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>
79 #include <net/pkt_sched.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV 0
88 #define BOND_LINK_ARP_INTERV 0
90 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
91 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
92 static int num_peer_notif
= 1;
93 static int miimon
= BOND_LINK_MON_INTERV
;
96 static int use_carrier
= 1;
99 static char *primary_reselect
;
100 static char *lacp_rate
;
101 static int min_links
;
102 static char *ad_select
;
103 static char *xmit_hash_policy
;
104 static int arp_interval
= BOND_LINK_ARP_INTERV
;
105 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
106 static char *arp_validate
;
107 static char *fail_over_mac
;
108 static int all_slaves_active
= 0;
109 static struct bond_params bonding_defaults
;
110 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
112 module_param(max_bonds
, int, 0);
113 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
114 module_param(tx_queues
, int, 0);
115 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
116 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
118 "failover event (alias of num_unsol_na)");
119 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
121 "failover event (alias of num_grat_arp)");
122 module_param(miimon
, int, 0);
123 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
124 module_param(updelay
, int, 0);
125 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
126 module_param(downdelay
, int, 0);
127 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
129 module_param(use_carrier
, int, 0);
130 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132 module_param(mode
, charp
, 0);
133 MODULE_PARM_DESC(mode
, "Mode of operation; 0 for balance-rr, "
134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
137 module_param(primary
, charp
, 0);
138 MODULE_PARM_DESC(primary
, "Primary network device to use");
139 module_param(primary_reselect
, charp
, 0);
140 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
145 "2 for only on active slave "
147 module_param(lacp_rate
, charp
, 0);
148 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner; "
149 "0 for slow, 1 for fast");
150 module_param(ad_select
, charp
, 0);
151 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
152 "0 for stable (default), 1 for bandwidth, "
154 module_param(min_links
, int, 0);
155 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
157 module_param(xmit_hash_policy
, charp
, 0);
158 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
159 "0 for layer 2 (default), 1 for layer 3+4, "
161 module_param(arp_interval
, int, 0);
162 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
163 module_param_array(arp_ip_target
, charp
, NULL
, 0);
164 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
165 module_param(arp_validate
, charp
, 0);
166 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
167 "0 for none (default), 1 for active, "
168 "2 for backup, 3 for all");
169 module_param(fail_over_mac
, charp
, 0);
170 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
171 "the same MAC; 0 for none (default), "
172 "1 for active, 2 for follow");
173 module_param(all_slaves_active
, int, 0);
174 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
175 "by setting active flag for all slaves; "
176 "0 for never (default), 1 for always.");
177 module_param(resend_igmp
, int, 0);
178 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
181 /*----------------------------- Global variables ----------------------------*/
183 #ifdef CONFIG_NET_POLL_CONTROLLER
184 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
187 int bond_net_id __read_mostly
;
189 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
190 static int arp_ip_count
;
191 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
192 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
193 static int lacp_fast
;
195 const struct bond_parm_tbl bond_lacp_tbl
[] = {
196 { "slow", AD_LACP_SLOW
},
197 { "fast", AD_LACP_FAST
},
201 const struct bond_parm_tbl bond_mode_tbl
[] = {
202 { "balance-rr", BOND_MODE_ROUNDROBIN
},
203 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
204 { "balance-xor", BOND_MODE_XOR
},
205 { "broadcast", BOND_MODE_BROADCAST
},
206 { "802.3ad", BOND_MODE_8023AD
},
207 { "balance-tlb", BOND_MODE_TLB
},
208 { "balance-alb", BOND_MODE_ALB
},
212 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
213 { "layer2", BOND_XMIT_POLICY_LAYER2
},
214 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
215 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
219 const struct bond_parm_tbl arp_validate_tbl
[] = {
220 { "none", BOND_ARP_VALIDATE_NONE
},
221 { "active", BOND_ARP_VALIDATE_ACTIVE
},
222 { "backup", BOND_ARP_VALIDATE_BACKUP
},
223 { "all", BOND_ARP_VALIDATE_ALL
},
227 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
228 { "none", BOND_FOM_NONE
},
229 { "active", BOND_FOM_ACTIVE
},
230 { "follow", BOND_FOM_FOLLOW
},
234 const struct bond_parm_tbl pri_reselect_tbl
[] = {
235 { "always", BOND_PRI_RESELECT_ALWAYS
},
236 { "better", BOND_PRI_RESELECT_BETTER
},
237 { "failure", BOND_PRI_RESELECT_FAILURE
},
241 struct bond_parm_tbl ad_select_tbl
[] = {
242 { "stable", BOND_AD_STABLE
},
243 { "bandwidth", BOND_AD_BANDWIDTH
},
244 { "count", BOND_AD_COUNT
},
248 /*-------------------------- Forward declarations ---------------------------*/
250 static int bond_init(struct net_device
*bond_dev
);
251 static void bond_uninit(struct net_device
*bond_dev
);
253 /*---------------------------- General routines -----------------------------*/
255 const char *bond_mode_name(int mode
)
257 static const char *names
[] = {
258 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
259 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
260 [BOND_MODE_XOR
] = "load balancing (xor)",
261 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
262 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
263 [BOND_MODE_TLB
] = "transmit load balancing",
264 [BOND_MODE_ALB
] = "adaptive load balancing",
267 if (mode
< 0 || mode
> BOND_MODE_ALB
)
273 /*---------------------------------- VLAN -----------------------------------*/
276 * bond_add_vlan - add a new vlan id on bond
277 * @bond: bond that got the notification
278 * @vlan_id: the vlan id to add
280 * Returns -ENOMEM if allocation failed.
282 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
284 struct vlan_entry
*vlan
;
286 pr_debug("bond: %s, vlan id %d\n",
287 (bond
? bond
->dev
->name
: "None"), vlan_id
);
289 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
293 INIT_LIST_HEAD(&vlan
->vlan_list
);
294 vlan
->vlan_id
= vlan_id
;
296 write_lock_bh(&bond
->lock
);
298 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
300 write_unlock_bh(&bond
->lock
);
302 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
308 * bond_del_vlan - delete a vlan id from bond
309 * @bond: bond that got the notification
310 * @vlan_id: the vlan id to delete
312 * returns -ENODEV if @vlan_id was not found in @bond.
314 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
316 struct vlan_entry
*vlan
;
319 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
322 write_lock_bh(&bond
->lock
);
324 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
325 if (vlan
->vlan_id
== vlan_id
) {
326 list_del(&vlan
->vlan_list
);
328 if (bond_is_lb(bond
))
329 bond_alb_clear_vlan(bond
, vlan_id
);
331 pr_debug("removed VLAN ID %d from bond %s\n",
332 vlan_id
, bond
->dev
->name
);
341 pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 vlan_id
, bond
->dev
->name
);
345 write_unlock_bh(&bond
->lock
);
346 unblock_netpoll_tx();
351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
358 * Caller must hold bond->lock
360 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
362 struct vlan_entry
*next
, *last
;
364 if (list_empty(&bond
->vlan_list
))
368 next
= list_entry(bond
->vlan_list
.next
,
369 struct vlan_entry
, vlan_list
);
371 last
= list_entry(bond
->vlan_list
.prev
,
372 struct vlan_entry
, vlan_list
);
374 next
= list_entry(bond
->vlan_list
.next
,
375 struct vlan_entry
, vlan_list
);
377 next
= list_entry(curr
->vlan_list
.next
,
378 struct vlan_entry
, vlan_list
);
386 * bond_dev_queue_xmit - Prepare skb for xmit.
388 * @bond: bond device that got this skb for tx.
389 * @skb: hw accel VLAN tagged skb to transmit
390 * @slave_dev: slave that is supposed to xmit this skbuff
392 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
393 struct net_device
*slave_dev
)
395 skb
->dev
= slave_dev
;
397 BUILD_BUG_ON(sizeof(skb
->queue_mapping
) !=
398 sizeof(qdisc_skb_cb(skb
)->slave_dev_queue_mapping
));
399 skb
->queue_mapping
= qdisc_skb_cb(skb
)->slave_dev_queue_mapping
;
401 if (unlikely(netpoll_tx_running(bond
->dev
)))
402 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
410 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411 * We don't protect the slave list iteration with a lock because:
412 * a. This operation is performed in IOCTL context,
413 * b. The operation is protected by the RTNL semaphore in the 8021q code,
414 * c. Holding a lock with BH disabled while directly calling a base driver
415 * entry point is generally a BAD idea.
417 * The design of synchronization/protection for this operation in the 8021q
418 * module is good for one or more VLAN devices over a single physical device
419 * and cannot be extended for a teaming solution like bonding, so there is a
420 * potential race condition here where a net device from the vlan group might
421 * be referenced (either by a base driver or the 8021q code) while it is being
422 * removed from the system. However, it turns out we're not making matters
423 * worse, and if it works for regular VLAN usage it will work here too.
427 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428 * @bond_dev: bonding net device that got called
429 * @vid: vlan id being added
431 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
433 struct bonding
*bond
= netdev_priv(bond_dev
);
434 struct slave
*slave
, *stop_at
;
437 bond_for_each_slave(bond
, slave
, i
) {
438 res
= vlan_vid_add(slave
->dev
, vid
);
443 res
= bond_add_vlan(bond
, vid
);
445 pr_err("%s: Error: Failed to add vlan id %d\n",
446 bond_dev
->name
, vid
);
453 /* unwind from head to the slave that failed */
455 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
)
456 vlan_vid_del(slave
->dev
, vid
);
462 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
463 * @bond_dev: bonding net device that got called
464 * @vid: vlan id being removed
466 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
468 struct bonding
*bond
= netdev_priv(bond_dev
);
472 bond_for_each_slave(bond
, slave
, i
)
473 vlan_vid_del(slave
->dev
, vid
);
475 res
= bond_del_vlan(bond
, vid
);
477 pr_err("%s: Error: Failed to remove vlan id %d\n",
478 bond_dev
->name
, vid
);
485 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
487 struct vlan_entry
*vlan
;
490 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
491 res
= vlan_vid_add(slave_dev
, vlan
->vlan_id
);
493 pr_warning("%s: Failed to add vlan id %d to device %s\n",
494 bond
->dev
->name
, vlan
->vlan_id
,
499 static void bond_del_vlans_from_slave(struct bonding
*bond
,
500 struct net_device
*slave_dev
)
502 struct vlan_entry
*vlan
;
504 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
507 vlan_vid_del(slave_dev
, vlan
->vlan_id
);
511 /*------------------------------- Link status -------------------------------*/
514 * Set the carrier state for the master according to the state of its
515 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
516 * do special 802.3ad magic.
518 * Returns zero if carrier state does not change, nonzero if it does.
520 static int bond_set_carrier(struct bonding
*bond
)
525 if (bond
->slave_cnt
== 0)
528 if (bond
->params
.mode
== BOND_MODE_8023AD
)
529 return bond_3ad_set_carrier(bond
);
531 bond_for_each_slave(bond
, slave
, i
) {
532 if (slave
->link
== BOND_LINK_UP
) {
533 if (!netif_carrier_ok(bond
->dev
)) {
534 netif_carrier_on(bond
->dev
);
542 if (netif_carrier_ok(bond
->dev
)) {
543 netif_carrier_off(bond
->dev
);
550 * Get link speed and duplex from the slave's base driver
551 * using ethtool. If for some reason the call fails or the
552 * values are invalid, set speed and duplex to -1,
555 static void bond_update_speed_duplex(struct slave
*slave
)
557 struct net_device
*slave_dev
= slave
->dev
;
558 struct ethtool_cmd ecmd
;
562 slave
->speed
= SPEED_UNKNOWN
;
563 slave
->duplex
= DUPLEX_UNKNOWN
;
565 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
569 slave_speed
= ethtool_cmd_speed(&ecmd
);
570 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
573 switch (ecmd
.duplex
) {
581 slave
->speed
= slave_speed
;
582 slave
->duplex
= ecmd
.duplex
;
588 * if <dev> supports MII link status reporting, check its link status.
590 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
591 * depending upon the setting of the use_carrier parameter.
593 * Return either BMSR_LSTATUS, meaning that the link is up (or we
594 * can't tell and just pretend it is), or 0, meaning that the link is
597 * If reporting is non-zero, instead of faking link up, return -1 if
598 * both ETHTOOL and MII ioctls fail (meaning the device does not
599 * support them). If use_carrier is set, return whatever it says.
600 * It'd be nice if there was a good way to tell if a driver supports
601 * netif_carrier, but there really isn't.
