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
,
432 __be16 proto
, u16 vid
)
434 struct bonding
*bond
= netdev_priv(bond_dev
);
435 struct slave
*slave
, *stop_at
;
438 bond_for_each_slave(bond
, slave
, i
) {
439 res
= vlan_vid_add(slave
->dev
, proto
, vid
);
444 res
= bond_add_vlan(bond
, vid
);
446 pr_err("%s: Error: Failed to add vlan id %d\n",
447 bond_dev
->name
, vid
);
454 /* unwind from head to the slave that failed */
456 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
)
457 vlan_vid_del(slave
->dev
, proto
, vid
);
463 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
464 * @bond_dev: bonding net device that got called
465 * @vid: vlan id being removed
467 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
,
468 __be16 proto
, u16 vid
)
470 struct bonding
*bond
= netdev_priv(bond_dev
);
474 bond_for_each_slave(bond
, slave
, i
)
475 vlan_vid_del(slave
->dev
, proto
, vid
);
477 res
= bond_del_vlan(bond
, vid
);
479 pr_err("%s: Error: Failed to remove vlan id %d\n",
480 bond_dev
->name
, vid
);
487 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
489 struct vlan_entry
*vlan
;
492 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
493 res
= vlan_vid_add(slave_dev
, htons(ETH_P_8021Q
),
496 pr_warning("%s: Failed to add vlan id %d to device %s\n",
497 bond
->dev
->name
, vlan
->vlan_id
,
502 static void bond_del_vlans_from_slave(struct bonding
*bond
,
503 struct net_device
*slave_dev
)
505 struct vlan_entry
*vlan
;
507 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
510 vlan_vid_del(slave_dev
, htons(ETH_P_8021Q
), vlan
->vlan_id
);
514 /*------------------------------- Link status -------------------------------*/
517 * Set the carrier state for the master according to the state of its
518 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
519 * do special 802.3ad magic.
521 * Returns zero if carrier state does not change, nonzero if it does.
523 static int bond_set_carrier(struct bonding
*bond
)
528 if (bond
->slave_cnt
== 0)
531 if (bond
->params
.mode
== BOND_MODE_8023AD
)
532 return bond_3ad_set_carrier(bond
);
534 bond_for_each_slave(bond
, slave
, i
) {
535 if (slave
->link
== BOND_LINK_UP
) {
536 if (!netif_carrier_ok(bond
->dev
)) {
537 netif_carrier_on(bond
->dev
);
545 if (netif_carrier_ok(bond
->dev
)) {
546 netif_carrier_off(bond
->dev
);
553 * Get link speed and duplex from the slave's base driver
554 * using ethtool. If for some reason the call fails or the
555 * values are invalid, set speed and duplex to -1,
558 static void bond_update_speed_duplex(struct slave
*slave
)
560 struct net_device
*slave_dev
= slave
->dev
;
561 struct ethtool_cmd ecmd
;
565 slave
->speed
= SPEED_UNKNOWN
;
566 slave
->duplex
= DUPLEX_UNKNOWN
;
568 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
572 slave_speed
= ethtool_cmd_speed(&ecmd
);
573 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
576 switch (ecmd
.duplex
) {
584 slave
->speed
= slave_speed
;
585 slave
->duplex
= ecmd
.duplex
;
591 * if <dev> supports MII link status reporting, check its link status.
593 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
594 * depending upon the setting of the use_carrier parameter.
596 * Return either BMSR_LSTATUS, meaning that the link is up (or we
597 * can't tell and just pretend it is), or 0, meaning that the link is
600 * If reporting is non-zero, instead of faking link up, return -1 if
601 * both ETHTOOL and MII ioctls fail (meaning the device does not
602 * support them). If use_carrier is set, return whatever it says.
603 * It'd be nice if there was a good way to tell if a driver supports
604 * netif_carrier, but there really isn't.
606 static int bond_check_dev_link(struct bonding
*bond
,
607 struct net_device
*slave_dev
, int reporting
)
609 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
610 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
612 struct mii_ioctl_data
*mii
;
614 if (!reporting
&& !netif_running(slave_dev
))
617 if (bond
->params
.use_carrier
)
618 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
620 /* Try to get link status using Ethtool first. */
621 if (slave_dev
->ethtool_ops
->get_link
)
622 return slave_dev
->ethtool_ops
->get_link(slave_dev
) ?
625 /* Ethtool can't be used, fallback to MII ioctls. */
626 ioctl
= slave_ops
->ndo_do_ioctl
;
628 /* TODO: set pointer to correct ioctl on a per team member */
629 /* bases to make this more efficient. that is, once */
630 /* we determine the correct ioctl, we will always */
631 /* call it and not the others for that team */
635 * We cannot assume that SIOCGMIIPHY will also read a
636 * register; not all network drivers (e.g., e100)
640 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
641 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
643 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
644 mii
->reg_num
= MII_BMSR
;
645 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
646 return mii
->val_out
& BMSR_LSTATUS
;
651 * If reporting, report that either there's no dev->do_ioctl,
652 * or both SIOCGMIIREG and get_link failed (meaning that we
653 * cannot report link status). If not reporting, pretend
656 return reporting
? -1 : BMSR_LSTATUS
;
659 /*----------------------------- Multicast list ------------------------------*/
662 * Push the promiscuity flag down to appropriate slaves
664 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
667 if (USES_PRIMARY(bond
->params
.mode
)) {
668 /* write lock already acquired */
669 if (bond
->curr_active_slave
) {
670 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
676 bond_for_each_slave(bond
, slave
, i
) {
677 err
= dev_set_promiscuity(slave
->dev
, inc
);
686 * Push the allmulti flag down to all slaves
688 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
691 if (USES_PRIMARY(bond
->params
.mode
)) {
692 /* write lock already acquired */
693 if (bond
->curr_active_slave
) {
694 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
700 bond_for_each_slave(bond
, slave
, i
) {
701 err
= dev_set_allmulti(slave
->dev
, inc
);
710 * Add a Multicast address to slaves
713 static void bond_mc_add(struct bonding
*bond
, void *addr
)
715 if (USES_PRIMARY(bond
->params
.mode
)) {
716 /* write lock already acquired */
717 if (bond
->curr_active_slave
)
718 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
723 bond_for_each_slave(bond
, slave
, i
)
724 dev_mc_add(slave
->dev
, addr
);
729 * Remove a multicast address from slave
732 static void bond_mc_del(struct bonding
*bond
, void *addr
)
734 if (USES_PRIMARY(bond
->params
.mode
)) {
735 /* write lock already acquired */
736 if (bond
->curr_active_slave
)
737 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
741 bond_for_each_slave(bond
, slave
, i
) {
742 dev_mc_del(slave
->dev
, addr
);
748 static void __bond_resend_igmp_join_requests(struct net_device
*dev
)
750 struct in_device
*in_dev
;
752 in_dev
= __in_dev_get_rcu(dev
);
754 ip_mc_rejoin_groups(in_dev
);
758 * Retrieve the list of registered multicast addresses for the bonding
759 * device and retransmit an IGMP JOIN request to the current active
762 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
764 struct net_device
*bond_dev
, *vlan_dev
, *upper_dev
;
765 struct vlan_entry
*vlan
;
767 read_lock(&bond
->lock
);
770 bond_dev
= bond
->dev
;
772 /* rejoin all groups on bond device */
773 __bond_resend_igmp_join_requests(bond_dev
);
776 * if bond is enslaved to a bridge,
777 * then rejoin all groups on its master
779 upper_dev
= netdev_master_upper_dev_get_rcu(bond_dev
);
780 if (upper_dev
&& upper_dev
->priv_flags
& IFF_EBRIDGE
)
781 __bond_resend_igmp_join_requests(upper_dev
);
783 /* rejoin all groups on vlan devices */
784 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
785 vlan_dev
= __vlan_find_dev_deep(bond_dev
, htons(ETH_P_8021Q
),
788 __bond_resend_igmp_join_requests(vlan_dev
);
792 /* We use curr_slave_lock to protect against concurrent access to
793 * igmp_retrans from multiple running instances of this function and
794 * bond_change_active_slave
796 write_lock_bh(&bond
->curr_slave_lock
);
797 if (bond
->igmp_retrans
> 1) {
798 bond
->igmp_retrans
--;
799 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
801 write_unlock_bh(&bond
->curr_slave_lock
);
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
,
810 bond_resend_igmp_join_requests(bond
);
814 * flush all members of flush->mc_list from device dev->mc_list
816 static void bond_mc_list_flush(struct net_device
*bond_dev
,
817 struct net_device
*slave_dev
)
819 struct bonding
*bond
= netdev_priv(bond_dev
);
820 struct netdev_hw_addr
*ha
;
822 netdev_for_each_mc_addr(ha
, bond_dev
)
823 dev_mc_del(slave_dev
, ha
->addr
);
825 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
826 /* del lacpdu mc addr from mc list */
827 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
829 dev_mc_del(slave_dev
, lacpdu_multicast
);
833 /*--------------------------- Active slave change ---------------------------*/
836 * Update the mc list and multicast-related flags for the new and
837 * old active slaves (if any) according to the multicast mode, and
838 * promiscuous flags unconditionally.
840 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
841 struct slave
*old_active
)
843 struct netdev_hw_addr
*ha
;
845 if (!USES_PRIMARY(bond
->params
.mode
))
846 /* nothing to do - mc list is already up-to-date on
852 if (bond
->dev
->flags
& IFF_PROMISC
)
853 dev_set_promiscuity(old_active
->dev
, -1);
855 if (bond
->dev
->flags
& IFF_ALLMULTI
)
856 dev_set_allmulti(old_active
->dev
, -1);
858 netif_addr_lock_bh(bond
->dev
);
859 netdev_for_each_mc_addr(ha
, bond
->dev
)
860 dev_mc_del(old_active
->dev
, ha
->addr
);
861 netif_addr_unlock_bh(bond
->dev
);
865 /* FIXME: Signal errors upstream. */
866 if (bond
->dev
->flags
& IFF_PROMISC
)
867 dev_set_promiscuity(new_active
->dev
, 1);
869 if (bond
->dev
->flags
& IFF_ALLMULTI
)
870 dev_set_allmulti(new_active
->dev
, 1);
872 netif_addr_lock_bh(bond
->dev
);
873 netdev_for_each_mc_addr(ha
, bond
->dev
)
874 dev_mc_add(new_active
->dev
, ha
->addr
);
875 netif_addr_unlock_bh(bond
->dev
);
879 static struct slave
*bond_get_old_active(struct bonding
*bond
,
880 struct slave
*new_active
)
885 bond_for_each_slave(bond
, slave
, i
) {
886 if (slave
== new_active
)
889 if (ether_addr_equal(bond
->dev
->dev_addr
, slave
->dev
->dev_addr
))
897 * bond_do_fail_over_mac
899 * Perform special MAC address swapping for fail_over_mac settings
901 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
903 static void bond_do_fail_over_mac(struct bonding
*bond
,
904 struct slave
*new_active
,
905 struct slave
*old_active
)
906 __releases(&bond
->curr_slave_lock
)
907 __releases(&bond
->lock
)
908 __acquires(&bond
->lock
)
909 __acquires(&bond
->curr_slave_lock
)
911 u8 tmp_mac
[ETH_ALEN
];
912 struct sockaddr saddr
;
915 switch (bond
->params
.fail_over_mac
) {
916 case BOND_FOM_ACTIVE
:
918 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
919 new_active
->dev
->addr_len
);
920 write_unlock_bh(&bond
->curr_slave_lock
);
921 read_unlock(&bond
->lock
);
922 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
923 read_lock(&bond
->lock
);
924 write_lock_bh(&bond
->curr_slave_lock
);
927 case BOND_FOM_FOLLOW
:
929 * if new_active && old_active, swap them
930 * if just old_active, do nothing (going to no active slave)
931 * if just new_active, set new_active to bond's MAC
936 write_unlock_bh(&bond
->curr_slave_lock
);
937 read_unlock(&bond
->lock
);
940 old_active
= bond_get_old_active(bond
, new_active
);
943 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
944 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
946 saddr
.sa_family
= new_active
->dev
->type
;
948 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
949 saddr
.sa_family
= bond
->dev
->type
;
952 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
954 pr_err("%s: Error %d setting MAC of slave %s\n",
955 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
962 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
963 saddr
.sa_family
= old_active
->dev
->type
;
965 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
967 pr_err("%s: Error %d setting MAC of slave %s\n",
968 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
970 read_lock(&bond
->lock
);
971 write_lock_bh(&bond
->curr_slave_lock
);
974 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
975 bond
->dev
->name
, bond
->params
.fail_over_mac
);
981 static bool bond_should_change_active(struct bonding
*bond
)
983 struct slave
*prim
= bond
->primary_slave
;
984 struct slave
*curr
= bond
->curr_active_slave
;
986 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
988 if (bond
->force_primary
) {
989 bond
->force_primary
= false;
992 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
993 (prim
->speed
< curr
->speed
||
994 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
996 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
1002 * find_best_interface - select the best available slave to be the active one
1003 * @bond: our bonding struct
1005 * Warning: Caller must hold curr_slave_lock for writing.
1007 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1009 struct slave
*new_active
, *old_active
;
1010 struct slave
*bestslave
= NULL
;
1011 int mintime
= bond
->params
.updelay
;
1014 new_active
= bond
->curr_active_slave
;
1016 if (!new_active
) { /* there were no active slaves left */
1017 if (bond
->slave_cnt
> 0) /* found one slave */
1018 new_active
= bond
->first_slave
;
1020 return NULL
; /* still no slave, return NULL */
1023 if ((bond
->primary_slave
) &&
1024 bond
->primary_slave
->link
== BOND_LINK_UP
&&
1025 bond_should_change_active(bond
)) {
1026 new_active
= bond
->primary_slave
;
1029 /* remember where to stop iterating over the slaves */
1030 old_active
= new_active
;
1032 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1033 if (new_active
->link
== BOND_LINK_UP
) {
1035 } else if (new_active
->link
== BOND_LINK_BACK
&&
1036 IS_UP(new_active
->dev
)) {
1037 /* link up, but waiting for stabilization */
1038 if (new_active
->delay
< mintime
) {
1039 mintime
= new_active
->delay
;
1040 bestslave
= new_active
;
1048 static bool bond_should_notify_peers(struct bonding
*bond
)
1050 struct slave
*slave
= bond
->curr_active_slave
;
1052 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1053 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
1055 if (!slave
|| !bond
->send_peer_notif
||
1056 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
1059 bond
->send_peer_notif
--;
1064 * change_active_interface - change the active slave into the specified one
1065 * @bond: our bonding struct
1066 * @new: the new slave to make the active one
1068 * Set the new slave to the bond's settings and unset them on the old
1069 * curr_active_slave.
