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>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.h>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
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;
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
;
105 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
106 static char *arp_validate
;
107 static char *arp_all_targets
;
108 static char *fail_over_mac
;
109 static int all_slaves_active
;
110 static struct bond_params bonding_defaults
;
111 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
112 static int packets_per_slave
= 1;
113 static int lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
115 module_param(max_bonds
, int, 0);
116 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
117 module_param(tx_queues
, int, 0);
118 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125 module_param(miimon
, int, 0);
126 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
127 module_param(updelay
, int, 0);
128 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
129 module_param(downdelay
, int, 0);
130 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
132 module_param(use_carrier
, int, 0);
133 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135 module_param(mode
, charp
, 0);
136 MODULE_PARM_DESC(mode
, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140 module_param(primary
, charp
, 0);
141 MODULE_PARM_DESC(primary
, "Primary network device to use");
142 module_param(primary_reselect
, charp
, 0);
143 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
148 "2 for only on active slave "
150 module_param(lacp_rate
, charp
, 0);
151 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153 module_param(ad_select
, charp
, 0);
154 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
157 module_param(min_links
, int, 0);
158 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
160 module_param(xmit_hash_policy
, charp
, 0);
161 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165 module_param(arp_interval
, int, 0);
166 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
167 module_param_array(arp_ip_target
, charp
, NULL
, 0);
168 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
169 module_param(arp_validate
, charp
, 0);
170 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173 module_param(arp_all_targets
, charp
, 0);
174 MODULE_PARM_DESC(arp_all_targets
, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac
, charp
, 0);
176 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179 module_param(all_slaves_active
, int, 0);
180 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183 module_param(resend_igmp
, int, 0);
184 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
186 module_param(packets_per_slave
, int, 0);
187 MODULE_PARM_DESC(packets_per_slave
, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190 module_param(lp_interval
, uint
, 0);
191 MODULE_PARM_DESC(lp_interval
, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
195 /*----------------------------- Global variables ----------------------------*/
197 #ifdef CONFIG_NET_POLL_CONTROLLER
198 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
201 int bond_net_id __read_mostly
;
203 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
204 static int arp_ip_count
;
205 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
206 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
207 static int lacp_fast
;
209 /*-------------------------- Forward declarations ---------------------------*/
211 static int bond_init(struct net_device
*bond_dev
);
212 static void bond_uninit(struct net_device
*bond_dev
);
214 /*---------------------------- General routines -----------------------------*/
216 const char *bond_mode_name(int mode
)
218 static const char *names
[] = {
219 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR
] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
223 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
224 [BOND_MODE_TLB
] = "transmit load balancing",
225 [BOND_MODE_ALB
] = "adaptive load balancing",
228 if (mode
< BOND_MODE_ROUNDROBIN
|| mode
> BOND_MODE_ALB
)
234 /*---------------------------------- VLAN -----------------------------------*/
237 * bond_dev_queue_xmit - Prepare skb for xmit.
239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
243 void bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
244 struct net_device
*slave_dev
)
246 skb
->dev
= slave_dev
;
248 BUILD_BUG_ON(sizeof(skb
->queue_mapping
) !=
249 sizeof(qdisc_skb_cb(skb
)->slave_dev_queue_mapping
));
250 skb
->queue_mapping
= qdisc_skb_cb(skb
)->slave_dev_queue_mapping
;
252 if (unlikely(netpoll_tx_running(bond
->dev
)))
253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
259 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
260 * We don't protect the slave list iteration with a lock because:
261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
280 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
,
281 __be16 proto
, u16 vid
)
283 struct bonding
*bond
= netdev_priv(bond_dev
);
284 struct slave
*slave
, *rollback_slave
;
285 struct list_head
*iter
;
288 bond_for_each_slave(bond
, slave
, iter
) {
289 res
= vlan_vid_add(slave
->dev
, proto
, vid
);
297 /* unwind to the slave that failed */
298 bond_for_each_slave(bond
, rollback_slave
, iter
) {
299 if (rollback_slave
== slave
)
302 vlan_vid_del(rollback_slave
->dev
, proto
, vid
);
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
313 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
,
314 __be16 proto
, u16 vid
)
316 struct bonding
*bond
= netdev_priv(bond_dev
);
317 struct list_head
*iter
;
320 bond_for_each_slave(bond
, slave
, iter
)
321 vlan_vid_del(slave
->dev
, proto
, vid
);
323 if (bond_is_lb(bond
))
324 bond_alb_clear_vlan(bond
, vid
);
329 /*------------------------------- Link status -------------------------------*/
332 * Set the carrier state for the master according to the state of its
333 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
334 * do special 802.3ad magic.