603 static int bond_check_dev_link(struct bonding
*bond
,
604 struct net_device
*slave_dev
, int reporting
)
606 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
607 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
609 struct mii_ioctl_data
*mii
;
611 if (!reporting
&& !netif_running(slave_dev
))
614 if (bond
->params
.use_carrier
)
615 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
617 /* Try to get link status using Ethtool first. */
618 if (slave_dev
->ethtool_ops
) {
619 if (slave_dev
->ethtool_ops
->get_link
) {
622 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
624 return link
? BMSR_LSTATUS
: 0;
628 /* Ethtool can't be used, fallback to MII ioctls. */
629 ioctl
= slave_ops
->ndo_do_ioctl
;
631 /* TODO: set pointer to correct ioctl on a per team member */
632 /* bases to make this more efficient. that is, once */
633 /* we determine the correct ioctl, we will always */
634 /* call it and not the others for that team */
638 * We cannot assume that SIOCGMIIPHY will also read a
639 * register; not all network drivers (e.g., e100)
643 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
644 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
646 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
647 mii
->reg_num
= MII_BMSR
;
648 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
649 return mii
->val_out
& BMSR_LSTATUS
;
654 * If reporting, report that either there's no dev->do_ioctl,
655 * or both SIOCGMIIREG and get_link failed (meaning that we
656 * cannot report link status). If not reporting, pretend
659 return reporting
? -1 : BMSR_LSTATUS
;
662 /*----------------------------- Multicast list ------------------------------*/
665 * Push the promiscuity flag down to appropriate slaves
667 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
670 if (USES_PRIMARY(bond
->params
.mode
)) {
671 /* write lock already acquired */
672 if (bond
->curr_active_slave
) {
673 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
679 bond_for_each_slave(bond
, slave
, i
) {
680 err
= dev_set_promiscuity(slave
->dev
, inc
);
689 * Push the allmulti flag down to all slaves
691 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
694 if (USES_PRIMARY(bond
->params
.mode
)) {
695 /* write lock already acquired */
696 if (bond
->curr_active_slave
) {
697 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
703 bond_for_each_slave(bond
, slave
, i
) {
704 err
= dev_set_allmulti(slave
->dev
, inc
);
713 * Add a Multicast address to slaves
716 static void bond_mc_add(struct bonding
*bond
, void *addr
)
718 if (USES_PRIMARY(bond
->params
.mode
)) {
719 /* write lock already acquired */
720 if (bond
->curr_active_slave
)
721 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
726 bond_for_each_slave(bond
, slave
, i
)
727 dev_mc_add(slave
->dev
, addr
);
732 * Remove a multicast address from slave
735 static void bond_mc_del(struct bonding
*bond
, void *addr
)
737 if (USES_PRIMARY(bond
->params
.mode
)) {
738 /* write lock already acquired */
739 if (bond
->curr_active_slave
)
740 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
744 bond_for_each_slave(bond
, slave
, i
) {
745 dev_mc_del(slave
->dev
, addr
);
751 static void __bond_resend_igmp_join_requests(struct net_device
*dev
)
753 struct in_device
*in_dev
;
756 in_dev
= __in_dev_get_rcu(dev
);
758 ip_mc_rejoin_groups(in_dev
);
763 * Retrieve the list of registered multicast addresses for the bonding
764 * device and retransmit an IGMP JOIN request to the current active
767 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
769 struct net_device
*bond_dev
, *vlan_dev
, *master_dev
;
770 struct vlan_entry
*vlan
;
772 read_lock(&bond
->lock
);
774 bond_dev
= bond
->dev
;
776 /* rejoin all groups on bond device */
777 __bond_resend_igmp_join_requests(bond_dev
);
780 * if bond is enslaved to a bridge,
781 * then rejoin all groups on its master
783 master_dev
= bond_dev
->master
;
785 if ((master_dev
->priv_flags
& IFF_EBRIDGE
)
786 && (bond_dev
->priv_flags
& IFF_BRIDGE_PORT
))
787 __bond_resend_igmp_join_requests(master_dev
);
789 /* rejoin all groups on vlan devices */
790 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
792 vlan_dev
= __vlan_find_dev_deep(bond_dev
,
796 __bond_resend_igmp_join_requests(vlan_dev
);
799 if (--bond
->igmp_retrans
> 0)
800 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
802 read_unlock(&bond
->lock
);
805 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
807 struct bonding
*bond
= container_of(work
, struct bonding
,
809 bond_resend_igmp_join_requests(bond
);
813 * flush all members of flush->mc_list from device dev->mc_list
815 static void bond_mc_list_flush(struct net_device
*bond_dev
,
816 struct net_device
*slave_dev
)
818 struct bonding
*bond
= netdev_priv(bond_dev
);
819 struct netdev_hw_addr
*ha
;
821 netdev_for_each_mc_addr(ha
, bond_dev
)
822 dev_mc_del(slave_dev
, ha
->addr
);
824 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
825 /* del lacpdu mc addr from mc list */
826 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
828 dev_mc_del(slave_dev
, lacpdu_multicast
);
832 /*--------------------------- Active slave change ---------------------------*/
835 * Update the mc list and multicast-related flags for the new and
836 * old active slaves (if any) according to the multicast mode, and
837 * promiscuous flags unconditionally.
839 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
840 struct slave
*old_active
)
842 struct netdev_hw_addr
*ha
;
844 if (!USES_PRIMARY(bond
->params
.mode
))
845 /* nothing to do - mc list is already up-to-date on
851 if (bond
->dev
->flags
& IFF_PROMISC
)
852 dev_set_promiscuity(old_active
->dev
, -1);
854 if (bond
->dev
->flags
& IFF_ALLMULTI
)
855 dev_set_allmulti(old_active
->dev
, -1);
857 netdev_for_each_mc_addr(ha
, bond
->dev
)
858 dev_mc_del(old_active
->dev
, ha
->addr
);
862 /* FIXME: Signal errors upstream. */
863 if (bond
->dev
->flags
& IFF_PROMISC
)
864 dev_set_promiscuity(new_active
->dev
, 1);
866 if (bond
->dev
->flags
& IFF_ALLMULTI
)
867 dev_set_allmulti(new_active
->dev
, 1);
869 netdev_for_each_mc_addr(ha
, bond
->dev
)
870 dev_mc_add(new_active
->dev
, ha
->addr
);
875 * bond_do_fail_over_mac
877 * Perform special MAC address swapping for fail_over_mac settings
879 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
881 static void bond_do_fail_over_mac(struct bonding
*bond
,
882 struct slave
*new_active
,
883 struct slave
*old_active
)
884 __releases(&bond
->curr_slave_lock
)
885 __releases(&bond
->lock
)
886 __acquires(&bond
->lock
)
887 __acquires(&bond
->curr_slave_lock
)
889 u8 tmp_mac
[ETH_ALEN
];
890 struct sockaddr saddr
;
893 switch (bond
->params
.fail_over_mac
) {
894 case BOND_FOM_ACTIVE
:
896 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
897 new_active
->dev
->addr_len
);
898 write_unlock_bh(&bond
->curr_slave_lock
);
899 read_unlock(&bond
->lock
);
900 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
901 read_lock(&bond
->lock
);
902 write_lock_bh(&bond
->curr_slave_lock
);
905 case BOND_FOM_FOLLOW
:
907 * if new_active && old_active, swap them
908 * if just old_active, do nothing (going to no active slave)
909 * if just new_active, set new_active to bond's MAC
914 write_unlock_bh(&bond
->curr_slave_lock
);
915 read_unlock(&bond
->lock
);
918 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
919 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
921 saddr
.sa_family
= new_active
->dev
->type
;
923 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
924 saddr
.sa_family
= bond
->dev
->type
;
927 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
929 pr_err("%s: Error %d setting MAC of slave %s\n",
930 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
937 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
938 saddr
.sa_family
= old_active
->dev
->type
;
940 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
942 pr_err("%s: Error %d setting MAC of slave %s\n",
943 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
945 read_lock(&bond
->lock
);
946 write_lock_bh(&bond
->curr_slave_lock
);
949 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
950 bond
->dev
->name
, bond
->params
.fail_over_mac
);
956 static bool bond_should_change_active(struct bonding
*bond
)
958 struct slave
*prim
= bond
->primary_slave
;
959 struct slave
*curr
= bond
->curr_active_slave
;
961 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
963 if (bond
->force_primary
) {
964 bond
->force_primary
= false;
967 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
968 (prim
->speed
< curr
->speed
||
969 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
971 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
977 * find_best_interface - select the best available slave to be the active one
978 * @bond: our bonding struct
980 * Warning: Caller must hold curr_slave_lock for writing.
982 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
984 struct slave
*new_active
, *old_active
;
985 struct slave
*bestslave
= NULL
;
986 int mintime
= bond
->params
.updelay
;
989 new_active
= bond
->curr_active_slave
;
991 if (!new_active
) { /* there were no active slaves left */
992 if (bond
->slave_cnt
> 0) /* found one slave */
993 new_active
= bond
->first_slave
;
995 return NULL
; /* still no slave, return NULL */
998 if ((bond
->primary_slave
) &&
999 bond
->primary_slave
->link
== BOND_LINK_UP
&&
1000 bond_should_change_active(bond
)) {
1001 new_active
= bond
->primary_slave
;
1004 /* remember where to stop iterating over the slaves */
1005 old_active
= new_active
;
1007 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1008 if (new_active
->link
== BOND_LINK_UP
) {
1010 } else if (new_active
->link
== BOND_LINK_BACK
&&
1011 IS_UP(new_active
->dev
)) {
1012 /* link up, but waiting for stabilization */
1013 if (new_active
->delay
< mintime
) {
1014 mintime
= new_active
->delay
;
1015 bestslave
= new_active
;
1023 static bool bond_should_notify_peers(struct bonding
*bond
)
1025 struct slave
*slave
= bond
->curr_active_slave
;
1027 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1028 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
1030 if (!slave
|| !bond
->send_peer_notif
||
1031 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
1034 bond
->send_peer_notif
--;
1039 * change_active_interface - change the active slave into the specified one
1040 * @bond: our bonding struct
1041 * @new: the new slave to make the active one
1043 * Set the new slave to the bond's settings and unset them on the old
1044 * curr_active_slave.
1045 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1047 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1048 * because it is apparently the best available slave we have, even though its
1049 * updelay hasn't timed out yet.
1051 * If new_active is not NULL, caller must hold bond->lock for read and
1052 * curr_slave_lock for write_bh.
1054 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1056 struct slave
*old_active
= bond
->curr_active_slave
;
1058 if (old_active
== new_active
)
1062 new_active
->jiffies
= jiffies
;
1064 if (new_active
->link
== BOND_LINK_BACK
) {
1065 if (USES_PRIMARY(bond
->params
.mode
)) {
1066 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1067 bond
->dev
->name
, new_active
->dev
->name
,
1068 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1071 new_active
->delay
= 0;
1072 new_active
->link
= BOND_LINK_UP
;
1074 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1075 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1077 if (bond_is_lb(bond
))
1078 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1080 if (USES_PRIMARY(bond
->params
.mode
)) {
1081 pr_info("%s: making interface %s the new active one.\n",
1082 bond
->dev
->name
, new_active
->dev
->name
);
1087 if (USES_PRIMARY(bond
->params
.mode
))
1088 bond_mc_swap(bond
, new_active
, old_active
);
1090 if (bond_is_lb(bond
)) {
1091 bond_alb_handle_active_change(bond
, new_active
);
1093 bond_set_slave_inactive_flags(old_active
);
1095 bond_set_slave_active_flags(new_active
);
1097 bond
->curr_active_slave
= new_active
;
1100 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1102 bond_set_slave_inactive_flags(old_active
);
1105 bool should_notify_peers
= false;
1107 bond_set_slave_active_flags(new_active
);
1109 if (bond
->params
.fail_over_mac
)
1110 bond_do_fail_over_mac(bond
, new_active
,
1113 if (netif_running(bond
->dev
)) {
1114 bond
->send_peer_notif
=
1115 bond
->params
.num_peer_notif
;
1116 should_notify_peers
=
1117 bond_should_notify_peers(bond
);
1120 write_unlock_bh(&bond
->curr_slave_lock
);
1121 read_unlock(&bond
->lock
);
1123 netdev_bonding_change(bond
->dev
, NETDEV_BONDING_FAILOVER
);
1124 if (should_notify_peers
)
1125 netdev_bonding_change(bond
->dev
,
1126 NETDEV_NOTIFY_PEERS
);
1128 read_lock(&bond
->lock
);
1129 write_lock_bh(&bond
->curr_slave_lock
);
1133 /* resend IGMP joins since active slave has changed or
1134 * all were sent on curr_active_slave.
1135 * resend only if bond is brought up with the affected
1136 * bonding modes and the retransmission is enabled */
1137 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
1138 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
1139 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
1140 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
1141 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 0);
1146 * bond_select_active_slave - select a new active slave, if needed
1147 * @bond: our bonding struct
1149 * This functions should be called when one of the following occurs:
1150 * - The old curr_active_slave has been released or lost its link.
1151 * - The primary_slave has got its link back.
1152 * - A slave has got its link back and there's no old curr_active_slave.
1154 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1156 void bond_select_active_slave(struct bonding
*bond
)
1158 struct slave
*best_slave
;
1161 best_slave
= bond_find_best_slave(bond
);
1162 if (best_slave
!= bond
->curr_active_slave
) {
1163 bond_change_active_slave(bond
, best_slave
);
1164 rv
= bond_set_carrier(bond
);
1168 if (netif_carrier_ok(bond
->dev
)) {
1169 pr_info("%s: first active interface up!\n",
1172 pr_info("%s: now running without any active interface !\n",
1178 /*--------------------------- slave list handling ---------------------------*/
1181 * This function attaches the slave to the end of list.
1183 * bond->lock held for writing by caller.
1185 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1187 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1188 new_slave
->next
= new_slave
;
1189 new_slave
->prev
= new_slave
;
1190 bond
->first_slave
= new_slave
;
1192 new_slave
->next
= bond
->first_slave
;
1193 new_slave
->prev
= bond
->first_slave
->prev
;
1194 new_slave
->next
->prev
= new_slave
;
1195 new_slave
->prev
->next
= new_slave
;
1202 * This function detaches the slave from the list.
1203 * WARNING: no check is made to verify if the slave effectively
1204 * belongs to <bond>.
1205 * Nothing is freed on return, structures are just unchained.
1206 * If any slave pointer in bond was pointing to <slave>,
1207 * it should be changed by the calling function.
1209 * bond->lock held for writing by caller.