1070 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1072 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1073 * because it is apparently the best available slave we have, even though its
1074 * updelay hasn't timed out yet.
1076 * If new_active is not NULL, caller must hold bond->lock for read and
1077 * curr_slave_lock for write_bh.
1079 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1081 struct slave
*old_active
= bond
->curr_active_slave
;
1083 if (old_active
== new_active
)
1087 new_active
->jiffies
= jiffies
;
1089 if (new_active
->link
== BOND_LINK_BACK
) {
1090 if (USES_PRIMARY(bond
->params
.mode
)) {
1091 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1092 bond
->dev
->name
, new_active
->dev
->name
,
1093 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1096 new_active
->delay
= 0;
1097 new_active
->link
= BOND_LINK_UP
;
1099 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1100 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1102 if (bond_is_lb(bond
))
1103 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1105 if (USES_PRIMARY(bond
->params
.mode
)) {
1106 pr_info("%s: making interface %s the new active one.\n",
1107 bond
->dev
->name
, new_active
->dev
->name
);
1112 if (USES_PRIMARY(bond
->params
.mode
))
1113 bond_mc_swap(bond
, new_active
, old_active
);
1115 if (bond_is_lb(bond
)) {
1116 bond_alb_handle_active_change(bond
, new_active
);
1118 bond_set_slave_inactive_flags(old_active
);
1120 bond_set_slave_active_flags(new_active
);
1122 bond
->curr_active_slave
= new_active
;
1125 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1127 bond_set_slave_inactive_flags(old_active
);
1130 bool should_notify_peers
= false;
1132 bond_set_slave_active_flags(new_active
);
1134 if (bond
->params
.fail_over_mac
)
1135 bond_do_fail_over_mac(bond
, new_active
,
1138 if (netif_running(bond
->dev
)) {
1139 bond
->send_peer_notif
=
1140 bond
->params
.num_peer_notif
;
1141 should_notify_peers
=
1142 bond_should_notify_peers(bond
);
1145 write_unlock_bh(&bond
->curr_slave_lock
);
1146 read_unlock(&bond
->lock
);
1148 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
1149 if (should_notify_peers
)
1150 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
1153 read_lock(&bond
->lock
);
1154 write_lock_bh(&bond
->curr_slave_lock
);
1158 /* resend IGMP joins since active slave has changed or
1159 * all were sent on curr_active_slave.
1160 * resend only if bond is brought up with the affected
1161 * bonding modes and the retransmission is enabled */
1162 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
1163 ((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
1164 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
)) {
1165 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
1166 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 0);
1171 * bond_select_active_slave - select a new active slave, if needed
1172 * @bond: our bonding struct
1174 * This functions should be called when one of the following occurs:
1175 * - The old curr_active_slave has been released or lost its link.
1176 * - The primary_slave has got its link back.
1177 * - A slave has got its link back and there's no old curr_active_slave.
1179 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1181 void bond_select_active_slave(struct bonding
*bond
)
1183 struct slave
*best_slave
;
1186 best_slave
= bond_find_best_slave(bond
);
1187 if (best_slave
!= bond
->curr_active_slave
) {
1188 bond_change_active_slave(bond
, best_slave
);
1189 rv
= bond_set_carrier(bond
);
1193 if (netif_carrier_ok(bond
->dev
)) {
1194 pr_info("%s: first active interface up!\n",
1197 pr_info("%s: now running without any active interface !\n",
1203 /*--------------------------- slave list handling ---------------------------*/
1206 * This function attaches the slave to the end of list.
1208 * bond->lock held for writing by caller.
1210 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1212 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1213 new_slave
->next
= new_slave
;
1214 new_slave
->prev
= new_slave
;
1215 bond
->first_slave
= new_slave
;
1217 new_slave
->next
= bond
->first_slave
;
1218 new_slave
->prev
= bond
->first_slave
->prev
;
1219 new_slave
->next
->prev
= new_slave
;
1220 new_slave
->prev
->next
= new_slave
;
1227 * This function detaches the slave from the list.
1228 * WARNING: no check is made to verify if the slave effectively
1229 * belongs to <bond>.
1230 * Nothing is freed on return, structures are just unchained.
1231 * If any slave pointer in bond was pointing to <slave>,
1232 * it should be changed by the calling function.
1234 * bond->lock held for writing by caller.
1236 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1239 slave
->next
->prev
= slave
->prev
;
1242 slave
->prev
->next
= slave
->next
;
1244 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1245 if (bond
->slave_cnt
> 1) { /* there are more slave */
1246 bond
->first_slave
= slave
->next
;
1248 bond
->first_slave
= NULL
; /* slave was the last one */
1257 #ifdef CONFIG_NET_POLL_CONTROLLER
1258 static inline int slave_enable_netpoll(struct slave
*slave
)
1263 np
= kzalloc(sizeof(*np
), GFP_ATOMIC
);
1268 err
= __netpoll_setup(np
, slave
->dev
, GFP_ATOMIC
);
1277 static inline void slave_disable_netpoll(struct slave
*slave
)
1279 struct netpoll
*np
= slave
->np
;
1285 __netpoll_free_async(np
);
1287 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1289 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1291 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1296 static void bond_poll_controller(struct net_device
*bond_dev
)
1300 static void __bond_netpoll_cleanup(struct bonding
*bond
)
1302 struct slave
*slave
;
1305 bond_for_each_slave(bond
, slave
, i
)
1306 if (IS_UP(slave
->dev
))
1307 slave_disable_netpoll(slave
);
1309 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1311 struct bonding
*bond
= netdev_priv(bond_dev
);
1313 read_lock(&bond
->lock
);
1314 __bond_netpoll_cleanup(bond
);
1315 read_unlock(&bond
->lock
);
1318 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
, gfp_t gfp
)
1320 struct bonding
*bond
= netdev_priv(dev
);
1321 struct slave
*slave
;
1324 read_lock(&bond
->lock
);
1325 bond_for_each_slave(bond
, slave
, i
) {
1326 err
= slave_enable_netpoll(slave
);
1328 __bond_netpoll_cleanup(bond
);
1332 read_unlock(&bond
->lock
);
1336 static struct netpoll_info
*bond_netpoll_info(struct bonding
*bond
)
1338 return bond
->dev
->npinfo
;
1342 static inline int slave_enable_netpoll(struct slave
*slave
)
1346 static inline void slave_disable_netpoll(struct slave
*slave
)
1349 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1354 /*---------------------------------- IOCTL ----------------------------------*/
1356 static void bond_set_dev_addr(struct net_device
*bond_dev
,
1357 struct net_device
*slave_dev
)
1359 pr_debug("bond_dev=%p\n", bond_dev
);
1360 pr_debug("slave_dev=%p\n", slave_dev
);
1361 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1362 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1363 bond_dev
->addr_assign_type
= NET_ADDR_SET
;
1364 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
1367 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1368 netdev_features_t features
)
1370 struct slave
*slave
;
1371 struct bonding
*bond
= netdev_priv(dev
);
1372 netdev_features_t mask
;
1375 read_lock(&bond
->lock
);
1377 if (!bond
->first_slave
) {
1378 /* Disable adding VLANs to empty bond. But why? --mq */
1379 features
|= NETIF_F_VLAN_CHALLENGED
;
1384 features
&= ~NETIF_F_ONE_FOR_ALL
;
1385 features
|= NETIF_F_ALL_FOR_ALL
;
1387 bond_for_each_slave(bond
, slave
, i
) {
1388 features
= netdev_increment_features(features
,
1389 slave
->dev
->features
,
1392 features
= netdev_add_tso_features(features
, mask
);
1395 read_unlock(&bond
->lock
);
1399 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1400 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1401 NETIF_F_HIGHDMA | NETIF_F_LRO)
1403 static void bond_compute_features(struct bonding
*bond
)
1405 struct slave
*slave
;
1406 struct net_device
*bond_dev
= bond
->dev
;
1407 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1408 unsigned short max_hard_header_len
= ETH_HLEN
;
1409 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1410 u16 gso_max_segs
= GSO_MAX_SEGS
;
1412 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1414 read_lock(&bond
->lock
);
1416 if (!bond
->first_slave
)
1419 bond_for_each_slave(bond
, slave
, i
) {
1420 vlan_features
= netdev_increment_features(vlan_features
,
1421 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1423 dst_release_flag
&= slave
->dev
->priv_flags
;
1424 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1425 max_hard_header_len
= slave
->dev
->hard_header_len
;
1427 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1428 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1432 bond_dev
->vlan_features
= vlan_features
;
1433 bond_dev
->hard_header_len
= max_hard_header_len
;
1434 bond_dev
->gso_max_segs
= gso_max_segs
;
1435 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1437 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1438 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1440 read_unlock(&bond
->lock
);
1442 netdev_change_features(bond_dev
);
1445 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1446 struct net_device
*slave_dev
)
1448 struct bonding
*bond
= netdev_priv(bond_dev
);
1450 bond_dev
->header_ops
= slave_dev
->header_ops
;
1452 bond_dev
->type
= slave_dev
->type
;
1453 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1454 bond_dev
->addr_len
= slave_dev
->addr_len
;
1456 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1457 slave_dev
->addr_len
);
1458 bond
->setup_by_slave
= 1;
1461 /* On bonding slaves other than the currently active slave, suppress
1462 * duplicates except for alb non-mcast/bcast.
1464 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1465 struct slave
*slave
,
1466 struct bonding
*bond
)
1468 if (bond_is_slave_inactive(slave
)) {
1469 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1470 skb
->pkt_type
!= PACKET_BROADCAST
&&
1471 skb
->pkt_type
!= PACKET_MULTICAST
)
1478 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1480 struct sk_buff
*skb
= *pskb
;
1481 struct slave
*slave
;
1482 struct bonding
*bond
;
1483 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1485 int ret
= RX_HANDLER_ANOTHER
;
1487 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1489 return RX_HANDLER_CONSUMED
;
1493 slave
= bond_slave_get_rcu(skb
->dev
);
1496 if (bond
->params
.arp_interval
)
1497 slave
->dev
->last_rx
= jiffies
;
1499 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1501 ret
= recv_probe(skb
, bond
, slave
);
1502 if (ret
== RX_HANDLER_CONSUMED
) {
1508 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1509 return RX_HANDLER_EXACT
;
1512 skb
->dev
= bond
->dev
;
1514 if (bond
->params
.mode
== BOND_MODE_ALB
&&
1515 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1516 skb
->pkt_type
== PACKET_HOST
) {
1518 if (unlikely(skb_cow_head(skb
,
1519 skb
->data
- skb_mac_header(skb
)))) {
1521 return RX_HANDLER_CONSUMED
;
1523 memcpy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
, ETH_ALEN
);
1529 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1530 struct net_device
*slave_dev
)
1534 err
= netdev_master_upper_dev_link(slave_dev
, bond_dev
);
1537 slave_dev
->flags
|= IFF_SLAVE
;
1538 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
);
1542 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1543 struct net_device
*slave_dev
)
1545 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1546 slave_dev
->flags
&= ~IFF_SLAVE
;
1547 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
);
1550 /* enslave device <slave> to bond device <master> */
1551 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1553 struct bonding
*bond
= netdev_priv(bond_dev
);
1554 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1555 struct slave
*new_slave
= NULL
;
1556 struct netdev_hw_addr
*ha
;
1557 struct sockaddr addr
;
1561 if (!bond
->params
.use_carrier
&&
1562 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1563 slave_ops
->ndo_do_ioctl
== NULL
) {
1564 pr_warning("%s: Warning: no link monitoring support for %s\n",
1565 bond_dev
->name
, slave_dev
->name
);
1568 /* already in-use? */
1569 if (netdev_is_rx_handler_busy(slave_dev
)) {
1570 netdev_err(bond_dev
,
1571 "Error: Device is in use and cannot be enslaved\n");
1575 /* vlan challenged mutual exclusion */
1576 /* no need to lock since we're protected by rtnl_lock */
1577 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1578 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1579 if (vlan_uses_dev(bond_dev
)) {
1580 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1581 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1584 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1585 bond_dev
->name
, slave_dev
->name
,
1586 slave_dev
->name
, bond_dev
->name
);
1589 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1593 * Old ifenslave binaries are no longer supported. These can
1594 * be identified with moderate accuracy by the state of the slave:
1595 * the current ifenslave will set the interface down prior to
1596 * enslaving it; the old ifenslave will not.
1598 if ((slave_dev
->flags
& IFF_UP
)) {
1599 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1602 goto err_undo_flags
;
1605 /* set bonding device ether type by slave - bonding netdevices are
1606 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1607 * there is a need to override some of the type dependent attribs/funcs.
1609 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1610 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1612 if (bond
->slave_cnt
== 0) {
1613 if (bond_dev
->type
!= slave_dev
->type
) {
1614 pr_debug("%s: change device type from %d to %d\n",
1616 bond_dev
->type
, slave_dev
->type
);
1618 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1620 res
= notifier_to_errno(res
);
1622 pr_err("%s: refused to change device type\n",
1625 goto err_undo_flags
;
1628 /* Flush unicast and multicast addresses */
1629 dev_uc_flush(bond_dev
);
1630 dev_mc_flush(bond_dev
);
1632 if (slave_dev
->type
!= ARPHRD_ETHER
)
1633 bond_setup_by_slave(bond_dev
, slave_dev
);
1635 ether_setup(bond_dev
);
1636 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1639 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1642 } else if (bond_dev
->type
!= slave_dev
->type
) {
1643 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1645 slave_dev
->type
, bond_dev
->type
);
1647 goto err_undo_flags
;
1650 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1651 if (bond
->slave_cnt
== 0) {
1652 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1654 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1655 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1656 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",
1659 goto err_undo_flags
;
1663 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1665 /* If this is the first slave, then we need to set the master's hardware
1666 * address to be the same as the slave's. */
1667 if (bond
->slave_cnt
== 0 && bond
->dev_addr_from_first
)
1668 bond_set_dev_addr(bond
->dev
, slave_dev
);
1670 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1673 goto err_undo_flags
;
1677 * Set the new_slave's queue_id to be zero. Queue ID mapping
1678 * is set via sysfs or module option if desired.
