
2432 bond_set_active_slave(slave
);
2433 } else if (slave
!= bond
->primary_slave
) {
2434 /* prevent it from being the active one */
2435 bond_set_backup_slave(slave
);
2438 bond_update_speed_duplex(slave
);
2440 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2441 bond
->dev
->name
, slave
->dev
->name
,
2442 slave
->speed
, slave
->duplex
? "full" : "half");
2444 /* notify ad that the link status has changed */
2445 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2446 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2448 if (bond_is_lb(bond
))
2449 bond_alb_handle_link_change(bond
, slave
,
2452 if (!bond
->curr_active_slave
||
2453 (slave
== bond
->primary_slave
))
2458 case BOND_LINK_DOWN
:
2459 if (slave
->link_failure_count
< UINT_MAX
)
2460 slave
->link_failure_count
++;
2462 slave
->link
= BOND_LINK_DOWN
;
2464 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2465 bond
->params
.mode
== BOND_MODE_8023AD
)
2466 bond_set_slave_inactive_flags(slave
);
2468 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2469 bond
->dev
->name
, slave
->dev
->name
);
2471 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2472 bond_3ad_handle_link_change(slave
,
2475 if (bond_is_lb(bond
))
2476 bond_alb_handle_link_change(bond
, slave
,
2479 if (slave
== bond
->curr_active_slave
)
2485 pr_err("%s: invalid new link %d on slave %s\n",
2486 bond
->dev
->name
, slave
->new_link
,
2488 slave
->new_link
= BOND_LINK_NOCHANGE
;
2496 write_lock_bh(&bond
->curr_slave_lock
);
2497 bond_select_active_slave(bond
);
2498 write_unlock_bh(&bond
->curr_slave_lock
);
2499 unblock_netpoll_tx();
2502 bond_set_carrier(bond
);
2508 * Really a wrapper that splits the mii monitor into two phases: an
2509 * inspection, then (if inspection indicates something needs to be done)
2510 * an acquisition of appropriate locks followed by a commit phase to
2511 * implement whatever link state changes are indicated.
2513 void bond_mii_monitor(struct work_struct
*work
)
2515 struct bonding
*bond
= container_of(work
, struct bonding
,
2517 bool should_notify_peers
= false;
2519 read_lock(&bond
->lock
);
2520 if (bond
->kill_timers
)
2523 if (bond
->slave_cnt
== 0)
2526 should_notify_peers
= bond_should_notify_peers(bond
);
2528 if (bond_miimon_inspect(bond
)) {
2529 read_unlock(&bond
->lock
);
2531 read_lock(&bond
->lock
);
2533 bond_miimon_commit(bond
);
2535 read_unlock(&bond
->lock
);
2536 rtnl_unlock(); /* might sleep, hold no other locks */
2537 read_lock(&bond
->lock
);
2541 if (bond
->params
.miimon
)
2542 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2543 msecs_to_jiffies(bond
->params
.miimon
));
2545 read_unlock(&bond
->lock
);
2547 if (should_notify_peers
) {
2549 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
2554 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2556 struct in_device
*idev
;
2557 struct in_ifaddr
*ifa
;
2564 idev
= __in_dev_get_rcu(dev
);
2568 ifa
= idev
->ifa_list
;
2572 addr
= ifa
->ifa_local
;
2578 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2580 struct vlan_entry
*vlan
;
2582 if (ip
== bond
->master_ip
)
2585 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2586 if (ip
== vlan
->vlan_ip
)
2594 * We go to the (large) trouble of VLAN tagging ARP frames because
2595 * switches in VLAN mode (especially if ports are configured as
2596 * "native" to a VLAN) might not pass non-tagged frames.
2598 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2600 struct sk_buff
*skb
;
2602 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2603 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2605 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2606 NULL
, slave_dev
->dev_addr
, NULL
);
2609 pr_err("ARP packet allocation failed\n");
2613 skb
= vlan_put_tag(skb
, vlan_id
);
2615 pr_err("failed to insert VLAN tag\n");
2623 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2626 __be32
*targets
= bond
->params
.arp_targets
;
2627 struct vlan_entry
*vlan
;
2628 struct net_device
*vlan_dev
;
2631 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2634 pr_debug("basa: target %x\n", targets
[i
]);
2635 if (!bond_vlan_used(bond
)) {
2636 pr_debug("basa: empty vlan: arp_send\n");
2637 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2638 bond
->master_ip
, 0);
2643 * If VLANs are configured, we do a route lookup to
2644 * determine which VLAN interface would be used, so we
2645 * can tag the ARP with the proper VLAN tag.
2647 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2650 if (net_ratelimit()) {
2651 pr_warning("%s: no route to arp_ip_target %pI4\n",
2652 bond
->dev
->name
, &targets
[i
]);
2658 * This target is not on a VLAN
2660 if (rt
->dst
.dev
== bond
->dev
) {
2662 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2663 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2664 bond
->master_ip
, 0);
2669 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2671 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
2674 if (vlan_dev
== rt
->dst
.dev
) {
2675 vlan_id
= vlan
->vlan_id
;
2676 pr_debug("basa: vlan match on %s %d\n",
2677 vlan_dev
->name
, vlan_id
);
2684 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2685 vlan
->vlan_ip
, vlan_id
);
2689 if (net_ratelimit()) {
2690 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2691 bond
->dev
->name
, &targets
[i
],
2692 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2698 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2701 __be32
*targets
= bond
->params
.arp_targets
;
2703 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2704 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2705 &sip
, &tip
, i
, &targets
[i
],
2706 bond_has_this_ip(bond
, tip
));
2707 if (sip
== targets
[i
]) {
2708 if (bond_has_this_ip(bond
, tip
))
2709 slave
->last_arp_rx
= jiffies
;
2715 static void bond_arp_rcv(struct sk_buff
*skb
, struct bonding
*bond
,
2716 struct slave
*slave
)
2719 unsigned char *arp_ptr
;
2722 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2725 read_lock(&bond
->lock
);
2727 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2728 bond
->dev
->name
, skb
->dev
->name
);
2730 if (!pskb_may_pull(skb
, arp_hdr_len(bond
->dev
)))
2734 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2735 skb
->pkt_type
== PACKET_OTHERHOST
||
2736 skb
->pkt_type
== PACKET_LOOPBACK
||
2737 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2738 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2742 arp_ptr
= (unsigned char *)(arp
+ 1);
2743 arp_ptr
+= bond
->dev
->addr_len
;
2744 memcpy(&sip
, arp_ptr
, 4);
2745 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2746 memcpy(&tip
, arp_ptr
, 4);
2748 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2749 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2750 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2754 * Backup slaves won't see the ARP reply, but do come through
2755 * here for each ARP probe (so we swap the sip/tip to validate
2756 * the probe). In a "redundant switch, common router" type of
2757 * configuration, the ARP probe will (hopefully) travel from
2758 * the active, through one switch, the router, then the other
2759 * switch before reaching the backup.
2761 if (bond_is_active_slave(slave
))
2762 bond_validate_arp(bond
, slave
, sip
, tip
);
2764 bond_validate_arp(bond
, slave
, tip
, sip
);
2767 read_unlock(&bond
->lock
);
2771 * this function is called regularly to monitor each slave's link
2772 * ensuring that traffic is being sent and received when arp monitoring
2773 * is used in load-balancing mode. if the adapter has been dormant, then an
2774 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2775 * arp monitoring in active backup mode.