1211 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1214 slave
->next
->prev
= slave
->prev
;
1217 slave
->prev
->next
= slave
->next
;
1219 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1220 if (bond
->slave_cnt
> 1) { /* there are more slave */
1221 bond
->first_slave
= slave
->next
;
1223 bond
->first_slave
= NULL
; /* slave was the last one */
1232 #ifdef CONFIG_NET_POLL_CONTROLLER
1233 static inline int slave_enable_netpoll(struct slave
*slave
)
1238 np
= kzalloc(sizeof(*np
), GFP_ATOMIC
);
1243 err
= __netpoll_setup(np
, slave
->dev
, GFP_ATOMIC
);
1252 static inline void slave_disable_netpoll(struct slave
*slave
)
1254 struct netpoll
*np
= slave
->np
;
1260 __netpoll_free_rcu(np
);
1262 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1264 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1266 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1271 static void bond_poll_controller(struct net_device
*bond_dev
)
1275 static void __bond_netpoll_cleanup(struct bonding
*bond
)
1277 struct slave
*slave
;
1280 bond_for_each_slave(bond
, slave
, i
)
1281 if (IS_UP(slave
->dev
))
1282 slave_disable_netpoll(slave
);
1284 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1286 struct bonding
*bond
= netdev_priv(bond_dev
);
1288 read_lock(&bond
->lock
);
1289 __bond_netpoll_cleanup(bond
);
1290 read_unlock(&bond
->lock
);
1293 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
, gfp_t gfp
)
1295 struct bonding
*bond
= netdev_priv(dev
);
1296 struct slave
*slave
;
1299 read_lock(&bond
->lock
);
1300 bond_for_each_slave(bond
, slave
, i
) {
1301 err
= slave_enable_netpoll(slave
);
1303 __bond_netpoll_cleanup(bond
);
1307 read_unlock(&bond
->lock
);
1311 static struct netpoll_info
*bond_netpoll_info(struct bonding
*bond
)
1313 return bond
->dev
->npinfo
;
1317 static inline int slave_enable_netpoll(struct slave
*slave
)
1321 static inline void slave_disable_netpoll(struct slave
*slave
)
1324 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1329 /*---------------------------------- IOCTL ----------------------------------*/
1331 static int bond_sethwaddr(struct net_device
*bond_dev
,
1332 struct net_device
*slave_dev
)
1334 pr_debug("bond_dev=%p\n", bond_dev
);
1335 pr_debug("slave_dev=%p\n", slave_dev
);
1336 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1337 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1341 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1342 netdev_features_t features
)
1344 struct slave
*slave
;
1345 struct bonding
*bond
= netdev_priv(dev
);
1346 netdev_features_t mask
;
1349 read_lock(&bond
->lock
);
1351 if (!bond
->first_slave
) {
1352 /* Disable adding VLANs to empty bond. But why? --mq */
1353 features
|= NETIF_F_VLAN_CHALLENGED
;
1358 features
&= ~NETIF_F_ONE_FOR_ALL
;
1359 features
|= NETIF_F_ALL_FOR_ALL
;
1361 bond_for_each_slave(bond
, slave
, i
) {
1362 features
= netdev_increment_features(features
,
1363 slave
->dev
->features
,
1368 read_unlock(&bond
->lock
);
1372 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1373 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1374 NETIF_F_HIGHDMA | NETIF_F_LRO)
1376 static void bond_compute_features(struct bonding
*bond
)
1378 struct slave
*slave
;
1379 struct net_device
*bond_dev
= bond
->dev
;
1380 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1381 unsigned short max_hard_header_len
= ETH_HLEN
;
1383 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1385 read_lock(&bond
->lock
);
1387 if (!bond
->first_slave
)
1390 bond_for_each_slave(bond
, slave
, i
) {
1391 vlan_features
= netdev_increment_features(vlan_features
,
1392 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1394 dst_release_flag
&= slave
->dev
->priv_flags
;
1395 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1396 max_hard_header_len
= slave
->dev
->hard_header_len
;
1400 bond_dev
->vlan_features
= vlan_features
;
1401 bond_dev
->hard_header_len
= max_hard_header_len
;
1403 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1404 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1406 read_unlock(&bond
->lock
);
1408 netdev_change_features(bond_dev
);
1411 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1412 struct net_device
*slave_dev
)
1414 struct bonding
*bond
= netdev_priv(bond_dev
);
1416 bond_dev
->header_ops
= slave_dev
->header_ops
;
1418 bond_dev
->type
= slave_dev
->type
;
1419 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1420 bond_dev
->addr_len
= slave_dev
->addr_len
;
1422 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1423 slave_dev
->addr_len
);
1424 bond
->setup_by_slave
= 1;
1427 /* On bonding slaves other than the currently active slave, suppress
1428 * duplicates except for alb non-mcast/bcast.
1430 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1431 struct slave
*slave
,
1432 struct bonding
*bond
)
1434 if (bond_is_slave_inactive(slave
)) {
1435 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1436 skb
->pkt_type
!= PACKET_BROADCAST
&&
1437 skb
->pkt_type
!= PACKET_MULTICAST
)
1444 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1446 struct sk_buff
*skb
= *pskb
;
1447 struct slave
*slave
;
1448 struct bonding
*bond
;
1449 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1451 int ret
= RX_HANDLER_ANOTHER
;
1453 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1455 return RX_HANDLER_CONSUMED
;
1459 slave
= bond_slave_get_rcu(skb
->dev
);
1462 if (bond
->params
.arp_interval
)
1463 slave
->dev
->last_rx
= jiffies
;
1465 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1467 ret
= recv_probe(skb
, bond
, slave
);
1468 if (ret
== RX_HANDLER_CONSUMED
) {
1474 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1475 return RX_HANDLER_EXACT
;
1478 skb
->dev
= bond
->dev
;
1480 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1481 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1482 skb
->pkt_type
== PACKET_HOST
) {
1484 if (unlikely(skb_cow_head(skb
,
1485 skb
->data
- skb_mac_header(skb
)))) {
1487 return RX_HANDLER_CONSUMED
;
1489 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1495 /* enslave device <slave> to bond device <master> */
1496 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1498 struct bonding
*bond
= netdev_priv(bond_dev
);
1499 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1500 struct slave
*new_slave
= NULL
;
1501 struct netdev_hw_addr
*ha
;
1502 struct sockaddr addr
;
1506 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1507 slave_ops
->ndo_do_ioctl
== NULL
) {
1508 pr_warning("%s: Warning: no link monitoring support for %s\n",
1509 bond_dev
->name
, slave_dev
->name
);
1512 /* already enslaved */
1513 if (slave_dev
->flags
& IFF_SLAVE
) {
1514 pr_debug("Error, Device was already enslaved\n");
1518 /* vlan challenged mutual exclusion */
1519 /* no need to lock since we're protected by rtnl_lock */
1520 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1521 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1522 if (bond_vlan_used(bond
)) {
1523 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1524 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1527 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1528 bond_dev
->name
, slave_dev
->name
,
1529 slave_dev
->name
, bond_dev
->name
);
1532 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1536 * Old ifenslave binaries are no longer supported. These can
1537 * be identified with moderate accuracy by the state of the slave:
1538 * the current ifenslave will set the interface down prior to
1539 * enslaving it; the old ifenslave will not.
1541 if ((slave_dev
->flags
& IFF_UP
)) {
1542 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1545 goto err_undo_flags
;
1548 /* set bonding device ether type by slave - bonding netdevices are
1549 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1550 * there is a need to override some of the type dependent attribs/funcs.
1552 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1553 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1555 if (bond
->slave_cnt
== 0) {
1556 if (bond_dev
->type
!= slave_dev
->type
) {
1557 pr_debug("%s: change device type from %d to %d\n",
1559 bond_dev
->type
, slave_dev
->type
);
1561 res
= netdev_bonding_change(bond_dev
,
1562 NETDEV_PRE_TYPE_CHANGE
);
1563 res
= notifier_to_errno(res
);
1565 pr_err("%s: refused to change device type\n",
1568 goto err_undo_flags
;
1571 /* Flush unicast and multicast addresses */
1572 dev_uc_flush(bond_dev
);
1573 dev_mc_flush(bond_dev
);
1575 if (slave_dev
->type
!= ARPHRD_ETHER
)
1576 bond_setup_by_slave(bond_dev
, slave_dev
);
1578 ether_setup(bond_dev
);
1579 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1582 netdev_bonding_change(bond_dev
,
1583 NETDEV_POST_TYPE_CHANGE
);
1585 } else if (bond_dev
->type
!= slave_dev
->type
) {
1586 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1588 slave_dev
->type
, bond_dev
->type
);
1590 goto err_undo_flags
;
1593 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1594 if (bond
->slave_cnt
== 0) {
1595 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1597 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1598 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1599 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",
1602 goto err_undo_flags
;
1606 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1608 /* If this is the first slave, then we need to set the master's hardware
1609 * address to be the same as the slave's. */
1610 if (is_zero_ether_addr(bond
->dev
->dev_addr
))
1611 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1612 slave_dev
->addr_len
);
1615 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1618 goto err_undo_flags
;
1622 * Set the new_slave's queue_id to be zero. Queue ID mapping
1623 * is set via sysfs or module option if desired.
1625 new_slave
->queue_id
= 0;
1627 /* Save slave's original mtu and then set it to match the bond */
1628 new_slave
->original_mtu
= slave_dev
->mtu
;
1629 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1631 pr_debug("Error %d calling dev_set_mtu\n", res
);
1636 * Save slave's original ("permanent") mac address for modes
1637 * that need it, and for restoring it upon release, and then
1638 * set it to the master's address
1640 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1642 if (!bond
->params
.fail_over_mac
) {
1644 * Set slave to master's mac address. The application already
1645 * set the master's mac address to that of the first slave
1647 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1648 addr
.sa_family
= slave_dev
->type
;
1649 res
= dev_set_mac_address(slave_dev
, &addr
);
1651 pr_debug("Error %d calling set_mac_address\n", res
);
1652 goto err_restore_mtu
;
1656 res
= netdev_set_bond_master(slave_dev
, bond_dev
);
1658 pr_debug("Error %d calling netdev_set_bond_master\n", res
);
1659 goto err_restore_mac
;
1662 /* open the slave since the application closed it */
1663 res
= dev_open(slave_dev
);
1665 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1666 goto err_unset_master
;
1669 new_slave
->bond
= bond
;
1670 new_slave
->dev
= slave_dev
;
1671 slave_dev
->priv_flags
|= IFF_BONDING
;
1673 if (bond_is_lb(bond
)) {
1674 /* bond_alb_init_slave() must be called before all other stages since
1675 * it might fail and we do not want to have to undo everything
1677 res
= bond_alb_init_slave(bond
, new_slave
);
1682 /* If the mode USES_PRIMARY, then the new slave gets the
1683 * master's promisc (and mc) settings only if it becomes the
1684 * curr_active_slave, and that is taken care of later when calling
1685 * bond_change_active()
1687 if (!USES_PRIMARY(bond
->params
.mode
)) {
1688 /* set promiscuity level to new slave */
1689 if (bond_dev
->flags
& IFF_PROMISC
) {
1690 res
= dev_set_promiscuity(slave_dev
, 1);
1695 /* set allmulti level to new slave */
1696 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1697 res
= dev_set_allmulti(slave_dev
, 1);
1702 netif_addr_lock_bh(bond_dev
);
1703 /* upload master's mc_list to new slave */
1704 netdev_for_each_mc_addr(ha
, bond_dev
)
1705 dev_mc_add(slave_dev
, ha
->addr
);
1706 netif_addr_unlock_bh(bond_dev
);
1709 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1710 /* add lacpdu mc addr to mc list */
1711 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1713 dev_mc_add(slave_dev
, lacpdu_multicast
);
1716 bond_add_vlans_on_slave(bond
, slave_dev
);
1718 write_lock_bh(&bond
->lock
);
1720 bond_attach_slave(bond
, new_slave
);
1722 new_slave
->delay
= 0;
1723 new_slave
->link_failure_count
= 0;
1725 write_unlock_bh(&bond
->lock
);
1727 bond_compute_features(bond
);
1729 read_lock(&bond
->lock
);
1731 new_slave
->last_arp_rx
= jiffies
-
1732 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1734 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1735 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1737 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1739 * miimon is set but a bonded network driver
1740 * does not support ETHTOOL/MII and
1741 * arp_interval is not set. Note: if
1742 * use_carrier is enabled, we will never go
1743 * here (because netif_carrier is always
1744 * supported); thus, we don't need to change
1745 * the messages for netif_carrier.
1747 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",
1748 bond_dev
->name
, slave_dev
->name
);
1749 } else if (link_reporting
== -1) {
1750 /* unable get link status using mii/ethtool */
1751 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",
1752 bond_dev
->name
, slave_dev
->name
);
1756 /* check for initial state */
1757 if (bond
->params
.miimon
) {
1758 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1759 if (bond
->params
.updelay
) {
1760 new_slave
->link
= BOND_LINK_BACK
;
1761 new_slave
->delay
= bond
->params
.updelay
;
1763 new_slave
->link
= BOND_LINK_UP
;
1766 new_slave
->link
= BOND_LINK_DOWN
;
1768 } else if (bond
->params
.arp_interval
) {
1769 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1770 BOND_LINK_UP
: BOND_LINK_DOWN
);
1772 new_slave
->link
= BOND_LINK_UP
;
1775 if (new_slave
->link
!= BOND_LINK_DOWN
)
1776 new_slave
->jiffies
= jiffies
;
1777 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1778 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1779 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1781 bond_update_speed_duplex(new_slave
);
1783 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1784 /* if there is a primary slave, remember it */
1785 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1786 bond
->primary_slave
= new_slave
;
1787 bond
->force_primary
= true;
1791 write_lock_bh(&bond
->curr_slave_lock
);
1793 switch (bond
->params
.mode
) {
1794 case BOND_MODE_ACTIVEBACKUP
:
1795 bond_set_slave_inactive_flags(new_slave
);
1796 bond_select_active_slave(bond
);
1798 case BOND_MODE_8023AD
:
1799 /* in 802.3ad mode, the internal mechanism
1800 * will activate the slaves in the selected
1803 bond_set_slave_inactive_flags(new_slave
);
1804 /* if this is the first slave */
1805 if (bond
->slave_cnt
== 1) {
1806 SLAVE_AD_INFO(new_slave
).id
= 1;
1807 /* Initialize AD with the number of times that the AD timer is called in 1 second
1808 * can be called only after the mac address of the bond is set
1810 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1812 SLAVE_AD_INFO(new_slave
).id
=
1813 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1816 bond_3ad_bind_slave(new_slave
);
1820 bond_set_active_slave(new_slave
);
1821 bond_set_slave_inactive_flags(new_slave
);
1822 bond_select_active_slave(bond
);
1825 pr_debug("This slave is always active in trunk mode\n");
1827 /* always active in trunk mode */
1828 bond_set_active_slave(new_slave
);
1830 /* In trunking mode there is little meaning to curr_active_slave
1831 * anyway (it holds no special properties of the bond device),
1832 * so we can change it without calling change_active_interface()
1834 if (!bond
->curr_active_slave
)
1835 bond
->curr_active_slave
= new_slave
;
1838 } /* switch(bond_mode) */
1840 write_unlock_bh(&bond
->curr_slave_lock
);
1842 bond_set_carrier(bond
);
1844 #ifdef CONFIG_NET_POLL_CONTROLLER
1845 slave_dev
->npinfo
= bond_netpoll_info(bond
);
1846 if (slave_dev
->npinfo
) {
1847 if (slave_enable_netpoll(new_slave
)) {
1848 read_unlock(&bond
->lock
);
1849 pr_info("Error, %s: master_dev is using netpoll, "
1850 "but new slave device does not support netpoll.\n",
1858 read_unlock(&bond
->lock
);
1860 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1864 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1867 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1868 goto err_dest_symlinks
;
1871 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1872 bond_dev
->name
, slave_dev
->name
,
1873 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1874 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1876 /* enslave is successful */
1879 /* Undo stages on error */
1881 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1884 write_lock_bh(&bond
->lock
);
1885 bond_detach_slave(bond
, new_slave
);
1886 write_unlock_bh(&bond
->lock
);
1889 dev_close(slave_dev
);
1892 netdev_set_bond_master(slave_dev
, NULL
);
1895 if (!bond
->params
.fail_over_mac
) {
1896 /* XXX TODO - fom follow mode needs to change master's
1897 * MAC if this slave's MAC is in use by the bond, or at
1898 * least print a warning.