1680 new_slave
->queue_id
= 0;
1682 /* Save slave's original mtu and then set it to match the bond */
1683 new_slave
->original_mtu
= slave_dev
->mtu
;
1684 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1686 pr_debug("Error %d calling dev_set_mtu\n", res
);
1691 * Save slave's original ("permanent") mac address for modes
1692 * that need it, and for restoring it upon release, and then
1693 * set it to the master's address
1695 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1697 if (!bond
->params
.fail_over_mac
) {
1699 * Set slave to master's mac address. The application already
1700 * set the master's mac address to that of the first slave
1702 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1703 addr
.sa_family
= slave_dev
->type
;
1704 res
= dev_set_mac_address(slave_dev
, &addr
);
1706 pr_debug("Error %d calling set_mac_address\n", res
);
1707 goto err_restore_mtu
;
1711 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
);
1713 pr_debug("Error %d calling bond_master_upper_dev_link\n", res
);
1714 goto err_restore_mac
;
1717 /* open the slave since the application closed it */
1718 res
= dev_open(slave_dev
);
1720 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1721 goto err_unset_master
;
1724 new_slave
->bond
= bond
;
1725 new_slave
->dev
= slave_dev
;
1726 slave_dev
->priv_flags
|= IFF_BONDING
;
1728 if (bond_is_lb(bond
)) {
1729 /* bond_alb_init_slave() must be called before all other stages since
1730 * it might fail and we do not want to have to undo everything
1732 res
= bond_alb_init_slave(bond
, new_slave
);
1737 /* If the mode USES_PRIMARY, then the new slave gets the
1738 * master's promisc (and mc) settings only if it becomes the
1739 * curr_active_slave, and that is taken care of later when calling
1740 * bond_change_active()
1742 if (!USES_PRIMARY(bond
->params
.mode
)) {
1743 /* set promiscuity level to new slave */
1744 if (bond_dev
->flags
& IFF_PROMISC
) {
1745 res
= dev_set_promiscuity(slave_dev
, 1);
1750 /* set allmulti level to new slave */
1751 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1752 res
= dev_set_allmulti(slave_dev
, 1);
1757 netif_addr_lock_bh(bond_dev
);
1758 /* upload master's mc_list to new slave */
1759 netdev_for_each_mc_addr(ha
, bond_dev
)
1760 dev_mc_add(slave_dev
, ha
->addr
);
1761 netif_addr_unlock_bh(bond_dev
);
1764 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1765 /* add lacpdu mc addr to mc list */
1766 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1768 dev_mc_add(slave_dev
, lacpdu_multicast
);
1771 bond_add_vlans_on_slave(bond
, slave_dev
);
1773 write_lock_bh(&bond
->lock
);
1775 bond_attach_slave(bond
, new_slave
);
1777 new_slave
->delay
= 0;
1778 new_slave
->link_failure_count
= 0;
1780 write_unlock_bh(&bond
->lock
);
1782 bond_compute_features(bond
);
1784 bond_update_speed_duplex(new_slave
);
1786 read_lock(&bond
->lock
);
1788 new_slave
->last_arp_rx
= jiffies
-
1789 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1791 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1792 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1794 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1796 * miimon is set but a bonded network driver
1797 * does not support ETHTOOL/MII and
1798 * arp_interval is not set. Note: if
1799 * use_carrier is enabled, we will never go
1800 * here (because netif_carrier is always
1801 * supported); thus, we don't need to change
1802 * the messages for netif_carrier.
1804 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",
1805 bond_dev
->name
, slave_dev
->name
);
1806 } else if (link_reporting
== -1) {
1807 /* unable get link status using mii/ethtool */
1808 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",
1809 bond_dev
->name
, slave_dev
->name
);
1813 /* check for initial state */
1814 if (bond
->params
.miimon
) {
1815 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1816 if (bond
->params
.updelay
) {
1817 new_slave
->link
= BOND_LINK_BACK
;
1818 new_slave
->delay
= bond
->params
.updelay
;
1820 new_slave
->link
= BOND_LINK_UP
;
1823 new_slave
->link
= BOND_LINK_DOWN
;
1825 } else if (bond
->params
.arp_interval
) {
1826 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1827 BOND_LINK_UP
: BOND_LINK_DOWN
);
1829 new_slave
->link
= BOND_LINK_UP
;
1832 if (new_slave
->link
!= BOND_LINK_DOWN
)
1833 new_slave
->jiffies
= jiffies
;
1834 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1835 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1836 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1838 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1839 /* if there is a primary slave, remember it */
1840 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1841 bond
->primary_slave
= new_slave
;
1842 bond
->force_primary
= true;
1846 write_lock_bh(&bond
->curr_slave_lock
);
1848 switch (bond
->params
.mode
) {
1849 case BOND_MODE_ACTIVEBACKUP
:
1850 bond_set_slave_inactive_flags(new_slave
);
1851 bond_select_active_slave(bond
);
1853 case BOND_MODE_8023AD
:
1854 /* in 802.3ad mode, the internal mechanism
1855 * will activate the slaves in the selected
1858 bond_set_slave_inactive_flags(new_slave
);
1859 /* if this is the first slave */
1860 if (bond
->slave_cnt
== 1) {
1861 SLAVE_AD_INFO(new_slave
).id
= 1;
1862 /* Initialize AD with the number of times that the AD timer is called in 1 second
1863 * can be called only after the mac address of the bond is set
1865 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1867 SLAVE_AD_INFO(new_slave
).id
=
1868 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1871 bond_3ad_bind_slave(new_slave
);
1875 bond_set_active_slave(new_slave
);
1876 bond_set_slave_inactive_flags(new_slave
);
1877 bond_select_active_slave(bond
);
1880 pr_debug("This slave is always active in trunk mode\n");
1882 /* always active in trunk mode */
1883 bond_set_active_slave(new_slave
);
1885 /* In trunking mode there is little meaning to curr_active_slave
1886 * anyway (it holds no special properties of the bond device),
1887 * so we can change it without calling change_active_interface()
1889 if (!bond
->curr_active_slave
&& new_slave
->link
== BOND_LINK_UP
)
1890 bond
->curr_active_slave
= new_slave
;
1893 } /* switch(bond_mode) */
1895 write_unlock_bh(&bond
->curr_slave_lock
);
1897 bond_set_carrier(bond
);
1899 #ifdef CONFIG_NET_POLL_CONTROLLER
1900 slave_dev
->npinfo
= bond_netpoll_info(bond
);
1901 if (slave_dev
->npinfo
) {
1902 if (slave_enable_netpoll(new_slave
)) {
1903 read_unlock(&bond
->lock
);
1904 pr_info("Error, %s: master_dev is using netpoll, "
1905 "but new slave device does not support netpoll.\n",
1913 read_unlock(&bond
->lock
);
1915 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1919 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1922 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1923 goto err_dest_symlinks
;
1926 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1927 bond_dev
->name
, slave_dev
->name
,
1928 bond_is_active_slave(new_slave
) ? "n active" : " backup",
1929 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1931 /* enslave is successful */
1934 /* Undo stages on error */
1936 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
1939 if (!USES_PRIMARY(bond
->params
.mode
)) {
1940 netif_addr_lock_bh(bond_dev
);
1941 bond_mc_list_flush(bond_dev
, slave_dev
);
1942 netif_addr_unlock_bh(bond_dev
);
1944 bond_del_vlans_from_slave(bond
, slave_dev
);
1945 write_lock_bh(&bond
->lock
);
1946 bond_detach_slave(bond
, new_slave
);
1947 if (bond
->primary_slave
== new_slave
)
1948 bond
->primary_slave
= NULL
;
1949 if (bond
->curr_active_slave
== new_slave
) {
1950 bond_change_active_slave(bond
, NULL
);
1951 write_unlock_bh(&bond
->lock
);
1952 read_lock(&bond
->lock
);
1953 write_lock_bh(&bond
->curr_slave_lock
);
1954 bond_select_active_slave(bond
);
1955 write_unlock_bh(&bond
->curr_slave_lock
);
1956 read_unlock(&bond
->lock
);
1958 write_unlock_bh(&bond
->lock
);
1960 slave_disable_netpoll(new_slave
);
1963 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1964 dev_close(slave_dev
);
1967 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1970 if (!bond
->params
.fail_over_mac
) {
1971 /* XXX TODO - fom follow mode needs to change master's
1972 * MAC if this slave's MAC is in use by the bond, or at
1973 * least print a warning.
1975 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1976 addr
.sa_family
= slave_dev
->type
;
1977 dev_set_mac_address(slave_dev
, &addr
);
1981 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1987 bond_compute_features(bond
);
1988 /* Enslave of first slave has failed and we need to fix master's mac */
1989 if (bond
->slave_cnt
== 0 &&
1990 ether_addr_equal(bond_dev
->dev_addr
, slave_dev
->dev_addr
))
1991 eth_hw_addr_random(bond_dev
);
1997 * Try to release the slave device <slave> from the bond device <master>
1998 * It is legal to access curr_active_slave without a lock because all the function
1999 * is write-locked. If "all" is true it means that the function is being called
2000 * while destroying a bond interface and all slaves are being released.
2002 * The rules for slave state should be:
2003 * for Active/Backup:
2004 * Active stays on all backups go down
2005 * for Bonded connections:
2006 * The first up interface should be left on and all others downed.
2008 static int __bond_release_one(struct net_device
*bond_dev
,
2009 struct net_device
*slave_dev
,
2012 struct bonding
*bond
= netdev_priv(bond_dev
);
2013 struct slave
*slave
, *oldcurrent
;
2014 struct sockaddr addr
;
2015 int old_flags
= bond_dev
->flags
;
2016 netdev_features_t old_features
= bond_dev
->features
;
2018 /* slave is not a slave or master is not master of this slave */
2019 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
2020 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
2021 pr_err("%s: Error: cannot release %s.\n",
2022 bond_dev
->name
, slave_dev
->name
);
2027 write_lock_bh(&bond
->lock
);
2029 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
2031 /* not a slave of this bond */
2032 pr_info("%s: %s not enslaved\n",
2033 bond_dev
->name
, slave_dev
->name
);
2034 write_unlock_bh(&bond
->lock
);
2035 unblock_netpoll_tx();
2039 write_unlock_bh(&bond
->lock
);
2040 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2041 * for this slave anymore.
2043 netdev_rx_handler_unregister(slave_dev
);
2044 write_lock_bh(&bond
->lock
);
2046 if (!all
&& !bond
->params
.fail_over_mac
) {
2047 if (ether_addr_equal(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
2048 bond
->slave_cnt
> 1)
2049 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",
2050 bond_dev
->name
, slave_dev
->name
,
2052 bond_dev
->name
, slave_dev
->name
);
2055 /* Inform AD package of unbinding of slave. */
2056 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2057 /* must be called before the slave is
2058 * detached from the list
2060 bond_3ad_unbind_slave(slave
);
2063 pr_info("%s: releasing %s interface %s\n",
2065 bond_is_active_slave(slave
) ? "active" : "backup",
2068 oldcurrent
= bond
->curr_active_slave
;
2070 bond
->current_arp_slave
= NULL
;
2072 /* release the slave from its bond */
2073 bond_detach_slave(bond
, slave
);
2075 if (bond
->primary_slave
== slave
)
2076 bond
->primary_slave
= NULL
;
2078 if (oldcurrent
== slave
)
2079 bond_change_active_slave(bond
, NULL
);
2081 if (bond_is_lb(bond
)) {
2082 /* Must be called only after the slave has been
2083 * detached from the list and the curr_active_slave
2084 * has been cleared (if our_slave == old_current),
2085 * but before a new active slave is selected.
2087 write_unlock_bh(&bond
->lock
);
2088 bond_alb_deinit_slave(bond
, slave
);
2089 write_lock_bh(&bond
->lock
);
2093 bond
->curr_active_slave
= NULL
;
2094 } else if (oldcurrent
== slave
) {
2096 * Note that we hold RTNL over this sequence, so there
2097 * is no concern that another slave add/remove event
2100 write_unlock_bh(&bond
->lock
);
2101 read_lock(&bond
->lock
);
2102 write_lock_bh(&bond
->curr_slave_lock
);
2104 bond_select_active_slave(bond
);
2106 write_unlock_bh(&bond
->curr_slave_lock
);
2107 read_unlock(&bond
->lock
);
2108 write_lock_bh(&bond
->lock
);
2111 if (bond
->slave_cnt
== 0) {
2112 bond_set_carrier(bond
);
2113 eth_hw_addr_random(bond_dev
);
2114 bond
->dev_addr_from_first
= true;
2116 if (bond_vlan_used(bond
)) {
2117 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2118 bond_dev
->name
, bond_dev
->name
);
2119 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2124 write_unlock_bh(&bond
->lock
);
2125 unblock_netpoll_tx();
2127 if (bond
->slave_cnt
== 0) {
2128 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
2129 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
2132 bond_compute_features(bond
);
2133 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
2134 (old_features
& NETIF_F_VLAN_CHALLENGED
))
2135 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2136 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
2138 /* must do this from outside any spinlocks */
2139 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2141 bond_del_vlans_from_slave(bond
, slave_dev
);
2143 /* If the mode USES_PRIMARY, then we should only remove its
2144 * promisc and mc settings if it was the curr_active_slave, but that was
2145 * already taken care of above when we detached the slave
2147 if (!USES_PRIMARY(bond
->params
.mode
)) {
2148 /* unset promiscuity level from slave
2149 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2150 * of the IFF_PROMISC flag in the bond_dev, but we need the
2151 * value of that flag before that change, as that was the value
2152 * when this slave was attached, so we cache at the start of the
2153 * function and use it here. Same goes for ALLMULTI below
2155 if (old_flags
& IFF_PROMISC
)
2156 dev_set_promiscuity(slave_dev
, -1);
2158 /* unset allmulti level from slave */
2159 if (old_flags
& IFF_ALLMULTI
)
2160 dev_set_allmulti(slave_dev
, -1);
2162 /* flush master's mc_list from slave */
2163 netif_addr_lock_bh(bond_dev
);
2164 bond_mc_list_flush(bond_dev
, slave_dev
);
2165 netif_addr_unlock_bh(bond_dev
);
2168 bond_upper_dev_unlink(bond_dev
, slave_dev
);
2170 slave_disable_netpoll(slave
);
2172 /* close slave before restoring its mac address */
2173 dev_close(slave_dev
);
2175 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2176 /* restore original ("permanent") mac address */
2177 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2178 addr
.sa_family
= slave_dev
->type
;
2179 dev_set_mac_address(slave_dev
, &addr
);
2182 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2184 slave_dev
->priv_flags
&= ~IFF_BONDING
;
2188 return 0; /* deletion OK */
2191 /* A wrapper used because of ndo_del_link */
2192 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2194 return __bond_release_one(bond_dev
, slave_dev
, false);
2198 * First release a slave and then destroy the bond if no more slaves are left.
