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>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
83 /*---------------------------- Module parameters ----------------------------*/
85 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
86 #define BOND_LINK_MON_INTERV 0
87 #define BOND_LINK_ARP_INTERV 0
89 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
90 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
91 static int num_peer_notif
= 1;
92 static int miimon
= BOND_LINK_MON_INTERV
;
95 static int use_carrier
= 1;
98 static char *primary_reselect
;
99 static char *lacp_rate
;
100 static int min_links
;
101 static char *ad_select
;
102 static char *xmit_hash_policy
;
103 static int arp_interval
= BOND_LINK_ARP_INTERV
;
104 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
105 static char *arp_validate
;
106 static char *fail_over_mac
;
107 static int all_slaves_active
= 0;
108 static struct bond_params bonding_defaults
;
109 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
111 module_param(max_bonds
, int, 0);
112 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
113 module_param(tx_queues
, int, 0);
114 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
115 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
116 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
117 "failover event (alias of num_unsol_na)");
118 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
119 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
120 "failover event (alias of num_grat_arp)");
121 module_param(miimon
, int, 0);
122 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
123 module_param(updelay
, int, 0);
124 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
125 module_param(downdelay
, int, 0);
126 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
128 module_param(use_carrier
, int, 0);
129 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
130 "0 for off, 1 for on (default)");
131 module_param(mode
, charp
, 0);
132 MODULE_PARM_DESC(mode
, "Mode of operation; 0 for balance-rr, "
133 "1 for active-backup, 2 for balance-xor, "
134 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
135 "6 for balance-alb");
136 module_param(primary
, charp
, 0);
137 MODULE_PARM_DESC(primary
, "Primary network device to use");
138 module_param(primary_reselect
, charp
, 0);
139 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
141 "0 for always (default), "
142 "1 for only if speed of primary is "
144 "2 for only on active slave "
146 module_param(lacp_rate
, charp
, 0);
147 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner; "
148 "0 for slow, 1 for fast");
149 module_param(ad_select
, charp
, 0);
150 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
151 "0 for stable (default), 1 for bandwidth, "
153 module_param(min_links
, int, 0);
154 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
156 module_param(xmit_hash_policy
, charp
, 0);
157 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
158 "0 for layer 2 (default), 1 for layer 3+4, "
160 module_param(arp_interval
, int, 0);
161 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
162 module_param_array(arp_ip_target
, charp
, NULL
, 0);
163 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
164 module_param(arp_validate
, charp
, 0);
165 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
166 "0 for none (default), 1 for active, "
167 "2 for backup, 3 for all");
168 module_param(fail_over_mac
, charp
, 0);
169 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
170 "the same MAC; 0 for none (default), "
171 "1 for active, 2 for follow");
172 module_param(all_slaves_active
, int, 0);
173 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
174 "by setting active flag for all slaves; "
175 "0 for never (default), 1 for always.");
176 module_param(resend_igmp
, int, 0);
177 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
180 /*----------------------------- Global variables ----------------------------*/
182 #ifdef CONFIG_NET_POLL_CONTROLLER
183 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
186 int bond_net_id __read_mostly
;
188 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
189 static int arp_ip_count
;
190 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
191 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
192 static int lacp_fast
;
194 const struct bond_parm_tbl bond_lacp_tbl
[] = {
195 { "slow", AD_LACP_SLOW
},
196 { "fast", AD_LACP_FAST
},
200 const struct bond_parm_tbl bond_mode_tbl
[] = {
201 { "balance-rr", BOND_MODE_ROUNDROBIN
},
202 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
203 { "balance-xor", BOND_MODE_XOR
},
204 { "broadcast", BOND_MODE_BROADCAST
},
205 { "802.3ad", BOND_MODE_8023AD
},
206 { "balance-tlb", BOND_MODE_TLB
},
207 { "balance-alb", BOND_MODE_ALB
},
211 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
212 { "layer2", BOND_XMIT_POLICY_LAYER2
},
213 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
214 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
218 const struct bond_parm_tbl arp_validate_tbl
[] = {
219 { "none", BOND_ARP_VALIDATE_NONE
},
220 { "active", BOND_ARP_VALIDATE_ACTIVE
},
221 { "backup", BOND_ARP_VALIDATE_BACKUP
},
222 { "all", BOND_ARP_VALIDATE_ALL
},
226 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
227 { "none", BOND_FOM_NONE
},
228 { "active", BOND_FOM_ACTIVE
},
229 { "follow", BOND_FOM_FOLLOW
},
233 const struct bond_parm_tbl pri_reselect_tbl
[] = {
234 { "always", BOND_PRI_RESELECT_ALWAYS
},
235 { "better", BOND_PRI_RESELECT_BETTER
},
236 { "failure", BOND_PRI_RESELECT_FAILURE
},
240 struct bond_parm_tbl ad_select_tbl
[] = {
241 { "stable", BOND_AD_STABLE
},
242 { "bandwidth", BOND_AD_BANDWIDTH
},
243 { "count", BOND_AD_COUNT
},
247 /*-------------------------- Forward declarations ---------------------------*/
249 static int bond_init(struct net_device
*bond_dev
);
250 static void bond_uninit(struct net_device
*bond_dev
);
252 /*---------------------------- General routines -----------------------------*/
254 const char *bond_mode_name(int mode
)
256 static const char *names
[] = {
257 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
258 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
259 [BOND_MODE_XOR
] = "load balancing (xor)",
260 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
261 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
262 [BOND_MODE_TLB
] = "transmit load balancing",
263 [BOND_MODE_ALB
] = "adaptive load balancing",
266 if (mode
< 0 || mode
> BOND_MODE_ALB
)
272 /*---------------------------------- VLAN -----------------------------------*/
275 * bond_add_vlan - add a new vlan id on bond
276 * @bond: bond that got the notification
277 * @vlan_id: the vlan id to add
279 * Returns -ENOMEM if allocation failed.
281 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
283 struct vlan_entry
*vlan
;
285 pr_debug("bond: %s, vlan id %d\n",
286 (bond
? bond
->dev
->name
: "None"), vlan_id
);
288 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
292 INIT_LIST_HEAD(&vlan
->vlan_list
);
293 vlan
->vlan_id
= vlan_id
;
295 write_lock_bh(&bond
->lock
);
297 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
299 write_unlock_bh(&bond
->lock
);
301 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
307 * bond_del_vlan - delete a vlan id from bond
308 * @bond: bond that got the notification
309 * @vlan_id: the vlan id to delete
311 * returns -ENODEV if @vlan_id was not found in @bond.
313 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
315 struct vlan_entry
*vlan
;
318 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
321 write_lock_bh(&bond
->lock
);
323 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
324 if (vlan
->vlan_id
== vlan_id
) {
325 list_del(&vlan
->vlan_list
);
327 if (bond_is_lb(bond
))
328 bond_alb_clear_vlan(bond
, vlan_id
);
330 pr_debug("removed VLAN ID %d from bond %s\n",
331 vlan_id
, bond
->dev
->name
);
340 pr_debug("couldn't find VLAN ID %d in bond %s\n",
341 vlan_id
, bond
->dev
->name
);
344 write_unlock_bh(&bond
->lock
);
345 unblock_netpoll_tx();
350 * bond_next_vlan - safely skip to the next item in the vlans list.
351 * @bond: the bond we're working on
352 * @curr: item we're advancing from
354 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
355 * or @curr->next otherwise (even if it is @curr itself again).
357 * Caller must hold bond->lock
359 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
361 struct vlan_entry
*next
, *last
;
363 if (list_empty(&bond
->vlan_list
))
367 next
= list_entry(bond
->vlan_list
.next
,
368 struct vlan_entry
, vlan_list
);
370 last
= list_entry(bond
->vlan_list
.prev
,
371 struct vlan_entry
, vlan_list
);
373 next
= list_entry(bond
->vlan_list
.next
,
374 struct vlan_entry
, vlan_list
);
376 next
= list_entry(curr
->vlan_list
.next
,
377 struct vlan_entry
, vlan_list
);
384 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
387 * bond_dev_queue_xmit - Prepare skb for xmit.
389 * @bond: bond device that got this skb for tx.
390 * @skb: hw accel VLAN tagged skb to transmit
391 * @slave_dev: slave that is supposed to xmit this skbuff
393 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
394 struct net_device
*slave_dev
)
396 skb
->dev
= slave_dev
;
398 skb
->queue_mapping
= bond_queue_mapping(skb
);
400 if (unlikely(netpoll_tx_running(slave_dev
)))
401 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
409 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
410 * We don't protect the slave list iteration with a lock because:
411 * a. This operation is performed in IOCTL context,
412 * b. The operation is protected by the RTNL semaphore in the 8021q code,
413 * c. Holding a lock with BH disabled while directly calling a base driver
414 * entry point is generally a BAD idea.
416 * The design of synchronization/protection for this operation in the 8021q
417 * module is good for one or more VLAN devices over a single physical device
418 * and cannot be extended for a teaming solution like bonding, so there is a
419 * potential race condition here where a net device from the vlan group might
420 * be referenced (either by a base driver or the 8021q code) while it is being
421 * removed from the system. However, it turns out we're not making matters
422 * worse, and if it works for regular VLAN usage it will work here too.
426 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
427 * @bond_dev: bonding net device that got called
428 * @vid: vlan id being added
430 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
432 struct bonding
*bond
= netdev_priv(bond_dev
);
433 struct slave
*slave
, *stop_at
;
436 bond_for_each_slave(bond
, slave
, i
) {
437 res
= vlan_vid_add(slave
->dev
, vid
);
442 res
= bond_add_vlan(bond
, vid
);
444 pr_err("%s: Error: Failed to add vlan id %d\n",
445 bond_dev
->name
, vid
);
452 /* unwind from head to the slave that failed */
454 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
)
455 vlan_vid_del(slave
->dev
, vid
);
461 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
462 * @bond_dev: bonding net device that got called
463 * @vid: vlan id being removed
465 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
467 struct bonding
*bond
= netdev_priv(bond_dev
);
471 bond_for_each_slave(bond
, slave
, i
)
472 vlan_vid_del(slave
->dev
, vid
);
474 res
= bond_del_vlan(bond
, vid
);
476 pr_err("%s: Error: Failed to remove vlan id %d\n",
477 bond_dev
->name
, vid
);
484 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
486 struct vlan_entry
*vlan
;
489 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
490 res
= vlan_vid_add(slave_dev
, vlan
->vlan_id
);
492 pr_warning("%s: Failed to add vlan id %d to device %s\n",
493 bond
->dev
->name
, vlan
->vlan_id
,
498 static void bond_del_vlans_from_slave(struct bonding
*bond
,
499 struct net_device
*slave_dev
)
501 struct vlan_entry
*vlan
;
503 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
506 vlan_vid_del(slave_dev
, vlan
->vlan_id
);
510 /*------------------------------- Link status -------------------------------*/
513 * Set the carrier state for the master according to the state of its
514 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
515 * do special 802.3ad magic.
517 * Returns zero if carrier state does not change, nonzero if it does.
519 static int bond_set_carrier(struct bonding
*bond
)
524 if (bond
->slave_cnt
== 0)
527 if (bond
->params
.mode
== BOND_MODE_8023AD
)
528 return bond_3ad_set_carrier(bond
);
530 bond_for_each_slave(bond
, slave
, i
) {
531 if (slave
->link
== BOND_LINK_UP
) {
532 if (!netif_carrier_ok(bond
->dev
)) {
533 netif_carrier_on(bond
->dev
);
541 if (netif_carrier_ok(bond
->dev
)) {
542 netif_carrier_off(bond
->dev
);
549 * Get link speed and duplex from the slave's base driver
550 * using ethtool. If for some reason the call fails or the
551 * values are invalid, set speed and duplex to -1,
554 static int bond_update_speed_duplex(struct slave
*slave
)
556 struct net_device
*slave_dev
= slave
->dev
;
557 struct ethtool_cmd ecmd
;
561 slave
->speed
= SPEED_UNKNOWN
;
562 slave
->duplex
= DUPLEX_UNKNOWN
;
564 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
568 slave_speed
= ethtool_cmd_speed(&ecmd
);
569 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
572 switch (ecmd
.duplex
) {
580 slave
->speed
= slave_speed
;
581 slave
->duplex
= ecmd
.duplex
;
587 * if <dev> supports MII link status reporting, check its link status.
589 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
590 * depending upon the setting of the use_carrier parameter.
592 * Return either BMSR_LSTATUS, meaning that the link is up (or we
593 * can't tell and just pretend it is), or 0, meaning that the link is
596 * If reporting is non-zero, instead of faking link up, return -1 if
597 * both ETHTOOL and MII ioctls fail (meaning the device does not
598 * support them). If use_carrier is set, return whatever it says.
599 * It'd be nice if there was a good way to tell if a driver supports
600 * netif_carrier, but there really isn't.