2777 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2779 struct bonding
*bond
= container_of(work
, struct bonding
,
2781 struct slave
*slave
, *oldcurrent
;
2782 int do_failover
= 0;
2786 read_lock(&bond
->lock
);
2788 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2790 if (bond
->kill_timers
)
2793 if (bond
->slave_cnt
== 0)
2796 read_lock(&bond
->curr_slave_lock
);
2797 oldcurrent
= bond
->curr_active_slave
;
2798 read_unlock(&bond
->curr_slave_lock
);
2800 /* see if any of the previous devices are up now (i.e. they have
2801 * xmt and rcv traffic). the curr_active_slave does not come into
2802 * the picture unless it is null. also, slave->jiffies is not needed
2803 * here because we send an arp on each slave and give a slave as
2804 * long as it needs to get the tx/rx within the delta.
2805 * TODO: what about up/down delay in arp mode? it wasn't here before
2808 bond_for_each_slave(bond
, slave
, i
) {
2809 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2811 if (slave
->link
!= BOND_LINK_UP
) {
2812 if (time_in_range(jiffies
,
2813 trans_start
- delta_in_ticks
,
2814 trans_start
+ delta_in_ticks
) &&
2815 time_in_range(jiffies
,
2816 slave
->dev
->last_rx
- delta_in_ticks
,
2817 slave
->dev
->last_rx
+ delta_in_ticks
)) {
2819 slave
->link
= BOND_LINK_UP
;
2820 bond_set_active_slave(slave
);
2822 /* primary_slave has no meaning in round-robin
2823 * mode. the window of a slave being up and
2824 * curr_active_slave being null after enslaving
2828 pr_info("%s: link status definitely up for interface %s, ",
2833 pr_info("%s: interface %s is now up\n",
2839 /* slave->link == BOND_LINK_UP */
2841 /* not all switches will respond to an arp request
2842 * when the source ip is 0, so don't take the link down
2843 * if we don't know our ip yet
2845 if (!time_in_range(jiffies
,
2846 trans_start
- delta_in_ticks
,
2847 trans_start
+ 2 * delta_in_ticks
) ||
2848 !time_in_range(jiffies
,
2849 slave
->dev
->last_rx
- delta_in_ticks
,
2850 slave
->dev
->last_rx
+ 2 * delta_in_ticks
)) {
2852 slave
->link
= BOND_LINK_DOWN
;
2853 bond_set_backup_slave(slave
);
2855 if (slave
->link_failure_count
< UINT_MAX
)
2856 slave
->link_failure_count
++;
2858 pr_info("%s: interface %s is now down.\n",
2862 if (slave
== oldcurrent
)
2867 /* note: if switch is in round-robin mode, all links
2868 * must tx arp to ensure all links rx an arp - otherwise
2869 * links may oscillate or not come up at all; if switch is
2870 * in something like xor mode, there is nothing we can
2871 * do - all replies will be rx'ed on same link causing slaves
2872 * to be unstable during low/no traffic periods
2874 if (IS_UP(slave
->dev
))
2875 bond_arp_send_all(bond
, slave
);
2880 write_lock_bh(&bond
->curr_slave_lock
);
2882 bond_select_active_slave(bond
);
2884 write_unlock_bh(&bond
->curr_slave_lock
);
2885 unblock_netpoll_tx();
2889 if (bond
->params
.arp_interval
)
2890 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2892 read_unlock(&bond
->lock
);
2896 * Called to inspect slaves for active-backup mode ARP monitor link state
2897 * changes. Sets new_link in slaves to specify what action should take
2898 * place for the slave. Returns 0 if no changes are found, >0 if changes
2899 * to link states must be committed.
2901 * Called with bond->lock held for read.
2903 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2905 struct slave
*slave
;
2907 unsigned long trans_start
;
2909 bond_for_each_slave(bond
, slave
, i
) {
2910 slave
->new_link
= BOND_LINK_NOCHANGE
;
2912 if (slave
->link
!= BOND_LINK_UP
) {
2913 if (time_in_range(jiffies
,
2914 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2915 slave_last_rx(bond
, slave
) + delta_in_ticks
)) {
2917 slave
->new_link
= BOND_LINK_UP
;
2925 * Give slaves 2*delta after being enslaved or made
2926 * active. This avoids bouncing, as the last receive
2927 * times need a full ARP monitor cycle to be updated.
2929 if (time_in_range(jiffies
,
2930 slave
->jiffies
- delta_in_ticks
,
2931 slave
->jiffies
+ 2 * delta_in_ticks
))
2935 * Backup slave is down if:
2936 * - No current_arp_slave AND
2937 * - more than 3*delta since last receive AND
2938 * - the bond has an IP address
2940 * Note: a non-null current_arp_slave indicates
2941 * the curr_active_slave went down and we are
2942 * searching for a new one; under this condition
2943 * we only take the curr_active_slave down - this
2944 * gives each slave a chance to tx/rx traffic
2945 * before being taken out
2947 if (!bond_is_active_slave(slave
) &&
2948 !bond
->current_arp_slave
&&
2949 !time_in_range(jiffies
,
2950 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2951 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
)) {
2953 slave
->new_link
= BOND_LINK_DOWN
;
2958 * Active slave is down if:
2959 * - more than 2*delta since transmitting OR
2960 * - (more than 2*delta since receive AND
2961 * the bond has an IP address)
2963 trans_start
= dev_trans_start(slave
->dev
);
2964 if (bond_is_active_slave(slave
) &&
2965 (!time_in_range(jiffies
,
2966 trans_start
- delta_in_ticks
,
2967 trans_start
+ 2 * delta_in_ticks
) ||
2968 !time_in_range(jiffies
,
2969 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2970 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
))) {
2972 slave
->new_link
= BOND_LINK_DOWN
;
2981 * Called to commit link state changes noted by inspection step of
2982 * active-backup mode ARP monitor.
2984 * Called with RTNL and bond->lock for read.
2986 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2988 struct slave
*slave
;
2990 unsigned long trans_start
;
2992 bond_for_each_slave(bond
, slave
, i
) {
2993 switch (slave
->new_link
) {
2994 case BOND_LINK_NOCHANGE
:
2998 trans_start
= dev_trans_start(slave
->dev
);
2999 if ((!bond
->curr_active_slave
&&
3000 time_in_range(jiffies
,
3001 trans_start
- delta_in_ticks
,
3002 trans_start
+ delta_in_ticks
)) ||
3003 bond
->curr_active_slave
!= slave
) {
3004 slave
->link
= BOND_LINK_UP
;
3005 bond
->current_arp_slave
= NULL
;
3007 pr_info("%s: link status definitely up for interface %s.\n",
3008 bond
->dev
->name
, slave
->dev
->name
);
3010 if (!bond
->curr_active_slave
||
3011 (slave
== bond
->primary_slave
))
3018 case BOND_LINK_DOWN
:
3019 if (slave
->link_failure_count
< UINT_MAX
)
3020 slave
->link_failure_count
++;
3022 slave
->link
= BOND_LINK_DOWN
;
3023 bond_set_slave_inactive_flags(slave
);
3025 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3026 bond
->dev
->name
, slave
->dev
->name
);
3028 if (slave
== bond
->curr_active_slave
) {
3029 bond
->current_arp_slave
= NULL
;
3036 pr_err("%s: impossible: new_link %d on slave %s\n",
3037 bond
->dev
->name
, slave
->new_link
,
3045 write_lock_bh(&bond
->curr_slave_lock
);
3046 bond_select_active_slave(bond
);
3047 write_unlock_bh(&bond
->curr_slave_lock
);
3048 unblock_netpoll_tx();
3051 bond_set_carrier(bond
);
3055 * Send ARP probes for active-backup mode ARP monitor.