2199 * Must be under rtnl_lock when this function is called.
2201 static int bond_release_and_destroy(struct net_device
*bond_dev
,
2202 struct net_device
*slave_dev
)
2204 struct bonding
*bond
= netdev_priv(bond_dev
);
2207 ret
= bond_release(bond_dev
, slave_dev
);
2208 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2209 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
2210 pr_info("%s: destroying bond %s.\n",
2211 bond_dev
->name
, bond_dev
->name
);
2212 bond_remove_proc_entry(bond
);
2213 unregister_netdevice(bond_dev
);
2219 * This function changes the active slave to slave <slave_dev>.
2220 * It returns -EINVAL in the following cases.
2221 * - <slave_dev> is not found in the list.
2222 * - There is not active slave now.
2223 * - <slave_dev> is already active.
2224 * - The link state of <slave_dev> is not BOND_LINK_UP.
2225 * - <slave_dev> is not running.
2226 * In these cases, this function does nothing.
2227 * In the other cases, current_slave pointer is changed and 0 is returned.
2229 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2231 struct bonding
*bond
= netdev_priv(bond_dev
);
2232 struct slave
*old_active
= NULL
;
2233 struct slave
*new_active
= NULL
;
2236 if (!USES_PRIMARY(bond
->params
.mode
))
2239 /* Verify that bond_dev is indeed the master of slave_dev */
2240 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
2241 !netdev_has_upper_dev(slave_dev
, bond_dev
))
2244 read_lock(&bond
->lock
);
2246 read_lock(&bond
->curr_slave_lock
);
2247 old_active
= bond
->curr_active_slave
;
2248 read_unlock(&bond
->curr_slave_lock
);
2250 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2253 * Changing to the current active: do nothing; return success.
2255 if (new_active
&& (new_active
== old_active
)) {
2256 read_unlock(&bond
->lock
);
2262 (new_active
->link
== BOND_LINK_UP
) &&
2263 IS_UP(new_active
->dev
)) {
2265 write_lock_bh(&bond
->curr_slave_lock
);
2266 bond_change_active_slave(bond
, new_active
);
2267 write_unlock_bh(&bond
->curr_slave_lock
);
2268 unblock_netpoll_tx();
2272 read_unlock(&bond
->lock
);
2277 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2279 struct bonding
*bond
= netdev_priv(bond_dev
);
2281 info
->bond_mode
= bond
->params
.mode
;
2282 info
->miimon
= bond
->params
.miimon
;
2284 read_lock(&bond
->lock
);
2285 info
->num_slaves
= bond
->slave_cnt
;
2286 read_unlock(&bond
->lock
);
2291 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2293 struct bonding
*bond
= netdev_priv(bond_dev
);
2294 struct slave
*slave
;
2295 int i
, res
= -ENODEV
;
2297 read_lock(&bond
->lock
);
2299 bond_for_each_slave(bond
, slave
, i
) {
2300 if (i
== (int)info
->slave_id
) {
2302 strcpy(info
->slave_name
, slave
->dev
->name
);
2303 info
->link
= slave
->link
;
2304 info
->state
= bond_slave_state(slave
);
2305 info
->link_failure_count
= slave
->link_failure_count
;
2310 read_unlock(&bond
->lock
);
2315 /*-------------------------------- Monitoring -------------------------------*/
2318 static int bond_miimon_inspect(struct bonding
*bond
)
2320 struct slave
*slave
;
2321 int i
, link_state
, commit
= 0;
2322 bool ignore_updelay
;
2324 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2326 bond_for_each_slave(bond
, slave
, i
) {
2327 slave
->new_link
= BOND_LINK_NOCHANGE
;
2329 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2331 switch (slave
->link
) {
2336 slave
->link
= BOND_LINK_FAIL
;
2337 slave
->delay
= bond
->params
.downdelay
;
2339 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2341 (bond
->params
.mode
==
2342 BOND_MODE_ACTIVEBACKUP
) ?
2343 (bond_is_active_slave(slave
) ?
2344 "active " : "backup ") : "",
2346 bond
->params
.downdelay
* bond
->params
.miimon
);
2349 case BOND_LINK_FAIL
:
2352 * recovered before downdelay expired
2354 slave
->link
= BOND_LINK_UP
;
2355 slave
->jiffies
= jiffies
;
2356 pr_info("%s: link status up again after %d ms for interface %s.\n",
2358 (bond
->params
.downdelay
- slave
->delay
) *
2359 bond
->params
.miimon
,
2364 if (slave
->delay
<= 0) {
2365 slave
->new_link
= BOND_LINK_DOWN
;
2373 case BOND_LINK_DOWN
:
2377 slave
->link
= BOND_LINK_BACK
;
2378 slave
->delay
= bond
->params
.updelay
;
2381 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2382 bond
->dev
->name
, slave
->dev
->name
,
2383 ignore_updelay
? 0 :
2384 bond
->params
.updelay
*
2385 bond
->params
.miimon
);
2388 case BOND_LINK_BACK
:
2390 slave
->link
= BOND_LINK_DOWN
;
2391 pr_info("%s: link status down again after %d ms for interface %s.\n",
2393 (bond
->params
.updelay
- slave
->delay
) *
2394 bond
->params
.miimon
,
2403 if (slave
->delay
<= 0) {
2404 slave
->new_link
= BOND_LINK_UP
;
2406 ignore_updelay
= false;
2418 static void bond_miimon_commit(struct bonding
*bond
)
2420 struct slave
*slave
;
2423 bond_for_each_slave(bond
, slave
, i
) {
2424 switch (slave
->new_link
) {
2425 case BOND_LINK_NOCHANGE
:
2429 slave
->link
= BOND_LINK_UP
;
2430 slave
->jiffies
= jiffies
;
2432 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2433 /* prevent it from being the active one */
2434 bond_set_backup_slave(slave
);
2435 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2436 /* make it immediately active */
2437 bond_set_active_slave(slave
);
2438 } else if (slave
!= bond
->primary_slave
) {
2439 /* prevent it from being the active one */
2440 bond_set_backup_slave(slave
);
2443 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2444 bond
->dev
->name
, slave
->dev
->name
,
2445 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
2446 slave
->duplex
? "full" : "half");
2448 /* notify ad that the link status has changed */
2449 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2450 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2452 if (bond_is_lb(bond
))
2453 bond_alb_handle_link_change(bond
, slave
,
2456 if (!bond
->curr_active_slave
||
2457 (slave
== bond
->primary_slave
))
2462 case BOND_LINK_DOWN
:
2463 if (slave
->link_failure_count
< UINT_MAX
)
2464 slave
->link_failure_count
++;
2466 slave
->link
= BOND_LINK_DOWN
;
2468 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2469 bond
->params
.mode
== BOND_MODE_8023AD
)
2470 bond_set_slave_inactive_flags(slave
);
2472 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2473 bond
->dev
->name
, slave
->dev
->name
);
2475 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2476 bond_3ad_handle_link_change(slave
,
2479 if (bond_is_lb(bond
))
2480 bond_alb_handle_link_change(bond
, slave
,
2483 if (slave
== bond
->curr_active_slave
)
2489 pr_err("%s: invalid new link %d on slave %s\n",
2490 bond
->dev
->name
, slave
->new_link
,
2492 slave
->new_link
= BOND_LINK_NOCHANGE
;
2500 write_lock_bh(&bond
->curr_slave_lock
);
2501 bond_select_active_slave(bond
);
2502 write_unlock_bh(&bond
->curr_slave_lock
);
2503 unblock_netpoll_tx();
2506 bond_set_carrier(bond
);
2512 * Really a wrapper that splits the mii monitor into two phases: an
2513 * inspection, then (if inspection indicates something needs to be done)
2514 * an acquisition of appropriate locks followed by a commit phase to
2515 * implement whatever link state changes are indicated.
2517 void bond_mii_monitor(struct work_struct
*work
)
2519 struct bonding
*bond
= container_of(work
, struct bonding
,
2521 bool should_notify_peers
= false;
2522 unsigned long delay
;
2524 read_lock(&bond
->lock
);
2526 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2528 if (bond
->slave_cnt
== 0)
2531 should_notify_peers
= bond_should_notify_peers(bond
);
2533 if (bond_miimon_inspect(bond
)) {
2534 read_unlock(&bond
->lock
);
2536 /* Race avoidance with bond_close cancel of workqueue */
2537 if (!rtnl_trylock()) {
2538 read_lock(&bond
->lock
);
2540 should_notify_peers
= false;
2544 read_lock(&bond
->lock
);
2546 bond_miimon_commit(bond
);
2548 read_unlock(&bond
->lock
);
2549 rtnl_unlock(); /* might sleep, hold no other locks */
2550 read_lock(&bond
->lock
);
2554 if (bond
->params
.miimon
)
2555 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2557 read_unlock(&bond
->lock
);
2559 if (should_notify_peers
) {
2560 if (!rtnl_trylock()) {
2561 read_lock(&bond
->lock
);
2562 bond
->send_peer_notif
++;
2563 read_unlock(&bond
->lock
);
2566 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2571 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2573 struct vlan_entry
*vlan
;
2574 struct net_device
*vlan_dev
;
2576 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2579 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2581 vlan_dev
= __vlan_find_dev_deep(bond
->dev
, htons(ETH_P_8021Q
),
2584 if (vlan_dev
&& ip
== bond_confirm_addr(vlan_dev
, 0, ip
))
2592 * We go to the (large) trouble of VLAN tagging ARP frames because
2593 * switches in VLAN mode (especially if ports are configured as
2594 * "native" to a VLAN) might not pass non-tagged frames.
2596 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2598 struct sk_buff
*skb
;
2600 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op
,
2601 slave_dev
->name
, &dest_ip
, &src_ip
, vlan_id
);
2603 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2604 NULL
, slave_dev
->dev_addr
, NULL
);
2607 pr_err("ARP packet allocation failed\n");
2611 skb
= vlan_put_tag(skb
, htons(ETH_P_8021Q
), vlan_id
);
2613 pr_err("failed to insert VLAN tag\n");
2621 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2624 __be32
*targets
= bond
->params
.arp_targets
;
2625 struct vlan_entry
*vlan
;
2626 struct net_device
*vlan_dev
= NULL
;
2629 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2633 pr_debug("basa: target %pI4\n", &targets
[i
]);
2634 if (!bond_vlan_used(bond
)) {
2635 pr_debug("basa: empty vlan: arp_send\n");
2636 addr
= bond_confirm_addr(bond
->dev
, targets
[i
], 0);
2637 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2643 * If VLANs are configured, we do a route lookup to
2644 * determine which VLAN interface would be used, so we
2645 * can tag the ARP with the proper VLAN tag.
2647 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2650 if (net_ratelimit()) {
2651 pr_warning("%s: no route to arp_ip_target %pI4\n",
2652 bond
->dev
->name
, &targets
[i
]);
2658 * This target is not on a VLAN
2660 if (rt
->dst
.dev
== bond
->dev
) {
2662 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2663 addr
= bond_confirm_addr(bond
->dev
, targets
[i
], 0);
2664 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2670 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2672 vlan_dev
= __vlan_find_dev_deep(bond
->dev
,
2676 if (vlan_dev
== rt
->dst
.dev
) {
2677 vlan_id
= vlan
->vlan_id
;
2678 pr_debug("basa: vlan match on %s %d\n",
2679 vlan_dev
->name
, vlan_id
);
2684 if (vlan_id
&& vlan_dev
) {
2686 addr
= bond_confirm_addr(vlan_dev
, targets
[i
], 0);
2687 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2692 if (net_ratelimit()) {
2693 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2694 bond
->dev
->name
, &targets
[i
],
2695 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2701 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2704 __be32
*targets
= bond
->params
.arp_targets
;
2706 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2707 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2708 &sip
, &tip
, i
, &targets
[i
],
2709 bond_has_this_ip(bond
, tip
));
2710 if (sip
== targets
[i
]) {
2711 if (bond_has_this_ip(bond
, tip
))
2712 slave
->last_arp_rx
= jiffies
;
2718 static int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2719 struct slave
*slave
)
2721 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2722 unsigned char *arp_ptr
;
2726 if (skb
->protocol
!= __cpu_to_be16(ETH_P_ARP
))
2727 return RX_HANDLER_ANOTHER
;
2729 read_lock(&bond
->lock
);
2730 alen
= arp_hdr_len(bond
->dev
);
2732 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2733 bond
->dev
->name
, skb
->dev
->name
);
2735 if (alen
> skb_headlen(skb
)) {
2736 arp
= kmalloc(alen
, GFP_ATOMIC
);
2739 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2743 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2744 skb
->pkt_type
== PACKET_OTHERHOST
||
2745 skb
->pkt_type
== PACKET_LOOPBACK
||
2746 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2747 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2751 arp_ptr
= (unsigned char *)(arp
+ 1);
2752 arp_ptr
+= bond
->dev
->addr_len
;
2753 memcpy(&sip
, arp_ptr
, 4);
2754 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2755 memcpy(&tip
, arp_ptr
, 4);
2757 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2758 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2759 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2763 * Backup slaves won't see the ARP reply, but do come through
2764 * here for each ARP probe (so we swap the sip/tip to validate
2765 * the probe). In a "redundant switch, common router" type of
2766 * configuration, the ARP probe will (hopefully) travel from
2767 * the active, through one switch, the router, then the other
2768 * switch before reaching the backup.