602 static int bond_check_dev_link(struct bonding
*bond
,
603 struct net_device
*slave_dev
, int reporting
)
605 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
606 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
608 struct mii_ioctl_data
*mii
;
610 if (!reporting
&& !netif_running(slave_dev
))
613 if (bond
->params
.use_carrier
)
614 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
616 /* Try to get link status using Ethtool first. */
617 if (slave_dev
->ethtool_ops
) {
618 if (slave_dev
->ethtool_ops
->get_link
) {
621 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
623 return link
? BMSR_LSTATUS
: 0;
627 /* Ethtool can't be used, fallback to MII ioctls. */
628 ioctl
= slave_ops
->ndo_do_ioctl
;
630 /* TODO: set pointer to correct ioctl on a per team member */
631 /* bases to make this more efficient. that is, once */
632 /* we determine the correct ioctl, we will always */
633 /* call it and not the others for that team */
637 * We cannot assume that SIOCGMIIPHY will also read a
638 * register; not all network drivers (e.g., e100)
642 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
643 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
645 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
646 mii
->reg_num
= MII_BMSR
;
647 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
648 return mii
->val_out
& BMSR_LSTATUS
;
653 * If reporting, report that either there's no dev->do_ioctl,
654 * or both SIOCGMIIREG and get_link failed (meaning that we
655 * cannot report link status). If not reporting, pretend
658 return reporting
? -1 : BMSR_LSTATUS
;
661 /*----------------------------- Multicast list ------------------------------*/
664 * Push the promiscuity flag down to appropriate slaves
666 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
669 if (USES_PRIMARY(bond
->params
.mode
)) {
670 /* write lock already acquired */
671 if (bond
->curr_active_slave
) {
672 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
678 bond_for_each_slave(bond
, slave
, i
) {
679 err
= dev_set_promiscuity(slave
->dev
, inc
);
688 * Push the allmulti flag down to all slaves
690 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
693 if (USES_PRIMARY(bond
->params
.mode
)) {
694 /* write lock already acquired */
695 if (bond
->curr_active_slave
) {
696 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
702 bond_for_each_slave(bond
, slave
, i
) {
703 err
= dev_set_allmulti(slave
->dev
, inc
);
712 * Add a Multicast address to slaves
715 static void bond_mc_add(struct bonding
*bond
, void *addr
)
717 if (USES_PRIMARY(bond
->params
.mode
)) {
718 /* write lock already acquired */
719 if (bond
->curr_active_slave
)
720 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
725 bond_for_each_slave(bond
, slave
, i
)
726 dev_mc_add(slave
->dev
, addr
);
731 * Remove a multicast address from slave
734 static void bond_mc_del(struct bonding
*bond
, void *addr
)
736 if (USES_PRIMARY(bond
->params
.mode
)) {
737 /* write lock already acquired */
738 if (bond
->curr_active_slave
)
739 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
743 bond_for_each_slave(bond
, slave
, i
) {
744 dev_mc_del(slave
->dev
, addr
);
750 static void __bond_resend_igmp_join_requests(struct net_device
*dev
)
752 struct in_device
*in_dev
;
755 in_dev
= __in_dev_get_rcu(dev
);
757 ip_mc_rejoin_groups(in_dev
);
762 * Retrieve the list of registered multicast addresses for the bonding
763 * device and retransmit an IGMP JOIN request to the current active
766 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
768 struct net_device
*bond_dev
, *vlan_dev
, *master_dev
;
769 struct vlan_entry
*vlan
;
771 read_lock(&bond
->lock
);
773 bond_dev
= bond
->dev
;
775 /* rejoin all groups on bond device */
776 __bond_resend_igmp_join_requests(bond_dev
);
779 * if bond is enslaved to a bridge,
780 * then rejoin all groups on its master
782 master_dev
= bond_dev
->master
;
784 if ((master_dev
->priv_flags
& IFF_EBRIDGE
)
785 && (bond_dev
->priv_flags
& IFF_BRIDGE_PORT
))
786 __bond_resend_igmp_join_requests(master_dev
);
788 /* rejoin all groups on vlan devices */
789 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
791 vlan_dev
= __vlan_find_dev_deep(bond_dev
,
795 __bond_resend_igmp_join_requests(vlan_dev
);
798 if (--bond
->igmp_retrans
> 0)
799 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
801 read_unlock(&bond
->lock
);
804 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
806 struct bonding
*bond
= container_of(work
, struct bonding
,
808 bond_resend_igmp_join_requests(bond
);
812 * flush all members of flush->mc_list from device dev->mc_list
814 static void bond_mc_list_flush(struct net_device
*bond_dev
,
815 struct net_device
*slave_dev
)
817 struct bonding
*bond
= netdev_priv(bond_dev
);
818 struct netdev_hw_addr
*ha
;
820 netdev_for_each_mc_addr(ha
, bond_dev
)
821 dev_mc_del(slave_dev
, ha
->addr
);
823 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
824 /* del lacpdu mc addr from mc list */
825 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
827 dev_mc_del(slave_dev
, lacpdu_multicast
);
831 /*--------------------------- Active slave change ---------------------------*/
834 * Update the mc list and multicast-related flags for the new and
835 * old active slaves (if any) according to the multicast mode, and
836 * promiscuous flags unconditionally.
838 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
839 struct slave
*old_active
)
841 struct netdev_hw_addr
*ha
;
843 if (!USES_PRIMARY(bond
->params
.mode
))
844 /* nothing to do - mc list is already up-to-date on
850 if (bond
->dev
->flags
& IFF_PROMISC
)
851 dev_set_promiscuity(old_active
->dev
, -1);
853 if (bond
->dev
->flags
& IFF_ALLMULTI
)
854 dev_set_allmulti(old_active
->dev
, -1);
856 netdev_for_each_mc_addr(ha
, bond
->dev
)
857 dev_mc_del(old_active
->dev
, ha
->addr
);
861 /* FIXME: Signal errors upstream. */
862 if (bond
->dev
->flags
& IFF_PROMISC
)
863 dev_set_promiscuity(new_active
->dev
, 1);
865 if (bond
->dev
->flags
& IFF_ALLMULTI
)
866 dev_set_allmulti(new_active
->dev
, 1);
868 netdev_for_each_mc_addr(ha
, bond
->dev
)
869 dev_mc_add(new_active
->dev
, ha
->addr
);
874 * bond_do_fail_over_mac
876 * Perform special MAC address swapping for fail_over_mac settings
878 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
880 static void bond_do_fail_over_mac(struct bonding
*bond
,
881 struct slave
*new_active
,
882 struct slave
*old_active
)
883 __releases(&bond
->curr_slave_lock
)
884 __releases(&bond
->lock
)
885 __acquires(&bond
->lock
)
886 __acquires(&bond
->curr_slave_lock
)
888 u8 tmp_mac
[ETH_ALEN
];
889 struct sockaddr saddr
;
892 switch (bond
->params
.fail_over_mac
) {
893 case BOND_FOM_ACTIVE
:
895 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
896 new_active
->dev
->addr_len
);
898 case BOND_FOM_FOLLOW
:
900 * if new_active && old_active, swap them
901 * if just old_active, do nothing (going to no active slave)
902 * if just new_active, set new_active to bond's MAC
907 write_unlock_bh(&bond
->curr_slave_lock
);
908 read_unlock(&bond
->lock
);
911 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
912 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
914 saddr
.sa_family
= new_active
->dev
->type
;
916 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
917 saddr
.sa_family
= bond
->dev
->type
;
920 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
922 pr_err("%s: Error %d setting MAC of slave %s\n",
923 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
930 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
931 saddr
.sa_family
= old_active
->dev
->type
;
933 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
935 pr_err("%s: Error %d setting MAC of slave %s\n",
936 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
938 read_lock(&bond
->lock
);
939 write_lock_bh(&bond
->curr_slave_lock
);
942 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
943 bond
->dev
->name
, bond
->params
.fail_over_mac
);
949 static bool bond_should_change_active(struct bonding
*bond
)
951 struct slave
*prim
= bond
->primary_slave
;
952 struct slave
*curr
= bond
->curr_active_slave
;
954 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
956 if (bond
->force_primary
) {
957 bond
->force_primary
= false;
960 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
961 (prim
->speed
< curr
->speed
||
962 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
964 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
970 * find_best_interface - select the best available slave to be the active one
971 * @bond: our bonding struct
973 * Warning: Caller must hold curr_slave_lock for writing.
975 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
977 struct slave
*new_active
, *old_active
;
978 struct slave
*bestslave
= NULL
;
979 int mintime
= bond
->params
.updelay
;
982 new_active
= bond
->curr_active_slave
;
984 if (!new_active
) { /* there were no active slaves left */
985 if (bond
->slave_cnt
> 0) /* found one slave */
986 new_active
= bond
->first_slave
;
988 return NULL
; /* still no slave, return NULL */
991 if ((bond
->primary_slave
) &&
992 bond
->primary_slave
->link
== BOND_LINK_UP
&&
993 bond_should_change_active(bond
)) {
994 new_active
= bond
->primary_slave
;
997 /* remember where to stop iterating over the slaves */
998 old_active
= new_active
;
1000 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1001 if (new_active
->link
== BOND_LINK_UP
) {
1003 } else if (new_active
->link
== BOND_LINK_BACK
&&
1004 IS_UP(new_active
->dev
)) {
1005 /* link up, but waiting for stabilization */
1006 if (new_active
->delay
< mintime
) {
1007 mintime
= new_active
->delay
;
1008 bestslave
= new_active
;
1016 static bool bond_should_notify_peers(struct bonding
*bond
)
1018 struct slave
*slave
= bond
->curr_active_slave
;
1020 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1021 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
1023 if (!slave
|| !bond
->send_peer_notif
||
1024 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
1027 bond
->send_peer_notif
--;
1032 * change_active_interface - change the active slave into the specified one
1033 * @bond: our bonding struct
1034 * @new: the new slave to make the active one
1036 * Set the new slave to the bond's settings and unset them on the old
1037 * curr_active_slave.
1038 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1040 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1041 * because it is apparently the best available slave we have, even though its
1042 * updelay hasn't timed out yet.
1044 * If new_active is not NULL, caller must hold bond->lock for read and
1045 * curr_slave_lock for write_bh.
1047 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1049 struct slave
*old_active
= bond
->curr_active_slave
;
1051 if (old_active
== new_active
)
1055 new_active
->jiffies
= jiffies
;
1057 if (new_active
->link
== BOND_LINK_BACK
) {
1058 if (USES_PRIMARY(bond
->params
.mode
)) {
1059 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1060 bond
->dev
->name
, new_active
->dev
->name
,
1061 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1064 new_active
->delay
= 0;
1065 new_active
->link
= BOND_LINK_UP
;
1067 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1068 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1070 if (bond_is_lb(bond
))
1071 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1073 if (USES_PRIMARY(bond
->params
.mode
)) {
1074 pr_info("%s: making interface %s the new active one.\n",
1075 bond
->dev
->name
, new_active
->dev
->name
);
1080 if (USES_PRIMARY(bond
->params
.mode
))
1081 bond_mc_swap(bond
, new_active
, old_active
);
1083 if (bond_is_lb(bond
)) {
1084 bond_alb_handle_active_change(bond
, new_active
);
1086 bond_set_slave_inactive_flags(old_active
);
1088 bond_set_slave_active_flags(new_active
);
1090 bond
->curr_active_slave
= new_active
;
1093 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1095 bond_set_slave_inactive_flags(old_active
);
1098 bool should_notify_peers
= false;
1100 bond_set_slave_active_flags(new_active
);
1102 if (bond
->params
.fail_over_mac
)
1103 bond_do_fail_over_mac(bond
, new_active
,
1106 if (netif_running(bond
->dev
)) {
1107 bond
->send_peer_notif
=
1108 bond
->params
.num_peer_notif
;
1109 should_notify_peers
=
1110 bond_should_notify_peers(bond
);
1113 write_unlock_bh(&bond
->curr_slave_lock
);
1114 read_unlock(&bond
->lock
);
1116 netdev_bonding_change(bond
->dev
, NETDEV_BONDING_FAILOVER
);
1117 if (should_notify_peers
)
1118 netdev_bonding_change(bond
->dev
,
1119 NETDEV_NOTIFY_PEERS
);
1121 read_lock(&bond
->lock
);
1122 write_lock_bh(&bond
->curr_slave_lock
);
1126 /* resend IGMP joins since active slave has changed or
1127 * all were sent on curr_active_slave.
1128 * resend only if bond is brought up with the affected
1129 * bonding modes and the retransmission is enabled */
1130 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
1131 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
1132 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
1133 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
1134 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 0);
1139 * bond_select_active_slave - select a new active slave, if needed
1140 * @bond: our bonding struct
1142 * This functions should be called when one of the following occurs:
1143 * - The old curr_active_slave has been released or lost its link.
1144 * - The primary_slave has got its link back.
1145 * - A slave has got its link back and there's no old curr_active_slave.
1147 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1149 void bond_select_active_slave(struct bonding
*bond
)
1151 struct slave
*best_slave
;
1154 best_slave
= bond_find_best_slave(bond
);
1155 if (best_slave
!= bond
->curr_active_slave
) {
1156 bond_change_active_slave(bond
, best_slave
);
1157 rv
= bond_set_carrier(bond
);
1161 if (netif_carrier_ok(bond
->dev
)) {
1162 pr_info("%s: first active interface up!\n",
1165 pr_info("%s: now running without any active interface !\n",
1171 /*--------------------------- slave list handling ---------------------------*/
1174 * This function attaches the slave to the end of list.
1176 * bond->lock held for writing by caller.
1178 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1180 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1181 new_slave
->next
= new_slave
;
1182 new_slave
->prev
= new_slave
;
1183 bond
->first_slave
= new_slave
;
1185 new_slave
->next
= bond
->first_slave
;
1186 new_slave
->prev
= bond
->first_slave
->prev
;
1187 new_slave
->next
->prev
= new_slave
;
1188 new_slave
->prev
->next
= new_slave
;
1195 * This function detaches the slave from the list.
1196 * WARNING: no check is made to verify if the slave effectively
1197 * belongs to <bond>.
1198 * Nothing is freed on return, structures are just unchained.
1199 * If any slave pointer in bond was pointing to <slave>,
1200 * it should be changed by the calling function.
1202 * bond->lock held for writing by caller.