3057 * Called with bond->lock held for read.
3059 static void bond_ab_arp_probe(struct bonding
*bond
)
3061 struct slave
*slave
;
3064 read_lock(&bond
->curr_slave_lock
);
3066 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3067 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3068 bond
->current_arp_slave
->dev
->name
,
3069 bond
->curr_active_slave
->dev
->name
);
3071 if (bond
->curr_active_slave
) {
3072 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3073 read_unlock(&bond
->curr_slave_lock
);
3077 read_unlock(&bond
->curr_slave_lock
);
3079 /* if we don't have a curr_active_slave, search for the next available
3080 * backup slave from the current_arp_slave and make it the candidate
3081 * for becoming the curr_active_slave
3084 if (!bond
->current_arp_slave
) {
3085 bond
->current_arp_slave
= bond
->first_slave
;
3086 if (!bond
->current_arp_slave
)
3090 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3092 /* search for next candidate */
3093 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3094 if (IS_UP(slave
->dev
)) {
3095 slave
->link
= BOND_LINK_BACK
;
3096 bond_set_slave_active_flags(slave
);
3097 bond_arp_send_all(bond
, slave
);
3098 slave
->jiffies
= jiffies
;
3099 bond
->current_arp_slave
= slave
;
3103 /* if the link state is up at this point, we
3104 * mark it down - this can happen if we have
3105 * simultaneous link failures and
3106 * reselect_active_interface doesn't make this
3107 * one the current slave so it is still marked
3108 * up when it is actually down
3110 if (slave
->link
== BOND_LINK_UP
) {
3111 slave
->link
= BOND_LINK_DOWN
;
3112 if (slave
->link_failure_count
< UINT_MAX
)
3113 slave
->link_failure_count
++;
3115 bond_set_slave_inactive_flags(slave
);
3117 pr_info("%s: backup interface %s is now down.\n",
3118 bond
->dev
->name
, slave
->dev
->name
);
3123 void bond_activebackup_arp_mon(struct work_struct
*work
)
3125 struct bonding
*bond
= container_of(work
, struct bonding
,
3127 bool should_notify_peers
= false;
3130 read_lock(&bond
->lock
);
3132 if (bond
->kill_timers
)
3135 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3137 if (bond
->slave_cnt
== 0)
3140 should_notify_peers
= bond_should_notify_peers(bond
);
3142 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3143 read_unlock(&bond
->lock
);
3145 read_lock(&bond
->lock
);
3147 bond_ab_arp_commit(bond
, delta_in_ticks
);
3149 read_unlock(&bond
->lock
);
3151 read_lock(&bond
->lock
);
3154 bond_ab_arp_probe(bond
);
3157 if (bond
->params
.arp_interval
)
3158 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3160 read_unlock(&bond
->lock
);
3162 if (should_notify_peers
) {
3164 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
3169 /*-------------------------- netdev event handling --------------------------*/
3172 * Change device name
3174 static int bond_event_changename(struct bonding
*bond
)
3176 bond_remove_proc_entry(bond
);
3177 bond_create_proc_entry(bond
);
3179 bond_debug_reregister(bond
);
3184 static int bond_master_netdev_event(unsigned long event
,
3185 struct net_device
*bond_dev
)
3187 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3190 case NETDEV_CHANGENAME
:
3191 return bond_event_changename(event_bond
);
3199 static int bond_slave_netdev_event(unsigned long event
,
3200 struct net_device
*slave_dev
)
3202 struct net_device
*bond_dev
= slave_dev
->master
;
3203 struct bonding
*bond
= netdev_priv(bond_dev
);
3206 case NETDEV_UNREGISTER
:
3208 if (bond
->setup_by_slave
)
3209 bond_release_and_destroy(bond_dev
, slave_dev
);
3211 bond_release(bond_dev
, slave_dev
);
3215 if (bond
->params
.mode
== BOND_MODE_8023AD
|| bond_is_lb(bond
)) {
3216 struct slave
*slave
;
3218 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3220 u32 old_speed
= slave
->speed
;
3221 u8 old_duplex
= slave
->duplex
;
3223 bond_update_speed_duplex(slave
);
3225 if (bond_is_lb(bond
))
3228 if (old_speed
!= slave
->speed
)
3229 bond_3ad_adapter_speed_changed(slave
);
3230 if (old_duplex
!= slave
->duplex
)
3231 bond_3ad_adapter_duplex_changed(slave
);
3238 * ... Or is it this?
3241 case NETDEV_CHANGEMTU
:
3243 * TODO: Should slaves be allowed to
3244 * independently alter their MTU? For
3245 * an active-backup bond, slaves need
3246 * not be the same type of device, so
3247 * MTUs may vary. For other modes,
3248 * slaves arguably should have the
3249 * same MTUs. To do this, we'd need to
3250 * take over the slave's change_mtu
3251 * function for the duration of their
3255 case NETDEV_CHANGENAME
:
3257 * TODO: handle changing the primary's name
3260 case NETDEV_FEAT_CHANGE
:
3261 bond_compute_features(bond
);
3271 * bond_netdev_event: handle netdev notifier chain events.
3273 * This function receives events for the netdev chain. The caller (an
3274 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3275 * locks for us to safely manipulate the slave devices (RTNL lock,
3278 static int bond_netdev_event(struct notifier_block
*this,
3279 unsigned long event
, void *ptr
)
3281 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3283 pr_debug("event_dev: %s, event: %lx\n",
3284 event_dev
? event_dev
->name
: "None",
3287 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3290 if (event_dev
->flags
& IFF_MASTER
) {
3291 pr_debug("IFF_MASTER\n");
3292 return bond_master_netdev_event(event
, event_dev
);
3295 if (event_dev
->flags
& IFF_SLAVE
) {
3296 pr_debug("IFF_SLAVE\n");
3297 return bond_slave_netdev_event(event
, event_dev
);
3304 * bond_inetaddr_event: handle inetaddr notifier chain events.
3306 * We keep track of device IPs primarily to use as source addresses in
3307 * ARP monitor probes (rather than spewing out broadcasts all the time).
3309 * We track one IP for the main device (if it has one), plus one per VLAN.