1900 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1901 addr
.sa_family
= slave_dev
->type
;
1902 dev_set_mac_address(slave_dev
, &addr
);
1906 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1912 bond_compute_features(bond
);
1918 * Try to release the slave device <slave> from the bond device <master>
1919 * It is legal to access curr_active_slave without a lock because all the function
1922 * The rules for slave state should be:
1923 * for Active/Backup:
1924 * Active stays on all backups go down
1925 * for Bonded connections:
1926 * The first up interface should be left on and all others downed.
1928 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1930 struct bonding
*bond
= netdev_priv(bond_dev
);
1931 struct slave
*slave
, *oldcurrent
;
1932 struct sockaddr addr
;
1933 netdev_features_t old_features
= bond_dev
->features
;
1935 /* slave is not a slave or master is not master of this slave */
1936 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1937 (slave_dev
->master
!= bond_dev
)) {
1938 pr_err("%s: Error: cannot release %s.\n",
1939 bond_dev
->name
, slave_dev
->name
);
1944 netdev_bonding_change(bond_dev
, NETDEV_RELEASE
);
1945 write_lock_bh(&bond
->lock
);
1947 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1949 /* not a slave of this bond */
1950 pr_info("%s: %s not enslaved\n",
1951 bond_dev
->name
, slave_dev
->name
);
1952 write_unlock_bh(&bond
->lock
);
1953 unblock_netpoll_tx();
1957 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1958 * for this slave anymore.
1960 netdev_rx_handler_unregister(slave_dev
);
1961 write_unlock_bh(&bond
->lock
);
1963 write_lock_bh(&bond
->lock
);
1965 if (!bond
->params
.fail_over_mac
) {
1966 if (ether_addr_equal(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1967 bond
->slave_cnt
> 1)
1968 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",
1969 bond_dev
->name
, slave_dev
->name
,
1971 bond_dev
->name
, slave_dev
->name
);
1974 /* Inform AD package of unbinding of slave. */
1975 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1976 /* must be called before the slave is
1977 * detached from the list
1979 bond_3ad_unbind_slave(slave
);
1982 pr_info("%s: releasing %s interface %s\n",
1984 bond_is_active_slave(slave
) ? "active" : "backup",
1987 oldcurrent
= bond
->curr_active_slave
;
1989 bond
->current_arp_slave
= NULL
;
1991 /* release the slave from its bond */
1992 bond_detach_slave(bond
, slave
);
1994 if (bond
->primary_slave
== slave
)
1995 bond
->primary_slave
= NULL
;
1997 if (oldcurrent
== slave
)
1998 bond_change_active_slave(bond
, NULL
);
2000 if (bond_is_lb(bond
)) {
2001 /* Must be called only after the slave has been
2002 * detached from the list and the curr_active_slave
2003 * has been cleared (if our_slave == old_current),
2004 * but before a new active slave is selected.
2006 write_unlock_bh(&bond
->lock
);
2007 bond_alb_deinit_slave(bond
, slave
);
2008 write_lock_bh(&bond
->lock
);
2011 if (oldcurrent
== slave
) {
2013 * Note that we hold RTNL over this sequence, so there
2014 * is no concern that another slave add/remove event
2017 write_unlock_bh(&bond
->lock
);
2018 read_lock(&bond
->lock
);
2019 write_lock_bh(&bond
->curr_slave_lock
);
2021 bond_select_active_slave(bond
);
2023 write_unlock_bh(&bond
->curr_slave_lock
);
2024 read_unlock(&bond
->lock
);
2025 write_lock_bh(&bond
->lock
);
2028 if (bond
->slave_cnt
== 0) {
2029 bond_set_carrier(bond
);
2031 /* if the last slave was removed, zero the mac address
2032 * of the master so it will be set by the application
2033 * to the mac address of the first slave
2035 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2037 if (bond_vlan_used(bond
)) {
2038 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2039 bond_dev
->name
, bond_dev
->name
);
2040 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2045 write_unlock_bh(&bond
->lock
);
2046 unblock_netpoll_tx();
2048 if (bond
->slave_cnt
== 0)
2049 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
2051 bond_compute_features(bond
);
2052 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
2053 (old_features
& NETIF_F_VLAN_CHALLENGED
))
2054 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2055 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
2057 /* must do this from outside any spinlocks */
2058 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2060 bond_del_vlans_from_slave(bond
, slave_dev
);
2062 /* If the mode USES_PRIMARY, then we should only remove its
2063 * promisc and mc settings if it was the curr_active_slave, but that was
2064 * already taken care of above when we detached the slave
2066 if (!USES_PRIMARY(bond
->params
.mode
)) {
2067 /* unset promiscuity level from slave */
2068 if (bond_dev
->flags
& IFF_PROMISC
)
2069 dev_set_promiscuity(slave_dev
, -1);
2071 /* unset allmulti level from slave */
2072 if (bond_dev
->flags
& IFF_ALLMULTI
)
2073 dev_set_allmulti(slave_dev
, -1);
2075 /* flush master's mc_list from slave */
2076 netif_addr_lock_bh(bond_dev
);
2077 bond_mc_list_flush(bond_dev
, slave_dev
);
2078 netif_addr_unlock_bh(bond_dev
);
2081 netdev_set_bond_master(slave_dev
, NULL
);
2083 slave_disable_netpoll(slave
);
2085 /* close slave before restoring its mac address */
2086 dev_close(slave_dev
);
2088 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2089 /* restore original ("permanent") mac address */
2090 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2091 addr
.sa_family
= slave_dev
->type
;
2092 dev_set_mac_address(slave_dev
, &addr
);
2095 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2097 slave_dev
->priv_flags
&= ~IFF_BONDING
;
2101 return 0; /* deletion OK */
2105 * First release a slave and then destroy the bond if no more slaves are left.
2106 * Must be under rtnl_lock when this function is called.
2108 static int bond_release_and_destroy(struct net_device
*bond_dev
,
2109 struct net_device
*slave_dev
)
2111 struct bonding
*bond
= netdev_priv(bond_dev
);
2114 ret
= bond_release(bond_dev
, slave_dev
);
2115 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2116 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
2117 pr_info("%s: destroying bond %s.\n",
2118 bond_dev
->name
, bond_dev
->name
);
2119 unregister_netdevice(bond_dev
);
2125 * This function releases all slaves.
2127 static int bond_release_all(struct net_device
*bond_dev
)
2129 struct bonding
*bond
= netdev_priv(bond_dev
);
2130 struct slave
*slave
;
2131 struct net_device
*slave_dev
;
2132 struct sockaddr addr
;
2134 write_lock_bh(&bond
->lock
);
2136 netif_carrier_off(bond_dev
);
2138 if (bond
->slave_cnt
== 0)
2141 bond
->current_arp_slave
= NULL
;
2142 bond
->primary_slave
= NULL
;
2143 bond_change_active_slave(bond
, NULL
);
2145 while ((slave
= bond
->first_slave
) != NULL
) {
2146 /* Inform AD package of unbinding of slave
2147 * before slave is detached from the list.
2149 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2150 bond_3ad_unbind_slave(slave
);
2152 slave_dev
= slave
->dev
;
2153 bond_detach_slave(bond
, slave
);
2155 /* now that the slave is detached, unlock and perform
2156 * all the undo steps that should not be called from
2159 write_unlock_bh(&bond
->lock
);
2161 /* unregister rx_handler early so bond_handle_frame wouldn't
2162 * be called for this slave anymore.
2164 netdev_rx_handler_unregister(slave_dev
);
2167 if (bond_is_lb(bond
)) {
2168 /* must be called only after the slave
2169 * has been detached from the list
2171 bond_alb_deinit_slave(bond
, slave
);
2174 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2175 bond_del_vlans_from_slave(bond
, slave_dev
);
2177 /* If the mode USES_PRIMARY, then we should only remove its
2178 * promisc and mc settings if it was the curr_active_slave, but that was
2179 * already taken care of above when we detached the slave
2181 if (!USES_PRIMARY(bond
->params
.mode
)) {
2182 /* unset promiscuity level from slave */
2183 if (bond_dev
->flags
& IFF_PROMISC
)
2184 dev_set_promiscuity(slave_dev
, -1);
2186 /* unset allmulti level from slave */
2187 if (bond_dev
->flags
& IFF_ALLMULTI
)
2188 dev_set_allmulti(slave_dev
, -1);
2190 /* flush master's mc_list from slave */
2191 netif_addr_lock_bh(bond_dev
);
2192 bond_mc_list_flush(bond_dev
, slave_dev
);
2193 netif_addr_unlock_bh(bond_dev
);
2196 netdev_set_bond_master(slave_dev
, NULL
);
2198 slave_disable_netpoll(slave
);
2200 /* close slave before restoring its mac address */
2201 dev_close(slave_dev
);
2203 if (!bond
->params
.fail_over_mac
) {
2204 /* restore original ("permanent") mac address*/
2205 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2206 addr
.sa_family
= slave_dev
->type
;
2207 dev_set_mac_address(slave_dev
, &addr
);
2212 /* re-acquire the lock before getting the next slave */
2213 write_lock_bh(&bond
->lock
);
2216 /* zero the mac address of the master so it will be
2217 * set by the application to the mac address of the
2220 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2222 if (bond_vlan_used(bond
)) {
2223 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2224 bond_dev
->name
, bond_dev
->name
);
2225 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2229 pr_info("%s: released all slaves\n", bond_dev
->name
);
2232 write_unlock_bh(&bond
->lock
);
2234 bond_compute_features(bond
);
2240 * This function changes the active slave to slave <slave_dev>.
2241 * It returns -EINVAL in the following cases.
2242 * - <slave_dev> is not found in the list.
2243 * - There is not active slave now.
2244 * - <slave_dev> is already active.
2245 * - The link state of <slave_dev> is not BOND_LINK_UP.
2246 * - <slave_dev> is not running.
2247 * In these cases, this function does nothing.
2248 * In the other cases, current_slave pointer is changed and 0 is returned.
2250 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2252 struct bonding
*bond
= netdev_priv(bond_dev
);
2253 struct slave
*old_active
= NULL
;
2254 struct slave
*new_active
= NULL
;
2257 if (!USES_PRIMARY(bond
->params
.mode
))
2260 /* Verify that master_dev is indeed the master of slave_dev */
2261 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2264 read_lock(&bond
->lock
);
2266 read_lock(&bond
->curr_slave_lock
);
2267 old_active
= bond
->curr_active_slave
;
2268 read_unlock(&bond
->curr_slave_lock
);
2270 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2273 * Changing to the current active: do nothing; return success.
2275 if (new_active
&& (new_active
== old_active
)) {
2276 read_unlock(&bond
->lock
);
2282 (new_active
->link
== BOND_LINK_UP
) &&
2283 IS_UP(new_active
->dev
)) {
2285 write_lock_bh(&bond
->curr_slave_lock
);
2286 bond_change_active_slave(bond
, new_active
);
2287 write_unlock_bh(&bond
->curr_slave_lock
);
2288 unblock_netpoll_tx();
2292 read_unlock(&bond
->lock
);
2297 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2299 struct bonding
*bond
= netdev_priv(bond_dev
);
2301 info
->bond_mode
= bond
->params
.mode
;
2302 info
->miimon
= bond
->params
.miimon
;
2304 read_lock(&bond
->lock
);
2305 info
->num_slaves
= bond
->slave_cnt
;
2306 read_unlock(&bond
->lock
);
2311 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2313 struct bonding
*bond
= netdev_priv(bond_dev
);
2314 struct slave
*slave
;
2315 int i
, res
= -ENODEV
;
2317 read_lock(&bond
->lock
);
2319 bond_for_each_slave(bond
, slave
, i
) {
2320 if (i
== (int)info
->slave_id
) {
2322 strcpy(info
->slave_name
, slave
->dev
->name
);
2323 info
->link
= slave
->link
;
2324 info
->state
= bond_slave_state(slave
);
2325 info
->link_failure_count
= slave
->link_failure_count
;
2330 read_unlock(&bond
->lock
);
2335 /*-------------------------------- Monitoring -------------------------------*/
2338 static int bond_miimon_inspect(struct bonding
*bond
)
2340 struct slave
*slave
;
2341 int i
, link_state
, commit
= 0;
2342 bool ignore_updelay
;
2344 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2346 bond_for_each_slave(bond
, slave
, i
) {
2347 slave
->new_link
= BOND_LINK_NOCHANGE
;
2349 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2351 switch (slave
->link
) {
2356 slave
->link
= BOND_LINK_FAIL
;
2357 slave
->delay
= bond
->params
.downdelay
;
2359 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2361 (bond
->params
.mode
==
2362 BOND_MODE_ACTIVEBACKUP
) ?
2363 (bond_is_active_slave(slave
) ?