1204 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1207 slave
->next
->prev
= slave
->prev
;
1210 slave
->prev
->next
= slave
->next
;
1212 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1213 if (bond
->slave_cnt
> 1) { /* there are more slave */
1214 bond
->first_slave
= slave
->next
;
1216 bond
->first_slave
= NULL
; /* slave was the last one */
1225 #ifdef CONFIG_NET_POLL_CONTROLLER
1226 static inline int slave_enable_netpoll(struct slave
*slave
)
1231 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
1236 np
->dev
= slave
->dev
;
1237 strlcpy(np
->dev_name
, slave
->dev
->name
, IFNAMSIZ
);
1238 err
= __netpoll_setup(np
);
1247 static inline void slave_disable_netpoll(struct slave
*slave
)
1249 struct netpoll
*np
= slave
->np
;
1255 synchronize_rcu_bh();
1256 __netpoll_cleanup(np
);
1259 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1261 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1263 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1268 static void bond_poll_controller(struct net_device
*bond_dev
)
1272 static void __bond_netpoll_cleanup(struct bonding
*bond
)
1274 struct slave
*slave
;
1277 bond_for_each_slave(bond
, slave
, i
)
1278 if (IS_UP(slave
->dev
))
1279 slave_disable_netpoll(slave
);
1281 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1283 struct bonding
*bond
= netdev_priv(bond_dev
);
1285 read_lock(&bond
->lock
);
1286 __bond_netpoll_cleanup(bond
);
1287 read_unlock(&bond
->lock
);
1290 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
1292 struct bonding
*bond
= netdev_priv(dev
);
1293 struct slave
*slave
;
1296 read_lock(&bond
->lock
);
1297 bond_for_each_slave(bond
, slave
, i
) {
1298 err
= slave_enable_netpoll(slave
);
1300 __bond_netpoll_cleanup(bond
);
1304 read_unlock(&bond
->lock
);
1308 static struct netpoll_info
*bond_netpoll_info(struct bonding
*bond
)
1310 return bond
->dev
->npinfo
;
1314 static inline int slave_enable_netpoll(struct slave
*slave
)
1318 static inline void slave_disable_netpoll(struct slave
*slave
)
1321 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1326 /*---------------------------------- IOCTL ----------------------------------*/
1328 static int bond_sethwaddr(struct net_device
*bond_dev
,
1329 struct net_device
*slave_dev
)
1331 pr_debug("bond_dev=%p\n", bond_dev
);
1332 pr_debug("slave_dev=%p\n", slave_dev
);
1333 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1334 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1338 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1339 netdev_features_t features
)
1341 struct slave
*slave
;
1342 struct bonding
*bond
= netdev_priv(dev
);
1343 netdev_features_t mask
;
1346 read_lock(&bond
->lock
);
1348 if (!bond
->first_slave
) {
1349 /* Disable adding VLANs to empty bond. But why? --mq */
1350 features
|= NETIF_F_VLAN_CHALLENGED
;
1355 features
&= ~NETIF_F_ONE_FOR_ALL
;
1356 features
|= NETIF_F_ALL_FOR_ALL
;
1358 bond_for_each_slave(bond
, slave
, i
) {
1359 features
= netdev_increment_features(features
,
1360 slave
->dev
->features
,
1365 read_unlock(&bond
->lock
);
1369 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1370 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1371 NETIF_F_HIGHDMA | NETIF_F_LRO)
1373 static void bond_compute_features(struct bonding
*bond
)
1375 struct slave
*slave
;
1376 struct net_device
*bond_dev
= bond
->dev
;
1377 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1378 unsigned short max_hard_header_len
= ETH_HLEN
;
1381 read_lock(&bond
->lock
);
1383 if (!bond
->first_slave
)
1386 bond_for_each_slave(bond
, slave
, i
) {
1387 vlan_features
= netdev_increment_features(vlan_features
,
1388 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1390 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1391 max_hard_header_len
= slave
->dev
->hard_header_len
;
1395 bond_dev
->vlan_features
= vlan_features
;
1396 bond_dev
->hard_header_len
= max_hard_header_len
;
1398 read_unlock(&bond
->lock
);
1400 netdev_change_features(bond_dev
);
1403 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1404 struct net_device
*slave_dev
)
1406 struct bonding
*bond
= netdev_priv(bond_dev
);
1408 bond_dev
->header_ops
= slave_dev
->header_ops
;
1410 bond_dev
->type
= slave_dev
->type
;
1411 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1412 bond_dev
->addr_len
= slave_dev
->addr_len
;
1414 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1415 slave_dev
->addr_len
);
1416 bond
->setup_by_slave
= 1;
1419 /* On bonding slaves other than the currently active slave, suppress
1420 * duplicates except for alb non-mcast/bcast.
1422 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1423 struct slave
*slave
,
1424 struct bonding
*bond
)
1426 if (bond_is_slave_inactive(slave
)) {
1427 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1428 skb
->pkt_type
!= PACKET_BROADCAST
&&
1429 skb
->pkt_type
!= PACKET_MULTICAST
)
1436 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1438 struct sk_buff
*skb
= *pskb
;
1439 struct slave
*slave
;
1440 struct bonding
*bond
;
1441 void (*recv_probe
)(struct sk_buff
*, struct bonding
*,
1444 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1446 return RX_HANDLER_CONSUMED
;
1450 slave
= bond_slave_get_rcu(skb
->dev
);
1453 if (bond
->params
.arp_interval
)
1454 slave
->dev
->last_rx
= jiffies
;
1456 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1458 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
1461 recv_probe(nskb
, bond
, slave
);
1462 dev_kfree_skb(nskb
);
1466 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1467 return RX_HANDLER_EXACT
;
1470 skb
->dev
= bond
->dev
;
1472 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1473 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1474 skb
->pkt_type
== PACKET_HOST
) {
1476 if (unlikely(skb_cow_head(skb
,
1477 skb
->data
- skb_mac_header(skb
)))) {
1479 return RX_HANDLER_CONSUMED
;
1481 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1484 return RX_HANDLER_ANOTHER
;
1487 /* enslave device <slave> to bond device <master> */
1488 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1490 struct bonding
*bond
= netdev_priv(bond_dev
);
1491 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1492 struct slave
*new_slave
= NULL
;
1493 struct netdev_hw_addr
*ha
;
1494 struct sockaddr addr
;
1498 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1499 slave_ops
->ndo_do_ioctl
== NULL
) {
1500 pr_warning("%s: Warning: no link monitoring support for %s\n",
1501 bond_dev
->name
, slave_dev
->name
);
1504 /* already enslaved */
1505 if (slave_dev
->flags
& IFF_SLAVE
) {
1506 pr_debug("Error, Device was already enslaved\n");
1510 /* vlan challenged mutual exclusion */
1511 /* no need to lock since we're protected by rtnl_lock */
1512 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1513 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1514 if (bond_vlan_used(bond
)) {
1515 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1516 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1519 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1520 bond_dev
->name
, slave_dev
->name
,
1521 slave_dev
->name
, bond_dev
->name
);
1524 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1528 * Old ifenslave binaries are no longer supported. These can
1529 * be identified with moderate accuracy by the state of the slave:
1530 * the current ifenslave will set the interface down prior to
1531 * enslaving it; the old ifenslave will not.
1533 if ((slave_dev
->flags
& IFF_UP
)) {
1534 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1537 goto err_undo_flags
;
1540 /* set bonding device ether type by slave - bonding netdevices are
1541 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1542 * there is a need to override some of the type dependent attribs/funcs.
1544 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1545 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1547 if (bond
->slave_cnt
== 0) {
1548 if (bond_dev
->type
!= slave_dev
->type
) {
1549 pr_debug("%s: change device type from %d to %d\n",
1551 bond_dev
->type
, slave_dev
->type
);
1553 res
= netdev_bonding_change(bond_dev
,
1554 NETDEV_PRE_TYPE_CHANGE
);
1555 res
= notifier_to_errno(res
);
1557 pr_err("%s: refused to change device type\n",
1560 goto err_undo_flags
;
1563 /* Flush unicast and multicast addresses */
1564 dev_uc_flush(bond_dev
);
1565 dev_mc_flush(bond_dev
);
1567 if (slave_dev
->type
!= ARPHRD_ETHER
)
1568 bond_setup_by_slave(bond_dev
, slave_dev
);
1570 ether_setup(bond_dev
);
1571 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1574 netdev_bonding_change(bond_dev
,
1575 NETDEV_POST_TYPE_CHANGE
);
1577 } else if (bond_dev
->type
!= slave_dev
->type
) {
1578 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1580 slave_dev
->type
, bond_dev
->type
);
1582 goto err_undo_flags
;
1585 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1586 if (bond
->slave_cnt
== 0) {
1587 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1589 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1590 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1591 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",
1594 goto err_undo_flags
;
1598 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1600 /* If this is the first slave, then we need to set the master's hardware
1601 * address to be the same as the slave's. */
1602 if (is_zero_ether_addr(bond
->dev
->dev_addr
))
1603 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1604 slave_dev
->addr_len
);
1607 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1610 goto err_undo_flags
;
1614 * Set the new_slave's queue_id to be zero. Queue ID mapping
1615 * is set via sysfs or module option if desired.
1617 new_slave
->queue_id
= 0;
1619 /* Save slave's original mtu and then set it to match the bond */
1620 new_slave
->original_mtu
= slave_dev
->mtu
;
1621 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1623 pr_debug("Error %d calling dev_set_mtu\n", res
);
1628 * Save slave's original ("permanent") mac address for modes
1629 * that need it, and for restoring it upon release, and then
1630 * set it to the master's address
1632 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1634 if (!bond
->params
.fail_over_mac
) {
1636 * Set slave to master's mac address. The application already
1637 * set the master's mac address to that of the first slave
1639 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1640 addr
.sa_family
= slave_dev
->type
;
1641 res
= dev_set_mac_address(slave_dev
, &addr
);
1643 pr_debug("Error %d calling set_mac_address\n", res
);
1644 goto err_restore_mtu
;
1648 res
= netdev_set_bond_master(slave_dev
, bond_dev
);
1650 pr_debug("Error %d calling netdev_set_bond_master\n", res
);
1651 goto err_restore_mac
;
1654 /* open the slave since the application closed it */
1655 res
= dev_open(slave_dev
);
1657 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1658 goto err_unset_master
;
1661 new_slave
->bond
= bond
;
1662 new_slave
->dev
= slave_dev
;
1663 slave_dev
->priv_flags
|= IFF_BONDING
;
1665 if (bond_is_lb(bond
)) {
1666 /* bond_alb_init_slave() must be called before all other stages since
1667 * it might fail and we do not want to have to undo everything
1669 res
= bond_alb_init_slave(bond
, new_slave
);
1674 /* If the mode USES_PRIMARY, then the new slave gets the
1675 * master's promisc (and mc) settings only if it becomes the
1676 * curr_active_slave, and that is taken care of later when calling
1677 * bond_change_active()
1679 if (!USES_PRIMARY(bond
->params
.mode
)) {
1680 /* set promiscuity level to new slave */
1681 if (bond_dev
->flags
& IFF_PROMISC
) {
1682 res
= dev_set_promiscuity(slave_dev
, 1);
1687 /* set allmulti level to new slave */
1688 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1689 res
= dev_set_allmulti(slave_dev
, 1);
1694 netif_addr_lock_bh(bond_dev
);
1695 /* upload master's mc_list to new slave */
1696 netdev_for_each_mc_addr(ha
, bond_dev
)
1697 dev_mc_add(slave_dev
, ha
->addr
);
1698 netif_addr_unlock_bh(bond_dev
);
1701 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1702 /* add lacpdu mc addr to mc list */
1703 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1705 dev_mc_add(slave_dev
, lacpdu_multicast
);
1708 bond_add_vlans_on_slave(bond
, slave_dev
);
1710 write_lock_bh(&bond
->lock
);
1712 bond_attach_slave(bond
, new_slave
);
1714 new_slave
->delay
= 0;
1715 new_slave
->link_failure_count
= 0;
1717 write_unlock_bh(&bond
->lock
);
1719 bond_compute_features(bond
);
1721 read_lock(&bond
->lock
);
1723 new_slave
->last_arp_rx
= jiffies
;
1725 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1726 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1728 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1730 * miimon is set but a bonded network driver
1731 * does not support ETHTOOL/MII and
1732 * arp_interval is not set. Note: if
1733 * use_carrier is enabled, we will never go
1734 * here (because netif_carrier is always
1735 * supported); thus, we don't need to change
1736 * the messages for netif_carrier.
1738 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",
1739 bond_dev
->name
, slave_dev
->name
);
1740 } else if (link_reporting
== -1) {
1741 /* unable get link status using mii/ethtool */
1742 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",
1743 bond_dev
->name
, slave_dev
->name
);
1747 /* check for initial state */
1748 if (!bond
->params
.miimon
||
1749 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1750 if (bond
->params
.updelay
) {
1751 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1752 new_slave
->link
= BOND_LINK_BACK
;
1753 new_slave
->delay
= bond
->params
.updelay
;
1755 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1756 new_slave
->link
= BOND_LINK_UP
;
1758 new_slave
->jiffies
= jiffies
;
1760 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1761 new_slave
->link
= BOND_LINK_DOWN
;
1764 bond_update_speed_duplex(new_slave
);
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 write_lock_bh(&bond
->lock
);
1868 bond_detach_slave(bond
, new_slave
);
1869 write_unlock_bh(&bond
->lock
);
1872 dev_close(slave_dev
);
1875 netdev_set_bond_master(slave_dev
, NULL
);
1878 if (!bond
->params
.fail_over_mac
) {
1879 /* XXX TODO - fom follow mode needs to change master's
1880 * MAC if this slave's MAC is in use by the bond, or at
1881 * least print a warning.
1883 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1884 addr
.sa_family
= slave_dev
->type
;
1885 dev_set_mac_address(slave_dev
, &addr
);
1889 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1895 bond_compute_features(bond
);
1901 * Try to release the slave device <slave> from the bond device <master>
1902 * It is legal to access curr_active_slave without a lock because all the function
1905 * The rules for slave state should be:
1906 * for Active/Backup:
1907 * Active stays on all backups go down
1908 * for Bonded connections:
1909 * The first up interface should be left on and all others downed.
1911 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1913 struct bonding
*bond
= netdev_priv(bond_dev
);
1914 struct slave
*slave
, *oldcurrent
;
1915 struct sockaddr addr
;
1916 netdev_features_t old_features
= bond_dev
->features
;
1918 /* slave is not a slave or master is not master of this slave */
1919 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1920 (slave_dev
->master
!= bond_dev
)) {
1921 pr_err("%s: Error: cannot release %s.\n",
1922 bond_dev
->name
, slave_dev
->name
);
1927 netdev_bonding_change(bond_dev
, NETDEV_RELEASE
);
1928 write_lock_bh(&bond
->lock
);
1930 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1932 /* not a slave of this bond */
1933 pr_info("%s: %s not enslaved\n",
1934 bond_dev
->name
, slave_dev
->name
);
1935 write_unlock_bh(&bond
->lock
);
1936 unblock_netpoll_tx();
1940 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1941 * for this slave anymore.