3311 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3313 struct in_ifaddr
*ifa
= ptr
;
3314 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3315 struct bond_net
*bn
= net_generic(dev_net(event_dev
), bond_net_id
);
3316 struct bonding
*bond
;
3317 struct vlan_entry
*vlan
;
3319 list_for_each_entry(bond
, &bn
->dev_list
, bond_list
) {
3320 if (bond
->dev
== event_dev
) {
3323 bond
->master_ip
= ifa
->ifa_local
;
3326 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3333 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3334 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
3336 if (vlan_dev
== event_dev
) {
3339 vlan
->vlan_ip
= ifa
->ifa_local
;
3343 bond_glean_dev_ip(vlan_dev
);
3354 static struct notifier_block bond_netdev_notifier
= {
3355 .notifier_call
= bond_netdev_event
,
3358 static struct notifier_block bond_inetaddr_notifier
= {
3359 .notifier_call
= bond_inetaddr_event
,
3362 /*---------------------------- Hashing Policies -----------------------------*/
3365 * Hash for the output device based upon layer 2 and layer 3 data. If
3366 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3368 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3370 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3371 struct iphdr
*iph
= ip_hdr(skb
);
3373 if (skb
->protocol
== htons(ETH_P_IP
)) {
3374 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3375 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3378 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3382 * Hash for the output device based upon layer 3 and layer 4 data. If
3383 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3384 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3386 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3388 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3389 struct iphdr
*iph
= ip_hdr(skb
);
3390 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3393 if (skb
->protocol
== htons(ETH_P_IP
)) {
3394 if (!ip_is_fragment(iph
) &&
3395 (iph
->protocol
== IPPROTO_TCP
||
3396 iph
->protocol
== IPPROTO_UDP
)) {
3397 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3399 return (layer4_xor
^
3400 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3404 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3408 * Hash for the output device based upon layer 2 data
3410 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3412 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3414 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3417 /*-------------------------- Device entry points ----------------------------*/
3419 static int bond_open(struct net_device
*bond_dev
)
3421 struct bonding
*bond
= netdev_priv(bond_dev
);
3423 bond
->kill_timers
= 0;
3425 INIT_DELAYED_WORK(&bond
->mcast_work
, bond_resend_igmp_join_requests_delayed
);
3427 if (bond_is_lb(bond
)) {
3428 /* bond_alb_initialize must be called before the timer
3431 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3432 /* something went wrong - fail the open operation */
3436 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3437 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3440 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3441 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3442 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3445 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3446 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3447 INIT_DELAYED_WORK(&bond
->arp_work
,
3448 bond_activebackup_arp_mon
);
3450 INIT_DELAYED_WORK(&bond
->arp_work
,
3451 bond_loadbalance_arp_mon
);
3453 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3454 if (bond
->params
.arp_validate
)
3455 bond
->recv_probe
= bond_arp_rcv
;
3458 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3459 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3460 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3461 /* register to receive LACPDUs */
3462 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3463 bond_3ad_initiate_agg_selection(bond
, 1);
3469 static int bond_close(struct net_device
*bond_dev
)
3471 struct bonding
*bond
= netdev_priv(bond_dev
);
3473 write_lock_bh(&bond
->lock
);
3475 bond
->send_peer_notif
= 0;
3477 /* signal timers not to re-arm */
3478 bond
->kill_timers
= 1;
3480 write_unlock_bh(&bond
->lock
);
3482 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3483 cancel_delayed_work(&bond
->mii_work
);
3486 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3487 cancel_delayed_work(&bond
->arp_work
);
3490 switch (bond
->params
.mode
) {
3491 case BOND_MODE_8023AD
:
3492 cancel_delayed_work(&bond
->ad_work
);
3496 cancel_delayed_work(&bond
->alb_work
);
3502 if (delayed_work_pending(&bond
->mcast_work
))
3503 cancel_delayed_work(&bond
->mcast_work
);
3505 if (bond_is_lb(bond
)) {
3506 /* Must be called only after all
3507 * slaves have been released
3509 bond_alb_deinitialize(bond
);
3511 bond
->recv_probe
= NULL
;
3516 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3517 struct rtnl_link_stats64
*stats
)
3519 struct bonding
*bond
= netdev_priv(bond_dev
);
3520 struct rtnl_link_stats64 temp
;
3521 struct slave
*slave
;
3524 memset(stats
, 0, sizeof(*stats
));
3526 read_lock_bh(&bond
->lock
);
3528 bond_for_each_slave(bond
, slave
, i
) {
3529 const struct rtnl_link_stats64
*sstats
=
3530 dev_get_stats(slave
->dev
, &temp
);
3532 stats
->rx_packets
+= sstats
->rx_packets
;
3533 stats
->rx_bytes
+= sstats
->rx_bytes
;
3534 stats
->rx_errors
+= sstats
->rx_errors
;
3535 stats
->rx_dropped
+= sstats
->rx_dropped
;
3537 stats
->tx_packets
+= sstats
->tx_packets
;
3538 stats
->tx_bytes
+= sstats
->tx_bytes
;
3539 stats
->tx_errors
+= sstats
->tx_errors
;
3540 stats
->tx_dropped
+= sstats
->tx_dropped
;
3542 stats
->multicast
+= sstats
->multicast
;
3543 stats
->collisions
+= sstats
->collisions
;
3545 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3546 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3547 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3548 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3549 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3550 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3552 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3553 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3554 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3555 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3556 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3559 read_unlock_bh(&bond
->lock
);
3564 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3566 struct net_device
*slave_dev
= NULL
;
3567 struct ifbond k_binfo
;
3568 struct ifbond __user
*u_binfo
= NULL
;
3569 struct ifslave k_sinfo
;
3570 struct ifslave __user
*u_sinfo
= NULL
;
3571 struct mii_ioctl_data
*mii
= NULL
;
3574 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3586 * We do this again just in case we were called by SIOCGMIIREG
3587 * instead of SIOCGMIIPHY.