2364 "active " : "backup ") : "",
2366 bond
->params
.downdelay
* bond
->params
.miimon
);
2369 case BOND_LINK_FAIL
:
2372 * recovered before downdelay expired
2374 slave
->link
= BOND_LINK_UP
;
2375 slave
->jiffies
= jiffies
;
2376 pr_info("%s: link status up again after %d ms for interface %s.\n",
2378 (bond
->params
.downdelay
- slave
->delay
) *
2379 bond
->params
.miimon
,
2384 if (slave
->delay
<= 0) {
2385 slave
->new_link
= BOND_LINK_DOWN
;
2393 case BOND_LINK_DOWN
:
2397 slave
->link
= BOND_LINK_BACK
;
2398 slave
->delay
= bond
->params
.updelay
;
2401 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2402 bond
->dev
->name
, slave
->dev
->name
,
2403 ignore_updelay
? 0 :
2404 bond
->params
.updelay
*
2405 bond
->params
.miimon
);
2408 case BOND_LINK_BACK
:
2410 slave
->link
= BOND_LINK_DOWN
;
2411 pr_info("%s: link status down again after %d ms for interface %s.\n",
2413 (bond
->params
.updelay
- slave
->delay
) *
2414 bond
->params
.miimon
,
2423 if (slave
->delay
<= 0) {
2424 slave
->new_link
= BOND_LINK_UP
;
2426 ignore_updelay
= false;
2438 static void bond_miimon_commit(struct bonding
*bond
)
2440 struct slave
*slave
;
2443 bond_for_each_slave(bond
, slave
, i
) {
2444 switch (slave
->new_link
) {
2445 case BOND_LINK_NOCHANGE
:
2449 slave
->link
= BOND_LINK_UP
;
2450 slave
->jiffies
= jiffies
;
2452 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2453 /* prevent it from being the active one */
2454 bond_set_backup_slave(slave
);
2455 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2456 /* make it immediately active */
2457 bond_set_active_slave(slave
);
2458 } else if (slave
!= bond
->primary_slave
) {
2459 /* prevent it from being the active one */
2460 bond_set_backup_slave(slave
);
2463 bond_update_speed_duplex(slave
);
2465 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2466 bond
->dev
->name
, slave
->dev
->name
,
2467 slave
->speed
, slave
->duplex
? "full" : "half");
2469 /* notify ad that the link status has changed */
2470 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2471 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2473 if (bond_is_lb(bond
))
2474 bond_alb_handle_link_change(bond
, slave
,
2477 if (!bond
->curr_active_slave
||
2478 (slave
== bond
->primary_slave
))
2483 case BOND_LINK_DOWN
:
2484 if (slave
->link_failure_count
< UINT_MAX
)
2485 slave
->link_failure_count
++;
2487 slave
->link
= BOND_LINK_DOWN
;
2489 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2490 bond
->params
.mode
== BOND_MODE_8023AD
)
2491 bond_set_slave_inactive_flags(slave
);
2493 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2494 bond
->dev
->name
, slave
->dev
->name
);
2496 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2497 bond_3ad_handle_link_change(slave
,
2500 if (bond_is_lb(bond
))
2501 bond_alb_handle_link_change(bond
, slave
,
2504 if (slave
== bond
->curr_active_slave
)
2510 pr_err("%s: invalid new link %d on slave %s\n",
2511 bond
->dev
->name
, slave
->new_link
,
2513 slave
->new_link
= BOND_LINK_NOCHANGE
;
2521 write_lock_bh(&bond
->curr_slave_lock
);
2522 bond_select_active_slave(bond
);
2523 write_unlock_bh(&bond
->curr_slave_lock
);
2524 unblock_netpoll_tx();
2527 bond_set_carrier(bond
);
2533 * Really a wrapper that splits the mii monitor into two phases: an
2534 * inspection, then (if inspection indicates something needs to be done)
2535 * an acquisition of appropriate locks followed by a commit phase to
2536 * implement whatever link state changes are indicated.
2538 void bond_mii_monitor(struct work_struct
*work
)
2540 struct bonding
*bond
= container_of(work
, struct bonding
,
2542 bool should_notify_peers
= false;
2543 unsigned long delay
;
2545 read_lock(&bond
->lock
);
2547 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2549 if (bond
->slave_cnt
== 0)
2552 should_notify_peers
= bond_should_notify_peers(bond
);
2554 if (bond_miimon_inspect(bond
)) {
2555 read_unlock(&bond
->lock
);
2557 /* Race avoidance with bond_close cancel of workqueue */
2558 if (!rtnl_trylock()) {
2559 read_lock(&bond
->lock
);
2561 should_notify_peers
= false;
2565 read_lock(&bond
->lock
);
2567 bond_miimon_commit(bond
);
2569 read_unlock(&bond
->lock
);
2570 rtnl_unlock(); /* might sleep, hold no other locks */
2571 read_lock(&bond
->lock
);
2575 if (bond
->params
.miimon
)
2576 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2578 read_unlock(&bond
->lock
);
2580 if (should_notify_peers
) {
2581 if (!rtnl_trylock()) {
2582 read_lock(&bond
->lock
);
2583 bond
->send_peer_notif
++;
2584 read_unlock(&bond
->lock
);
2587 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
2592 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2594 struct vlan_entry
*vlan
;
2595 struct net_device
*vlan_dev
;
2597 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2600 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2602 vlan_dev
= __vlan_find_dev_deep(bond
->dev
, vlan
->vlan_id
);
2604 if (vlan_dev
&& ip
== bond_confirm_addr(vlan_dev
, 0, ip
))
2612 * We go to the (large) trouble of VLAN tagging ARP frames because
2613 * switches in VLAN mode (especially if ports are configured as
2614 * "native" to a VLAN) might not pass non-tagged frames.
2616 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2618 struct sk_buff
*skb
;
2620 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2621 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2623 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2624 NULL
, slave_dev
->dev_addr
, NULL
);
2627 pr_err("ARP packet allocation failed\n");
2631 skb
= vlan_put_tag(skb
, vlan_id
);
2633 pr_err("failed to insert VLAN tag\n");
2641 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2644 __be32
*targets
= bond
->params
.arp_targets
;
2645 struct vlan_entry
*vlan
;
2646 struct net_device
*vlan_dev
= NULL
;
2649 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2653 pr_debug("basa: target %x\n", targets
[i
]);
2654 if (!bond_vlan_used(bond
)) {
2655 pr_debug("basa: empty vlan: arp_send\n");
2656 addr
= bond_confirm_addr(bond
->dev
, targets
[i
], 0);
2657 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2663 * If VLANs are configured, we do a route lookup to
2664 * determine which VLAN interface would be used, so we
2665 * can tag the ARP with the proper VLAN tag.
2667 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2670 if (net_ratelimit()) {
2671 pr_warning("%s: no route to arp_ip_target %pI4\n",
2672 bond
->dev
->name
, &targets
[i
]);
2678 * This target is not on a VLAN
2680 if (rt
->dst
.dev
== bond
->dev
) {
2682 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2683 addr
= bond_confirm_addr(bond
->dev
, targets
[i
], 0);
2684 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2690 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2692 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
2695 if (vlan_dev
== rt
->dst
.dev
) {
2696 vlan_id
= vlan
->vlan_id
;
2697 pr_debug("basa: vlan match on %s %d\n",
2698 vlan_dev
->name
, vlan_id
);
2703 if (vlan_id
&& vlan_dev
) {
2705 addr
= bond_confirm_addr(vlan_dev
, targets
[i
], 0);
2706 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2711 if (net_ratelimit()) {
2712 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2713 bond
->dev
->name
, &targets
[i
],
2714 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2720 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2723 __be32
*targets
= bond
->params
.arp_targets
;
2725 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2726 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2727 &sip
, &tip
, i
, &targets
[i
],
2728 bond_has_this_ip(bond
, tip
));
2729 if (sip
== targets
[i
]) {
2730 if (bond_has_this_ip(bond
, tip
))
2731 slave
->last_arp_rx
= jiffies
;
2737 static int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2738 struct slave
*slave
)
2740 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2741 unsigned char *arp_ptr
;
2745 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2746 return RX_HANDLER_ANOTHER
;
2748 read_lock(&bond
->lock
);
2749 alen
= arp_hdr_len(bond
->dev
);
2751 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2752 bond
->dev
->name
, skb
->dev
->name
);
2754 if (alen
> skb_headlen(skb
)) {
2755 arp
= kmalloc(alen
, GFP_ATOMIC
);
2758 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2762 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2763 skb
->pkt_type
== PACKET_OTHERHOST
||
2764 skb
->pkt_type
== PACKET_LOOPBACK
||
2765 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2766 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2770 arp_ptr
= (unsigned char *)(arp
+ 1);
2771 arp_ptr
+= bond
->dev
->addr_len
;
2772 memcpy(&sip
, arp_ptr
, 4);
2773 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2774 memcpy(&tip
, arp_ptr
, 4);
2776 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2777 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2778 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2782 * Backup slaves won't see the ARP reply, but do come through
2783 * here for each ARP probe (so we swap the sip/tip to validate
2784 * the probe). In a "redundant switch, common router" type of
2785 * configuration, the ARP probe will (hopefully) travel from
2786 * the active, through one switch, the router, then the other
2787 * switch before reaching the backup.
2789 if (bond_is_active_slave(slave
))
2790 bond_validate_arp(bond
, slave
, sip
, tip
);
2792 bond_validate_arp(bond
, slave
, tip
, sip
);
2795 read_unlock(&bond
->lock
);
2796 if (arp
!= (struct arphdr
*)skb
->data
)
2798 return RX_HANDLER_ANOTHER
;
2802 * this function is called regularly to monitor each slave's link
2803 * ensuring that traffic is being sent and received when arp monitoring
2804 * is used in load-balancing mode. if the adapter has been dormant, then an
2805 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2806 * arp monitoring in active backup mode.
2808 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2810 struct bonding
*bond
= container_of(work
, struct bonding
,
2812 struct slave
*slave
, *oldcurrent
;
2813 int do_failover
= 0;
2817 read_lock(&bond
->lock
);
2819 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2821 if (bond
->slave_cnt
== 0)
2824 read_lock(&bond
->curr_slave_lock
);
2825 oldcurrent
= bond
->curr_active_slave
;
2826 read_unlock(&bond
->curr_slave_lock
);
2828 /* see if any of the previous devices are up now (i.e. they have
2829 * xmt and rcv traffic). the curr_active_slave does not come into
2830 * the picture unless it is null. also, slave->jiffies is not needed
2831 * here because we send an arp on each slave and give a slave as
2832 * long as it needs to get the tx/rx within the delta.
2833 * TODO: what about up/down delay in arp mode? it wasn't here before
2836 bond_for_each_slave(bond
, slave
, i
) {
2837 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2839 if (slave
->link
!= BOND_LINK_UP
) {
2840 if (time_in_range(jiffies
,
2841 trans_start
- delta_in_ticks
,
2842 trans_start
+ delta_in_ticks
) &&
2843 time_in_range(jiffies
,
2844 slave
->dev
->last_rx
- delta_in_ticks
,
2845 slave
->dev
->last_rx
+ delta_in_ticks
)) {
2847 slave
->link
= BOND_LINK_UP
;
2848 bond_set_active_slave(slave
);
2850 /* primary_slave has no meaning in round-robin
2851 * mode. the window of a slave being up and
2852 * curr_active_slave being null after enslaving
2856 pr_info("%s: link status definitely up for interface %s, ",
2861 pr_info("%s: interface %s is now up\n",
2867 /* slave->link == BOND_LINK_UP */
2869 /* not all switches will respond to an arp request
2870 * when the source ip is 0, so don't take the link down
2871 * if we don't know our ip yet
2873 if (!time_in_range(jiffies
,
2874 trans_start
- delta_in_ticks
,
2875 trans_start
+ 2 * delta_in_ticks
) ||
2876 !time_in_range(jiffies
,
2877 slave
->dev
->last_rx
- delta_in_ticks
,
2878 slave
->dev
->last_rx
+ 2 * delta_in_ticks
)) {
2880 slave
->link
= BOND_LINK_DOWN
;
2881 bond_set_backup_slave(slave
);
2883 if (slave
->link_failure_count
< UINT_MAX
)
2884 slave
->link_failure_count
++;
2886 pr_info("%s: interface %s is now down.\n",
2890 if (slave
== oldcurrent
)
2895 /* note: if switch is in round-robin mode, all links
2896 * must tx arp to ensure all links rx an arp - otherwise
2897 * links may oscillate or not come up at all; if switch is
2898 * in something like xor mode, there is nothing we can
2899 * do - all replies will be rx'ed on same link causing slaves
2900 * to be unstable during low/no traffic periods
2902 if (IS_UP(slave
->dev
))
2903 bond_arp_send_all(bond
, slave
);
2908 write_lock_bh(&bond
->curr_slave_lock
);
2910 bond_select_active_slave(bond
);
2912 write_unlock_bh(&bond
->curr_slave_lock
);
2913 unblock_netpoll_tx();
2917 if (bond
->params
.arp_interval
)
2918 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2920 read_unlock(&bond
->lock
);
2924 * Called to inspect slaves for active-backup mode ARP monitor link state
2925 * changes. Sets new_link in slaves to specify what action should take
2926 * place for the slave. Returns 0 if no changes are found, >0 if changes
2927 * to link states must be committed.
2929 * Called with bond->lock held for read.
2931 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2933 struct slave
*slave
;
2935 unsigned long trans_start
;
2937 bond_for_each_slave(bond
, slave
, i
) {
2938 slave
->new_link
= BOND_LINK_NOCHANGE
;
2940 if (slave
->link
!= BOND_LINK_UP
) {
2941 if (time_in_range(jiffies
,
2942 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2943 slave_last_rx(bond
, slave
) + delta_in_ticks
)) {
2945 slave
->new_link
= BOND_LINK_UP
;
2953 * Give slaves 2*delta after being enslaved or made
2954 * active. This avoids bouncing, as the last receive
2955 * times need a full ARP monitor cycle to be updated.
2957 if (time_in_range(jiffies
,
2958 slave
->jiffies
- delta_in_ticks
,
2959 slave
->jiffies
+ 2 * delta_in_ticks
))
2963 * Backup slave is down if:
2964 * - No current_arp_slave AND
2965 * - more than 3*delta since last receive AND
2966 * - the bond has an IP address
2968 * Note: a non-null current_arp_slave indicates
2969 * the curr_active_slave went down and we are
2970 * searching for a new one; under this condition
2971 * we only take the curr_active_slave down - this
2972 * gives each slave a chance to tx/rx traffic
2973 * before being taken out
2975 if (!bond_is_active_slave(slave
) &&
2976 !bond
->current_arp_slave
&&
2977 !time_in_range(jiffies
,
2978 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2979 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
)) {
2981 slave
->new_link
= BOND_LINK_DOWN
;
2986 * Active slave is down if:
2987 * - more than 2*delta since transmitting OR
2988 * - (more than 2*delta since receive AND
2989 * the bond has an IP address)
2991 trans_start
= dev_trans_start(slave
->dev
);
2992 if (bond_is_active_slave(slave
) &&
2993 (!time_in_range(jiffies
,
2994 trans_start
- delta_in_ticks
,
2995 trans_start
+ 2 * delta_in_ticks
) ||
2996 !time_in_range(jiffies
,
2997 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2998 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
))) {
3000 slave
->new_link
= BOND_LINK_DOWN
;
3009 * Called to commit link state changes noted by inspection step of
3010 * active-backup mode ARP monitor.
3012 * Called with RTNL and bond->lock for read.