336 * Returns zero if carrier state does not change, nonzero if it does.
338 static int bond_set_carrier(struct bonding
*bond
)
340 struct list_head
*iter
;
343 if (!bond_has_slaves(bond
))
346 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
347 return bond_3ad_set_carrier(bond
);
349 bond_for_each_slave(bond
, slave
, iter
) {
350 if (slave
->link
== BOND_LINK_UP
) {
351 if (!netif_carrier_ok(bond
->dev
)) {
352 netif_carrier_on(bond
->dev
);
360 if (netif_carrier_ok(bond
->dev
)) {
361 netif_carrier_off(bond
->dev
);
368 * Get link speed and duplex from the slave's base driver
369 * using ethtool. If for some reason the call fails or the
370 * values are invalid, set speed and duplex to -1,
373 static void bond_update_speed_duplex(struct slave
*slave
)
375 struct net_device
*slave_dev
= slave
->dev
;
376 struct ethtool_cmd ecmd
;
380 slave
->speed
= SPEED_UNKNOWN
;
381 slave
->duplex
= DUPLEX_UNKNOWN
;
383 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
387 slave_speed
= ethtool_cmd_speed(&ecmd
);
388 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
391 switch (ecmd
.duplex
) {
399 slave
->speed
= slave_speed
;
400 slave
->duplex
= ecmd
.duplex
;
405 const char *bond_slave_link_status(s8 link
)
422 * if <dev> supports MII link status reporting, check its link status.
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
425 * depending upon the setting of the use_carrier parameter.
427 * Return either BMSR_LSTATUS, meaning that the link is up (or we
428 * can't tell and just pretend it is), or 0, meaning that the link is
431 * If reporting is non-zero, instead of faking link up, return -1 if
432 * both ETHTOOL and MII ioctls fail (meaning the device does not
433 * support them). If use_carrier is set, return whatever it says.
434 * It'd be nice if there was a good way to tell if a driver supports
435 * netif_carrier, but there really isn't.
437 static int bond_check_dev_link(struct bonding
*bond
,
438 struct net_device
*slave_dev
, int reporting
)
440 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
441 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
443 struct mii_ioctl_data
*mii
;
445 if (!reporting
&& !netif_running(slave_dev
))
448 if (bond
->params
.use_carrier
)
449 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
451 /* Try to get link status using Ethtool first. */
452 if (slave_dev
->ethtool_ops
->get_link
)
453 return slave_dev
->ethtool_ops
->get_link(slave_dev
) ?