3594 if (mii
->reg_num
== 1) {
3595 struct bonding
*bond
= netdev_priv(bond_dev
);
3597 read_lock(&bond
->lock
);
3598 read_lock(&bond
->curr_slave_lock
);
3599 if (netif_carrier_ok(bond
->dev
))
3600 mii
->val_out
= BMSR_LSTATUS
;
3602 read_unlock(&bond
->curr_slave_lock
);
3603 read_unlock(&bond
->lock
);
3607 case BOND_INFO_QUERY_OLD
:
3608 case SIOCBONDINFOQUERY
:
3609 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3611 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3614 res
= bond_info_query(bond_dev
, &k_binfo
);
3616 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3620 case BOND_SLAVE_INFO_QUERY_OLD
:
3621 case SIOCBONDSLAVEINFOQUERY
:
3622 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3624 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3627 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3629 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3638 if (!capable(CAP_NET_ADMIN
))
3641 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
3643 pr_debug("slave_dev=%p:\n", slave_dev
);
3648 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3650 case BOND_ENSLAVE_OLD
:
3651 case SIOCBONDENSLAVE
:
3652 res
= bond_enslave(bond_dev
, slave_dev
);
3654 case BOND_RELEASE_OLD
:
3655 case SIOCBONDRELEASE
:
3656 res
= bond_release(bond_dev
, slave_dev
);
3658 case BOND_SETHWADDR_OLD
:
3659 case SIOCBONDSETHWADDR
:
3660 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3662 case BOND_CHANGE_ACTIVE_OLD
:
3663 case SIOCBONDCHANGEACTIVE
:
3664 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3676 static bool bond_addr_in_mc_list(unsigned char *addr
,
3677 struct netdev_hw_addr_list
*list
,
3680 struct netdev_hw_addr
*ha
;
3682 netdev_hw_addr_list_for_each(ha
, list
)
3683 if (!memcmp(ha
->addr
, addr
, addrlen
))
3689 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3691 struct bonding
*bond
= netdev_priv(bond_dev
);
3692 struct netdev_hw_addr
*ha
;
3696 * Do promisc before checking multicast_mode
3698 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
))
3700 * FIXME: Need to handle the error when one of the multi-slaves
3703 bond_set_promiscuity(bond
, 1);
3706 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
))
3707 bond_set_promiscuity(bond
, -1);
3710 /* set allmulti flag to slaves */
3711 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
))
3713 * FIXME: Need to handle the error when one of the multi-slaves
3716 bond_set_allmulti(bond
, 1);
3719 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
))
3720 bond_set_allmulti(bond
, -1);
3723 read_lock(&bond
->lock
);
3725 bond
->flags
= bond_dev
->flags
;
3727 /* looking for addresses to add to slaves' mc list */
3728 netdev_for_each_mc_addr(ha
, bond_dev
) {
3729 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
3730 bond_dev
->addr_len
);
3732 bond_mc_add(bond
, ha
->addr
);
3735 /* looking for addresses to delete from slaves' list */
3736 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
3737 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
3738 bond_dev
->addr_len
);
3740 bond_mc_del(bond
, ha
->addr
);
3743 /* save master's multicast list */
3744 __hw_addr_flush(&bond
->mc_list
);
3745 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
3746 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
3748 read_unlock(&bond
->lock
);
3751 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
3753 struct bonding
*bond
= netdev_priv(dev
);
3754 struct slave
*slave
= bond
->first_slave
;
3757 const struct net_device_ops
*slave_ops
3758 = slave
->dev
->netdev_ops
;
3759 if (slave_ops
->ndo_neigh_setup
)
3760 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
3766 * Change the MTU of all of a master's slaves to match the master
3768 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3770 struct bonding
*bond
= netdev_priv(bond_dev
);
3771 struct slave
*slave
, *stop_at
;
3775 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3776 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3778 /* Can't hold bond->lock with bh disabled here since
3779 * some base drivers panic. On the other hand we can't
3780 * hold bond->lock without bh disabled because we'll
3781 * deadlock. The only solution is to rely on the fact
3782 * that we're under rtnl_lock here, and the slaves
3783 * list won't change. This doesn't solve the problem
3784 * of setting the slave's MTU while it is
3785 * transmitting, but the assumption is that the base
3786 * driver can handle that.
3788 * TODO: figure out a way to safely iterate the slaves
3789 * list, but without holding a lock around the actual
3790 * call to the base driver.
3793 bond_for_each_slave(bond
, slave
, i
) {
3794 pr_debug("s %p s->p %p c_m %p\n",
3797 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3799 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3802 /* If we failed to set the slave's mtu to the new value
3803 * we must abort the operation even in ACTIVE_BACKUP
3804 * mode, because if we allow the backup slaves to have
3805 * different mtu values than the active slave we'll
3806 * need to change their mtu when doing a failover. That
3807 * means changing their mtu from timer context, which
3808 * is probably not a good idea.
3810 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3815 bond_dev
->mtu
= new_mtu
;
3820 /* unwind from head to the slave that failed */
3822 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3825 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3827 pr_debug("unwind err %d dev %s\n",
3828 tmp_res
, slave
->dev
->name
);
3838 * Note that many devices must be down to change the HW address, and
3839 * downing the master releases all slaves. We can make bonds full of
3840 * bonding devices to test this, however.
3842 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3844 struct bonding
*bond
= netdev_priv(bond_dev
);
3845 struct sockaddr
*sa
= addr
, tmp_sa
;
3846 struct slave
*slave
, *stop_at
;
3850 if (bond
->params
.mode
== BOND_MODE_ALB
)
3851 return bond_alb_set_mac_address(bond_dev
, addr
);
3854 pr_debug("bond=%p, name=%s\n",
3855 bond
, bond_dev
? bond_dev
->name
: "None");
3858 * If fail_over_mac is set to active, do nothing and return
3859 * success. Returning an error causes ifenslave to fail.
3861 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
3864 if (!is_valid_ether_addr(sa
->sa_data
))
3865 return -EADDRNOTAVAIL
;
3867 /* Can't hold bond->lock with bh disabled here since
3868 * some base drivers panic. On the other hand we can't
3869 * hold bond->lock without bh disabled because we'll
3870 * deadlock. The only solution is to rely on the fact
3871 * that we're under rtnl_lock here, and the slaves
3872 * list won't change. This doesn't solve the problem
3873 * of setting the slave's hw address while it is
3874 * transmitting, but the assumption is that the base
3875 * driver can handle that.
3877 * TODO: figure out a way to safely iterate the slaves
3878 * list, but without holding a lock around the actual
3879 * call to the base driver.
3882 bond_for_each_slave(bond
, slave
, i
) {
3883 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3884 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3886 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3888 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3892 res
= dev_set_mac_address(slave
->dev
, addr
);
3894 /* TODO: consider downing the slave
3896 * User should expect communications
3897 * breakage anyway until ARP finish
3900 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3906 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3910 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3911 tmp_sa
.sa_family
= bond_dev
->type
;
3913 /* unwind from head to the slave that failed */
3915 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3918 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3920 pr_debug("unwind err %d dev %s\n",
3921 tmp_res
, slave
->dev
->name
);
3928 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3930 struct bonding
*bond
= netdev_priv(bond_dev
);
3931 struct slave
*slave
, *start_at
;
3932 int i
, slave_no
, res
= 1;
3933 struct iphdr
*iph
= ip_hdr(skb
);
3936 * Start with the curr_active_slave that joined the bond as the
3937 * default for sending IGMP traffic. For failover purposes one
3938 * needs to maintain some consistency for the interface that will
3939 * send the join/membership reports. The curr_active_slave found
3940 * will send all of this type of traffic.
3942 if ((iph
->protocol
== IPPROTO_IGMP
) &&
3943 (skb
->protocol
== htons(ETH_P_IP
))) {
3945 read_lock(&bond
->curr_slave_lock
);
3946 slave
= bond
->curr_active_slave
;
3947 read_unlock(&bond
->curr_slave_lock
);
3953 * Concurrent TX may collide on rr_tx_counter; we accept
3954 * that as being rare enough not to justify using an
3957 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
3959 bond_for_each_slave(bond
, slave
, i
) {
3967 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3968 if (IS_UP(slave
->dev
) &&
3969 (slave
->link
== BOND_LINK_UP
) &&
3970 bond_is_active_slave(slave
)) {
3971 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3978 /* no suitable interface, frame not sent */
3982 return NETDEV_TX_OK
;
3987 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3988 * the bond has a usable interface.
3990 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3992 struct bonding
*bond
= netdev_priv(bond_dev
);
3995 read_lock(&bond
->curr_slave_lock
);
3997 if (bond
->curr_active_slave
)
3998 res
= bond_dev_queue_xmit(bond
, skb
,
3999 bond
->curr_active_slave
->dev
);
4002 /* no suitable interface, frame not sent */
4005 read_unlock(&bond
->curr_slave_lock
);
4007 return NETDEV_TX_OK
;
4011 * In bond_xmit_xor() , we determine the output device by using a pre-
4012 * determined xmit_hash_policy(), If the selected device is not enabled,
4013 * find the next active slave.