3014 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
3016 struct slave
*slave
;
3018 unsigned long trans_start
;
3020 bond_for_each_slave(bond
, slave
, i
) {
3021 switch (slave
->new_link
) {
3022 case BOND_LINK_NOCHANGE
:
3026 trans_start
= dev_trans_start(slave
->dev
);
3027 if ((!bond
->curr_active_slave
&&
3028 time_in_range(jiffies
,
3029 trans_start
- delta_in_ticks
,
3030 trans_start
+ delta_in_ticks
)) ||
3031 bond
->curr_active_slave
!= slave
) {
3032 slave
->link
= BOND_LINK_UP
;
3033 if (bond
->current_arp_slave
) {
3034 bond_set_slave_inactive_flags(
3035 bond
->current_arp_slave
);
3036 bond
->current_arp_slave
= NULL
;
3039 pr_info("%s: link status definitely up for interface %s.\n",
3040 bond
->dev
->name
, slave
->dev
->name
);
3042 if (!bond
->curr_active_slave
||
3043 (slave
== bond
->primary_slave
))
3050 case BOND_LINK_DOWN
:
3051 if (slave
->link_failure_count
< UINT_MAX
)
3052 slave
->link_failure_count
++;
3054 slave
->link
= BOND_LINK_DOWN
;
3055 bond_set_slave_inactive_flags(slave
);
3057 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3058 bond
->dev
->name
, slave
->dev
->name
);
3060 if (slave
== bond
->curr_active_slave
) {
3061 bond
->current_arp_slave
= NULL
;
3068 pr_err("%s: impossible: new_link %d on slave %s\n",
3069 bond
->dev
->name
, slave
->new_link
,
3077 write_lock_bh(&bond
->curr_slave_lock
);
3078 bond_select_active_slave(bond
);
3079 write_unlock_bh(&bond
->curr_slave_lock
);
3080 unblock_netpoll_tx();
3083 bond_set_carrier(bond
);
3087 * Send ARP probes for active-backup mode ARP monitor.
3089 * Called with bond->lock held for read.
3091 static void bond_ab_arp_probe(struct bonding
*bond
)
3093 struct slave
*slave
;
3096 read_lock(&bond
->curr_slave_lock
);
3098 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3099 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3100 bond
->current_arp_slave
->dev
->name
,
3101 bond
->curr_active_slave
->dev
->name
);
3103 if (bond
->curr_active_slave
) {
3104 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3105 read_unlock(&bond
->curr_slave_lock
);
3109 read_unlock(&bond
->curr_slave_lock
);
3111 /* if we don't have a curr_active_slave, search for the next available
3112 * backup slave from the current_arp_slave and make it the candidate
3113 * for becoming the curr_active_slave
3116 if (!bond
->current_arp_slave
) {
3117 bond
->current_arp_slave
= bond
->first_slave
;
3118 if (!bond
->current_arp_slave
)
3122 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3124 /* search for next candidate */
3125 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3126 if (IS_UP(slave
->dev
)) {
3127 slave
->link
= BOND_LINK_BACK
;
3128 bond_set_slave_active_flags(slave
);
3129 bond_arp_send_all(bond
, slave
);
3130 slave
->jiffies
= jiffies
;
3131 bond
->current_arp_slave
= slave
;
3135 /* if the link state is up at this point, we
3136 * mark it down - this can happen if we have
3137 * simultaneous link failures and
3138 * reselect_active_interface doesn't make this
3139 * one the current slave so it is still marked
3140 * up when it is actually down
3142 if (slave
->link
== BOND_LINK_UP
) {
3143 slave
->link
= BOND_LINK_DOWN
;
3144 if (slave
->link_failure_count
< UINT_MAX
)
3145 slave
->link_failure_count
++;
3147 bond_set_slave_inactive_flags(slave
);
3149 pr_info("%s: backup interface %s is now down.\n",
3150 bond
->dev
->name
, slave
->dev
->name
);
3155 void bond_activebackup_arp_mon(struct work_struct
*work
)
3157 struct bonding
*bond
= container_of(work
, struct bonding
,
3159 bool should_notify_peers
= false;
3162 read_lock(&bond
->lock
);
3164 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3166 if (bond
->slave_cnt
== 0)
3169 should_notify_peers
= bond_should_notify_peers(bond
);
3171 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3172 read_unlock(&bond
->lock
);
3174 /* Race avoidance with bond_close flush of workqueue */
3175 if (!rtnl_trylock()) {
3176 read_lock(&bond
->lock
);
3178 should_notify_peers
= false;
3182 read_lock(&bond
->lock
);
3184 bond_ab_arp_commit(bond
, delta_in_ticks
);
3186 read_unlock(&bond
->lock
);
3188 read_lock(&bond
->lock
);
3191 bond_ab_arp_probe(bond
);
3194 if (bond
->params
.arp_interval
)
3195 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3197 read_unlock(&bond
->lock
);
3199 if (should_notify_peers
) {
3200 if (!rtnl_trylock()) {
3201 read_lock(&bond
->lock
);
3202 bond
->send_peer_notif
++;
3203 read_unlock(&bond
->lock
);
3206 netdev_bonding_change(bond
->dev
, NETDEV_NOTIFY_PEERS
);
3211 /*-------------------------- netdev event handling --------------------------*/
3214 * Change device name
3216 static int bond_event_changename(struct bonding
*bond
)
3218 bond_remove_proc_entry(bond
);
3219 bond_create_proc_entry(bond
);
3221 bond_debug_reregister(bond
);
3226 static int bond_master_netdev_event(unsigned long event
,
3227 struct net_device
*bond_dev
)
3229 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3232 case NETDEV_CHANGENAME
:
3233 return bond_event_changename(event_bond
);
3234 case NETDEV_UNREGISTER
:
3235 bond_remove_proc_entry(event_bond
);
3237 case NETDEV_REGISTER
:
3238 bond_create_proc_entry(event_bond
);
3247 static int bond_slave_netdev_event(unsigned long event
,
3248 struct net_device
*slave_dev
)
3250 struct net_device
*bond_dev
= slave_dev
->master
;
3251 struct bonding
*bond
= netdev_priv(bond_dev
);
3252 struct slave
*slave
= NULL
;
3255 case NETDEV_UNREGISTER
:
3257 if (bond
->setup_by_slave
)
3258 bond_release_and_destroy(bond_dev
, slave_dev
);
3260 bond_release(bond_dev
, slave_dev
);
3265 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3267 u32 old_speed
= slave
->speed
;
3268 u8 old_duplex
= slave
->duplex
;
3270 bond_update_speed_duplex(slave
);
3272 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3273 if (old_speed
!= slave
->speed
)
3274 bond_3ad_adapter_speed_changed(slave
);
3275 if (old_duplex
!= slave
->duplex
)
3276 bond_3ad_adapter_duplex_changed(slave
);
3283 * ... Or is it this?
3286 case NETDEV_CHANGEMTU
:
3288 * TODO: Should slaves be allowed to
3289 * independently alter their MTU? For
3290 * an active-backup bond, slaves need
3291 * not be the same type of device, so
3292 * MTUs may vary. For other modes,
3293 * slaves arguably should have the
3294 * same MTUs. To do this, we'd need to
3295 * take over the slave's change_mtu
3296 * function for the duration of their
3300 case NETDEV_CHANGENAME
:
3302 * TODO: handle changing the primary's name
3305 case NETDEV_FEAT_CHANGE
:
3306 bond_compute_features(bond
);
3316 * bond_netdev_event: handle netdev notifier chain events.
3318 * This function receives events for the netdev chain. The caller (an
3319 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3320 * locks for us to safely manipulate the slave devices (RTNL lock,
3323 static int bond_netdev_event(struct notifier_block
*this,
3324 unsigned long event
, void *ptr
)
3326 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3328 pr_debug("event_dev: %s, event: %lx\n",
3329 event_dev
? event_dev
->name
: "None",
3332 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3335 if (event_dev
->flags
& IFF_MASTER
) {
3336 pr_debug("IFF_MASTER\n");
3337 return bond_master_netdev_event(event
, event_dev
);
3340 if (event_dev
->flags
& IFF_SLAVE
) {
3341 pr_debug("IFF_SLAVE\n");
3342 return bond_slave_netdev_event(event
, event_dev
);
3348 static struct notifier_block bond_netdev_notifier
= {
3349 .notifier_call
= bond_netdev_event
,
3352 /*---------------------------- Hashing Policies -----------------------------*/
3355 * Hash for the output device based upon layer 2 and layer 3 data. If
3356 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3358 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3360 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3361 struct iphdr
*iph
= ip_hdr(skb
);
3363 if (skb
->protocol
== htons(ETH_P_IP
)) {
3364 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3365 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3368 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3372 * Hash for the output device based upon layer 3 and layer 4 data. If
3373 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3374 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3376 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3378 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3379 struct iphdr
*iph
= ip_hdr(skb
);
3380 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3383 if (skb
->protocol
== htons(ETH_P_IP
)) {
3384 if (!ip_is_fragment(iph
) &&
3385 (iph
->protocol
== IPPROTO_TCP
||
3386 iph
->protocol
== IPPROTO_UDP
)) {
3387 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3389 return (layer4_xor
^
3390 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3394 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3398 * Hash for the output device based upon layer 2 data
3400 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3402 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3404 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3407 /*-------------------------- Device entry points ----------------------------*/
3409 static int bond_open(struct net_device
*bond_dev
)
3411 struct bonding
*bond
= netdev_priv(bond_dev
);
3412 struct slave
*slave
;
3415 /* reset slave->backup and slave->inactive */
3416 read_lock(&bond
->lock
);
3417 if (bond
->slave_cnt
> 0) {
3418 read_lock(&bond
->curr_slave_lock
);
3419 bond_for_each_slave(bond
, slave
, i
) {
3420 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3421 && (slave
!= bond
->curr_active_slave
)) {
3422 bond_set_slave_inactive_flags(slave
);
3424 bond_set_slave_active_flags(slave
);
3427 read_unlock(&bond
->curr_slave_lock
);
3429 read_unlock(&bond
->lock
);
3431 INIT_DELAYED_WORK(&bond
->mcast_work
, bond_resend_igmp_join_requests_delayed
);
3433 if (bond_is_lb(bond
)) {
3434 /* bond_alb_initialize must be called before the timer
3437 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3438 /* something went wrong - fail the open operation */
3442 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3443 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3446 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3447 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3448 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3451 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3452 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3453 INIT_DELAYED_WORK(&bond
->arp_work
,
3454 bond_activebackup_arp_mon
);
3456 INIT_DELAYED_WORK(&bond
->arp_work
,
3457 bond_loadbalance_arp_mon
);
3459 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3460 if (bond
->params
.arp_validate
)
3461 bond
->recv_probe
= bond_arp_rcv
;
3464 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3465 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3466 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3467 /* register to receive LACPDUs */
3468 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3469 bond_3ad_initiate_agg_selection(bond
, 1);
3475 static int bond_close(struct net_device
*bond_dev
)
3477 struct bonding
*bond
= netdev_priv(bond_dev
);
3479 write_lock_bh(&bond
->lock
);
3481 bond
->send_peer_notif
= 0;
3483 write_unlock_bh(&bond
->lock
);
3485 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3486 cancel_delayed_work_sync(&bond
->mii_work
);
3489 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3490 cancel_delayed_work_sync(&bond
->arp_work
);
3493 switch (bond
->params
.mode
) {
3494 case BOND_MODE_8023AD
:
3495 cancel_delayed_work_sync(&bond
->ad_work
);
3499 cancel_delayed_work_sync(&bond
->alb_work
);
3505 if (delayed_work_pending(&bond
->mcast_work
))
3506 cancel_delayed_work_sync(&bond
->mcast_work
);
3508 if (bond_is_lb(bond
)) {
3509 /* Must be called only after all
3510 * slaves have been released
3512 bond_alb_deinitialize(bond
);
3514 bond
->recv_probe
= NULL
;
3519 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3520 struct rtnl_link_stats64
*stats
)
3522 struct bonding
*bond
= netdev_priv(bond_dev
);
3523 struct rtnl_link_stats64 temp
;
3524 struct slave
*slave
;
3527 memset(stats
, 0, sizeof(*stats
));
3529 read_lock_bh(&bond
->lock
);
3531 bond_for_each_slave(bond
, slave
, i
) {
3532 const struct rtnl_link_stats64
*sstats
=
3533 dev_get_stats(slave
->dev
, &temp
);
3535 stats
->rx_packets
+= sstats
->rx_packets
;
3536 stats
->rx_bytes
+= sstats
->rx_bytes
;
3537 stats
->rx_errors
+= sstats
->rx_errors
;
3538 stats
->rx_dropped
+= sstats
->rx_dropped
;
3540 stats
->tx_packets
+= sstats
->tx_packets
;
3541 stats
->tx_bytes
+= sstats
->tx_bytes
;
3542 stats
->tx_errors
+= sstats
->tx_errors
;
3543 stats
->tx_dropped
+= sstats
->tx_dropped
;
3545 stats
->multicast
+= sstats
->multicast
;
3546 stats
->collisions
+= sstats
->collisions
;
3548 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3549 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3550 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3551 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3552 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3553 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3555 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3556 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3557 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3558 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3559 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3562 read_unlock_bh(&bond
->lock
);
3567 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3569 struct net_device
*slave_dev
= NULL
;
3570 struct ifbond k_binfo
;
3571 struct ifbond __user
*u_binfo
= NULL
;
3572 struct ifslave k_sinfo
;
3573 struct ifslave __user
*u_sinfo
= NULL
;
3574 struct mii_ioctl_data
*mii
= NULL
;
3577 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3589 * We do this again just in case we were called by SIOCGMIIREG
3590 * instead of SIOCGMIIPHY.