1943 netdev_rx_handler_unregister(slave_dev
);
1944 write_unlock_bh(&bond
->lock
);
1946 write_lock_bh(&bond
->lock
);
1948 if (!bond
->params
.fail_over_mac
) {
1949 if (!compare_ether_addr(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1950 bond
->slave_cnt
> 1)
1951 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",
1952 bond_dev
->name
, slave_dev
->name
,
1954 bond_dev
->name
, slave_dev
->name
);
1957 /* Inform AD package of unbinding of slave. */
1958 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1959 /* must be called before the slave is
1960 * detached from the list
1962 bond_3ad_unbind_slave(slave
);
1965 pr_info("%s: releasing %s interface %s\n",
1967 bond_is_active_slave(slave
) ? "active" : "backup",
1970 oldcurrent
= bond
->curr_active_slave
;
1972 bond
->current_arp_slave
= NULL
;
1974 /* release the slave from its bond */
1975 bond_detach_slave(bond
, slave
);
1977 if (bond
->primary_slave
== slave
)
1978 bond
->primary_slave
= NULL
;
1980 if (oldcurrent
== slave
)
1981 bond_change_active_slave(bond
, NULL
);
1983 if (bond_is_lb(bond
)) {
1984 /* Must be called only after the slave has been
1985 * detached from the list and the curr_active_slave
1986 * has been cleared (if our_slave == old_current),
1987 * but before a new active slave is selected.
1989 write_unlock_bh(&bond
->lock
);
1990 bond_alb_deinit_slave(bond
, slave
);
1991 write_lock_bh(&bond
->lock
);
1994 if (oldcurrent
== slave
) {
1996 * Note that we hold RTNL over this sequence, so there
1997 * is no concern that another slave add/remove event
2000 write_unlock_bh(&bond
->lock
);
2001 read_lock(&bond
->lock
);
2002 write_lock_bh(&bond
->curr_slave_lock
);
2004 bond_select_active_slave(bond
);
2006 write_unlock_bh(&bond
->curr_slave_lock
);
2007 read_unlock(&bond
->lock
);
2008 write_lock_bh(&bond
->lock
);
2011 if (bond
->slave_cnt
== 0) {
2012 bond_set_carrier(bond
);
2014 /* if the last slave was removed, zero the mac address
2015 * of the master so it will be set by the application
2016 * to the mac address of the first slave
2018 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2020 if (bond_vlan_used(bond
)) {
2021 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2022 bond_dev
->name
, bond_dev
->name
);
2023 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2028 write_unlock_bh(&bond
->lock
);
2029 unblock_netpoll_tx();
2031 bond_compute_features(bond
);
2032 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
2033 (old_features
& NETIF_F_VLAN_CHALLENGED
))
2034 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2035 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
2037 /* must do this from outside any spinlocks */
2038 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2040 bond_del_vlans_from_slave(bond
, slave_dev
);
2042 /* If the mode USES_PRIMARY, then we should only remove its
2043 * promisc and mc settings if it was the curr_active_slave, but that was
2044 * already taken care of above when we detached the slave
2046 if (!USES_PRIMARY(bond
->params
.mode
)) {
2047 /* unset promiscuity level from slave */
2048 if (bond_dev
->flags
& IFF_PROMISC
)
2049 dev_set_promiscuity(slave_dev
, -1);
2051 /* unset allmulti level from slave */
2052 if (bond_dev
->flags
& IFF_ALLMULTI
)
2053 dev_set_allmulti(slave_dev
, -1);
2055 /* flush master's mc_list from slave */
2056 netif_addr_lock_bh(bond_dev
);
2057 bond_mc_list_flush(bond_dev
, slave_dev
);
2058 netif_addr_unlock_bh(bond_dev
);
2061 netdev_set_bond_master(slave_dev
, NULL
);
2063 slave_disable_netpoll(slave
);
2065 /* close slave before restoring its mac address */
2066 dev_close(slave_dev
);
2068 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2069 /* restore original ("permanent") mac address */
2070 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2071 addr
.sa_family
= slave_dev
->type
;
2072 dev_set_mac_address(slave_dev
, &addr
);
2075 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2077 slave_dev
->priv_flags
&= ~IFF_BONDING
;
2081 return 0; /* deletion OK */
2085 * First release a slave and then destroy the bond if no more slaves are left.
2086 * Must be under rtnl_lock when this function is called.
2088 static int bond_release_and_destroy(struct net_device
*bond_dev
,
2089 struct net_device
*slave_dev
)
2091 struct bonding
*bond
= netdev_priv(bond_dev
);
2094 ret
= bond_release(bond_dev
, slave_dev
);
2095 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2096 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
2097 pr_info("%s: destroying bond %s.\n",
2098 bond_dev
->name
, bond_dev
->name
);
2099 unregister_netdevice(bond_dev
);
2105 * This function releases all slaves.
2107 static int bond_release_all(struct net_device
*bond_dev
)
2109 struct bonding
*bond
= netdev_priv(bond_dev
);
2110 struct slave
*slave
;
2111 struct net_device
*slave_dev
;
2112 struct sockaddr addr
;
2114 write_lock_bh(&bond
->lock
);
2116 netif_carrier_off(bond_dev
);
2118 if (bond
->slave_cnt
== 0)
2121 bond
->current_arp_slave
= NULL
;
2122 bond
->primary_slave
= NULL
;
2123 bond_change_active_slave(bond
, NULL
);
2125 while ((slave
= bond
->first_slave
) != NULL
) {
2126 /* Inform AD package of unbinding of slave
2127 * before slave is detached from the list.
2129 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2130 bond_3ad_unbind_slave(slave
);
2132 slave_dev
= slave
->dev
;
2133 bond_detach_slave(bond
, slave
);
2135 /* now that the slave is detached, unlock and perform
2136 * all the undo steps that should not be called from
2139 write_unlock_bh(&bond
->lock
);
2141 /* unregister rx_handler early so bond_handle_frame wouldn't
2142 * be called for this slave anymore.
2144 netdev_rx_handler_unregister(slave_dev
);
2147 if (bond_is_lb(bond
)) {
2148 /* must be called only after the slave
2149 * has been detached from the list
2151 bond_alb_deinit_slave(bond
, slave
);
2154 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2155 bond_del_vlans_from_slave(bond
, slave_dev
);
2157 /* If the mode USES_PRIMARY, then we should only remove its
2158 * promisc and mc settings if it was the curr_active_slave, but that was
2159 * already taken care of above when we detached the slave
2161 if (!USES_PRIMARY(bond
->params
.mode
)) {
2162 /* unset promiscuity level from slave */
2163 if (bond_dev
->flags
& IFF_PROMISC
)
2164 dev_set_promiscuity(slave_dev
, -1);
2166 /* unset allmulti level from slave */
2167 if (bond_dev
->flags
& IFF_ALLMULTI
)
2168 dev_set_allmulti(slave_dev
, -1);
2170 /* flush master's mc_list from slave */
2171 netif_addr_lock_bh(bond_dev
);
2172 bond_mc_list_flush(bond_dev
, slave_dev
);
2173 netif_addr_unlock_bh(bond_dev
);
2176 netdev_set_bond_master(slave_dev
, NULL
);
2178 slave_disable_netpoll(slave
);
2180 /* close slave before restoring its mac address */
2181 dev_close(slave_dev
);
2183 if (!bond
->params
.fail_over_mac
) {
2184 /* restore original ("permanent") mac address*/
2185 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2186 addr
.sa_family
= slave_dev
->type
;
2187 dev_set_mac_address(slave_dev
, &addr
);
2192 /* re-acquire the lock before getting the next slave */
2193 write_lock_bh(&bond
->lock
);
2196 /* zero the mac address of the master so it will be
2197 * set by the application to the mac address of the
2200 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2202 if (bond_vlan_used(bond
)) {
2203 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2204 bond_dev
->name
, bond_dev
->name
);
2205 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2209 pr_info("%s: released all slaves\n", bond_dev
->name
);
2212 write_unlock_bh(&bond
->lock
);
2214 bond_compute_features(bond
);
2220 * This function changes the active slave to slave <slave_dev>.
2221 * It returns -EINVAL in the following cases.
2222 * - <slave_dev> is not found in the list.
2223 * - There is not active slave now.
2224 * - <slave_dev> is already active.
2225 * - The link state of <slave_dev> is not BOND_LINK_UP.
2226 * - <slave_dev> is not running.
2227 * In these cases, this function does nothing.
2228 * In the other cases, current_slave pointer is changed and 0 is returned.
2230 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2232 struct bonding
*bond
= netdev_priv(bond_dev
);
2233 struct slave
*old_active
= NULL
;
2234 struct slave
*new_active
= NULL
;
2237 if (!USES_PRIMARY(bond
->params
.mode
))
2240 /* Verify that master_dev is indeed the master of slave_dev */
2241 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2244 read_lock(&bond
->lock
);
2246 read_lock(&bond
->curr_slave_lock
);
2247 old_active
= bond
->curr_active_slave
;
2248 read_unlock(&bond
->curr_slave_lock
);
2250 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2253 * Changing to the current active: do nothing; return success.
2255 if (new_active
&& (new_active
== old_active
)) {
2256 read_unlock(&bond
->lock
);
2262 (new_active
->link
== BOND_LINK_UP
) &&
2263 IS_UP(new_active
->dev
)) {
2265 write_lock_bh(&bond
->curr_slave_lock
);
2266 bond_change_active_slave(bond
, new_active
);
2267 write_unlock_bh(&bond
->curr_slave_lock
);
2268 unblock_netpoll_tx();
2272 read_unlock(&bond
->lock
);
2277 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2279 struct bonding
*bond
= netdev_priv(bond_dev
);
2281 info
->bond_mode
= bond
->params
.mode
;
2282 info
->miimon
= bond
->params
.miimon
;
2284 read_lock(&bond
->lock
);
2285 info
->num_slaves
= bond
->slave_cnt
;
2286 read_unlock(&bond
->lock
);
2291 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2293 struct bonding
*bond
= netdev_priv(bond_dev
);
2294 struct slave
*slave
;
2295 int i
, res
= -ENODEV
;
2297 read_lock(&bond
->lock
);
2299 bond_for_each_slave(bond
, slave
, i
) {
2300 if (i
== (int)info
->slave_id
) {
2302 strcpy(info
->slave_name
, slave
->dev
->name
);
2303 info
->link
= slave
->link
;
2304 info
->state
= bond_slave_state(slave
);
2305 info
->link_failure_count
= slave
->link_failure_count
;
2310 read_unlock(&bond
->lock
);
2315 /*-------------------------------- Monitoring -------------------------------*/
2318 static int bond_miimon_inspect(struct bonding
*bond
)
2320 struct slave
*slave
;
2321 int i
, link_state
, commit
= 0;
2322 bool ignore_updelay
;
2324 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2326 bond_for_each_slave(bond
, slave
, i
) {
2327 slave
->new_link
= BOND_LINK_NOCHANGE
;
2329 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2331 switch (slave
->link
) {
2336 slave
->link
= BOND_LINK_FAIL
;
2337 slave
->delay
= bond
->params
.downdelay
;
2339 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2341 (bond
->params
.mode
==
2342 BOND_MODE_ACTIVEBACKUP
) ?
2343 (bond_is_active_slave(slave
) ?
2344 "active " : "backup ") : "",
2346 bond
->params
.downdelay
* bond
->params
.miimon
);
2349 case BOND_LINK_FAIL
:
2352 * recovered before downdelay expired
2354 slave
->link
= BOND_LINK_UP
;
2355 slave
->jiffies
= jiffies
;
2356 pr_info("%s: link status up again after %d ms for interface %s.\n",
2358 (bond
->params
.downdelay
- slave
->delay
) *
2359 bond
->params
.miimon
,
2364 if (slave
->delay
<= 0) {
2365 slave
->new_link
= BOND_LINK_DOWN
;
2373 case BOND_LINK_DOWN
:
2377 slave
->link
= BOND_LINK_BACK
;
2378 slave
->delay
= bond
->params
.updelay
;
2381 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2382 bond
->dev
->name
, slave
->dev
->name
,
2383 ignore_updelay
? 0 :
2384 bond
->params
.updelay
*
2385 bond
->params
.miimon
);
2388 case BOND_LINK_BACK
:
2390 slave
->link
= BOND_LINK_DOWN
;
2391 pr_info("%s: link status down again after %d ms for interface %s.\n",
2393 (bond
->params
.updelay
- slave
->delay
) *
2394 bond
->params
.miimon
,
2403 if (slave
->delay
<= 0) {
2404 slave
->new_link
= BOND_LINK_UP
;
2406 ignore_updelay
= false;
2418 static void bond_miimon_commit(struct bonding
*bond
)
2420 struct slave
*slave
;
2423 bond_for_each_slave(bond
, slave
, i
) {
2424 switch (slave
->new_link
) {
2425 case BOND_LINK_NOCHANGE
:
2429 slave
->link
= BOND_LINK_UP
;
2430 slave
->jiffies
= jiffies
;
2432 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2433 /* prevent it from being the active one */
2434 bond_set_backup_slave(slave
);
2435 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2436 /* make it immediately active */
2437 bond_set_active_slave(slave
);
2438 } else if (slave
!= bond
->primary_slave
) {
2439 /* prevent it from being the active one */
2440 bond_set_backup_slave(slave
);
2443 bond_update_speed_duplex(slave
);
2445 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2446 bond
->dev
->name
, slave
->dev
->name
,
2447 slave
->speed
, slave
->duplex
? "full" : "half");
2449 /* notify ad that the link status has changed */
2450 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2451 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2453 if (bond_is_lb(bond
))
2454 bond_alb_handle_link_change(bond
, slave
,
2457 if (!bond
->curr_active_slave
||
2458 (slave
== bond
->primary_slave
))
2463 case BOND_LINK_DOWN
:
2464 if (slave
->link_failure_count
< UINT_MAX
)
2465 slave
->link_failure_count
++;
2467 slave
->link
= BOND_LINK_DOWN
;
2469 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2470 bond
->params
.mode
== BOND_MODE_8023AD
)
2471 bond_set_slave_inactive_flags(slave
);
2473 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2474 bond
->dev
->name
, slave
->dev
->name
);
2476 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2477 bond_3ad_handle_link_change(slave
,
2480 if (bond_is_lb(bond
))
2481 bond_alb_handle_link_change(bond
, slave
,
2484 if (slave
== bond
->curr_active_slave
)
2490 pr_err("%s: invalid new link %d on slave %s\n",
2491 bond
->dev
->name
, slave
->new_link
,
2493 slave
->new_link
= BOND_LINK_NOCHANGE
;
2501 write_lock_bh(&bond
->curr_slave_lock
);
2502 bond_select_active_slave(bond
);
2503 write_unlock_bh(&bond
->curr_slave_lock
);
2504 unblock_netpoll_tx();
2507 bond_set_carrier(bond
);
2513 * Really a wrapper that splits the mii monitor into two phases: an
2514 * inspection, then (if inspection indicates something needs to be done)
2515 * an acquisition of appropriate locks followed by a commit phase to
2516 * implement whatever link state changes are indicated.