4015 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4017 struct bonding
*bond
= netdev_priv(bond_dev
);
4018 struct slave
*slave
, *start_at
;
4023 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4025 bond_for_each_slave(bond
, slave
, i
) {
4033 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4034 if (IS_UP(slave
->dev
) &&
4035 (slave
->link
== BOND_LINK_UP
) &&
4036 bond_is_active_slave(slave
)) {
4037 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4043 /* no suitable interface, frame not sent */
4047 return NETDEV_TX_OK
;
4051 * in broadcast mode, we send everything to all usable interfaces.
4053 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4055 struct bonding
*bond
= netdev_priv(bond_dev
);
4056 struct slave
*slave
, *start_at
;
4057 struct net_device
*tx_dev
= NULL
;
4061 read_lock(&bond
->curr_slave_lock
);
4062 start_at
= bond
->curr_active_slave
;
4063 read_unlock(&bond
->curr_slave_lock
);
4068 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4069 if (IS_UP(slave
->dev
) &&
4070 (slave
->link
== BOND_LINK_UP
) &&
4071 bond_is_active_slave(slave
)) {
4073 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4075 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4080 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4082 dev_kfree_skb(skb2
);
4086 tx_dev
= slave
->dev
;
4091 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4095 /* no suitable interface, frame not sent */
4098 /* frame sent to all suitable interfaces */
4099 return NETDEV_TX_OK
;
4102 /*------------------------- Device initialization ---------------------------*/
4104 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4106 switch (bond
->params
.xmit_policy
) {
4107 case BOND_XMIT_POLICY_LAYER23
:
4108 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4110 case BOND_XMIT_POLICY_LAYER34
:
4111 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4113 case BOND_XMIT_POLICY_LAYER2
:
4115 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4121 * Lookup the slave that corresponds to a qid
4123 static inline int bond_slave_override(struct bonding
*bond
,
4124 struct sk_buff
*skb
)
4127 struct slave
*slave
= NULL
;
4128 struct slave
*check_slave
;
4130 if (!skb
->queue_mapping
)
4133 /* Find out if any slaves have the same mapping as this skb. */
4134 bond_for_each_slave(bond
, check_slave
, i
) {
4135 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4136 slave
= check_slave
;
4141 /* If the slave isn't UP, use default transmit policy. */
4142 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4143 (slave
->link
== BOND_LINK_UP
)) {
4144 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4151 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4154 * This helper function exists to help dev_pick_tx get the correct
4155 * destination queue. Using a helper function skips a call to
4156 * skb_tx_hash and will put the skbs in the queue we expect on their
4157 * way down to the bonding driver.
4159 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
4162 * Save the original txq to restore before passing to the driver
4164 bond_queue_mapping(skb
) = skb
->queue_mapping
;
4166 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
4168 txq
-= dev
->real_num_tx_queues
;
4169 } while (txq
>= dev
->real_num_tx_queues
);
4174 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4176 struct bonding
*bond
= netdev_priv(dev
);
4178 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4179 if (!bond_slave_override(bond
, skb
))
4180 return NETDEV_TX_OK
;
4183 switch (bond
->params
.mode
) {
4184 case BOND_MODE_ROUNDROBIN
:
4185 return bond_xmit_roundrobin(skb
, dev
);
4186 case BOND_MODE_ACTIVEBACKUP
:
4187 return bond_xmit_activebackup(skb
, dev
);
4189 return bond_xmit_xor(skb
, dev
);
4190 case BOND_MODE_BROADCAST
:
4191 return bond_xmit_broadcast(skb
, dev
);
4192 case BOND_MODE_8023AD
:
4193 return bond_3ad_xmit_xor(skb
, dev
);
4196 return bond_alb_xmit(skb
, dev
);
4198 /* Should never happen, mode already checked */
4199 pr_err("%s: Error: Unknown bonding mode %d\n",
4200 dev
->name
, bond
->params
.mode
);
4203 return NETDEV_TX_OK
;
4207 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4209 struct bonding
*bond
= netdev_priv(dev
);
4210 netdev_tx_t ret
= NETDEV_TX_OK
;
4213 * If we risk deadlock from transmitting this in the
4214 * netpoll path, tell netpoll to queue the frame for later tx
4216 if (is_netpoll_tx_blocked(dev
))
4217 return NETDEV_TX_BUSY
;
4219 read_lock(&bond
->lock
);
4221 if (bond
->slave_cnt
)
4222 ret
= __bond_start_xmit(skb
, dev
);
4226 read_unlock(&bond
->lock
);
4232 * set bond mode specific net device operations
4234 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4236 struct net_device
*bond_dev
= bond
->dev
;
4239 case BOND_MODE_ROUNDROBIN
:
4241 case BOND_MODE_ACTIVEBACKUP
:
4244 bond_set_xmit_hash_policy(bond
);
4246 case BOND_MODE_BROADCAST
:
4248 case BOND_MODE_8023AD
:
4249 bond_set_xmit_hash_policy(bond
);
4256 /* Should never happen, mode already checked */
4257 pr_err("%s: Error: Unknown bonding mode %d\n",
4258 bond_dev
->name
, mode
);
4263 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4264 struct ethtool_drvinfo
*drvinfo
)
4266 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4267 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4268 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4271 static const struct ethtool_ops bond_ethtool_ops
= {
4272 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4273 .get_link
= ethtool_op_get_link
,
4276 static const struct net_device_ops bond_netdev_ops
= {
4277 .ndo_init
= bond_init
,
4278 .ndo_uninit
= bond_uninit
,
4279 .ndo_open
= bond_open
,
4280 .ndo_stop
= bond_close
,
4281 .ndo_start_xmit
= bond_start_xmit
,
4282 .ndo_select_queue
= bond_select_queue
,
4283 .ndo_get_stats64
= bond_get_stats
,
4284 .ndo_do_ioctl
= bond_do_ioctl
,
4285 .ndo_set_multicast_list
= bond_set_multicast_list
,
4286 .ndo_change_mtu
= bond_change_mtu
,
4287 .ndo_set_mac_address
= bond_set_mac_address
,
4288 .ndo_neigh_setup
= bond_neigh_setup
,
4289 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4290 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4291 #ifdef CONFIG_NET_POLL_CONTROLLER
4292 .ndo_netpoll_setup
= bond_netpoll_setup
,
4293 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4294 .ndo_poll_controller
= bond_poll_controller
,
4296 .ndo_add_slave
= bond_enslave
,
4297 .ndo_del_slave
= bond_release
,
4298 .ndo_fix_features
= bond_fix_features
,
4301 static void bond_destructor(struct net_device
*bond_dev
)
4303 struct bonding
*bond
= netdev_priv(bond_dev
);
4305 destroy_workqueue(bond
->wq
);
4306 free_netdev(bond_dev
);
4309 static void bond_setup(struct net_device
*bond_dev
)
4311 struct bonding
*bond
= netdev_priv(bond_dev
);
4313 /* initialize rwlocks */
4314 rwlock_init(&bond
->lock
);
4315 rwlock_init(&bond
->curr_slave_lock
);
4317 bond
->params
= bonding_defaults
;
4319 /* Initialize pointers */
4320 bond
->dev
= bond_dev
;
4321 INIT_LIST_HEAD(&bond
->vlan_list
);
4323 /* Initialize the device entry points */
4324 ether_setup(bond_dev
);
4325 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4326 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4327 bond_set_mode_ops(bond
, bond
->params
.mode
);
4329 bond_dev
->destructor
= bond_destructor
;
4331 /* Initialize the device options */
4332 bond_dev
->tx_queue_len
= 0;
4333 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4334 bond_dev
->priv_flags
|= IFF_BONDING
;
4335 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
4337 /* At first, we block adding VLANs. That's the only way to
4338 * prevent problems that occur when adding VLANs over an
4339 * empty bond. The block will be removed once non-challenged
4340 * slaves are enslaved.