456 /* Ethtool can't be used, fallback to MII ioctls. */
457 ioctl
= slave_ops
->ndo_do_ioctl
;
459 /* TODO: set pointer to correct ioctl on a per team member */
460 /* bases to make this more efficient. that is, once */
461 /* we determine the correct ioctl, we will always */
462 /* call it and not the others for that team */
466 * We cannot assume that SIOCGMIIPHY will also read a
467 * register; not all network drivers (e.g., e100)
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
474 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
475 mii
->reg_num
= MII_BMSR
;
476 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
477 return mii
->val_out
& BMSR_LSTATUS
;
482 * If reporting, report that either there's no dev->do_ioctl,
483 * or both SIOCGMIIREG and get_link failed (meaning that we
484 * cannot report link status). If not reporting, pretend
487 return reporting
? -1 : BMSR_LSTATUS
;
490 /*----------------------------- Multicast list ------------------------------*/
493 * Push the promiscuity flag down to appropriate slaves
495 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
497 struct list_head
*iter
;
500 if (bond_uses_primary(bond
)) {
501 /* write lock already acquired */
502 if (bond
->curr_active_slave
) {
503 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
509 bond_for_each_slave(bond
, slave
, iter
) {
510 err
= dev_set_promiscuity(slave
->dev
, inc
);
519 * Push the allmulti flag down to all slaves
521 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
523 struct list_head
*iter
;
526 if (bond_uses_primary(bond
)) {
527 /* write lock already acquired */
528 if (bond
->curr_active_slave
) {
529 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
535 bond_for_each_slave(bond
, slave
, iter
) {
536 err
= dev_set_allmulti(slave
->dev
, inc
);
545 * Retrieve the list of registered multicast addresses for the bonding
546 * device and retransmit an IGMP JOIN request to the current active
549 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
551 struct bonding
*bond
= container_of(work
, struct bonding
,
554 if (!rtnl_trylock()) {
555 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
558 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, bond
->dev
);
560 if (bond
->igmp_retrans
> 1) {
561 bond
->igmp_retrans
--;
562 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
567 /* Flush bond's hardware addresses from slave
569 static void bond_hw_addr_flush(struct net_device
*bond_dev
,
570 struct net_device
*slave_dev
)
572 struct bonding
*bond
= netdev_priv(bond_dev
);
574 dev_uc_unsync(slave_dev
, bond_dev
);
575 dev_mc_unsync(slave_dev
, bond_dev
);
577 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
578 /* del lacpdu mc addr from mc list */
579 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
581 dev_mc_del(slave_dev
, lacpdu_multicast
);
585 /*--------------------------- Active slave change ---------------------------*/
587 /* Update the hardware address list and promisc/allmulti for the new and
588 * old active slaves (if any). Modes that are not using primary keep all
589 * slaves up date at all times; only the modes that use primary need to call
590 * this function to swap these settings during a failover.
592 static void bond_hw_addr_swap(struct bonding
*bond
, struct slave
*new_active
,
593 struct slave
*old_active
)
598 if (bond
->dev
->flags
& IFF_PROMISC
)
599 dev_set_promiscuity(old_active
->dev
, -1);
601 if (bond
->dev
->flags
& IFF_ALLMULTI
)
602 dev_set_allmulti(old_active
->dev
, -1);
604 bond_hw_addr_flush(bond
->dev
, old_active
->dev
);
608 /* FIXME: Signal errors upstream. */
609 if (bond
->dev
->flags
& IFF_PROMISC
)
610 dev_set_promiscuity(new_active
->dev
, 1);
612 if (bond
->dev
->flags
& IFF_ALLMULTI
)
613 dev_set_allmulti(new_active
->dev
, 1);
615 netif_addr_lock_bh(bond
->dev
);
616 dev_uc_sync(new_active
->dev
, bond
->dev
);
617 dev_mc_sync(new_active
->dev
, bond
->dev
);
618 netif_addr_unlock_bh(bond
->dev
);
623 * bond_set_dev_addr - clone slave's address to bond
624 * @bond_dev: bond net device
625 * @slave_dev: slave net device
627 * Should be called with RTNL held.
629 static void bond_set_dev_addr(struct net_device
*bond_dev
,
630 struct net_device
*slave_dev
)
632 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
633 bond_dev
, slave_dev
, slave_dev
->addr_len
);
634 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
635 bond_dev
->addr_assign_type
= NET_ADDR_STOLEN
;
636 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
640 * bond_do_fail_over_mac
642 * Perform special MAC address swapping for fail_over_mac settings
644 * Called with RTNL, curr_slave_lock for write_bh.