2770 if (bond_is_active_slave(slave
))
2771 bond_validate_arp(bond
, slave
, sip
, tip
);
2773 bond_validate_arp(bond
, slave
, tip
, sip
);
2776 read_unlock(&bond
->lock
);
2777 if (arp
!= (struct arphdr
*)skb
->data
)
2779 return RX_HANDLER_ANOTHER
;
2783 * this function is called regularly to monitor each slave's link
2784 * ensuring that traffic is being sent and received when arp monitoring
2785 * is used in load-balancing mode. if the adapter has been dormant, then an
2786 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2787 * arp monitoring in active backup mode.
2789 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2791 struct bonding
*bond
= container_of(work
, struct bonding
,
2793 struct slave
*slave
, *oldcurrent
;
2794 int do_failover
= 0;
2795 int delta_in_ticks
, extra_ticks
;
2798 read_lock(&bond
->lock
);
2800 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2801 extra_ticks
= delta_in_ticks
/ 2;
2803 if (bond
->slave_cnt
== 0)
2806 read_lock(&bond
->curr_slave_lock
);
2807 oldcurrent
= bond
->curr_active_slave
;
2808 read_unlock(&bond
->curr_slave_lock
);
2810 /* see if any of the previous devices are up now (i.e. they have
2811 * xmt and rcv traffic). the curr_active_slave does not come into
2812 * the picture unless it is null. also, slave->jiffies is not needed
2813 * here because we send an arp on each slave and give a slave as
2814 * long as it needs to get the tx/rx within the delta.
2815 * TODO: what about up/down delay in arp mode? it wasn't here before
2818 bond_for_each_slave(bond
, slave
, i
) {
2819 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2821 if (slave
->link
!= BOND_LINK_UP
) {
2822 if (time_in_range(jiffies
,
2823 trans_start
- delta_in_ticks
,
2824 trans_start
+ delta_in_ticks
+ extra_ticks
) &&
2825 time_in_range(jiffies
,
2826 slave
->dev
->last_rx
- delta_in_ticks
,
2827 slave
->dev
->last_rx
+ delta_in_ticks
+ extra_ticks
)) {
2829 slave
->link
= BOND_LINK_UP
;
2830 bond_set_active_slave(slave
);
2832 /* primary_slave has no meaning in round-robin
2833 * mode. the window of a slave being up and
2834 * curr_active_slave being null after enslaving
2838 pr_info("%s: link status definitely up for interface %s, ",
2843 pr_info("%s: interface %s is now up\n",
2849 /* slave->link == BOND_LINK_UP */
2851 /* not all switches will respond to an arp request
2852 * when the source ip is 0, so don't take the link down
2853 * if we don't know our ip yet
2855 if (!time_in_range(jiffies
,
2856 trans_start
- delta_in_ticks
,
2857 trans_start
+ 2 * delta_in_ticks
+ extra_ticks
) ||
2858 !time_in_range(jiffies
,
2859 slave
->dev
->last_rx
- delta_in_ticks
,
2860 slave
->dev
->last_rx
+ 2 * delta_in_ticks
+ extra_ticks
)) {
2862 slave
->link
= BOND_LINK_DOWN
;
2863 bond_set_backup_slave(slave
);
2865 if (slave
->link_failure_count
< UINT_MAX
)
2866 slave
->link_failure_count
++;
2868 pr_info("%s: interface %s is now down.\n",
2872 if (slave
== oldcurrent
)
2877 /* note: if switch is in round-robin mode, all links
2878 * must tx arp to ensure all links rx an arp - otherwise
2879 * links may oscillate or not come up at all; if switch is
2880 * in something like xor mode, there is nothing we can
2881 * do - all replies will be rx'ed on same link causing slaves
2882 * to be unstable during low/no traffic periods
2884 if (IS_UP(slave
->dev
))
2885 bond_arp_send_all(bond
, slave
);
2890 write_lock_bh(&bond
->curr_slave_lock
);
2892 bond_select_active_slave(bond
);
2894 write_unlock_bh(&bond
->curr_slave_lock
);
2895 unblock_netpoll_tx();
2899 if (bond
->params
.arp_interval
)
2900 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2902 read_unlock(&bond
->lock
);
2906 * Called to inspect slaves for active-backup mode ARP monitor link state
2907 * changes. Sets new_link in slaves to specify what action should take
2908 * place for the slave. Returns 0 if no changes are found, >0 if changes
2909 * to link states must be committed.
2911 * Called with bond->lock held for read.
2913 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
2915 struct slave
*slave
;
2917 unsigned long trans_start
;
2920 /* All the time comparisons below need some extra time. Otherwise, on
2921 * fast networks the ARP probe/reply may arrive within the same jiffy
2922 * as it was sent. Then, the next time the ARP monitor is run, one
2923 * arp_interval will already have passed in the comparisons.
2925 extra_ticks
= delta_in_ticks
/ 2;
2927 bond_for_each_slave(bond
, slave
, i
) {
2928 slave
->new_link
= BOND_LINK_NOCHANGE
;
2930 if (slave
->link
!= BOND_LINK_UP
) {
2931 if (time_in_range(jiffies
,
2932 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2933 slave_last_rx(bond
, slave
) + delta_in_ticks
+ extra_ticks
)) {
2935 slave
->new_link
= BOND_LINK_UP
;
2943 * Give slaves 2*delta after being enslaved or made
2944 * active. This avoids bouncing, as the last receive
2945 * times need a full ARP monitor cycle to be updated.
2947 if (time_in_range(jiffies
,
2948 slave
->jiffies
- delta_in_ticks
,
2949 slave
->jiffies
+ 2 * delta_in_ticks
+ extra_ticks
))
2953 * Backup slave is down if:
2954 * - No current_arp_slave AND
2955 * - more than 3*delta since last receive AND
2956 * - the bond has an IP address
2958 * Note: a non-null current_arp_slave indicates
2959 * the curr_active_slave went down and we are
2960 * searching for a new one; under this condition
2961 * we only take the curr_active_slave down - this
2962 * gives each slave a chance to tx/rx traffic
2963 * before being taken out
2965 if (!bond_is_active_slave(slave
) &&
2966 !bond
->current_arp_slave
&&
2967 !time_in_range(jiffies
,
2968 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2969 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
+ extra_ticks
)) {
2971 slave
->new_link
= BOND_LINK_DOWN
;
2976 * Active slave is down if:
2977 * - more than 2*delta since transmitting OR
2978 * - (more than 2*delta since receive AND
2979 * the bond has an IP address)
2981 trans_start
= dev_trans_start(slave
->dev
);
2982 if (bond_is_active_slave(slave
) &&
2983 (!time_in_range(jiffies
,
2984 trans_start
- delta_in_ticks
,
2985 trans_start
+ 2 * delta_in_ticks
+ extra_ticks
) ||
2986 !time_in_range(jiffies
,
2987 slave_last_rx(bond
, slave
) - delta_in_ticks
,
2988 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
+ extra_ticks
))) {
2990 slave
->new_link
= BOND_LINK_DOWN
;
2999 * Called to commit link state changes noted by inspection step of
3000 * active-backup mode ARP monitor.
3002 * Called with RTNL and bond->lock for read.
3004 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
3006 struct slave
*slave
;
3008 unsigned long trans_start
;
3010 bond_for_each_slave(bond
, slave
, i
) {
3011 switch (slave
->new_link
) {
3012 case BOND_LINK_NOCHANGE
:
3016 trans_start
= dev_trans_start(slave
->dev
);
3017 if ((!bond
->curr_active_slave
&&
3018 time_in_range(jiffies
,
3019 trans_start
- delta_in_ticks
,
3020 trans_start
+ delta_in_ticks
+ delta_in_ticks
/ 2)) ||
3021 bond
->curr_active_slave
!= slave
) {
3022 slave
->link
= BOND_LINK_UP
;
3023 if (bond
->current_arp_slave
) {
3024 bond_set_slave_inactive_flags(
3025 bond
->current_arp_slave
);
3026 bond
->current_arp_slave
= NULL
;
3029 pr_info("%s: link status definitely up for interface %s.\n",
3030 bond
->dev
->name
, slave
->dev
->name
);
3032 if (!bond
->curr_active_slave
||
3033 (slave
== bond
->primary_slave
))
3040 case BOND_LINK_DOWN
:
3041 if (slave
->link_failure_count
< UINT_MAX
)
3042 slave
->link_failure_count
++;
3044 slave
->link
= BOND_LINK_DOWN
;
3045 bond_set_slave_inactive_flags(slave
);
3047 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3048 bond
->dev
->name
, slave
->dev
->name
);
3050 if (slave
== bond
->curr_active_slave
) {
3051 bond
->current_arp_slave
= NULL
;
3058 pr_err("%s: impossible: new_link %d on slave %s\n",
3059 bond
->dev
->name
, slave
->new_link
,
3067 write_lock_bh(&bond
->curr_slave_lock
);
3068 bond_select_active_slave(bond
);
3069 write_unlock_bh(&bond
->curr_slave_lock
);
3070 unblock_netpoll_tx();
3073 bond_set_carrier(bond
);
3077 * Send ARP probes for active-backup mode ARP monitor.
3079 * Called with bond->lock held for read.
3081 static void bond_ab_arp_probe(struct bonding
*bond
)
3083 struct slave
*slave
;
3086 read_lock(&bond
->curr_slave_lock
);
3088 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3089 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3090 bond
->current_arp_slave
->dev
->name
,
3091 bond
->curr_active_slave
->dev
->name
);
3093 if (bond
->curr_active_slave
) {
3094 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3095 read_unlock(&bond
->curr_slave_lock
);
3099 read_unlock(&bond
->curr_slave_lock
);
3101 /* if we don't have a curr_active_slave, search for the next available
3102 * backup slave from the current_arp_slave and make it the candidate
3103 * for becoming the curr_active_slave
3106 if (!bond
->current_arp_slave
) {
3107 bond
->current_arp_slave
= bond
->first_slave
;
3108 if (!bond
->current_arp_slave
)
3112 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3114 /* search for next candidate */
3115 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3116 if (IS_UP(slave
->dev
)) {
3117 slave
->link
= BOND_LINK_BACK
;
3118 bond_set_slave_active_flags(slave
);
3119 bond_arp_send_all(bond
, slave
);
3120 slave
->jiffies
= jiffies
;
3121 bond
->current_arp_slave
= slave
;
3125 /* if the link state is up at this point, we
3126 * mark it down - this can happen if we have
3127 * simultaneous link failures and
3128 * reselect_active_interface doesn't make this
3129 * one the current slave so it is still marked
3130 * up when it is actually down
3132 if (slave
->link
== BOND_LINK_UP
) {
3133 slave
->link
= BOND_LINK_DOWN
;
3134 if (slave
->link_failure_count
< UINT_MAX
)
3135 slave
->link_failure_count
++;
3137 bond_set_slave_inactive_flags(slave
);
3139 pr_info("%s: backup interface %s is now down.\n",
3140 bond
->dev
->name
, slave
->dev
->name
);
3145 void bond_activebackup_arp_mon(struct work_struct
*work
)
3147 struct bonding
*bond
= container_of(work
, struct bonding
,
3149 bool should_notify_peers
= false;
3152 read_lock(&bond
->lock
);
3154 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3156 if (bond
->slave_cnt
== 0)
3159 should_notify_peers
= bond_should_notify_peers(bond
);
3161 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3162 read_unlock(&bond
->lock
);
3164 /* Race avoidance with bond_close flush of workqueue */
3165 if (!rtnl_trylock()) {
3166 read_lock(&bond
->lock
);
3168 should_notify_peers
= false;
3172 read_lock(&bond
->lock
);
3174 bond_ab_arp_commit(bond
, delta_in_ticks
);
3176 read_unlock(&bond
->lock
);
3178 read_lock(&bond
->lock
);
3181 bond_ab_arp_probe(bond
);
3184 if (bond
->params
.arp_interval
)
3185 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3187 read_unlock(&bond
->lock
);
3189 if (should_notify_peers
) {
3190 if (!rtnl_trylock()) {
3191 read_lock(&bond
->lock
);
3192 bond
->send_peer_notif
++;
3193 read_unlock(&bond
->lock
);
3196 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
3201 /*-------------------------- netdev event handling --------------------------*/
3204 * Change device name
3206 static int bond_event_changename(struct bonding
*bond
)
3208 bond_remove_proc_entry(bond
);
3209 bond_create_proc_entry(bond
);
3211 bond_debug_reregister(bond
);
3216 static int bond_master_netdev_event(unsigned long event
,
3217 struct net_device
*bond_dev
)
3219 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3222 case NETDEV_CHANGENAME
:
3223 return bond_event_changename(event_bond
);
3224 case NETDEV_UNREGISTER
:
3225 bond_remove_proc_entry(event_bond
);
3227 case NETDEV_REGISTER
:
3228 bond_create_proc_entry(event_bond
);
3237 static int bond_slave_netdev_event(unsigned long event
,
3238 struct net_device
*slave_dev
)
3240 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
);
3241 struct bonding
*bond
;
3242 struct net_device
*bond_dev
;
3246 /* A netdev event can be generated while enslaving a device
3247 * before netdev_rx_handler_register is called in which case
3248 * slave will be NULL
3252 bond_dev
= slave
->bond
->dev
;
3256 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
);
3264 old_speed
= slave
->speed
;
3265 old_duplex
= slave
->duplex
;
3267 bond_update_speed_duplex(slave
);
3269 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3270 if (old_speed
!= slave
->speed
)
3271 bond_3ad_adapter_speed_changed(slave
);
3272 if (old_duplex
!= slave
->duplex
)
3273 bond_3ad_adapter_duplex_changed(slave
);
3278 * ... Or is it this?