4342 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4344 /* don't acquire bond device's netif_tx_lock when
4346 bond_dev
->features
|= NETIF_F_LLTX
;
4348 /* By default, we declare the bond to be fully
4349 * VLAN hardware accelerated capable. Special
4350 * care is taken in the various xmit functions
4351 * when there are slaves that are not hw accel
4355 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4356 NETIF_F_HW_VLAN_TX
|
4357 NETIF_F_HW_VLAN_RX
|
4358 NETIF_F_HW_VLAN_FILTER
;
4360 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_NO_CSUM
);
4361 bond_dev
->features
|= bond_dev
->hw_features
;
4364 static void bond_work_cancel_all(struct bonding
*bond
)
4366 write_lock_bh(&bond
->lock
);
4367 bond
->kill_timers
= 1;
4368 write_unlock_bh(&bond
->lock
);
4370 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4371 cancel_delayed_work(&bond
->mii_work
);
4373 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4374 cancel_delayed_work(&bond
->arp_work
);
4376 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4377 delayed_work_pending(&bond
->alb_work
))
4378 cancel_delayed_work(&bond
->alb_work
);
4380 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4381 delayed_work_pending(&bond
->ad_work
))
4382 cancel_delayed_work(&bond
->ad_work
);
4384 if (delayed_work_pending(&bond
->mcast_work
))
4385 cancel_delayed_work(&bond
->mcast_work
);
4389 * Destroy a bonding device.
4390 * Must be under rtnl_lock when this function is called.
4392 static void bond_uninit(struct net_device
*bond_dev
)
4394 struct bonding
*bond
= netdev_priv(bond_dev
);
4395 struct vlan_entry
*vlan
, *tmp
;
4397 bond_netpoll_cleanup(bond_dev
);
4399 /* Release the bonded slaves */
4400 bond_release_all(bond_dev
);
4402 list_del(&bond
->bond_list
);
4404 bond_work_cancel_all(bond
);
4406 bond_remove_proc_entry(bond
);
4408 bond_debug_unregister(bond
);
4410 __hw_addr_flush(&bond
->mc_list
);
4412 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4413 list_del(&vlan
->vlan_list
);
4418 /*------------------------- Module initialization ---------------------------*/
4421 * Convert string input module parms. Accept either the
4422 * number of the mode or its string name. A bit complicated because
4423 * some mode names are substrings of other names, and calls from sysfs
4424 * may have whitespace in the name (trailing newlines, for example).
4426 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4428 int modeint
= -1, i
, rv
;
4429 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4431 for (p
= (char *)buf
; *p
; p
++)
4432 if (!(isdigit(*p
) || isspace(*p
)))
4436 rv
= sscanf(buf
, "%20s", modestr
);
4438 rv
= sscanf(buf
, "%d", &modeint
);
4443 for (i
= 0; tbl
[i
].modename
; i
++) {
4444 if (modeint
== tbl
[i
].mode
)
4446 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4453 static int bond_check_params(struct bond_params
*params
)
4455 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4458 * Convert string parameters.
4461 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4462 if (bond_mode
== -1) {
4463 pr_err("Error: Invalid bonding mode \"%s\"\n",
4464 mode
== NULL
? "NULL" : mode
);
4469 if (xmit_hash_policy
) {
4470 if ((bond_mode
!= BOND_MODE_XOR
) &&
4471 (bond_mode
!= BOND_MODE_8023AD
)) {
4472 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4473 bond_mode_name(bond_mode
));
4475 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4477 if (xmit_hashtype
== -1) {
4478 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4479 xmit_hash_policy
== NULL
? "NULL" :
4487 if (bond_mode
!= BOND_MODE_8023AD
) {
4488 pr_info("lacp_rate param is irrelevant in mode %s\n",
4489 bond_mode_name(bond_mode
));
4491 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4492 if (lacp_fast
== -1) {
4493 pr_err("Error: Invalid lacp rate \"%s\"\n",
4494 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4501 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4502 if (params
->ad_select
== -1) {
4503 pr_err("Error: Invalid ad_select \"%s\"\n",
4504 ad_select
== NULL
? "NULL" : ad_select
);
4508 if (bond_mode
!= BOND_MODE_8023AD
) {
4509 pr_warning("ad_select param only affects 802.3ad mode\n");
4512 params
->ad_select
= BOND_AD_STABLE
;
4515 if (max_bonds
< 0) {
4516 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4517 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4518 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4522 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4523 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4524 miimon
= BOND_LINK_MON_INTERV
;
4528 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4533 if (downdelay
< 0) {
4534 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4535 downdelay
, INT_MAX
);
4539 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4540 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4545 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4546 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4551 /* reset values for 802.3ad */
4552 if (bond_mode
== BOND_MODE_8023AD
) {
4554 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");
4555 pr_warning("Forcing miimon to 100msec\n");
4560 if (tx_queues
< 1 || tx_queues
> 255) {
4561 pr_warning("Warning: tx_queues (%d) should be between "
4562 "1 and 255, resetting to %d\n",
4563 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4564 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4567 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4568 pr_warning("Warning: all_slaves_active module parameter (%d), "
4569 "not of valid value (0/1), so it was set to "
4570 "0\n", all_slaves_active
);
4571 all_slaves_active
= 0;
4574 if (resend_igmp
< 0 || resend_igmp
> 255) {
4575 pr_warning("Warning: resend_igmp (%d) should be between "
4576 "0 and 255, resetting to %d\n",
4577 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4578 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4581 /* reset values for TLB/ALB */
4582 if ((bond_mode
== BOND_MODE_TLB
) ||
4583 (bond_mode
== BOND_MODE_ALB
)) {
4585 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");
4586 pr_warning("Forcing miimon to 100msec\n");
4591 if (bond_mode
== BOND_MODE_ALB
) {
4592 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",
4597 if (updelay
|| downdelay
) {
4598 /* just warn the user the up/down delay will have
4599 * no effect since miimon is zero...