2518 void bond_mii_monitor(struct work_struct
*work
)
2520 struct bonding
*bond
= container_of(work
, struct bonding
,
2522 bool should_notify_peers
= false;
2523 unsigned long delay
;
2525 read_lock(&bond
->lock
);
2527 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2529 if (bond
->slave_cnt
== 0)
2532 should_notify_peers
= bond_should_notify_peers(bond
);
2534 if (bond_miimon_inspect(bond
)) {
2535 read_unlock(&bond
->lock
);
2537 /* Race avoidance with bond_close cancel of workqueue */
2538 if (!rtnl_trylock()) {
2539 read_lock(&bond
->lock
);
2541 should_notify_peers
= false;
2545 read_lock(&bond
->lock
);
2547 bond_miimon_commit(bond
);
2549 read_unlock(&bond
->lock
);
2550 rtnl_unlock(); /* might sleep, hold no other locks */
2551 read_lock(&bond
->lock
);
2555 if (bond
->params
.miimon
)
2556 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2558 read_unlock(&bond
->lock
);
2560 if (should_notify_peers
) {
2561 if (!rtnl_trylock()) {
2562 read_lock(&bond
->lock
);
2563 bond
->send_peer_notif
++;
2564 read_unlock(&bond
->lock
);
2567 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
2572 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2574 struct vlan_entry
*vlan
;
2576 if (ip
== bond
->master_ip
)
2579 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2580 if (ip
== vlan
->vlan_ip
)
2588 * We go to the (large) trouble of VLAN tagging ARP frames because
2589 * switches in VLAN mode (especially if ports are configured as
2590 * "native" to a VLAN) might not pass non-tagged frames.
2592 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2594 struct sk_buff
*skb
;
2596 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2597 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2599 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2600 NULL
, slave_dev
->dev_addr
, NULL
);
2603 pr_err("ARP packet allocation failed\n");
2607 skb
= vlan_put_tag(skb
, vlan_id
);
2609 pr_err("failed to insert VLAN tag\n");
2617 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2620 __be32
*targets
= bond
->params
.arp_targets
;
2621 struct vlan_entry
*vlan
;
2622 struct net_device
*vlan_dev
;
2625 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2628 pr_debug("basa: target %x\n", targets
[i
]);
2629 if (!bond_vlan_used(bond
)) {
2630 pr_debug("basa: empty vlan: arp_send\n");
2631 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2632 bond
->master_ip
, 0);
2637 * If VLANs are configured, we do a route lookup to
2638 * determine which VLAN interface would be used, so we
2639 * can tag the ARP with the proper VLAN tag.
2641 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2644 if (net_ratelimit()) {
2645 pr_warning("%s: no route to arp_ip_target %pI4\n",
2646 bond
->dev
->name
, &targets
[i
]);
2652 * This target is not on a VLAN
2654 if (rt
->dst
.dev
== bond
->dev
) {
2656 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2657 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2658 bond
->master_ip
, 0);
2663 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2665 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
2668 if (vlan_dev
== rt
->dst
.dev
) {
2669 vlan_id
= vlan
->vlan_id
;
2670 pr_debug("basa: vlan match on %s %d\n",
2671 vlan_dev
->name
, vlan_id
);
2678 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2679 vlan
->vlan_ip
, vlan_id
);
2683 if (net_ratelimit()) {
2684 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2685 bond
->dev
->name
, &targets
[i
],
2686 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2692 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2695 __be32
*targets
= bond
->params
.arp_targets
;
2697 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2698 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2699 &sip
, &tip
, i
, &targets
[i
],
2700 bond_has_this_ip(bond
, tip
));
2701 if (sip
== targets
[i
]) {
2702 if (bond_has_this_ip(bond
, tip
))
2703 slave
->last_arp_rx
= jiffies
;
2709 static void bond_arp_rcv(struct sk_buff
*skb
, struct bonding
*bond
,
2710 struct slave
*slave
)
2713 unsigned char *arp_ptr
;
2716 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2719 read_lock(&bond
->lock
);
2721 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2722 bond
->dev
->name
, skb
->dev
->name
);
2724 if (!pskb_may_pull(skb
, arp_hdr_len(bond
->dev
)))
2728 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2729 skb
->pkt_type
== PACKET_OTHERHOST
||
2730 skb
->pkt_type
== PACKET_LOOPBACK
||
2731 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2732 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2736 arp_ptr
= (unsigned char *)(arp
+ 1);
2737 arp_ptr
+= bond
->dev
->addr_len
;
2738 memcpy(&sip
, arp_ptr
, 4);
2739 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2740 memcpy(&tip
, arp_ptr
, 4);
2742 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2743 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2744 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2748 * Backup slaves won't see the ARP reply, but do come through
2749 * here for each ARP probe (so we swap the sip/tip to validate
2750 * the probe). In a "redundant switch, common router" type of
2751 * configuration, the ARP probe will (hopefully) travel from
2752 * the active, through one switch, the router, then the other
2753 * switch before reaching the backup.
2755 if (bond_is_active_slave(slave
))
2756 bond_validate_arp(bond
, slave
, sip
, tip
);
2758 bond_validate_arp(bond
, slave
, tip
, sip
);
2761 read_unlock(&bond
->lock
);
2765 * this function is called regularly to monitor each slave's link
2766 * ensuring that traffic is being sent and received when arp monitoring
2767 * is used in load-balancing mode. if the adapter has been dormant, then an
2768 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2769 * arp monitoring in active backup mode.
2771 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2773 struct bonding
*bond
= container_of(work
, struct bonding
,
2775 struct slave
*slave
, *oldcurrent
;
2776 int do_failover
= 0;
2780 read_lock(&bond
->lock
);
2782 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2784 if (bond
->slave_cnt
== 0)
2787 read_lock(&bond
->curr_slave_lock
);
2788 oldcurrent
= bond
->curr_active_slave
;
2789 read_unlock(&bond
->curr_slave_lock
);
2791 /* see if any of the previous devices are up now (i.e. they have
2792 * xmt and rcv traffic). the curr_active_slave does not come into
2793 * the picture unless it is null. also, slave->jiffies is not needed
2794 * here because we send an arp on each slave and give a slave as
2795 * long as it needs to get the tx/rx within the delta.
2796 * TODO: what about up/down delay in arp mode? it wasn't here before
2799 bond_for_each_slave(bond
, slave
, i
) {
2800 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2802 if (slave
->link
!= BOND_LINK_UP
) {
2803 if (time_in_range(jiffies
,
2804 trans_start
- delta_in_ticks
,
2805 trans_start
+ delta_in_ticks
) &&
2806 time_in_range(jiffies
,
2807 slave
->dev
->last_rx
- delta_in_ticks
,
2808 slave
->dev
->last_rx
+ delta_in_ticks
)) {
2810 slave
->link
= BOND_LINK_UP
;
2811 bond_set_active_slave(slave
);
2813 /* primary_slave has no meaning in round-robin
2814 * mode. the window of a slave being up and
2815 * curr_active_slave being null after enslaving
2819 pr_info("%s: link status definitely up for interface %s, ",
2824 pr_info("%s: interface %s is now up\n",
2830 /* slave->link == BOND_LINK_UP */
2832 /* not all switches will respond to an arp request
2833 * when the source ip is 0, so don't take the link down
2834 * if we don't know our ip yet
2836 if (!time_in_range(jiffies
,
2837 trans_start
- delta_in_ticks
,
2838 trans_start
+ 2 * delta_in_ticks
) ||
2839 !time_in_range(jiffies
,
2840 slave
->dev
->last_rx
- delta_in_ticks
,
2841 slave
->dev
->last_rx
+ 2 * delta_in_ticks
)) {
2843 slave
->link
= BOND_LINK_DOWN
;
2844 bond_set_backup_slave(slave
);
2846 if (slave
->link_failure_count
< UINT_MAX
)
2847 slave
->link_failure_count
++;
2849 pr_info("%s: interface %s is now down.\n",
2853 if (slave
== oldcurrent
)
2858 /* note: if switch is in round-robin mode, all links
2859 * must tx arp to ensure all links rx an arp - otherwise
2860 * links may oscillate or not come up at all; if switch is
2861 * in something like xor mode, there is nothing we can
2862 * do - all replies will be rx'ed on same link causing slaves
2863 * to be unstable during low/no traffic periods
2865 if (IS_UP(slave
->dev
))
2866 bond_arp_send_all(bond
, slave
);
2871 write_lock_bh(&bond
->curr_slave_lock
);
2873 bond_select_active_slave(bond
);
2875 write_unlock_bh(&bond
->curr_slave_lock
);
2876 unblock_netpoll_tx();
2880 if (bond
->params
.arp_interval
)
2881 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2883 read_unlock(&bond
->lock
);
2887 * Called to inspect slaves for active-backup mode ARP monitor link state
2888 * changes. Sets new_link in slaves to specify what action should take
2889 * place for the slave. Returns 0 if no changes are found, >0 if changes
2890 * to link states must be committed.
2892 * Called with bond->lock held for read.
2894 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2896 struct slave
*slave
;
2898 unsigned long trans_start
;
2900 bond_for_each_slave(bond
, slave
, i
) {
2901 slave
->new_link
= BOND_LINK_NOCHANGE
;
2903 if (slave
->link
!= BOND_LINK_UP
) {
2904 if (time_in_range(jiffies
,
2905 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2906 slave_last_rx(bond
, slave
) + delta_in_ticks
)) {
2908 slave
->new_link
= BOND_LINK_UP
;
2916 * Give slaves 2*delta after being enslaved or made
2917 * active. This avoids bouncing, as the last receive
2918 * times need a full ARP monitor cycle to be updated.
2920 if (time_in_range(jiffies
,
2921 slave
->jiffies
- delta_in_ticks
,
2922 slave
->jiffies
+ 2 * delta_in_ticks
))
2926 * Backup slave is down if:
2927 * - No current_arp_slave AND
2928 * - more than 3*delta since last receive AND
2929 * - the bond has an IP address
2931 * Note: a non-null current_arp_slave indicates
2932 * the curr_active_slave went down and we are
2933 * searching for a new one; under this condition
2934 * we only take the curr_active_slave down - this
2935 * gives each slave a chance to tx/rx traffic
2936 * before being taken out
2938 if (!bond_is_active_slave(slave
) &&
2939 !bond
->current_arp_slave
&&
2940 !time_in_range(jiffies
,
2941 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2942 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
)) {
2944 slave
->new_link
= BOND_LINK_DOWN
;
2949 * Active slave is down if:
2950 * - more than 2*delta since transmitting OR
2951 * - (more than 2*delta since receive AND
2952 * the bond has an IP address)
2954 trans_start
= dev_trans_start(slave
->dev
);
2955 if (bond_is_active_slave(slave
) &&
2956 (!time_in_range(jiffies
,
2957 trans_start
- delta_in_ticks
,
2958 trans_start
+ 2 * delta_in_ticks
) ||
2959 !time_in_range(jiffies
,
2960 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2961 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
))) {
2963 slave
->new_link
= BOND_LINK_DOWN
;
2972 * Called to commit link state changes noted by inspection step of
2973 * active-backup mode ARP monitor.
2975 * Called with RTNL and bond->lock for read.
2977 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
2979 struct slave
*slave
;
2981 unsigned long trans_start
;
2983 bond_for_each_slave(bond
, slave
, i
) {
2984 switch (slave
->new_link
) {
2985 case BOND_LINK_NOCHANGE
:
2989 trans_start
= dev_trans_start(slave
->dev
);
2990 if ((!bond
->curr_active_slave
&&
2991 time_in_range(jiffies
,
2992 trans_start
- delta_in_ticks
,
2993 trans_start
+ delta_in_ticks
)) ||
2994 bond
->curr_active_slave
!= slave
) {
2995 slave
->link
= BOND_LINK_UP
;
2996 bond
->current_arp_slave
= NULL
;
2998 pr_info("%s: link status definitely up for interface %s.\n",
2999 bond
->dev
->name
, slave
->dev
->name
);
3001 if (!bond
->curr_active_slave
||
3002 (slave
== bond
->primary_slave
))
3009 case BOND_LINK_DOWN
:
3010 if (slave
->link_failure_count
< UINT_MAX
)
3011 slave
->link_failure_count
++;
3013 slave
->link
= BOND_LINK_DOWN
;
3014 bond_set_slave_inactive_flags(slave
);
3016 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3017 bond
->dev
->name
, slave
->dev
->name
);
3019 if (slave
== bond
->curr_active_slave
) {
3020 bond
->current_arp_slave
= NULL
;
3027 pr_err("%s: impossible: new_link %d on slave %s\n",
3028 bond
->dev
->name
, slave
->new_link
,
3036 write_lock_bh(&bond
->curr_slave_lock
);
3037 bond_select_active_slave(bond
);
3038 write_unlock_bh(&bond
->curr_slave_lock
);
3039 unblock_netpoll_tx();
3042 bond_set_carrier(bond
);
3046 * Send ARP probes for active-backup mode ARP monitor.