3597 if (mii
->reg_num
== 1) {
3598 struct bonding
*bond
= netdev_priv(bond_dev
);
3600 read_lock(&bond
->lock
);
3601 read_lock(&bond
->curr_slave_lock
);
3602 if (netif_carrier_ok(bond
->dev
))
3603 mii
->val_out
= BMSR_LSTATUS
;
3605 read_unlock(&bond
->curr_slave_lock
);
3606 read_unlock(&bond
->lock
);
3610 case BOND_INFO_QUERY_OLD
:
3611 case SIOCBONDINFOQUERY
:
3612 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3614 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3617 res
= bond_info_query(bond_dev
, &k_binfo
);
3619 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3623 case BOND_SLAVE_INFO_QUERY_OLD
:
3624 case SIOCBONDSLAVEINFOQUERY
:
3625 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3627 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3630 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3632 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3641 if (!capable(CAP_NET_ADMIN
))
3644 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
3646 pr_debug("slave_dev=%p:\n", slave_dev
);
3651 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3653 case BOND_ENSLAVE_OLD
:
3654 case SIOCBONDENSLAVE
:
3655 res
= bond_enslave(bond_dev
, slave_dev
);
3657 case BOND_RELEASE_OLD
:
3658 case SIOCBONDRELEASE
:
3659 res
= bond_release(bond_dev
, slave_dev
);
3661 case BOND_SETHWADDR_OLD
:
3662 case SIOCBONDSETHWADDR
:
3663 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3665 case BOND_CHANGE_ACTIVE_OLD
:
3666 case SIOCBONDCHANGEACTIVE
:
3667 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3679 static bool bond_addr_in_mc_list(unsigned char *addr
,
3680 struct netdev_hw_addr_list
*list
,
3683 struct netdev_hw_addr
*ha
;
3685 netdev_hw_addr_list_for_each(ha
, list
)
3686 if (!memcmp(ha
->addr
, addr
, addrlen
))
3692 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3694 struct bonding
*bond
= netdev_priv(bond_dev
);
3696 if (change
& IFF_PROMISC
)
3697 bond_set_promiscuity(bond
,
3698 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3700 if (change
& IFF_ALLMULTI
)
3701 bond_set_allmulti(bond
,
3702 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3705 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3707 struct bonding
*bond
= netdev_priv(bond_dev
);
3708 struct netdev_hw_addr
*ha
;
3711 read_lock(&bond
->lock
);
3713 /* looking for addresses to add to slaves' mc list */
3714 netdev_for_each_mc_addr(ha
, bond_dev
) {
3715 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
3716 bond_dev
->addr_len
);
3718 bond_mc_add(bond
, ha
->addr
);
3721 /* looking for addresses to delete from slaves' list */
3722 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
3723 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
3724 bond_dev
->addr_len
);
3726 bond_mc_del(bond
, ha
->addr
);
3729 /* save master's multicast list */
3730 __hw_addr_flush(&bond
->mc_list
);
3731 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
3732 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
3734 read_unlock(&bond
->lock
);
3737 static int bond_neigh_init(struct neighbour
*n
)
3739 struct bonding
*bond
= netdev_priv(n
->dev
);
3740 struct slave
*slave
= bond
->first_slave
;
3741 const struct net_device_ops
*slave_ops
;
3742 struct neigh_parms parms
;
3748 slave_ops
= slave
->dev
->netdev_ops
;
3750 if (!slave_ops
->ndo_neigh_setup
)
3753 parms
.neigh_setup
= NULL
;
3754 parms
.neigh_cleanup
= NULL
;
3755 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3760 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3761 * after the last slave has been detached. Assumes that all slaves
3762 * utilize the same neigh_cleanup (true at this writing as only user
3765 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3767 if (!parms
.neigh_setup
)
3770 return parms
.neigh_setup(n
);
3774 * The bonding ndo_neigh_setup is called at init time beofre any
3775 * slave exists. So we must declare proxy setup function which will
3776 * be used at run time to resolve the actual slave neigh param setup.
3778 static int bond_neigh_setup(struct net_device
*dev
,
3779 struct neigh_parms
*parms
)
3781 parms
->neigh_setup
= bond_neigh_init
;
3787 * Change the MTU of all of a master's slaves to match the master
3789 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3791 struct bonding
*bond
= netdev_priv(bond_dev
);
3792 struct slave
*slave
, *stop_at
;
3796 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3797 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3799 /* Can't hold bond->lock with bh disabled here since
3800 * some base drivers panic. On the other hand we can't
3801 * hold bond->lock without bh disabled because we'll
3802 * deadlock. The only solution is to rely on the fact
3803 * that we're under rtnl_lock here, and the slaves
3804 * list won't change. This doesn't solve the problem
3805 * of setting the slave's MTU while it is
3806 * transmitting, but the assumption is that the base
3807 * driver can handle that.
3809 * TODO: figure out a way to safely iterate the slaves
3810 * list, but without holding a lock around the actual
3811 * call to the base driver.
3814 bond_for_each_slave(bond
, slave
, i
) {
3815 pr_debug("s %p s->p %p c_m %p\n",
3818 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3820 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3823 /* If we failed to set the slave's mtu to the new value
3824 * we must abort the operation even in ACTIVE_BACKUP
3825 * mode, because if we allow the backup slaves to have
3826 * different mtu values than the active slave we'll
3827 * need to change their mtu when doing a failover. That
3828 * means changing their mtu from timer context, which
3829 * is probably not a good idea.
3831 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3836 bond_dev
->mtu
= new_mtu
;
3841 /* unwind from head to the slave that failed */
3843 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3846 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3848 pr_debug("unwind err %d dev %s\n",
3849 tmp_res
, slave
->dev
->name
);
3859 * Note that many devices must be down to change the HW address, and
3860 * downing the master releases all slaves. We can make bonds full of
3861 * bonding devices to test this, however.
3863 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3865 struct bonding
*bond
= netdev_priv(bond_dev
);
3866 struct sockaddr
*sa
= addr
, tmp_sa
;
3867 struct slave
*slave
, *stop_at
;
3871 if (bond
->params
.mode
== BOND_MODE_ALB
)
3872 return bond_alb_set_mac_address(bond_dev
, addr
);
3875 pr_debug("bond=%p, name=%s\n",
3876 bond
, bond_dev
? bond_dev
->name
: "None");
3879 * If fail_over_mac is set to active, do nothing and return
3880 * success. Returning an error causes ifenslave to fail.
3882 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
3885 if (!is_valid_ether_addr(sa
->sa_data
))
3886 return -EADDRNOTAVAIL
;
3888 /* Can't hold bond->lock with bh disabled here since
3889 * some base drivers panic. On the other hand we can't
3890 * hold bond->lock without bh disabled because we'll
3891 * deadlock. The only solution is to rely on the fact
3892 * that we're under rtnl_lock here, and the slaves
3893 * list won't change. This doesn't solve the problem
3894 * of setting the slave's hw address while it is
3895 * transmitting, but the assumption is that the base
3896 * driver can handle that.
3898 * TODO: figure out a way to safely iterate the slaves
3899 * list, but without holding a lock around the actual
3900 * call to the base driver.
3903 bond_for_each_slave(bond
, slave
, i
) {
3904 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3905 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3907 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3909 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3913 res
= dev_set_mac_address(slave
->dev
, addr
);
3915 /* TODO: consider downing the slave
3917 * User should expect communications
3918 * breakage anyway until ARP finish
3921 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3927 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3931 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3932 tmp_sa
.sa_family
= bond_dev
->type
;
3934 /* unwind from head to the slave that failed */
3936 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3939 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3941 pr_debug("unwind err %d dev %s\n",
3942 tmp_res
, slave
->dev
->name
);
3949 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3951 struct bonding
*bond
= netdev_priv(bond_dev
);
3952 struct slave
*slave
, *start_at
;
3953 int i
, slave_no
, res
= 1;
3954 struct iphdr
*iph
= ip_hdr(skb
);
3957 * Start with the curr_active_slave that joined the bond as the
3958 * default for sending IGMP traffic. For failover purposes one
3959 * needs to maintain some consistency for the interface that will
3960 * send the join/membership reports. The curr_active_slave found
3961 * will send all of this type of traffic.
3963 if ((iph
->protocol
== IPPROTO_IGMP
) &&
3964 (skb
->protocol
== htons(ETH_P_IP
))) {
3966 read_lock(&bond
->curr_slave_lock
);
3967 slave
= bond
->curr_active_slave
;
3968 read_unlock(&bond
->curr_slave_lock
);
3974 * Concurrent TX may collide on rr_tx_counter; we accept
3975 * that as being rare enough not to justify using an
3978 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
3980 bond_for_each_slave(bond
, slave
, i
) {
3988 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
3989 if (IS_UP(slave
->dev
) &&
3990 (slave
->link
== BOND_LINK_UP
) &&
3991 bond_is_active_slave(slave
)) {
3992 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3999 /* no suitable interface, frame not sent */
4003 return NETDEV_TX_OK
;
4008 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4009 * the bond has a usable interface.
4011 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4013 struct bonding
*bond
= netdev_priv(bond_dev
);
4016 read_lock(&bond
->curr_slave_lock
);
4018 if (bond
->curr_active_slave
)
4019 res
= bond_dev_queue_xmit(bond
, skb
,
4020 bond
->curr_active_slave
->dev
);
4022 read_unlock(&bond
->curr_slave_lock
);
4025 /* no suitable interface, frame not sent */
4028 return NETDEV_TX_OK
;
4032 * In bond_xmit_xor() , we determine the output device by using a pre-
4033 * determined xmit_hash_policy(), If the selected device is not enabled,
4034 * find the next active slave.
4036 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4038 struct bonding
*bond
= netdev_priv(bond_dev
);
4039 struct slave
*slave
, *start_at
;
4044 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4046 bond_for_each_slave(bond
, slave
, i
) {
4054 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4055 if (IS_UP(slave
->dev
) &&
4056 (slave
->link
== BOND_LINK_UP
) &&
4057 bond_is_active_slave(slave
)) {
4058 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4064 /* no suitable interface, frame not sent */
4068 return NETDEV_TX_OK
;
4072 * in broadcast mode, we send everything to all usable interfaces.
4074 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4076 struct bonding
*bond
= netdev_priv(bond_dev
);
4077 struct slave
*slave
, *start_at
;
4078 struct net_device
*tx_dev
= NULL
;
4082 read_lock(&bond
->curr_slave_lock
);
4083 start_at
= bond
->curr_active_slave
;
4084 read_unlock(&bond
->curr_slave_lock
);
4089 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4090 if (IS_UP(slave
->dev
) &&
4091 (slave
->link
== BOND_LINK_UP
) &&
4092 bond_is_active_slave(slave
)) {
4094 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4096 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4101 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4107 tx_dev
= slave
->dev
;
4112 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4116 /* no suitable interface, frame not sent */
4119 /* frame sent to all suitable interfaces */
4120 return NETDEV_TX_OK
;
4123 /*------------------------- Device initialization ---------------------------*/
4125 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4127 switch (bond
->params
.xmit_policy
) {
4128 case BOND_XMIT_POLICY_LAYER23
:
4129 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4131 case BOND_XMIT_POLICY_LAYER34
:
4132 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4134 case BOND_XMIT_POLICY_LAYER2
:
4136 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4142 * Lookup the slave that corresponds to a qid
4144 static inline int bond_slave_override(struct bonding
*bond
,
4145 struct sk_buff
*skb
)
4148 struct slave
*slave
= NULL
;
4149 struct slave
*check_slave
;
4151 if (!skb
->queue_mapping
)
4154 /* Find out if any slaves have the same mapping as this skb. */
4155 bond_for_each_slave(bond
, check_slave
, i
) {
4156 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4157 slave
= check_slave
;
4162 /* If the slave isn't UP, use default transmit policy. */
4163 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4164 (slave
->link
== BOND_LINK_UP
)) {
4165 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4172 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4175 * This helper function exists to help dev_pick_tx get the correct
4176 * destination queue. Using a helper function skips a call to
4177 * skb_tx_hash and will put the skbs in the queue we expect on their
4178 * way down to the bonding driver.
4180 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
4183 * Save the original txq to restore before passing to the driver
4185 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
4187 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
4189 txq
-= dev
->real_num_tx_queues
;
4190 } while (txq
>= dev
->real_num_tx_queues
);
4195 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4197 struct bonding
*bond
= netdev_priv(dev
);
4199 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4200 if (!bond_slave_override(bond
, skb
))
4201 return NETDEV_TX_OK
;
4204 switch (bond
->params
.mode
) {
4205 case BOND_MODE_ROUNDROBIN
:
4206 return bond_xmit_roundrobin(skb
, dev
);
4207 case BOND_MODE_ACTIVEBACKUP
:
4208 return bond_xmit_activebackup(skb
, dev
);
4210 return bond_xmit_xor(skb
, dev
);
4211 case BOND_MODE_BROADCAST
:
4212 return bond_xmit_broadcast(skb
, dev
);
4213 case BOND_MODE_8023AD
:
4214 return bond_3ad_xmit_xor(skb
, dev
);
4217 return bond_alb_xmit(skb
, dev
);
4219 /* Should never happen, mode already checked */
4220 pr_err("%s: Error: Unknown bonding mode %d\n",
4221 dev
->name
, bond
->params
.mode
);
4224 return NETDEV_TX_OK
;
4228 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4230 struct bonding
*bond
= netdev_priv(dev
);
4231 netdev_tx_t ret
= NETDEV_TX_OK
;
4234 * If we risk deadlock from transmitting this in the
4235 * netpoll path, tell netpoll to queue the frame for later tx
4237 if (is_netpoll_tx_blocked(dev
))
4238 return NETDEV_TX_BUSY
;
4240 read_lock(&bond
->lock
);
4242 if (bond
->slave_cnt
)
4243 ret
= __bond_start_xmit(skb
, dev
);
4247 read_unlock(&bond
->lock
);
4253 * set bond mode specific net device operations
4255 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4257 struct net_device
*bond_dev
= bond
->dev
;
4260 case BOND_MODE_ROUNDROBIN
:
4262 case BOND_MODE_ACTIVEBACKUP
:
4265 bond_set_xmit_hash_policy(bond
);
4267 case BOND_MODE_BROADCAST
:
4269 case BOND_MODE_8023AD
:
4270 bond_set_xmit_hash_policy(bond
);
4277 /* Should never happen, mode already checked */
4278 pr_err("%s: Error: Unknown bonding mode %d\n",
4279 bond_dev
->name
, mode
);
4284 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4285 struct ethtool_drvinfo
*drvinfo
)
4287 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4288 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4289 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4292 static const struct ethtool_ops bond_ethtool_ops
= {
4293 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4294 .get_link
= ethtool_op_get_link
,
4297 static const struct net_device_ops bond_netdev_ops
= {
4298 .ndo_init
= bond_init
,
4299 .ndo_uninit
= bond_uninit
,
4300 .ndo_open
= bond_open
,
4301 .ndo_stop
= bond_close
,
4302 .ndo_start_xmit
= bond_start_xmit
,
4303 .ndo_select_queue
= bond_select_queue
,
4304 .ndo_get_stats64
= bond_get_stats
,
4305 .ndo_do_ioctl
= bond_do_ioctl
,
4306 .ndo_change_rx_flags
= bond_change_rx_flags
,
4307 .ndo_set_rx_mode
= bond_set_multicast_list
,
4308 .ndo_change_mtu
= bond_change_mtu
,
4309 .ndo_set_mac_address
= bond_set_mac_address
,
4310 .ndo_neigh_setup
= bond_neigh_setup
,
4311 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4312 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4313 #ifdef CONFIG_NET_POLL_CONTROLLER
4314 .ndo_netpoll_setup
= bond_netpoll_setup
,
4315 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4316 .ndo_poll_controller
= bond_poll_controller
,
4318 .ndo_add_slave
= bond_enslave
,
4319 .ndo_del_slave
= bond_release
,
4320 .ndo_fix_features
= bond_fix_features
,
4323 static void bond_destructor(struct net_device
*bond_dev
)
4325 struct bonding
*bond
= netdev_priv(bond_dev
);
4327 destroy_workqueue(bond
->wq
);
4328 free_netdev(bond_dev
);
4331 static void bond_setup(struct net_device
*bond_dev
)
4333 struct bonding
*bond
= netdev_priv(bond_dev
);
4335 /* initialize rwlocks */
4336 rwlock_init(&bond
->lock
);
4337 rwlock_init(&bond
->curr_slave_lock
);
4339 bond
->params
= bonding_defaults
;
4341 /* Initialize pointers */
4342 bond
->dev
= bond_dev
;
4343 INIT_LIST_HEAD(&bond
->vlan_list
);
4345 /* Initialize the device entry points */
4346 ether_setup(bond_dev
);
4347 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4348 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4349 bond_set_mode_ops(bond
, bond
->params
.mode
);
4351 bond_dev
->destructor
= bond_destructor
;
4353 /* Initialize the device options */
4354 bond_dev
->tx_queue_len
= 0;
4355 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4356 bond_dev
->priv_flags
|= IFF_BONDING
;
4357 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
4359 /* At first, we block adding VLANs. That's the only way to
4360 * prevent problems that occur when adding VLANs over an
4361 * empty bond. The block will be removed once non-challenged
4362 * slaves are enslaved.