4601 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",
4602 updelay
, downdelay
);
4605 /* don't allow arp monitoring */
4607 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4608 miimon
, arp_interval
);
4612 if ((updelay
% miimon
) != 0) {
4613 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4615 (updelay
/ miimon
) * miimon
);
4620 if ((downdelay
% miimon
) != 0) {
4621 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4623 (downdelay
/ miimon
) * miimon
);
4626 downdelay
/= miimon
;
4629 if (arp_interval
< 0) {
4630 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4631 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4632 arp_interval
= BOND_LINK_ARP_INTERV
;
4635 for (arp_ip_count
= 0;
4636 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4638 /* not complete check, but should be good enough to
4640 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4641 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4642 arp_ip_target
[arp_ip_count
]);
4645 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4646 arp_target
[arp_ip_count
] = ip
;
4650 if (arp_interval
&& !arp_ip_count
) {
4651 /* don't allow arping if no arp_ip_target given... */
4652 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4658 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4659 pr_err("arp_validate only supported in active-backup mode\n");
4662 if (!arp_interval
) {
4663 pr_err("arp_validate requires arp_interval\n");
4667 arp_validate_value
= bond_parse_parm(arp_validate
,
4669 if (arp_validate_value
== -1) {
4670 pr_err("Error: invalid arp_validate \"%s\"\n",
4671 arp_validate
== NULL
? "NULL" : arp_validate
);
4675 arp_validate_value
= 0;
4678 pr_info("MII link monitoring set to %d ms\n", miimon
);
4679 } else if (arp_interval
) {
4682 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4684 arp_validate_tbl
[arp_validate_value
].modename
,
4687 for (i
= 0; i
< arp_ip_count
; i
++)
4688 pr_info(" %s", arp_ip_target
[i
]);
4692 } else if (max_bonds
) {
4693 /* miimon and arp_interval not set, we need one so things
4694 * work as expected, see bonding.txt for details
4696 pr_debug("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");
4699 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4700 /* currently, using a primary only makes sense
4701 * in active backup, TLB or ALB modes
4703 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4704 primary
, bond_mode_name(bond_mode
));
4708 if (primary
&& primary_reselect
) {
4709 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4711 if (primary_reselect_value
== -1) {
4712 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4714 NULL
? "NULL" : primary_reselect
);
4718 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4721 if (fail_over_mac
) {
4722 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4724 if (fail_over_mac_value
== -1) {
4725 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4726 arp_validate
== NULL
? "NULL" : arp_validate
);
4730 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4731 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4733 fail_over_mac_value
= BOND_FOM_NONE
;
4736 /* fill params struct with the proper values */
4737 params
->mode
= bond_mode
;
4738 params
->xmit_policy
= xmit_hashtype
;
4739 params
->miimon
= miimon
;
4740 params
->num_peer_notif
= num_peer_notif
;
4741 params
->arp_interval
= arp_interval
;
4742 params
->arp_validate
= arp_validate_value
;
4743 params
->updelay
= updelay
;
4744 params
->downdelay
= downdelay
;
4745 params
->use_carrier
= use_carrier
;
4746 params
->lacp_fast
= lacp_fast
;
4747 params
->primary
[0] = 0;
4748 params
->primary_reselect
= primary_reselect_value
;
4749 params
->fail_over_mac
= fail_over_mac_value
;
4750 params
->tx_queues
= tx_queues
;
4751 params
->all_slaves_active
= all_slaves_active
;
4752 params
->resend_igmp
= resend_igmp
;
4753 params
->min_links
= min_links
;
4756 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4757 params
->primary
[IFNAMSIZ
- 1] = 0;
4760 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4765 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4766 static struct lock_class_key bonding_netdev_addr_lock_key
;
4768 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4769 struct netdev_queue
*txq
,
4772 lockdep_set_class(&txq
->_xmit_lock
,
4773 &bonding_netdev_xmit_lock_key
);
4776 static void bond_set_lockdep_class(struct net_device
*dev
)
4778 lockdep_set_class(&dev
->addr_list_lock
,
4779 &bonding_netdev_addr_lock_key
);
4780 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4784 * Called from registration process
4786 static int bond_init(struct net_device
*bond_dev
)
4788 struct bonding
*bond
= netdev_priv(bond_dev
);
4789 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4790 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4792 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4795 * Initialize locks that may be required during
4796 * en/deslave operations. All of the bond_open work
4797 * (of which this is part) should really be moved to
4798 * a phase prior to dev_open
4800 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4801 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4803 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4807 bond_set_lockdep_class(bond_dev
);
4809 bond_create_proc_entry(bond
);
4810 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4812 bond_prepare_sysfs_group(bond
);
4814 bond_debug_register(bond
);
4816 __hw_addr_init(&bond
->mc_list
);
4820 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4822 if (tb
[IFLA_ADDRESS
]) {
4823 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4825 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4826 return -EADDRNOTAVAIL
;
4831 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
4833 .priv_size
= sizeof(struct bonding
),
4834 .setup
= bond_setup
,
4835 .validate
= bond_validate
,
4838 /* Create a new bond based on the specified name and bonding parameters.
4839 * If name is NULL, obtain a suitable "bond%d" name for us.
4840 * Caller must NOT hold rtnl_lock; we need to release it here before we
4841 * set up our sysfs entries.
4843 int bond_create(struct net
*net
, const char *name
)
4845 struct net_device
*bond_dev
;
4850 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4851 name
? name
: "bond%d",
4852 bond_setup
, tx_queues
);
4854 pr_err("%s: eek! can't alloc netdev!\n", name
);
4859 dev_net_set(bond_dev
, net
);
4860 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4862 res
= register_netdevice(bond_dev
);
4864 netif_carrier_off(bond_dev
);
4868 bond_destructor(bond_dev
);
4872 static int __net_init
bond_net_init(struct net
*net
)
4874 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4877 INIT_LIST_HEAD(&bn
->dev_list
);
4879 bond_create_proc_dir(bn
);
4884 static void __net_exit
bond_net_exit(struct net
*net
)
4886 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4888 bond_destroy_proc_dir(bn
);
4891 static struct pernet_operations bond_net_ops
= {
4892 .init
= bond_net_init
,
4893 .exit
= bond_net_exit
,
4895 .size
= sizeof(struct bond_net
),
4898 static int __init
bonding_init(void)
4903 pr_info("%s", bond_version
);
4905 res
= bond_check_params(&bonding_defaults
);
4909 res
= register_pernet_subsys(&bond_net_ops
);
4913 res
= rtnl_link_register(&bond_link_ops
);
4917 bond_create_debugfs();
4919 for (i
= 0; i
< max_bonds
; i
++) {
4920 res
= bond_create(&init_net
, NULL
);
4925 res
= bond_create_sysfs();
4929 register_netdevice_notifier(&bond_netdev_notifier
);
4930 register_inetaddr_notifier(&bond_inetaddr_notifier
);
4934 rtnl_link_unregister(&bond_link_ops
);
4936 unregister_pernet_subsys(&bond_net_ops
);
4941 static void __exit
bonding_exit(void)
4943 unregister_netdevice_notifier(&bond_netdev_notifier
);
4944 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
4946 bond_destroy_sysfs();
4947 bond_destroy_debugfs();
4949 rtnl_link_unregister(&bond_link_ops
);
4950 unregister_pernet_subsys(&bond_net_ops
);
4952 #ifdef CONFIG_NET_POLL_CONTROLLER
4954 * Make sure we don't have an imbalance on our netpoll blocking
4956 WARN_ON(atomic_read(&netpoll_block_tx
));
4960 module_init(bonding_init
);
4961 module_exit(bonding_exit
);
4962 MODULE_LICENSE("GPL");
4963 MODULE_VERSION(DRV_VERSION
);
4964 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4965 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4966 MODULE_ALIAS_RTNL_LINK("bond");