3048 * Called with bond->lock held for read.
3050 static void bond_ab_arp_probe(struct bonding
*bond
)
3052 struct slave
*slave
;
3055 read_lock(&bond
->curr_slave_lock
);
3057 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3058 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3059 bond
->current_arp_slave
->dev
->name
,
3060 bond
->curr_active_slave
->dev
->name
);
3062 if (bond
->curr_active_slave
) {
3063 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3064 read_unlock(&bond
->curr_slave_lock
);
3068 read_unlock(&bond
->curr_slave_lock
);
3070 /* if we don't have a curr_active_slave, search for the next available
3071 * backup slave from the current_arp_slave and make it the candidate
3072 * for becoming the curr_active_slave
3075 if (!bond
->current_arp_slave
) {
3076 bond
->current_arp_slave
= bond
->first_slave
;
3077 if (!bond
->current_arp_slave
)
3081 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3083 /* search for next candidate */
3084 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3085 if (IS_UP(slave
->dev
)) {
3086 slave
->link
= BOND_LINK_BACK
;
3087 bond_set_slave_active_flags(slave
);
3088 bond_arp_send_all(bond
, slave
);
3089 slave
->jiffies
= jiffies
;
3090 bond
->current_arp_slave
= slave
;
3094 /* if the link state is up at this point, we
3095 * mark it down - this can happen if we have
3096 * simultaneous link failures and
3097 * reselect_active_interface doesn't make this
3098 * one the current slave so it is still marked
3099 * up when it is actually down
3101 if (slave
->link
== BOND_LINK_UP
) {
3102 slave
->link
= BOND_LINK_DOWN
;
3103 if (slave
->link_failure_count
< UINT_MAX
)
3104 slave
->link_failure_count
++;
3106 bond_set_slave_inactive_flags(slave
);
3108 pr_info("%s: backup interface %s is now down.\n",
3109 bond
->dev
->name
, slave
->dev
->name
);
3114 void bond_activebackup_arp_mon(struct work_struct
*work
)
3116 struct bonding
*bond
= container_of(work
, struct bonding
,
3118 bool should_notify_peers
= false;
3121 read_lock(&bond
->lock
);
3123 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3125 if (bond
->slave_cnt
== 0)
3128 should_notify_peers
= bond_should_notify_peers(bond
);
3130 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3131 read_unlock(&bond
->lock
);
3133 /* Race avoidance with bond_close flush of workqueue */
3134 if (!rtnl_trylock()) {
3135 read_lock(&bond
->lock
);
3137 should_notify_peers
= false;
3141 read_lock(&bond
->lock
);
3143 bond_ab_arp_commit(bond
, delta_in_ticks
);
3145 read_unlock(&bond
->lock
);
3147 read_lock(&bond
->lock
);
3150 bond_ab_arp_probe(bond
);
3153 if (bond
->params
.arp_interval
)
3154 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3156 read_unlock(&bond
->lock
);
3158 if (should_notify_peers
) {
3159 if (!rtnl_trylock()) {
3160 read_lock(&bond
->lock
);
3161 bond
->send_peer_notif
++;
3162 read_unlock(&bond
->lock
);
3165 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
3170 /*-------------------------- netdev event handling --------------------------*/
3173 * Change device name
3175 static int bond_event_changename(struct bonding
*bond
)
3177 bond_remove_proc_entry(bond
);
3178 bond_create_proc_entry(bond
);
3180 bond_debug_reregister(bond
);
3185 static int bond_master_netdev_event(unsigned long event
,
3186 struct net_device
*bond_dev
)
3188 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3191 case NETDEV_CHANGENAME
:
3192 return bond_event_changename(event_bond
);
3200 static int bond_slave_netdev_event(unsigned long event
,
3201 struct net_device
*slave_dev
)
3203 struct net_device
*bond_dev
= slave_dev
->master
;
3204 struct bonding
*bond
= netdev_priv(bond_dev
);
3205 struct slave
*slave
= NULL
;
3208 case NETDEV_UNREGISTER
:
3210 if (bond
->setup_by_slave
)
3211 bond_release_and_destroy(bond_dev
, slave_dev
);
3213 bond_release(bond_dev
, slave_dev
);
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
->params
.mode
== BOND_MODE_8023AD
) {
3226 if (old_speed
!= slave
->speed
)
3227 bond_3ad_adapter_speed_changed(slave
);
3228 if (old_duplex
!= slave
->duplex
)
3229 bond_3ad_adapter_duplex_changed(slave
);
3236 * ... Or is it this?
3239 case NETDEV_CHANGEMTU
:
3241 * TODO: Should slaves be allowed to
3242 * independently alter their MTU? For
3243 * an active-backup bond, slaves need
3244 * not be the same type of device, so
3245 * MTUs may vary. For other modes,
3246 * slaves arguably should have the
3247 * same MTUs. To do this, we'd need to
3248 * take over the slave's change_mtu
3249 * function for the duration of their
3253 case NETDEV_CHANGENAME
:
3255 * TODO: handle changing the primary's name
3258 case NETDEV_FEAT_CHANGE
:
3259 bond_compute_features(bond
);
3269 * bond_netdev_event: handle netdev notifier chain events.
3271 * This function receives events for the netdev chain. The caller (an
3272 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3273 * locks for us to safely manipulate the slave devices (RTNL lock,
3276 static int bond_netdev_event(struct notifier_block
*this,
3277 unsigned long event
, void *ptr
)
3279 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3281 pr_debug("event_dev: %s, event: %lx\n",
3282 event_dev
? event_dev
->name
: "None",
3285 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3288 if (event_dev
->flags
& IFF_MASTER
) {
3289 pr_debug("IFF_MASTER\n");
3290 return bond_master_netdev_event(event
, event_dev
);
3293 if (event_dev
->flags
& IFF_SLAVE
) {
3294 pr_debug("IFF_SLAVE\n");
3295 return bond_slave_netdev_event(event
, event_dev
);
3302 * bond_inetaddr_event: handle inetaddr notifier chain events.
3304 * We keep track of device IPs primarily to use as source addresses in
3305 * ARP monitor probes (rather than spewing out broadcasts all the time).
3307 * We track one IP for the main device (if it has one), plus one per VLAN.
3309 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3311 struct in_ifaddr
*ifa
= ptr
;
3312 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3313 struct bond_net
*bn
= net_generic(dev_net(event_dev
), bond_net_id
);
3314 struct bonding
*bond
;
3315 struct vlan_entry
*vlan
;
3317 /* we only care about primary address */
3318 if(ifa
->ifa_flags
& IFA_F_SECONDARY
)
3321 list_for_each_entry(bond
, &bn
->dev_list
, bond_list
) {
3322 if (bond
->dev
== event_dev
) {
3325 bond
->master_ip
= ifa
->ifa_local
;
3328 bond
->master_ip
= 0;
3335 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3336 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
3338 if (vlan_dev
== event_dev
) {
3341 vlan
->vlan_ip
= ifa
->ifa_local
;
3355 static struct notifier_block bond_netdev_notifier
= {
3356 .notifier_call
= bond_netdev_event
,
3359 static struct notifier_block bond_inetaddr_notifier
= {
3360 .notifier_call
= bond_inetaddr_event
,
3363 /*---------------------------- Hashing Policies -----------------------------*/
3366 * Hash for the output device based upon layer 2 and layer 3 data. If
3367 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3369 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3371 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3372 struct iphdr
*iph
= ip_hdr(skb
);
3374 if (skb
->protocol
== htons(ETH_P_IP
)) {
3375 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3376 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3379 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3383 * Hash for the output device based upon layer 3 and layer 4 data. If
3384 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3385 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3387 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3389 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3390 struct iphdr
*iph
= ip_hdr(skb
);
3391 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3394 if (skb
->protocol
== htons(ETH_P_IP
)) {
3395 if (!ip_is_fragment(iph
) &&
3396 (iph
->protocol
== IPPROTO_TCP
||
3397 iph
->protocol
== IPPROTO_UDP
)) {
3398 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3400 return (layer4_xor
^
3401 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3405 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3409 * Hash for the output device based upon layer 2 data
3411 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3413 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3415 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3418 /*-------------------------- Device entry points ----------------------------*/
3420 static int bond_open(struct net_device
*bond_dev
)
3422 struct bonding
*bond
= netdev_priv(bond_dev
);
3423 struct slave
*slave
;
3426 /* reset slave->backup and slave->inactive */
3427 read_lock(&bond
->lock
);
3428 if (bond
->slave_cnt
> 0) {
3429 read_lock(&bond
->curr_slave_lock
);
3430 bond_for_each_slave(bond
, slave
, i
) {
3431 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3432 && (slave
!= bond
->curr_active_slave
)) {
3433 bond_set_slave_inactive_flags(slave
);
3435 bond_set_slave_active_flags(slave
);
3438 read_unlock(&bond
->curr_slave_lock
);
3440 read_unlock(&bond
->lock
);
3442 INIT_DELAYED_WORK(&bond
->mcast_work
, bond_resend_igmp_join_requests_delayed
);
3444 if (bond_is_lb(bond
)) {
3445 /* bond_alb_initialize must be called before the timer
3448 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3449 /* something went wrong - fail the open operation */
3453 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3454 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3457 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3458 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3459 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3462 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3463 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3464 INIT_DELAYED_WORK(&bond
->arp_work
,
3465 bond_activebackup_arp_mon
);
3467 INIT_DELAYED_WORK(&bond
->arp_work
,
3468 bond_loadbalance_arp_mon
);
3470 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3471 if (bond
->params
.arp_validate
)
3472 bond
->recv_probe
= bond_arp_rcv
;
3475 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3476 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3477 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3478 /* register to receive LACPDUs */
3479 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3480 bond_3ad_initiate_agg_selection(bond
, 1);
3486 static int bond_close(struct net_device
*bond_dev
)
3488 struct bonding
*bond
= netdev_priv(bond_dev
);
3490 write_lock_bh(&bond
->lock
);
3492 bond
->send_peer_notif
= 0;
3494 write_unlock_bh(&bond
->lock
);
3496 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3497 cancel_delayed_work_sync(&bond
->mii_work
);
3500 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3501 cancel_delayed_work_sync(&bond
->arp_work
);
3504 switch (bond
->params
.mode
) {
3505 case BOND_MODE_8023AD
:
3506 cancel_delayed_work_sync(&bond
->ad_work
);
3510 cancel_delayed_work_sync(&bond
->alb_work
);
3516 if (delayed_work_pending(&bond
->mcast_work
))
3517 cancel_delayed_work_sync(&bond
->mcast_work
);
3519 if (bond_is_lb(bond
)) {
3520 /* Must be called only after all
3521 * slaves have been released
3523 bond_alb_deinitialize(bond
);
3525 bond
->recv_probe
= NULL
;
3530 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3531 struct rtnl_link_stats64
*stats
)
3533 struct bonding
*bond
= netdev_priv(bond_dev
);
3534 struct rtnl_link_stats64 temp
;
3535 struct slave
*slave
;
3538 memset(stats
, 0, sizeof(*stats
));
3540 read_lock_bh(&bond
->lock
);
3542 bond_for_each_slave(bond
, slave
, i
) {
3543 const struct rtnl_link_stats64
*sstats
=
3544 dev_get_stats(slave
->dev
, &temp
);
3546 stats
->rx_packets
+= sstats
->rx_packets
;
3547 stats
->rx_bytes
+= sstats
->rx_bytes
;
3548 stats
->rx_errors
+= sstats
->rx_errors
;
3549 stats
->rx_dropped
+= sstats
->rx_dropped
;
3551 stats
->tx_packets
+= sstats
->tx_packets
;
3552 stats
->tx_bytes
+= sstats
->tx_bytes
;
3553 stats
->tx_errors
+= sstats
->tx_errors
;
3554 stats
->tx_dropped
+= sstats
->tx_dropped
;
3556 stats
->multicast
+= sstats
->multicast
;
3557 stats
->collisions
+= sstats
->collisions
;
3559 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3560 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3561 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3562 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3563 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3564 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3566 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3567 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3568 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3569 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3570 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3573 read_unlock_bh(&bond
->lock
);
3578 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3580 struct net_device
*slave_dev
= NULL
;
3581 struct ifbond k_binfo
;
3582 struct ifbond __user
*u_binfo
= NULL
;
3583 struct ifslave k_sinfo
;
3584 struct ifslave __user
*u_sinfo
= NULL
;
3585 struct mii_ioctl_data
*mii
= NULL
;
3588 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3600 * We do this again just in case we were called by SIOCGMIIREG
3601 * instead of SIOCGMIIPHY.