646 static void bond_do_fail_over_mac(struct bonding
*bond
,
647 struct slave
*new_active
,
648 struct slave
*old_active
)
649 __releases(&bond
->curr_slave_lock
)
650 __acquires(&bond
->curr_slave_lock
)
652 u8 tmp_mac
[ETH_ALEN
];
653 struct sockaddr saddr
;
656 switch (bond
->params
.fail_over_mac
) {
657 case BOND_FOM_ACTIVE
:
659 write_unlock_bh(&bond
->curr_slave_lock
);
660 bond_set_dev_addr(bond
->dev
, new_active
->dev
);
661 write_lock_bh(&bond
->curr_slave_lock
);
664 case BOND_FOM_FOLLOW
:
666 * if new_active && old_active, swap them
667 * if just old_active, do nothing (going to no active slave)
668 * if just new_active, set new_active to bond's MAC
673 write_unlock_bh(&bond
->curr_slave_lock
);
676 ether_addr_copy(tmp_mac
, new_active
->dev
->dev_addr
);
677 ether_addr_copy(saddr
.sa_data
,
678 old_active
->dev
->dev_addr
);
679 saddr
.sa_family
= new_active
->dev
->type
;
681 ether_addr_copy(saddr
.sa_data
, bond
->dev
->dev_addr
);
682 saddr
.sa_family
= bond
->dev
->type
;
685 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
687 pr_err("%s: Error %d setting MAC of slave %s\n",
688 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
695 ether_addr_copy(saddr
.sa_data
, tmp_mac
);
696 saddr
.sa_family
= old_active
->dev
->type
;
698 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
700 pr_err("%s: Error %d setting MAC of slave %s\n",
701 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
703 write_lock_bh(&bond
->curr_slave_lock
);
706 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
707 bond
->dev
->name
, bond
->params
.fail_over_mac
);
713 static bool bond_should_change_active(struct bonding
*bond
)
715 struct slave
*prim
= bond
->primary_slave
;
716 struct slave
*curr
= bond
->curr_active_slave
;
718 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
720 if (bond
->force_primary
) {
721 bond
->force_primary
= false;
724 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
725 (prim
->speed
< curr
->speed
||
726 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
728 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
734 * find_best_interface - select the best available slave to be the active one
735 * @bond: our bonding struct
737 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
739 struct slave
*slave
, *bestslave
= NULL
;
740 struct list_head
*iter
;
741 int mintime
= bond
->params
.updelay
;
743 if (bond
->primary_slave
&& bond
->primary_slave
->link
== BOND_LINK_UP
&&
744 bond_should_change_active(bond
))
745 return bond
->primary_slave
;
747 bond_for_each_slave(bond
, slave
, iter
) {
748 if (slave
->link
== BOND_LINK_UP
)
750 if (slave
->link
== BOND_LINK_BACK
&& bond_slave_is_up(slave
) &&
751 slave
->delay
< mintime
) {
752 mintime
= slave
->delay
;
760 static bool bond_should_notify_peers(struct bonding
*bond
)
765 slave
= rcu_dereference(bond
->curr_active_slave
);
768 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
769 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
771 if (!slave
|| !bond
->send_peer_notif
||
772 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
779 * change_active_interface - change the active slave into the specified one
780 * @bond: our bonding struct
781 * @new: the new slave to make the active one
783 * Set the new slave to the bond's settings and unset them on the old
785 * Setting include flags, mc-list, promiscuity, allmulti, etc.
787 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
788 * because it is apparently the best available slave we have, even though its
789 * updelay hasn't timed out yet.