3281 case NETDEV_CHANGEMTU
:
3283 * TODO: Should slaves be allowed to
3284 * independently alter their MTU? For
3285 * an active-backup bond, slaves need
3286 * not be the same type of device, so
3287 * MTUs may vary. For other modes,
3288 * slaves arguably should have the
3289 * same MTUs. To do this, we'd need to
3290 * take over the slave's change_mtu
3291 * function for the duration of their
3295 case NETDEV_CHANGENAME
:
3297 * TODO: handle changing the primary's name
3300 case NETDEV_FEAT_CHANGE
:
3301 bond_compute_features(bond
);
3311 * bond_netdev_event: handle netdev notifier chain events.
3313 * This function receives events for the netdev chain. The caller (an
3314 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3315 * locks for us to safely manipulate the slave devices (RTNL lock,
3318 static int bond_netdev_event(struct notifier_block
*this,
3319 unsigned long event
, void *ptr
)
3321 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3323 pr_debug("event_dev: %s, event: %lx\n",
3324 event_dev
? event_dev
->name
: "None",
3327 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3330 if (event_dev
->flags
& IFF_MASTER
) {
3331 pr_debug("IFF_MASTER\n");
3332 return bond_master_netdev_event(event
, event_dev
);
3335 if (event_dev
->flags
& IFF_SLAVE
) {
3336 pr_debug("IFF_SLAVE\n");
3337 return bond_slave_netdev_event(event
, event_dev
);
3343 static struct notifier_block bond_netdev_notifier
= {
3344 .notifier_call
= bond_netdev_event
,
3347 /*---------------------------- Hashing Policies -----------------------------*/
3350 * Hash for the output device based upon layer 2 data
3352 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3354 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3356 if (skb_headlen(skb
) >= offsetof(struct ethhdr
, h_proto
))
3357 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3363 * Hash for the output device based upon layer 2 and layer 3 data. If
3364 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3366 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3368 const struct ethhdr
*data
;
3369 const struct iphdr
*iph
;
3370 const struct ipv6hdr
*ipv6h
;
3372 const __be32
*s
, *d
;
3374 if (skb
->protocol
== htons(ETH_P_IP
) &&
3375 pskb_network_may_pull(skb
, sizeof(*iph
))) {
3377 data
= (struct ethhdr
*)skb
->data
;
3378 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3379 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3380 } else if (skb
->protocol
== htons(ETH_P_IPV6
) &&
3381 pskb_network_may_pull(skb
, sizeof(*ipv6h
))) {
3382 ipv6h
= ipv6_hdr(skb
);
3383 data
= (struct ethhdr
*)skb
->data
;
3384 s
= &ipv6h
->saddr
.s6_addr32
[0];
3385 d
= &ipv6h
->daddr
.s6_addr32
[0];
3386 v6hash
= (s
[1] ^ d
[1]) ^ (s
[2] ^ d
[2]) ^ (s
[3] ^ d
[3]);
3387 v6hash
^= (v6hash
>> 24) ^ (v6hash
>> 16) ^ (v6hash
>> 8);
3388 return (v6hash
^ data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3391 return bond_xmit_hash_policy_l2(skb
, count
);
3395 * Hash for the output device based upon layer 3 and layer 4 data. If
3396 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3397 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3399 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3402 const struct iphdr
*iph
;
3403 const struct ipv6hdr
*ipv6h
;
3404 const __be32
*s
, *d
;
3405 const __be16
*l4
= NULL
;
3407 int noff
= skb_network_offset(skb
);
3410 if (skb
->protocol
== htons(ETH_P_IP
) &&
3411 pskb_may_pull(skb
, noff
+ sizeof(*iph
))) {
3413 poff
= proto_ports_offset(iph
->protocol
);
3415 if (!ip_is_fragment(iph
) && poff
>= 0) {
3416 l4
= skb_header_pointer(skb
, noff
+ (iph
->ihl
<< 2) + poff
,
3419 layer4_xor
= ntohs(l4
[0] ^ l4
[1]);
3421 return (layer4_xor
^
3422 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3423 } else if (skb
->protocol
== htons(ETH_P_IPV6
) &&
3424 pskb_may_pull(skb
, noff
+ sizeof(*ipv6h
))) {
3425 ipv6h
= ipv6_hdr(skb
);
3426 poff
= proto_ports_offset(ipv6h
->nexthdr
);
3428 l4
= skb_header_pointer(skb
, noff
+ sizeof(*ipv6h
) + poff
,
3431 layer4_xor
= ntohs(l4
[0] ^ l4
[1]);
3433 s
= &ipv6h
->saddr
.s6_addr32
[0];
3434 d
= &ipv6h
->daddr
.s6_addr32
[0];
3435 layer4_xor
^= (s
[1] ^ d
[1]) ^ (s
[2] ^ d
[2]) ^ (s
[3] ^ d
[3]);
3436 layer4_xor
^= (layer4_xor
>> 24) ^ (layer4_xor
>> 16) ^
3438 return layer4_xor
% count
;
3441 return bond_xmit_hash_policy_l2(skb
, count
);
3444 /*-------------------------- Device entry points ----------------------------*/
3446 static void bond_work_init_all(struct bonding
*bond
)
3448 INIT_DELAYED_WORK(&bond
->mcast_work
,
3449 bond_resend_igmp_join_requests_delayed
);
3450 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3451 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3452 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3453 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3455 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3456 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3459 static void bond_work_cancel_all(struct bonding
*bond
)
3461 cancel_delayed_work_sync(&bond
->mii_work
);
3462 cancel_delayed_work_sync(&bond
->arp_work
);
3463 cancel_delayed_work_sync(&bond
->alb_work
);
3464 cancel_delayed_work_sync(&bond
->ad_work
);
3465 cancel_delayed_work_sync(&bond
->mcast_work
);
3468 static int bond_open(struct net_device
*bond_dev
)
3470 struct bonding
*bond
= netdev_priv(bond_dev
);
3471 struct slave
*slave
;
3474 /* reset slave->backup and slave->inactive */
3475 read_lock(&bond
->lock
);
3476 if (bond
->slave_cnt
> 0) {
3477 read_lock(&bond
->curr_slave_lock
);
3478 bond_for_each_slave(bond
, slave
, i
) {
3479 if ((bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3480 && (slave
!= bond
->curr_active_slave
)) {
3481 bond_set_slave_inactive_flags(slave
);
3483 bond_set_slave_active_flags(slave
);
3486 read_unlock(&bond
->curr_slave_lock
);
3488 read_unlock(&bond
->lock
);
3490 bond_work_init_all(bond
);
3492 if (bond_is_lb(bond
)) {
3493 /* bond_alb_initialize must be called before the timer
3496 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
)))
3498 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3501 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3502 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3504 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3505 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3506 if (bond
->params
.arp_validate
)
3507 bond
->recv_probe
= bond_arp_rcv
;
3510 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3511 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3512 /* register to receive LACPDUs */
3513 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3514 bond_3ad_initiate_agg_selection(bond
, 1);
3520 static int bond_close(struct net_device
*bond_dev
)
3522 struct bonding
*bond
= netdev_priv(bond_dev
);
3524 write_lock_bh(&bond
->lock
);
3525 bond
->send_peer_notif
= 0;
3526 write_unlock_bh(&bond
->lock
);
3528 bond_work_cancel_all(bond
);
3529 if (bond_is_lb(bond
)) {
3530 /* Must be called only after all
3531 * slaves have been released
3533 bond_alb_deinitialize(bond
);
3535 bond
->recv_probe
= NULL
;
3540 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3541 struct rtnl_link_stats64
*stats
)
3543 struct bonding
*bond
= netdev_priv(bond_dev
);
3544 struct rtnl_link_stats64 temp
;
3545 struct slave
*slave
;
3548 memset(stats
, 0, sizeof(*stats
));
3550 read_lock_bh(&bond
->lock
);
3552 bond_for_each_slave(bond
, slave
, i
) {
3553 const struct rtnl_link_stats64
*sstats
=
3554 dev_get_stats(slave
->dev
, &temp
);
3556 stats
->rx_packets
+= sstats
->rx_packets
;
3557 stats
->rx_bytes
+= sstats
->rx_bytes
;
3558 stats
->rx_errors
+= sstats
->rx_errors
;
3559 stats
->rx_dropped
+= sstats
->rx_dropped
;
3561 stats
->tx_packets
+= sstats
->tx_packets
;
3562 stats
->tx_bytes
+= sstats
->tx_bytes
;
3563 stats
->tx_errors
+= sstats
->tx_errors
;
3564 stats
->tx_dropped
+= sstats
->tx_dropped
;
3566 stats
->multicast
+= sstats
->multicast
;
3567 stats
->collisions
+= sstats
->collisions
;
3569 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3570 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3571 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3572 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3573 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3574 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3576 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3577 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3578 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3579 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3580 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3583 read_unlock_bh(&bond
->lock
);
3588 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3590 struct net_device
*slave_dev
= NULL
;
3591 struct ifbond k_binfo
;
3592 struct ifbond __user
*u_binfo
= NULL
;
3593 struct ifslave k_sinfo
;
3594 struct ifslave __user
*u_sinfo
= NULL
;
3595 struct mii_ioctl_data
*mii
= NULL
;
3599 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3611 * We do this again just in case we were called by SIOCGMIIREG
3612 * instead of SIOCGMIIPHY.
3619 if (mii
->reg_num
== 1) {
3620 struct bonding
*bond
= netdev_priv(bond_dev
);
3622 read_lock(&bond
->lock
);
3623 read_lock(&bond
->curr_slave_lock
);
3624 if (netif_carrier_ok(bond
->dev
))
3625 mii
->val_out
= BMSR_LSTATUS
;
3627 read_unlock(&bond
->curr_slave_lock
);
3628 read_unlock(&bond
->lock
);
3632 case BOND_INFO_QUERY_OLD
:
3633 case SIOCBONDINFOQUERY
:
3634 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3636 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3639 res
= bond_info_query(bond_dev
, &k_binfo
);
3641 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3645 case BOND_SLAVE_INFO_QUERY_OLD
:
3646 case SIOCBONDSLAVEINFOQUERY
:
3647 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3649 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3652 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3654 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3663 net
= dev_net(bond_dev
);
3665 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3668 slave_dev
= dev_get_by_name(net
, ifr
->ifr_slave
);
3670 pr_debug("slave_dev=%p:\n", slave_dev
);
3675 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3677 case BOND_ENSLAVE_OLD
:
3678 case SIOCBONDENSLAVE
:
3679 res
= bond_enslave(bond_dev
, slave_dev
);
3681 case BOND_RELEASE_OLD
:
3682 case SIOCBONDRELEASE
:
3683 res
= bond_release(bond_dev
, slave_dev
);
3685 case BOND_SETHWADDR_OLD
:
3686 case SIOCBONDSETHWADDR
:
3687 bond_set_dev_addr(bond_dev
, slave_dev
);
3690 case BOND_CHANGE_ACTIVE_OLD
:
3691 case SIOCBONDCHANGEACTIVE
:
3692 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3704 static bool bond_addr_in_mc_list(unsigned char *addr
,
3705 struct netdev_hw_addr_list
*list
,
3708 struct netdev_hw_addr
*ha
;
3710 netdev_hw_addr_list_for_each(ha
, list
)
3711 if (!memcmp(ha
->addr
, addr
, addrlen
))
3717 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3719 struct bonding
*bond
= netdev_priv(bond_dev
);
3721 if (change
& IFF_PROMISC
)
3722 bond_set_promiscuity(bond
,
3723 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3725 if (change
& IFF_ALLMULTI
)
3726 bond_set_allmulti(bond
,
3727 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3730 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3732 struct bonding
*bond
= netdev_priv(bond_dev
);
3733 struct netdev_hw_addr
*ha
;
3736 read_lock(&bond
->lock
);
3738 /* looking for addresses to add to slaves' mc list */
3739 netdev_for_each_mc_addr(ha
, bond_dev
) {
3740 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
3741 bond_dev
->addr_len
);
3743 bond_mc_add(bond
, ha
->addr
);
3746 /* looking for addresses to delete from slaves' list */
3747 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
3748 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
3749 bond_dev
->addr_len
);
3751 bond_mc_del(bond
, ha
->addr
);
3754 /* save master's multicast list */
3755 __hw_addr_flush(&bond
->mc_list
);
3756 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
3757 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
3759 read_unlock(&bond
->lock
);
3762 static int bond_neigh_init(struct neighbour
*n
)
3764 struct bonding
*bond
= netdev_priv(n
->dev
);
3765 struct slave
*slave
= bond
->first_slave
;
3766 const struct net_device_ops
*slave_ops
;
3767 struct neigh_parms parms
;
3773 slave_ops
= slave
->dev
->netdev_ops
;
3775 if (!slave_ops
->ndo_neigh_setup
)
3778 parms
.neigh_setup
= NULL
;
3779 parms
.neigh_cleanup
= NULL
;
3780 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3785 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3786 * after the last slave has been detached. Assumes that all slaves
3787 * utilize the same neigh_cleanup (true at this writing as only user
3790 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3792 if (!parms
.neigh_setup
)
3795 return parms
.neigh_setup(n
);
3799 * The bonding ndo_neigh_setup is called at init time beofre any
3800 * slave exists. So we must declare proxy setup function which will
3801 * be used at run time to resolve the actual slave neigh param setup.