3608 if (mii
->reg_num
== 1) {
3609 struct bonding
*bond
= netdev_priv(bond_dev
);
3611 read_lock(&bond
->lock
);
3612 read_lock(&bond
->curr_slave_lock
);
3613 if (netif_carrier_ok(bond
->dev
))
3614 mii
->val_out
= BMSR_LSTATUS
;
3616 read_unlock(&bond
->curr_slave_lock
);
3617 read_unlock(&bond
->lock
);
3621 case BOND_INFO_QUERY_OLD
:
3622 case SIOCBONDINFOQUERY
:
3623 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3625 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3628 res
= bond_info_query(bond_dev
, &k_binfo
);
3630 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3634 case BOND_SLAVE_INFO_QUERY_OLD
:
3635 case SIOCBONDSLAVEINFOQUERY
:
3636 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3638 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3641 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3643 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3652 if (!capable(CAP_NET_ADMIN
))
3655 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
3657 pr_debug("slave_dev=%p:\n", slave_dev
);
3662 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3664 case BOND_ENSLAVE_OLD
:
3665 case SIOCBONDENSLAVE
:
3666 res
= bond_enslave(bond_dev
, slave_dev
);
3668 case BOND_RELEASE_OLD
:
3669 case SIOCBONDRELEASE
:
3670 res
= bond_release(bond_dev
, slave_dev
);
3672 case BOND_SETHWADDR_OLD
:
3673 case SIOCBONDSETHWADDR
:
3674 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3676 case BOND_CHANGE_ACTIVE_OLD
:
3677 case SIOCBONDCHANGEACTIVE
:
3678 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3690 static bool bond_addr_in_mc_list(unsigned char *addr
,
3691 struct netdev_hw_addr_list
*list
,
3694 struct netdev_hw_addr
*ha
;
3696 netdev_hw_addr_list_for_each(ha
, list
)
3697 if (!memcmp(ha
->addr
, addr
, addrlen
))
3703 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3705 struct bonding
*bond
= netdev_priv(bond_dev
);
3707 if (change
& IFF_PROMISC
)
3708 bond_set_promiscuity(bond
,
3709 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3711 if (change
& IFF_ALLMULTI
)
3712 bond_set_allmulti(bond
,
3713 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3716 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3718 struct bonding
*bond
= netdev_priv(bond_dev
);
3719 struct netdev_hw_addr
*ha
;
3722 read_lock(&bond
->lock
);
3724 /* looking for addresses to add to slaves' mc list */
3725 netdev_for_each_mc_addr(ha
, bond_dev
) {
3726 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
3727 bond_dev
->addr_len
);
3729 bond_mc_add(bond
, ha
->addr
);
3732 /* looking for addresses to delete from slaves' list */
3733 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
3734 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
3735 bond_dev
->addr_len
);
3737 bond_mc_del(bond
, ha
->addr
);
3740 /* save master's multicast list */
3741 __hw_addr_flush(&bond
->mc_list
);
3742 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
3743 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
3745 read_unlock(&bond
->lock
);
3748 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
3750 struct bonding
*bond
= netdev_priv(dev
);
3751 struct slave
*slave
= bond
->first_slave
;
3754 const struct net_device_ops
*slave_ops
3755 = slave
->dev
->netdev_ops
;
3756 if (slave_ops
->ndo_neigh_setup
)
3757 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
3763 * Change the MTU of all of a master's slaves to match the master
3765 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3767 struct bonding
*bond
= netdev_priv(bond_dev
);
3768 struct slave
*slave
, *stop_at
;
3772 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3773 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3775 /* Can't hold bond->lock with bh disabled here since
3776 * some base drivers panic. On the other hand we can't
3777 * hold bond->lock without bh disabled because we'll
3778 * deadlock. The only solution is to rely on the fact
3779 * that we're under rtnl_lock here, and the slaves
3780 * list won't change. This doesn't solve the problem
3781 * of setting the slave's MTU while it is
3782 * transmitting, but the assumption is that the base
3783 * driver can handle that.
3785 * TODO: figure out a way to safely iterate the slaves
3786 * list, but without holding a lock around the actual
3787 * call to the base driver.
3790 bond_for_each_slave(bond
, slave
, i
) {
3791 pr_debug("s %p s->p %p c_m %p\n",
3794 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3796 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3799 /* If we failed to set the slave's mtu to the new value
3800 * we must abort the operation even in ACTIVE_BACKUP
3801 * mode, because if we allow the backup slaves to have
3802 * different mtu values than the active slave we'll
3803 * need to change their mtu when doing a failover. That
3804 * means changing their mtu from timer context, which
3805 * is probably not a good idea.
3807 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3812 bond_dev
->mtu
= new_mtu
;
3817 /* unwind from head to the slave that failed */
3819 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3822 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3824 pr_debug("unwind err %d dev %s\n",
3825 tmp_res
, slave
->dev
->name
);
3835 * Note that many devices must be down to change the HW address, and
3836 * downing the master releases all slaves. We can make bonds full of
3837 * bonding devices to test this, however.
3839 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3841 struct bonding
*bond
= netdev_priv(bond_dev
);
3842 struct sockaddr
*sa
= addr
, tmp_sa
;
3843 struct slave
*slave
, *stop_at
;
3847 if (bond
->params
.mode
== BOND_MODE_ALB
)
3848 return bond_alb_set_mac_address(bond_dev
, addr
);
3851 pr_debug("bond=%p, name=%s\n",
3852 bond
, bond_dev
? bond_dev
->name
: "None");
3855 * If fail_over_mac is set to active, do nothing and return
3856 * success. Returning an error causes ifenslave to fail.
3858 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
3861 if (!is_valid_ether_addr(sa
->sa_data
))
3862 return -EADDRNOTAVAIL
;
3864 /* Can't hold bond->lock with bh disabled here since
3865 * some base drivers panic. On the other hand we can't
3866 * hold bond->lock without bh disabled because we'll
3867 * deadlock. The only solution is to rely on the fact
3868 * that we're under rtnl_lock here, and the slaves
3869 * list won't change. This doesn't solve the problem
3870 * of setting the slave's hw address while it is
3871 * transmitting, but the assumption is that the base
3872 * driver can handle that.
3874 * TODO: figure out a way to safely iterate the slaves
3875 * list, but without holding a lock around the actual
3876 * call to the base driver.
3879 bond_for_each_slave(bond
, slave
, i
) {
3880 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3881 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3883 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3885 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3889 res
= dev_set_mac_address(slave
->dev
, addr
);
3891 /* TODO: consider downing the slave
3893 * User should expect communications
3894 * breakage anyway until ARP finish
3897 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3903 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3907 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3908 tmp_sa
.sa_family
= bond_dev
->type
;
3910 /* unwind from head to the slave that failed */
3912 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3915 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3917 pr_debug("unwind err %d dev %s\n",
3918 tmp_res
, slave
->dev
->name
);
3925 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3927 struct bonding
*bond
= netdev_priv(bond_dev
);
3928 struct slave
*slave
, *start_at
;
3929 int i
, slave_no
, res
= 1;
3930 struct iphdr
*iph
= ip_hdr(skb
);
3933 * Start with the curr_active_slave that joined the bond as the
3934 * default for sending IGMP traffic. For failover purposes one
3935 * needs to maintain some consistency for the interface that will
3936 * send the join/membership reports. The curr_active_slave found
3937 * will send all of this type of traffic.
3939 if ((iph
->protocol
== IPPROTO_IGMP
) &&
3940 (skb
->protocol
== htons(ETH_P_IP
))) {
3942 read_lock(&bond
->curr_slave_lock
);
3943 slave
= bond
->curr_active_slave
;
3944 read_unlock(&bond
->curr_slave_lock
);
3950 * Concurrent TX may collide on rr_tx_counter; we accept
3951 * that as being rare enough not to justify using an
3954 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
3956 bond_for_each_slave(bond
, slave
, i
) {
3964 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3965 if (IS_UP(slave
->dev
) &&
3966 (slave
->link
== BOND_LINK_UP
) &&
3967 bond_is_active_slave(slave
)) {
3968 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3975 /* no suitable interface, frame not sent */
3979 return NETDEV_TX_OK
;
3984 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3985 * the bond has a usable interface.
3987 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3989 struct bonding
*bond
= netdev_priv(bond_dev
);
3992 read_lock(&bond
->curr_slave_lock
);
3994 if (bond
->curr_active_slave
)
3995 res
= bond_dev_queue_xmit(bond
, skb
,
3996 bond
->curr_active_slave
->dev
);
3999 /* no suitable interface, frame not sent */
4002 read_unlock(&bond
->curr_slave_lock
);
4004 return NETDEV_TX_OK
;
4008 * In bond_xmit_xor() , we determine the output device by using a pre-
4009 * determined xmit_hash_policy(), If the selected device is not enabled,
4010 * find the next active slave.
4012 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4014 struct bonding
*bond
= netdev_priv(bond_dev
);
4015 struct slave
*slave
, *start_at
;
4020 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4022 bond_for_each_slave(bond
, slave
, i
) {
4030 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4031 if (IS_UP(slave
->dev
) &&
4032 (slave
->link
== BOND_LINK_UP
) &&
4033 bond_is_active_slave(slave
)) {
4034 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4040 /* no suitable interface, frame not sent */
4044 return NETDEV_TX_OK
;
4048 * in broadcast mode, we send everything to all usable interfaces.
4050 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4052 struct bonding
*bond
= netdev_priv(bond_dev
);
4053 struct slave
*slave
, *start_at
;
4054 struct net_device
*tx_dev
= NULL
;
4058 read_lock(&bond
->curr_slave_lock
);
4059 start_at
= bond
->curr_active_slave
;
4060 read_unlock(&bond
->curr_slave_lock
);
4065 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4066 if (IS_UP(slave
->dev
) &&
4067 (slave
->link
== BOND_LINK_UP
) &&
4068 bond_is_active_slave(slave
)) {
4070 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4072 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4077 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4079 dev_kfree_skb(skb2
);
4083 tx_dev
= slave
->dev
;
4088 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4092 /* no suitable interface, frame not sent */
4095 /* frame sent to all suitable interfaces */
4096 return NETDEV_TX_OK
;
4099 /*------------------------- Device initialization ---------------------------*/
4101 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4103 switch (bond
->params
.xmit_policy
) {
4104 case BOND_XMIT_POLICY_LAYER23
:
4105 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4107 case BOND_XMIT_POLICY_LAYER34
:
4108 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4110 case BOND_XMIT_POLICY_LAYER2
:
4112 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4118 * Lookup the slave that corresponds to a qid
4120 static inline int bond_slave_override(struct bonding
*bond
,
4121 struct sk_buff
*skb
)
4124 struct slave
*slave
= NULL
;
4125 struct slave
*check_slave
;
4127 if (!skb
->queue_mapping
)
4130 /* Find out if any slaves have the same mapping as this skb. */
4131 bond_for_each_slave(bond
, check_slave
, i
) {
4132 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4133 slave
= check_slave
;
4138 /* If the slave isn't UP, use default transmit policy. */
4139 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4140 (slave
->link
== BOND_LINK_UP
)) {
4141 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4148 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4151 * This helper function exists to help dev_pick_tx get the correct
4152 * destination queue. Using a helper function skips a call to
4153 * skb_tx_hash and will put the skbs in the queue we expect on their
4154 * way down to the bonding driver.
4156 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
4159 * Save the original txq to restore before passing to the driver
4161 bond_queue_mapping(skb
) = skb
->queue_mapping
;
4163 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
4165 txq
-= dev
->real_num_tx_queues
;
4166 } while (txq
>= dev
->real_num_tx_queues
);
4171 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4173 struct bonding
*bond
= netdev_priv(dev
);
4175 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4176 if (!bond_slave_override(bond
, skb
))
4177 return NETDEV_TX_OK
;
4180 switch (bond
->params
.mode
) {
4181 case BOND_MODE_ROUNDROBIN
:
4182 return bond_xmit_roundrobin(skb
, dev
);
4183 case BOND_MODE_ACTIVEBACKUP
:
4184 return bond_xmit_activebackup(skb
, dev
);
4186 return bond_xmit_xor(skb
, dev
);
4187 case BOND_MODE_BROADCAST
:
4188 return bond_xmit_broadcast(skb
, dev
);
4189 case BOND_MODE_8023AD
:
4190 return bond_3ad_xmit_xor(skb
, dev
);
4193 return bond_alb_xmit(skb
, dev
);
4195 /* Should never happen, mode already checked */
4196 pr_err("%s: Error: Unknown bonding mode %d\n",
4197 dev
->name
, bond
->params
.mode
);
4200 return NETDEV_TX_OK
;
4204 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4206 struct bonding
*bond
= netdev_priv(dev
);
4207 netdev_tx_t ret
= NETDEV_TX_OK
;
4210 * If we risk deadlock from transmitting this in the
4211 * netpoll path, tell netpoll to queue the frame for later tx
4213 if (is_netpoll_tx_blocked(dev
))
4214 return NETDEV_TX_BUSY
;
4216 read_lock(&bond
->lock
);
4218 if (bond
->slave_cnt
)
4219 ret
= __bond_start_xmit(skb
, dev
);
4223 read_unlock(&bond
->lock
);
4229 * set bond mode specific net device operations
4231 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4233 struct net_device
*bond_dev
= bond
->dev
;
4236 case BOND_MODE_ROUNDROBIN
:
4238 case BOND_MODE_ACTIVEBACKUP
:
4241 bond_set_xmit_hash_policy(bond
);
4243 case BOND_MODE_BROADCAST
:
4245 case BOND_MODE_8023AD
:
4246 bond_set_xmit_hash_policy(bond
);
4253 /* Should never happen, mode already checked */
4254 pr_err("%s: Error: Unknown bonding mode %d\n",
4255 bond_dev
->name
, mode
);
4260 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4261 struct ethtool_drvinfo
*drvinfo
)
4263 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4264 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4265 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4268 static const struct ethtool_ops bond_ethtool_ops
= {
4269 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4270 .get_link
= ethtool_op_get_link
,
4273 static const struct net_device_ops bond_netdev_ops
= {
4274 .ndo_init
= bond_init
,
4275 .ndo_uninit
= bond_uninit
,
4276 .ndo_open
= bond_open
,
4277 .ndo_stop
= bond_close
,
4278 .ndo_start_xmit
= bond_start_xmit
,
4279 .ndo_select_queue
= bond_select_queue
,
4280 .ndo_get_stats64
= bond_get_stats
,
4281 .ndo_do_ioctl
= bond_do_ioctl
,
4282 .ndo_change_rx_flags
= bond_change_rx_flags
,
4283 .ndo_set_rx_mode
= bond_set_multicast_list
,
4284 .ndo_change_mtu
= bond_change_mtu
,
4285 .ndo_set_mac_address
= bond_set_mac_address
,
4286 .ndo_neigh_setup
= bond_neigh_setup
,
4287 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4288 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4289 #ifdef CONFIG_NET_POLL_CONTROLLER
4290 .ndo_netpoll_setup
= bond_netpoll_setup
,
4291 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4292 .ndo_poll_controller
= bond_poll_controller
,
4294 .ndo_add_slave
= bond_enslave
,
4295 .ndo_del_slave
= bond_release
,
4296 .ndo_fix_features
= bond_fix_features
,
4299 static void bond_destructor(struct net_device
*bond_dev
)
4301 struct bonding
*bond
= netdev_priv(bond_dev
);
4303 destroy_workqueue(bond
->wq
);
4304 free_netdev(bond_dev
);
4307 static void bond_setup(struct net_device
*bond_dev
)
4309 struct bonding
*bond
= netdev_priv(bond_dev
);
4311 /* initialize rwlocks */
4312 rwlock_init(&bond
->lock
);
4313 rwlock_init(&bond
->curr_slave_lock
);
4315 bond
->params
= bonding_defaults
;
4317 /* Initialize pointers */
4318 bond
->dev
= bond_dev
;
4319 INIT_LIST_HEAD(&bond
->vlan_list
);
4321 /* Initialize the device entry points */
4322 ether_setup(bond_dev
);
4323 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4324 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4325 bond_set_mode_ops(bond
, bond
->params
.mode
);
4327 bond_dev
->destructor
= bond_destructor
;
4329 /* Initialize the device options */
4330 bond_dev
->tx_queue_len
= 0;
4331 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4332 bond_dev
->priv_flags
|= IFF_BONDING
;
4333 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
4335 /* At first, we block adding VLANs. That's the only way to
4336 * prevent problems that occur when adding VLANs over an
4337 * empty bond. The block will be removed once non-challenged
4338 * slaves are enslaved.