4364 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4366 /* don't acquire bond device's netif_tx_lock when
4368 bond_dev
->features
|= NETIF_F_LLTX
;
4370 /* By default, we declare the bond to be fully
4371 * VLAN hardware accelerated capable. Special
4372 * care is taken in the various xmit functions
4373 * when there are slaves that are not hw accel
4377 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4378 NETIF_F_HW_VLAN_TX
|
4379 NETIF_F_HW_VLAN_RX
|
4380 NETIF_F_HW_VLAN_FILTER
;
4382 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
4383 bond_dev
->features
|= bond_dev
->hw_features
;
4386 static void bond_work_cancel_all(struct bonding
*bond
)
4388 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4389 cancel_delayed_work_sync(&bond
->mii_work
);
4391 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4392 cancel_delayed_work_sync(&bond
->arp_work
);
4394 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4395 delayed_work_pending(&bond
->alb_work
))
4396 cancel_delayed_work_sync(&bond
->alb_work
);
4398 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4399 delayed_work_pending(&bond
->ad_work
))
4400 cancel_delayed_work_sync(&bond
->ad_work
);
4402 if (delayed_work_pending(&bond
->mcast_work
))
4403 cancel_delayed_work_sync(&bond
->mcast_work
);
4407 * Destroy a bonding device.
4408 * Must be under rtnl_lock when this function is called.
4410 static void bond_uninit(struct net_device
*bond_dev
)
4412 struct bonding
*bond
= netdev_priv(bond_dev
);
4413 struct vlan_entry
*vlan
, *tmp
;
4415 bond_netpoll_cleanup(bond_dev
);
4417 /* Release the bonded slaves */
4418 bond_release_all(bond_dev
);
4420 list_del(&bond
->bond_list
);
4422 bond_work_cancel_all(bond
);
4424 bond_debug_unregister(bond
);
4426 __hw_addr_flush(&bond
->mc_list
);
4428 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4429 list_del(&vlan
->vlan_list
);
4434 /*------------------------- Module initialization ---------------------------*/
4437 * Convert string input module parms. Accept either the
4438 * number of the mode or its string name. A bit complicated because
4439 * some mode names are substrings of other names, and calls from sysfs
4440 * may have whitespace in the name (trailing newlines, for example).
4442 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4444 int modeint
= -1, i
, rv
;
4445 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4447 for (p
= (char *)buf
; *p
; p
++)
4448 if (!(isdigit(*p
) || isspace(*p
)))
4452 rv
= sscanf(buf
, "%20s", modestr
);
4454 rv
= sscanf(buf
, "%d", &modeint
);
4459 for (i
= 0; tbl
[i
].modename
; i
++) {
4460 if (modeint
== tbl
[i
].mode
)
4462 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4469 static int bond_check_params(struct bond_params
*params
)
4471 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4474 * Convert string parameters.
4477 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4478 if (bond_mode
== -1) {
4479 pr_err("Error: Invalid bonding mode \"%s\"\n",
4480 mode
== NULL
? "NULL" : mode
);
4485 if (xmit_hash_policy
) {
4486 if ((bond_mode
!= BOND_MODE_XOR
) &&
4487 (bond_mode
!= BOND_MODE_8023AD
)) {
4488 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4489 bond_mode_name(bond_mode
));
4491 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4493 if (xmit_hashtype
== -1) {
4494 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4495 xmit_hash_policy
== NULL
? "NULL" :
4503 if (bond_mode
!= BOND_MODE_8023AD
) {
4504 pr_info("lacp_rate param is irrelevant in mode %s\n",
4505 bond_mode_name(bond_mode
));
4507 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4508 if (lacp_fast
== -1) {
4509 pr_err("Error: Invalid lacp rate \"%s\"\n",
4510 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4517 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4518 if (params
->ad_select
== -1) {
4519 pr_err("Error: Invalid ad_select \"%s\"\n",
4520 ad_select
== NULL
? "NULL" : ad_select
);
4524 if (bond_mode
!= BOND_MODE_8023AD
) {
4525 pr_warning("ad_select param only affects 802.3ad mode\n");
4528 params
->ad_select
= BOND_AD_STABLE
;
4531 if (max_bonds
< 0) {
4532 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4533 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4534 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4538 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4539 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4540 miimon
= BOND_LINK_MON_INTERV
;
4544 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4549 if (downdelay
< 0) {
4550 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4551 downdelay
, INT_MAX
);
4555 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4556 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4561 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4562 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4567 /* reset values for 802.3ad */
4568 if (bond_mode
== BOND_MODE_8023AD
) {
4570 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");
4571 pr_warning("Forcing miimon to 100msec\n");
4576 if (tx_queues
< 1 || tx_queues
> 255) {
4577 pr_warning("Warning: tx_queues (%d) should be between "
4578 "1 and 255, resetting to %d\n",
4579 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4580 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4583 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4584 pr_warning("Warning: all_slaves_active module parameter (%d), "
4585 "not of valid value (0/1), so it was set to "
4586 "0\n", all_slaves_active
);
4587 all_slaves_active
= 0;
4590 if (resend_igmp
< 0 || resend_igmp
> 255) {
4591 pr_warning("Warning: resend_igmp (%d) should be between "
4592 "0 and 255, resetting to %d\n",
4593 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4594 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4597 /* reset values for TLB/ALB */
4598 if ((bond_mode
== BOND_MODE_TLB
) ||
4599 (bond_mode
== BOND_MODE_ALB
)) {
4601 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");
4602 pr_warning("Forcing miimon to 100msec\n");
4607 if (bond_mode
== BOND_MODE_ALB
) {
4608 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",
4613 if (updelay
|| downdelay
) {
4614 /* just warn the user the up/down delay will have
4615 * no effect since miimon is zero...
4617 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",
4618 updelay
, downdelay
);
4621 /* don't allow arp monitoring */
4623 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4624 miimon
, arp_interval
);
4628 if ((updelay
% miimon
) != 0) {
4629 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4631 (updelay
/ miimon
) * miimon
);
4636 if ((downdelay
% miimon
) != 0) {
4637 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4639 (downdelay
/ miimon
) * miimon
);
4642 downdelay
/= miimon
;
4645 if (arp_interval
< 0) {
4646 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4647 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4648 arp_interval
= BOND_LINK_ARP_INTERV
;
4651 for (arp_ip_count
= 0;
4652 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4654 /* not complete check, but should be good enough to
4656 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4657 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4658 arp_ip_target
[arp_ip_count
]);
4661 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4662 arp_target
[arp_ip_count
] = ip
;
4666 if (arp_interval
&& !arp_ip_count
) {
4667 /* don't allow arping if no arp_ip_target given... */
4668 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4674 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4675 pr_err("arp_validate only supported in active-backup mode\n");
4678 if (!arp_interval
) {
4679 pr_err("arp_validate requires arp_interval\n");
4683 arp_validate_value
= bond_parse_parm(arp_validate
,
4685 if (arp_validate_value
== -1) {
4686 pr_err("Error: invalid arp_validate \"%s\"\n",
4687 arp_validate
== NULL
? "NULL" : arp_validate
);
4691 arp_validate_value
= 0;
4694 pr_info("MII link monitoring set to %d ms\n", miimon
);
4695 } else if (arp_interval
) {
4698 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4700 arp_validate_tbl
[arp_validate_value
].modename
,
4703 for (i
= 0; i
< arp_ip_count
; i
++)
4704 pr_info(" %s", arp_ip_target
[i
]);
4708 } else if (max_bonds
) {
4709 /* miimon and arp_interval not set, we need one so things
4710 * work as expected, see bonding.txt for details
4712 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");
4715 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4716 /* currently, using a primary only makes sense
4717 * in active backup, TLB or ALB modes
4719 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4720 primary
, bond_mode_name(bond_mode
));
4724 if (primary
&& primary_reselect
) {
4725 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4727 if (primary_reselect_value
== -1) {
4728 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4730 NULL
? "NULL" : primary_reselect
);
4734 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4737 if (fail_over_mac
) {
4738 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4740 if (fail_over_mac_value
== -1) {
4741 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4742 arp_validate
== NULL
? "NULL" : arp_validate
);
4746 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4747 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4749 fail_over_mac_value
= BOND_FOM_NONE
;
4752 /* fill params struct with the proper values */
4753 params
->mode
= bond_mode
;
4754 params
->xmit_policy
= xmit_hashtype
;
4755 params
->miimon
= miimon
;
4756 params
->num_peer_notif
= num_peer_notif
;
4757 params
->arp_interval
= arp_interval
;
4758 params
->arp_validate
= arp_validate_value
;
4759 params
->updelay
= updelay
;
4760 params
->downdelay
= downdelay
;
4761 params
->use_carrier
= use_carrier
;
4762 params
->lacp_fast
= lacp_fast
;
4763 params
->primary
[0] = 0;
4764 params
->primary_reselect
= primary_reselect_value
;
4765 params
->fail_over_mac
= fail_over_mac_value
;
4766 params
->tx_queues
= tx_queues
;
4767 params
->all_slaves_active
= all_slaves_active
;
4768 params
->resend_igmp
= resend_igmp
;
4769 params
->min_links
= min_links
;
4772 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4773 params
->primary
[IFNAMSIZ
- 1] = 0;
4776 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4781 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4782 static struct lock_class_key bonding_netdev_addr_lock_key
;
4784 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4785 struct netdev_queue
*txq
,
4788 lockdep_set_class(&txq
->_xmit_lock
,
4789 &bonding_netdev_xmit_lock_key
);
4792 static void bond_set_lockdep_class(struct net_device
*dev
)
4794 lockdep_set_class(&dev
->addr_list_lock
,
4795 &bonding_netdev_addr_lock_key
);
4796 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4800 * Called from registration process
4802 static int bond_init(struct net_device
*bond_dev
)
4804 struct bonding
*bond
= netdev_priv(bond_dev
);
4805 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4806 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4808 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4811 * Initialize locks that may be required during
4812 * en/deslave operations. All of the bond_open work
4813 * (of which this is part) should really be moved to
4814 * a phase prior to dev_open
4816 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4817 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4819 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4823 bond_set_lockdep_class(bond_dev
);
4825 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4827 bond_prepare_sysfs_group(bond
);
4829 bond_debug_register(bond
);
4831 __hw_addr_init(&bond
->mc_list
);
4835 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4837 if (tb
[IFLA_ADDRESS
]) {
4838 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4840 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4841 return -EADDRNOTAVAIL
;
4846 static unsigned int bond_get_num_tx_queues(void)
4851 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
4853 .priv_size
= sizeof(struct bonding
),
4854 .setup
= bond_setup
,
4855 .validate
= bond_validate
,
4856 .get_num_tx_queues
= bond_get_num_tx_queues
,
4857 .get_num_rx_queues
= bond_get_num_tx_queues
, /* Use the same number
4861 /* Create a new bond based on the specified name and bonding parameters.
4862 * If name is NULL, obtain a suitable "bond%d" name for us.
4863 * Caller must NOT hold rtnl_lock; we need to release it here before we
4864 * set up our sysfs entries.
4866 int bond_create(struct net
*net
, const char *name
)
4868 struct net_device
*bond_dev
;
4873 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4874 name
? name
: "bond%d",
4875 bond_setup
, tx_queues
);
4877 pr_err("%s: eek! can't alloc netdev!\n", name
);
4882 dev_net_set(bond_dev
, net
);
4883 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4885 res
= register_netdevice(bond_dev
);
4887 netif_carrier_off(bond_dev
);
4891 bond_destructor(bond_dev
);
4895 static int __net_init
bond_net_init(struct net
*net
)
4897 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4900 INIT_LIST_HEAD(&bn
->dev_list
);
4902 bond_create_proc_dir(bn
);
4903 bond_create_sysfs(bn
);
4908 static void __net_exit
bond_net_exit(struct net
*net
)
4910 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4912 bond_destroy_sysfs(bn
);
4913 bond_destroy_proc_dir(bn
);
4916 static struct pernet_operations bond_net_ops
= {
4917 .init
= bond_net_init
,
4918 .exit
= bond_net_exit
,
4920 .size
= sizeof(struct bond_net
),
4923 static int __init
bonding_init(void)
4928 pr_info("%s", bond_version
);
4930 res
= bond_check_params(&bonding_defaults
);
4934 res
= register_pernet_subsys(&bond_net_ops
);
4938 res
= rtnl_link_register(&bond_link_ops
);
4942 bond_create_debugfs();
4944 for (i
= 0; i
< max_bonds
; i
++) {
4945 res
= bond_create(&init_net
, NULL
);
4950 register_netdevice_notifier(&bond_netdev_notifier
);
4954 rtnl_link_unregister(&bond_link_ops
);
4956 unregister_pernet_subsys(&bond_net_ops
);
4961 static void __exit
bonding_exit(void)
4963 unregister_netdevice_notifier(&bond_netdev_notifier
);
4965 bond_destroy_debugfs();
4967 rtnl_link_unregister(&bond_link_ops
);
4968 unregister_pernet_subsys(&bond_net_ops
);
4970 #ifdef CONFIG_NET_POLL_CONTROLLER
4972 * Make sure we don't have an imbalance on our netpoll blocking
4974 WARN_ON(atomic_read(&netpoll_block_tx
));
4978 module_init(bonding_init
);
4979 module_exit(bonding_exit
);
4980 MODULE_LICENSE("GPL");
4981 MODULE_VERSION(DRV_VERSION
);
4982 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4983 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4984 MODULE_ALIAS_RTNL_LINK("bond");