3803 * It's also called by master devices (such as vlans) to setup their
3804 * underlying devices. In that case - do nothing, we're already set up from
3807 static int bond_neigh_setup(struct net_device
*dev
,
3808 struct neigh_parms
*parms
)
3810 /* modify only our neigh_parms */
3811 if (parms
->dev
== dev
)
3812 parms
->neigh_setup
= bond_neigh_init
;
3818 * Change the MTU of all of a master's slaves to match the master
3820 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3822 struct bonding
*bond
= netdev_priv(bond_dev
);
3823 struct slave
*slave
, *stop_at
;
3827 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3828 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3830 /* Can't hold bond->lock with bh disabled here since
3831 * some base drivers panic. On the other hand we can't
3832 * hold bond->lock without bh disabled because we'll
3833 * deadlock. The only solution is to rely on the fact
3834 * that we're under rtnl_lock here, and the slaves
3835 * list won't change. This doesn't solve the problem
3836 * of setting the slave's MTU while it is
3837 * transmitting, but the assumption is that the base
3838 * driver can handle that.
3840 * TODO: figure out a way to safely iterate the slaves
3841 * list, but without holding a lock around the actual
3842 * call to the base driver.
3845 bond_for_each_slave(bond
, slave
, i
) {
3846 pr_debug("s %p s->p %p c_m %p\n",
3849 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3851 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3854 /* If we failed to set the slave's mtu to the new value
3855 * we must abort the operation even in ACTIVE_BACKUP
3856 * mode, because if we allow the backup slaves to have
3857 * different mtu values than the active slave we'll
3858 * need to change their mtu when doing a failover. That
3859 * means changing their mtu from timer context, which
3860 * is probably not a good idea.
3862 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3867 bond_dev
->mtu
= new_mtu
;
3872 /* unwind from head to the slave that failed */
3874 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3877 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3879 pr_debug("unwind err %d dev %s\n",
3880 tmp_res
, slave
->dev
->name
);
3890 * Note that many devices must be down to change the HW address, and
3891 * downing the master releases all slaves. We can make bonds full of
3892 * bonding devices to test this, however.
3894 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3896 struct bonding
*bond
= netdev_priv(bond_dev
);
3897 struct sockaddr
*sa
= addr
, tmp_sa
;
3898 struct slave
*slave
, *stop_at
;
3902 if (bond
->params
.mode
== BOND_MODE_ALB
)
3903 return bond_alb_set_mac_address(bond_dev
, addr
);
3906 pr_debug("bond=%p, name=%s\n",
3907 bond
, bond_dev
? bond_dev
->name
: "None");
3910 * If fail_over_mac is set to active, do nothing and return
3911 * success. Returning an error causes ifenslave to fail.
3913 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
3916 if (!is_valid_ether_addr(sa
->sa_data
))
3917 return -EADDRNOTAVAIL
;
3919 /* Can't hold bond->lock with bh disabled here since
3920 * some base drivers panic. On the other hand we can't
3921 * hold bond->lock without bh disabled because we'll
3922 * deadlock. The only solution is to rely on the fact
3923 * that we're under rtnl_lock here, and the slaves
3924 * list won't change. This doesn't solve the problem
3925 * of setting the slave's hw address while it is
3926 * transmitting, but the assumption is that the base
3927 * driver can handle that.
3929 * TODO: figure out a way to safely iterate the slaves
3930 * list, but without holding a lock around the actual
3931 * call to the base driver.
3934 bond_for_each_slave(bond
, slave
, i
) {
3935 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
3936 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3938 if (slave_ops
->ndo_set_mac_address
== NULL
) {
3940 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
3944 res
= dev_set_mac_address(slave
->dev
, addr
);
3946 /* TODO: consider downing the slave
3948 * User should expect communications
3949 * breakage anyway until ARP finish
3952 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3958 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3962 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3963 tmp_sa
.sa_family
= bond_dev
->type
;
3965 /* unwind from head to the slave that failed */
3967 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3970 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
3972 pr_debug("unwind err %d dev %s\n",
3973 tmp_res
, slave
->dev
->name
);
3980 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3982 struct bonding
*bond
= netdev_priv(bond_dev
);
3983 struct slave
*slave
, *start_at
;
3984 int i
, slave_no
, res
= 1;
3985 struct iphdr
*iph
= ip_hdr(skb
);
3988 * Start with the curr_active_slave that joined the bond as the
3989 * default for sending IGMP traffic. For failover purposes one
3990 * needs to maintain some consistency for the interface that will
3991 * send the join/membership reports. The curr_active_slave found
3992 * will send all of this type of traffic.
3994 if ((iph
->protocol
== IPPROTO_IGMP
) &&
3995 (skb
->protocol
== htons(ETH_P_IP
))) {
3997 read_lock(&bond
->curr_slave_lock
);
3998 slave
= bond
->curr_active_slave
;
3999 read_unlock(&bond
->curr_slave_lock
);
4005 * Concurrent TX may collide on rr_tx_counter; we accept
4006 * that as being rare enough not to justify using an
4009 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4011 bond_for_each_slave(bond
, slave
, i
) {
4019 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4020 if (IS_UP(slave
->dev
) &&
4021 (slave
->link
== BOND_LINK_UP
) &&
4022 bond_is_active_slave(slave
)) {
4023 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4030 /* no suitable interface, frame not sent */
4034 return NETDEV_TX_OK
;
4039 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4040 * the bond has a usable interface.
4042 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4044 struct bonding
*bond
= netdev_priv(bond_dev
);
4047 read_lock(&bond
->curr_slave_lock
);
4049 if (bond
->curr_active_slave
)
4050 res
= bond_dev_queue_xmit(bond
, skb
,
4051 bond
->curr_active_slave
->dev
);
4053 read_unlock(&bond
->curr_slave_lock
);
4056 /* no suitable interface, frame not sent */
4059 return NETDEV_TX_OK
;
4063 * In bond_xmit_xor() , we determine the output device by using a pre-
4064 * determined xmit_hash_policy(), If the selected device is not enabled,
4065 * find the next active slave.
4067 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4069 struct bonding
*bond
= netdev_priv(bond_dev
);
4070 struct slave
*slave
, *start_at
;
4075 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4077 bond_for_each_slave(bond
, slave
, i
) {
4085 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4086 if (IS_UP(slave
->dev
) &&
4087 (slave
->link
== BOND_LINK_UP
) &&
4088 bond_is_active_slave(slave
)) {
4089 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4095 /* no suitable interface, frame not sent */
4099 return NETDEV_TX_OK
;
4103 * in broadcast mode, we send everything to all usable interfaces.
4105 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4107 struct bonding
*bond
= netdev_priv(bond_dev
);
4108 struct slave
*slave
, *start_at
;
4109 struct net_device
*tx_dev
= NULL
;
4113 read_lock(&bond
->curr_slave_lock
);
4114 start_at
= bond
->curr_active_slave
;
4115 read_unlock(&bond
->curr_slave_lock
);
4120 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4121 if (IS_UP(slave
->dev
) &&
4122 (slave
->link
== BOND_LINK_UP
) &&
4123 bond_is_active_slave(slave
)) {
4125 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4127 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4132 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4138 tx_dev
= slave
->dev
;
4143 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4147 /* no suitable interface, frame not sent */
4150 /* frame sent to all suitable interfaces */
4151 return NETDEV_TX_OK
;
4154 /*------------------------- Device initialization ---------------------------*/
4156 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4158 switch (bond
->params
.xmit_policy
) {
4159 case BOND_XMIT_POLICY_LAYER23
:
4160 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4162 case BOND_XMIT_POLICY_LAYER34
:
4163 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4165 case BOND_XMIT_POLICY_LAYER2
:
4167 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4173 * Lookup the slave that corresponds to a qid
4175 static inline int bond_slave_override(struct bonding
*bond
,
4176 struct sk_buff
*skb
)
4179 struct slave
*slave
= NULL
;
4180 struct slave
*check_slave
;
4182 if (!skb
->queue_mapping
)
4185 /* Find out if any slaves have the same mapping as this skb. */
4186 bond_for_each_slave(bond
, check_slave
, i
) {
4187 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4188 slave
= check_slave
;
4193 /* If the slave isn't UP, use default transmit policy. */
4194 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4195 (slave
->link
== BOND_LINK_UP
)) {
4196 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4203 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4206 * This helper function exists to help dev_pick_tx get the correct
4207 * destination queue. Using a helper function skips a call to
4208 * skb_tx_hash and will put the skbs in the queue we expect on their
4209 * way down to the bonding driver.
4211 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
4214 * Save the original txq to restore before passing to the driver
4216 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
4218 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
4220 txq
-= dev
->real_num_tx_queues
;
4221 } while (txq
>= dev
->real_num_tx_queues
);
4226 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4228 struct bonding
*bond
= netdev_priv(dev
);
4230 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4231 if (!bond_slave_override(bond
, skb
))
4232 return NETDEV_TX_OK
;
4235 switch (bond
->params
.mode
) {
4236 case BOND_MODE_ROUNDROBIN
:
4237 return bond_xmit_roundrobin(skb
, dev
);
4238 case BOND_MODE_ACTIVEBACKUP
:
4239 return bond_xmit_activebackup(skb
, dev
);
4241 return bond_xmit_xor(skb
, dev
);
4242 case BOND_MODE_BROADCAST
:
4243 return bond_xmit_broadcast(skb
, dev
);
4244 case BOND_MODE_8023AD
:
4245 return bond_3ad_xmit_xor(skb
, dev
);
4248 return bond_alb_xmit(skb
, dev
);
4250 /* Should never happen, mode already checked */
4251 pr_err("%s: Error: Unknown bonding mode %d\n",
4252 dev
->name
, bond
->params
.mode
);
4255 return NETDEV_TX_OK
;
4259 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4261 struct bonding
*bond
= netdev_priv(dev
);
4262 netdev_tx_t ret
= NETDEV_TX_OK
;
4265 * If we risk deadlock from transmitting this in the
4266 * netpoll path, tell netpoll to queue the frame for later tx
4268 if (is_netpoll_tx_blocked(dev
))
4269 return NETDEV_TX_BUSY
;
4271 read_lock(&bond
->lock
);
4273 if (bond
->slave_cnt
)
4274 ret
= __bond_start_xmit(skb
, dev
);
4278 read_unlock(&bond
->lock
);
4284 * set bond mode specific net device operations
4286 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4288 struct net_device
*bond_dev
= bond
->dev
;
4291 case BOND_MODE_ROUNDROBIN
:
4293 case BOND_MODE_ACTIVEBACKUP
:
4296 bond_set_xmit_hash_policy(bond
);
4298 case BOND_MODE_BROADCAST
:
4300 case BOND_MODE_8023AD
:
4301 bond_set_xmit_hash_policy(bond
);
4308 /* Should never happen, mode already checked */
4309 pr_err("%s: Error: Unknown bonding mode %d\n",
4310 bond_dev
->name
, mode
);
4315 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
4316 struct ethtool_cmd
*ecmd
)
4318 struct bonding
*bond
= netdev_priv(bond_dev
);
4319 struct slave
*slave
;
4321 unsigned long speed
= 0;
4323 ecmd
->duplex
= DUPLEX_UNKNOWN
;
4324 ecmd
->port
= PORT_OTHER
;
4326 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4327 * do not need to check mode. Though link speed might not represent
4328 * the true receive or transmit bandwidth (not all modes are symmetric)
4329 * this is an accurate maximum.
4331 read_lock(&bond
->lock
);
4332 bond_for_each_slave(bond
, slave
, i
) {
4333 if (SLAVE_IS_OK(slave
)) {
4334 if (slave
->speed
!= SPEED_UNKNOWN
)
4335 speed
+= slave
->speed
;
4336 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
4337 slave
->duplex
!= DUPLEX_UNKNOWN
)
4338 ecmd
->duplex
= slave
->duplex
;
4341 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
4342 read_unlock(&bond
->lock
);
4346 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4347 struct ethtool_drvinfo
*drvinfo
)
4349 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
4350 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
4351 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
4355 static const struct ethtool_ops bond_ethtool_ops
= {
4356 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4357 .get_settings
= bond_ethtool_get_settings
,
4358 .get_link
= ethtool_op_get_link
,
4361 static const struct net_device_ops bond_netdev_ops
= {
4362 .ndo_init
= bond_init
,
4363 .ndo_uninit
= bond_uninit
,
4364 .ndo_open
= bond_open
,
4365 .ndo_stop
= bond_close
,
4366 .ndo_start_xmit
= bond_start_xmit
,
4367 .ndo_select_queue
= bond_select_queue
,
4368 .ndo_get_stats64
= bond_get_stats
,
4369 .ndo_do_ioctl
= bond_do_ioctl
,
4370 .ndo_change_rx_flags
= bond_change_rx_flags
,
4371 .ndo_set_rx_mode
= bond_set_multicast_list
,
4372 .ndo_change_mtu
= bond_change_mtu
,
4373 .ndo_set_mac_address
= bond_set_mac_address
,
4374 .ndo_neigh_setup
= bond_neigh_setup
,
4375 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4376 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4377 #ifdef CONFIG_NET_POLL_CONTROLLER
4378 .ndo_netpoll_setup
= bond_netpoll_setup
,
4379 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4380 .ndo_poll_controller
= bond_poll_controller
,
4382 .ndo_add_slave
= bond_enslave
,
4383 .ndo_del_slave
= bond_release
,
4384 .ndo_fix_features
= bond_fix_features
,
4387 static const struct device_type bond_type
= {
4391 static void bond_destructor(struct net_device
*bond_dev
)
4393 struct bonding
*bond
= netdev_priv(bond_dev
);
4395 destroy_workqueue(bond
->wq
);
4396 free_netdev(bond_dev
);
4399 static void bond_setup(struct net_device
*bond_dev
)
4401 struct bonding
*bond
= netdev_priv(bond_dev
);
4403 /* initialize rwlocks */
4404 rwlock_init(&bond
->lock
);
4405 rwlock_init(&bond
->curr_slave_lock
);
4407 bond
->params
= bonding_defaults
;
4409 /* Initialize pointers */
4410 bond
->dev
= bond_dev
;
4411 INIT_LIST_HEAD(&bond
->vlan_list
);
4413 /* Initialize the device entry points */
4414 ether_setup(bond_dev
);
4415 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4416 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4417 bond_set_mode_ops(bond
, bond
->params
.mode
);
4419 bond_dev
->destructor
= bond_destructor
;
4421 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
4423 /* Initialize the device options */
4424 bond_dev
->tx_queue_len
= 0;
4425 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4426 bond_dev
->priv_flags
|= IFF_BONDING
;
4427 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
4429 /* At first, we block adding VLANs. That's the only way to
4430 * prevent problems that occur when adding VLANs over an
4431 * empty bond. The block will be removed once non-challenged
4432 * slaves are enslaved.