4340 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4342 /* don't acquire bond device's netif_tx_lock when
4344 bond_dev
->features
|= NETIF_F_LLTX
;
4346 /* By default, we declare the bond to be fully
4347 * VLAN hardware accelerated capable. Special
4348 * care is taken in the various xmit functions
4349 * when there are slaves that are not hw accel
4353 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4354 NETIF_F_HW_VLAN_TX
|
4355 NETIF_F_HW_VLAN_RX
|
4356 NETIF_F_HW_VLAN_FILTER
;
4358 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
4359 bond_dev
->features
|= bond_dev
->hw_features
;
4362 static void bond_work_cancel_all(struct bonding
*bond
)
4364 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4365 cancel_delayed_work_sync(&bond
->mii_work
);
4367 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4368 cancel_delayed_work_sync(&bond
->arp_work
);
4370 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4371 delayed_work_pending(&bond
->alb_work
))
4372 cancel_delayed_work_sync(&bond
->alb_work
);
4374 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4375 delayed_work_pending(&bond
->ad_work
))
4376 cancel_delayed_work_sync(&bond
->ad_work
);
4378 if (delayed_work_pending(&bond
->mcast_work
))
4379 cancel_delayed_work_sync(&bond
->mcast_work
);
4383 * Destroy a bonding device.
4384 * Must be under rtnl_lock when this function is called.
4386 static void bond_uninit(struct net_device
*bond_dev
)
4388 struct bonding
*bond
= netdev_priv(bond_dev
);
4389 struct vlan_entry
*vlan
, *tmp
;
4391 bond_netpoll_cleanup(bond_dev
);
4393 /* Release the bonded slaves */
4394 bond_release_all(bond_dev
);
4396 list_del(&bond
->bond_list
);
4398 bond_work_cancel_all(bond
);
4400 bond_remove_proc_entry(bond
);
4402 bond_debug_unregister(bond
);
4404 __hw_addr_flush(&bond
->mc_list
);
4406 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4407 list_del(&vlan
->vlan_list
);
4412 /*------------------------- Module initialization ---------------------------*/
4415 * Convert string input module parms. Accept either the
4416 * number of the mode or its string name. A bit complicated because
4417 * some mode names are substrings of other names, and calls from sysfs
4418 * may have whitespace in the name (trailing newlines, for example).
4420 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4422 int modeint
= -1, i
, rv
;
4423 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4425 for (p
= (char *)buf
; *p
; p
++)
4426 if (!(isdigit(*p
) || isspace(*p
)))
4430 rv
= sscanf(buf
, "%20s", modestr
);
4432 rv
= sscanf(buf
, "%d", &modeint
);
4437 for (i
= 0; tbl
[i
].modename
; i
++) {
4438 if (modeint
== tbl
[i
].mode
)
4440 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4447 static int bond_check_params(struct bond_params
*params
)
4449 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4452 * Convert string parameters.
4455 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4456 if (bond_mode
== -1) {
4457 pr_err("Error: Invalid bonding mode \"%s\"\n",
4458 mode
== NULL
? "NULL" : mode
);
4463 if (xmit_hash_policy
) {
4464 if ((bond_mode
!= BOND_MODE_XOR
) &&
4465 (bond_mode
!= BOND_MODE_8023AD
)) {
4466 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4467 bond_mode_name(bond_mode
));
4469 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4471 if (xmit_hashtype
== -1) {
4472 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4473 xmit_hash_policy
== NULL
? "NULL" :
4481 if (bond_mode
!= BOND_MODE_8023AD
) {
4482 pr_info("lacp_rate param is irrelevant in mode %s\n",
4483 bond_mode_name(bond_mode
));
4485 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4486 if (lacp_fast
== -1) {
4487 pr_err("Error: Invalid lacp rate \"%s\"\n",
4488 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4495 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4496 if (params
->ad_select
== -1) {
4497 pr_err("Error: Invalid ad_select \"%s\"\n",
4498 ad_select
== NULL
? "NULL" : ad_select
);
4502 if (bond_mode
!= BOND_MODE_8023AD
) {
4503 pr_warning("ad_select param only affects 802.3ad mode\n");
4506 params
->ad_select
= BOND_AD_STABLE
;
4509 if (max_bonds
< 0) {
4510 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4511 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4512 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4516 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4517 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4518 miimon
= BOND_LINK_MON_INTERV
;
4522 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4527 if (downdelay
< 0) {
4528 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4529 downdelay
, INT_MAX
);
4533 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4534 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4539 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4540 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4545 /* reset values for 802.3ad */
4546 if (bond_mode
== BOND_MODE_8023AD
) {
4548 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");
4549 pr_warning("Forcing miimon to 100msec\n");
4554 if (tx_queues
< 1 || tx_queues
> 255) {
4555 pr_warning("Warning: tx_queues (%d) should be between "
4556 "1 and 255, resetting to %d\n",
4557 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4558 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4561 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4562 pr_warning("Warning: all_slaves_active module parameter (%d), "
4563 "not of valid value (0/1), so it was set to "
4564 "0\n", all_slaves_active
);
4565 all_slaves_active
= 0;
4568 if (resend_igmp
< 0 || resend_igmp
> 255) {
4569 pr_warning("Warning: resend_igmp (%d) should be between "
4570 "0 and 255, resetting to %d\n",
4571 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4572 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4575 /* reset values for TLB/ALB */
4576 if ((bond_mode
== BOND_MODE_TLB
) ||
4577 (bond_mode
== BOND_MODE_ALB
)) {
4579 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");
4580 pr_warning("Forcing miimon to 100msec\n");
4585 if (bond_mode
== BOND_MODE_ALB
) {
4586 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",
4591 if (updelay
|| downdelay
) {
4592 /* just warn the user the up/down delay will have
4593 * no effect since miimon is zero...
4595 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",
4596 updelay
, downdelay
);
4599 /* don't allow arp monitoring */
4601 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4602 miimon
, arp_interval
);
4606 if ((updelay
% miimon
) != 0) {
4607 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4609 (updelay
/ miimon
) * miimon
);
4614 if ((downdelay
% miimon
) != 0) {
4615 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4617 (downdelay
/ miimon
) * miimon
);
4620 downdelay
/= miimon
;
4623 if (arp_interval
< 0) {
4624 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4625 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4626 arp_interval
= BOND_LINK_ARP_INTERV
;
4629 for (arp_ip_count
= 0;
4630 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4632 /* not complete check, but should be good enough to
4634 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4635 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4636 arp_ip_target
[arp_ip_count
]);
4639 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4640 arp_target
[arp_ip_count
] = ip
;
4644 if (arp_interval
&& !arp_ip_count
) {
4645 /* don't allow arping if no arp_ip_target given... */
4646 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4652 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4653 pr_err("arp_validate only supported in active-backup mode\n");
4656 if (!arp_interval
) {
4657 pr_err("arp_validate requires arp_interval\n");
4661 arp_validate_value
= bond_parse_parm(arp_validate
,
4663 if (arp_validate_value
== -1) {
4664 pr_err("Error: invalid arp_validate \"%s\"\n",
4665 arp_validate
== NULL
? "NULL" : arp_validate
);
4669 arp_validate_value
= 0;
4672 pr_info("MII link monitoring set to %d ms\n", miimon
);
4673 } else if (arp_interval
) {
4676 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4678 arp_validate_tbl
[arp_validate_value
].modename
,
4681 for (i
= 0; i
< arp_ip_count
; i
++)
4682 pr_info(" %s", arp_ip_target
[i
]);
4686 } else if (max_bonds
) {
4687 /* miimon and arp_interval not set, we need one so things
4688 * work as expected, see bonding.txt for details
4690 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");
4693 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4694 /* currently, using a primary only makes sense
4695 * in active backup, TLB or ALB modes
4697 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4698 primary
, bond_mode_name(bond_mode
));
4702 if (primary
&& primary_reselect
) {
4703 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4705 if (primary_reselect_value
== -1) {
4706 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4708 NULL
? "NULL" : primary_reselect
);
4712 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4715 if (fail_over_mac
) {
4716 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4718 if (fail_over_mac_value
== -1) {
4719 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4720 arp_validate
== NULL
? "NULL" : arp_validate
);
4724 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4725 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4727 fail_over_mac_value
= BOND_FOM_NONE
;
4730 /* fill params struct with the proper values */
4731 params
->mode
= bond_mode
;
4732 params
->xmit_policy
= xmit_hashtype
;
4733 params
->miimon
= miimon
;
4734 params
->num_peer_notif
= num_peer_notif
;
4735 params
->arp_interval
= arp_interval
;
4736 params
->arp_validate
= arp_validate_value
;
4737 params
->updelay
= updelay
;
4738 params
->downdelay
= downdelay
;
4739 params
->use_carrier
= use_carrier
;
4740 params
->lacp_fast
= lacp_fast
;
4741 params
->primary
[0] = 0;
4742 params
->primary_reselect
= primary_reselect_value
;
4743 params
->fail_over_mac
= fail_over_mac_value
;
4744 params
->tx_queues
= tx_queues
;
4745 params
->all_slaves_active
= all_slaves_active
;
4746 params
->resend_igmp
= resend_igmp
;
4747 params
->min_links
= min_links
;
4750 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4751 params
->primary
[IFNAMSIZ
- 1] = 0;
4754 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4759 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4760 static struct lock_class_key bonding_netdev_addr_lock_key
;
4762 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4763 struct netdev_queue
*txq
,
4766 lockdep_set_class(&txq
->_xmit_lock
,
4767 &bonding_netdev_xmit_lock_key
);
4770 static void bond_set_lockdep_class(struct net_device
*dev
)
4772 lockdep_set_class(&dev
->addr_list_lock
,
4773 &bonding_netdev_addr_lock_key
);
4774 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4778 * Called from registration process
4780 static int bond_init(struct net_device
*bond_dev
)
4782 struct bonding
*bond
= netdev_priv(bond_dev
);
4783 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4784 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4786 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4789 * Initialize locks that may be required during
4790 * en/deslave operations. All of the bond_open work
4791 * (of which this is part) should really be moved to
4792 * a phase prior to dev_open
4794 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4795 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4797 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4801 bond_set_lockdep_class(bond_dev
);
4803 bond_create_proc_entry(bond
);
4804 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4806 bond_prepare_sysfs_group(bond
);
4808 bond_debug_register(bond
);
4810 __hw_addr_init(&bond
->mc_list
);
4814 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4816 if (tb
[IFLA_ADDRESS
]) {
4817 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4819 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4820 return -EADDRNOTAVAIL
;
4825 static int bond_get_tx_queues(struct net
*net
, struct nlattr
*tb
[],
4826 unsigned int *num_queues
,
4827 unsigned int *real_num_queues
)
4829 *num_queues
= tx_queues
;
4833 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
4835 .priv_size
= sizeof(struct bonding
),
4836 .setup
= bond_setup
,
4837 .validate
= bond_validate
,
4838 .get_tx_queues
= bond_get_tx_queues
,
4841 /* Create a new bond based on the specified name and bonding parameters.
4842 * If name is NULL, obtain a suitable "bond%d" name for us.
4843 * Caller must NOT hold rtnl_lock; we need to release it here before we
4844 * set up our sysfs entries.
4846 int bond_create(struct net
*net
, const char *name
)
4848 struct net_device
*bond_dev
;
4853 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4854 name
? name
: "bond%d",
4855 bond_setup
, tx_queues
);
4857 pr_err("%s: eek! can't alloc netdev!\n", name
);
4862 dev_net_set(bond_dev
, net
);
4863 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4865 res
= register_netdevice(bond_dev
);
4867 netif_carrier_off(bond_dev
);
4871 bond_destructor(bond_dev
);
4875 static int __net_init
bond_net_init(struct net
*net
)
4877 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4880 INIT_LIST_HEAD(&bn
->dev_list
);
4882 bond_create_proc_dir(bn
);
4883 bond_create_sysfs(bn
);
4888 static void __net_exit
bond_net_exit(struct net
*net
)
4890 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4892 bond_destroy_sysfs(bn
);
4893 bond_destroy_proc_dir(bn
);
4896 static struct pernet_operations bond_net_ops
= {
4897 .init
= bond_net_init
,
4898 .exit
= bond_net_exit
,
4900 .size
= sizeof(struct bond_net
),
4903 static int __init
bonding_init(void)
4908 pr_info("%s", bond_version
);
4910 res
= bond_check_params(&bonding_defaults
);
4914 res
= register_pernet_subsys(&bond_net_ops
);
4918 res
= rtnl_link_register(&bond_link_ops
);
4922 bond_create_debugfs();
4924 for (i
= 0; i
< max_bonds
; i
++) {
4925 res
= bond_create(&init_net
, NULL
);
4930 register_netdevice_notifier(&bond_netdev_notifier
);
4931 register_inetaddr_notifier(&bond_inetaddr_notifier
);
4935 rtnl_link_unregister(&bond_link_ops
);
4937 unregister_pernet_subsys(&bond_net_ops
);
4942 static void __exit
bonding_exit(void)
4944 unregister_netdevice_notifier(&bond_netdev_notifier
);
4945 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
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");