4434 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4436 /* don't acquire bond device's netif_tx_lock when
4438 bond_dev
->features
|= NETIF_F_LLTX
;
4440 /* By default, we declare the bond to be fully
4441 * VLAN hardware accelerated capable. Special
4442 * care is taken in the various xmit functions
4443 * when there are slaves that are not hw accel
4447 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4448 NETIF_F_HW_VLAN_CTAG_TX
|
4449 NETIF_F_HW_VLAN_CTAG_RX
|
4450 NETIF_F_HW_VLAN_CTAG_FILTER
;
4452 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
4453 bond_dev
->features
|= bond_dev
->hw_features
;
4457 * Destroy a bonding device.
4458 * Must be under rtnl_lock when this function is called.
4460 static void bond_uninit(struct net_device
*bond_dev
)
4462 struct bonding
*bond
= netdev_priv(bond_dev
);
4463 struct vlan_entry
*vlan
, *tmp
;
4465 bond_netpoll_cleanup(bond_dev
);
4467 /* Release the bonded slaves */
4468 while (bond
->first_slave
!= NULL
)
4469 __bond_release_one(bond_dev
, bond
->first_slave
->dev
, true);
4470 pr_info("%s: released all slaves\n", bond_dev
->name
);
4472 list_del(&bond
->bond_list
);
4474 bond_debug_unregister(bond
);
4476 __hw_addr_flush(&bond
->mc_list
);
4478 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4479 list_del(&vlan
->vlan_list
);
4484 /*------------------------- Module initialization ---------------------------*/
4487 * Convert string input module parms. Accept either the
4488 * number of the mode or its string name. A bit complicated because
4489 * some mode names are substrings of other names, and calls from sysfs
4490 * may have whitespace in the name (trailing newlines, for example).
4492 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4494 int modeint
= -1, i
, rv
;
4495 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4497 for (p
= (char *)buf
; *p
; p
++)
4498 if (!(isdigit(*p
) || isspace(*p
)))
4502 rv
= sscanf(buf
, "%20s", modestr
);
4504 rv
= sscanf(buf
, "%d", &modeint
);
4509 for (i
= 0; tbl
[i
].modename
; i
++) {
4510 if (modeint
== tbl
[i
].mode
)
4512 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4519 static int bond_check_params(struct bond_params
*params
)
4521 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
4524 * Convert string parameters.
4527 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4528 if (bond_mode
== -1) {
4529 pr_err("Error: Invalid bonding mode \"%s\"\n",
4530 mode
== NULL
? "NULL" : mode
);
4535 if (xmit_hash_policy
) {
4536 if ((bond_mode
!= BOND_MODE_XOR
) &&
4537 (bond_mode
!= BOND_MODE_8023AD
)) {
4538 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4539 bond_mode_name(bond_mode
));
4541 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4543 if (xmit_hashtype
== -1) {
4544 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4545 xmit_hash_policy
== NULL
? "NULL" :
4553 if (bond_mode
!= BOND_MODE_8023AD
) {
4554 pr_info("lacp_rate param is irrelevant in mode %s\n",
4555 bond_mode_name(bond_mode
));
4557 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4558 if (lacp_fast
== -1) {
4559 pr_err("Error: Invalid lacp rate \"%s\"\n",
4560 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4567 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4568 if (params
->ad_select
== -1) {
4569 pr_err("Error: Invalid ad_select \"%s\"\n",
4570 ad_select
== NULL
? "NULL" : ad_select
);
4574 if (bond_mode
!= BOND_MODE_8023AD
) {
4575 pr_warning("ad_select param only affects 802.3ad mode\n");
4578 params
->ad_select
= BOND_AD_STABLE
;
4581 if (max_bonds
< 0) {
4582 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4583 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4584 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4588 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4589 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4590 miimon
= BOND_LINK_MON_INTERV
;
4594 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4599 if (downdelay
< 0) {
4600 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4601 downdelay
, INT_MAX
);
4605 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4606 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4611 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4612 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4617 /* reset values for 802.3ad */
4618 if (bond_mode
== BOND_MODE_8023AD
) {
4620 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");
4621 pr_warning("Forcing miimon to 100msec\n");
4626 if (tx_queues
< 1 || tx_queues
> 255) {
4627 pr_warning("Warning: tx_queues (%d) should be between "
4628 "1 and 255, resetting to %d\n",
4629 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4630 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4633 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4634 pr_warning("Warning: all_slaves_active module parameter (%d), "
4635 "not of valid value (0/1), so it was set to "
4636 "0\n", all_slaves_active
);
4637 all_slaves_active
= 0;
4640 if (resend_igmp
< 0 || resend_igmp
> 255) {
4641 pr_warning("Warning: resend_igmp (%d) should be between "
4642 "0 and 255, resetting to %d\n",
4643 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4644 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4647 /* reset values for TLB/ALB */
4648 if ((bond_mode
== BOND_MODE_TLB
) ||
4649 (bond_mode
== BOND_MODE_ALB
)) {
4651 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");
4652 pr_warning("Forcing miimon to 100msec\n");
4657 if (bond_mode
== BOND_MODE_ALB
) {
4658 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",
4663 if (updelay
|| downdelay
) {
4664 /* just warn the user the up/down delay will have
4665 * no effect since miimon is zero...
4667 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",
4668 updelay
, downdelay
);
4671 /* don't allow arp monitoring */
4673 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4674 miimon
, arp_interval
);
4678 if ((updelay
% miimon
) != 0) {
4679 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4681 (updelay
/ miimon
) * miimon
);
4686 if ((downdelay
% miimon
) != 0) {
4687 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4689 (downdelay
/ miimon
) * miimon
);
4692 downdelay
/= miimon
;
4695 if (arp_interval
< 0) {
4696 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4697 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4698 arp_interval
= BOND_LINK_ARP_INTERV
;
4701 for (arp_ip_count
= 0, i
= 0;
4702 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4703 /* not complete check, but should be good enough to
4705 __be32 ip
= in_aton(arp_ip_target
[i
]);
4706 if (!isdigit(arp_ip_target
[i
][0]) || ip
== 0 ||
4707 ip
== htonl(INADDR_BROADCAST
)) {
4708 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4712 arp_target
[arp_ip_count
++] = ip
;
4716 if (arp_interval
&& !arp_ip_count
) {
4717 /* don't allow arping if no arp_ip_target given... */
4718 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4724 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4725 pr_err("arp_validate only supported in active-backup mode\n");
4728 if (!arp_interval
) {
4729 pr_err("arp_validate requires arp_interval\n");
4733 arp_validate_value
= bond_parse_parm(arp_validate
,
4735 if (arp_validate_value
== -1) {
4736 pr_err("Error: invalid arp_validate \"%s\"\n",
4737 arp_validate
== NULL
? "NULL" : arp_validate
);
4741 arp_validate_value
= 0;
4744 pr_info("MII link monitoring set to %d ms\n", miimon
);
4745 } else if (arp_interval
) {
4746 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4748 arp_validate_tbl
[arp_validate_value
].modename
,
4751 for (i
= 0; i
< arp_ip_count
; i
++)
4752 pr_info(" %s", arp_ip_target
[i
]);
4756 } else if (max_bonds
) {
4757 /* miimon and arp_interval not set, we need one so things
4758 * work as expected, see bonding.txt for details
4760 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");
4763 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4764 /* currently, using a primary only makes sense
4765 * in active backup, TLB or ALB modes
4767 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4768 primary
, bond_mode_name(bond_mode
));
4772 if (primary
&& primary_reselect
) {
4773 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4775 if (primary_reselect_value
== -1) {
4776 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4778 NULL
? "NULL" : primary_reselect
);
4782 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4785 if (fail_over_mac
) {
4786 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4788 if (fail_over_mac_value
== -1) {
4789 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4790 arp_validate
== NULL
? "NULL" : arp_validate
);
4794 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4795 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4797 fail_over_mac_value
= BOND_FOM_NONE
;
4800 /* fill params struct with the proper values */
4801 params
->mode
= bond_mode
;
4802 params
->xmit_policy
= xmit_hashtype
;
4803 params
->miimon
= miimon
;
4804 params
->num_peer_notif
= num_peer_notif
;
4805 params
->arp_interval
= arp_interval
;
4806 params
->arp_validate
= arp_validate_value
;
4807 params
->updelay
= updelay
;
4808 params
->downdelay
= downdelay
;
4809 params
->use_carrier
= use_carrier
;
4810 params
->lacp_fast
= lacp_fast
;
4811 params
->primary
[0] = 0;
4812 params
->primary_reselect
= primary_reselect_value
;
4813 params
->fail_over_mac
= fail_over_mac_value
;
4814 params
->tx_queues
= tx_queues
;
4815 params
->all_slaves_active
= all_slaves_active
;
4816 params
->resend_igmp
= resend_igmp
;
4817 params
->min_links
= min_links
;
4820 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4821 params
->primary
[IFNAMSIZ
- 1] = 0;
4824 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4829 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4830 static struct lock_class_key bonding_netdev_addr_lock_key
;
4831 static struct lock_class_key bonding_tx_busylock_key
;
4833 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4834 struct netdev_queue
*txq
,
4837 lockdep_set_class(&txq
->_xmit_lock
,
4838 &bonding_netdev_xmit_lock_key
);
4841 static void bond_set_lockdep_class(struct net_device
*dev
)
4843 lockdep_set_class(&dev
->addr_list_lock
,
4844 &bonding_netdev_addr_lock_key
);
4845 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4846 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4850 * Called from registration process
4852 static int bond_init(struct net_device
*bond_dev
)
4854 struct bonding
*bond
= netdev_priv(bond_dev
);
4855 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4856 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4858 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4861 * Initialize locks that may be required during
4862 * en/deslave operations. All of the bond_open work
4863 * (of which this is part) should really be moved to
4864 * a phase prior to dev_open
4866 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4867 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4869 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4873 bond_set_lockdep_class(bond_dev
);
4875 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4877 bond_prepare_sysfs_group(bond
);
4879 bond_debug_register(bond
);
4881 /* Ensure valid dev_addr */
4882 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4883 bond_dev
->addr_assign_type
== NET_ADDR_PERM
) {
4884 eth_hw_addr_random(bond_dev
);
4885 bond
->dev_addr_from_first
= true;
4888 __hw_addr_init(&bond
->mc_list
);
4892 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4894 if (tb
[IFLA_ADDRESS
]) {
4895 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4897 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4898 return -EADDRNOTAVAIL
;
4903 static unsigned int bond_get_num_tx_queues(void)
4908 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
4910 .priv_size
= sizeof(struct bonding
),
4911 .setup
= bond_setup
,
4912 .validate
= bond_validate
,
4913 .get_num_tx_queues
= bond_get_num_tx_queues
,
4914 .get_num_rx_queues
= bond_get_num_tx_queues
, /* Use the same number
4918 /* Create a new bond based on the specified name and bonding parameters.
4919 * If name is NULL, obtain a suitable "bond%d" name for us.
4920 * Caller must NOT hold rtnl_lock; we need to release it here before we
4921 * set up our sysfs entries.
4923 int bond_create(struct net
*net
, const char *name
)
4925 struct net_device
*bond_dev
;
4930 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4931 name
? name
: "bond%d",
4932 bond_setup
, tx_queues
);
4934 pr_err("%s: eek! can't alloc netdev!\n", name
);
4939 dev_net_set(bond_dev
, net
);
4940 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4942 res
= register_netdevice(bond_dev
);
4944 netif_carrier_off(bond_dev
);
4948 bond_destructor(bond_dev
);
4952 static int __net_init
bond_net_init(struct net
*net
)
4954 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4957 INIT_LIST_HEAD(&bn
->dev_list
);
4959 bond_create_proc_dir(bn
);
4960 bond_create_sysfs(bn
);
4965 static void __net_exit
bond_net_exit(struct net
*net
)
4967 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4968 struct bonding
*bond
, *tmp_bond
;
4971 bond_destroy_sysfs(bn
);
4972 bond_destroy_proc_dir(bn
);
4974 /* Kill off any bonds created after unregistering bond rtnl ops */
4976 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4977 unregister_netdevice_queue(bond
->dev
, &list
);
4978 unregister_netdevice_many(&list
);
4982 static struct pernet_operations bond_net_ops
= {
4983 .init
= bond_net_init
,
4984 .exit
= bond_net_exit
,
4986 .size
= sizeof(struct bond_net
),
4989 static int __init
bonding_init(void)
4994 pr_info("%s", bond_version
);
4996 res
= bond_check_params(&bonding_defaults
);
5000 res
= register_pernet_subsys(&bond_net_ops
);
5004 res
= rtnl_link_register(&bond_link_ops
);
5008 bond_create_debugfs();
5010 for (i
= 0; i
< max_bonds
; i
++) {
5011 res
= bond_create(&init_net
, NULL
);
5016 register_netdevice_notifier(&bond_netdev_notifier
);
5020 bond_destroy_debugfs();
5021 rtnl_link_unregister(&bond_link_ops
);
5023 unregister_pernet_subsys(&bond_net_ops
);
5028 static void __exit
bonding_exit(void)
5030 unregister_netdevice_notifier(&bond_netdev_notifier
);
5032 bond_destroy_debugfs();
5034 rtnl_link_unregister(&bond_link_ops
);
5035 unregister_pernet_subsys(&bond_net_ops
);
5037 #ifdef CONFIG_NET_POLL_CONTROLLER
5039 * Make sure we don't have an imbalance on our netpoll blocking
5041 WARN_ON(atomic_read(&netpoll_block_tx
));
5045 module_init(bonding_init
);
5046 module_exit(bonding_exit
);
5047 MODULE_LICENSE("GPL");
5048 MODULE_VERSION(DRV_VERSION
);
5049 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5050 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5051 MODULE_ALIAS_RTNL_LINK("bond");