791 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
793 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
795 struct slave
*old_active
= bond
->curr_active_slave
;
797 if (old_active
== new_active
)
801 new_active
->last_link_up
= jiffies
;
803 if (new_active
->link
== BOND_LINK_BACK
) {
804 if (bond_uses_primary(bond
)) {
805 pr_info("%s: making interface %s the new active one %d ms earlier\n",
806 bond
->dev
->name
, new_active
->dev
->name
,
807 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
810 new_active
->delay
= 0;
811 new_active
->link
= BOND_LINK_UP
;
813 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
814 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
816 if (bond_is_lb(bond
))
817 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
819 if (bond_uses_primary(bond
)) {
820 pr_info("%s: making interface %s the new active one\n",
821 bond
->dev
->name
, new_active
->dev
->name
);
826 if (bond_uses_primary(bond
))
827 bond_hw_addr_swap(bond
, new_active
, old_active
);
829 if (bond_is_lb(bond
)) {
830 bond_alb_handle_active_change(bond
, new_active
);
832 bond_set_slave_inactive_flags(old_active
,
833 BOND_SLAVE_NOTIFY_NOW
);
835 bond_set_slave_active_flags(new_active
,
836 BOND_SLAVE_NOTIFY_NOW
);
838 rcu_assign_pointer(bond
->curr_active_slave
, new_active
);
841 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
) {
843 bond_set_slave_inactive_flags(old_active
,
844 BOND_SLAVE_NOTIFY_NOW
);
847 bool should_notify_peers
= false;
849 bond_set_slave_active_flags(new_active
,
850 BOND_SLAVE_NOTIFY_NOW
);
852 if (bond
->params
.fail_over_mac
)
853 bond_do_fail_over_mac(bond
, new_active
,
856 if (netif_running(bond
->dev
)) {
857 bond
->send_peer_notif
=
858 bond
->params
.num_peer_notif
;
859 should_notify_peers
=
860 bond_should_notify_peers(bond
);
863 write_unlock_bh(&bond
->curr_slave_lock
);
865 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
866 if (should_notify_peers
)
867 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
870 write_lock_bh(&bond
->curr_slave_lock
);
874 /* resend IGMP joins since active slave has changed or
875 * all were sent on curr_active_slave.
876 * resend only if bond is brought up with the affected
877 * bonding modes and the retransmission is enabled */
878 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
879 ((bond_uses_primary(bond
) && new_active
) ||
880 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
)) {
881 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
882 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
887 * bond_select_active_slave - select a new active slave, if needed
888 * @bond: our bonding struct
890 * This functions should be called when one of the following occurs:
891 * - The old curr_active_slave has been released or lost its link.
892 * - The primary_slave has got its link back.
893 * - A slave has got its link back and there's no old curr_active_slave.
895 * Caller must hold curr_slave_lock for write_bh.
897 void bond_select_active_slave(struct bonding
*bond
)
899 struct slave
*best_slave
;
902 best_slave
= bond_find_best_slave(bond
);
903 if (best_slave
!= bond
->curr_active_slave
) {
904 bond_change_active_slave(bond
, best_slave
);
905 rv
= bond_set_carrier(bond
);
909 if (netif_carrier_ok(bond
->dev
)) {
910 pr_info("%s: first active interface up!\n",
913 pr_info("%s: now running without any active interface!\n",
919 #ifdef CONFIG_NET_POLL_CONTROLLER
920 static inline int slave_enable_netpoll(struct slave
*slave
)
925 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
930 err
= __netpoll_setup(np
, slave
->dev
);
939 static inline void slave_disable_netpoll(struct slave
*slave
)
941 struct netpoll
*np
= slave
->np
;
947 __netpoll_free_async(np
);
950 static void bond_poll_controller(struct net_device
*bond_dev
)
954 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
956 struct bonding
*bond
= netdev_priv(bond_dev
);
957 struct list_head
*iter
;
960 bond_for_each_slave(bond
, slave
, iter
)
961 if (bond_slave_is_up(slave
))
962 slave_disable_netpoll(slave
);
965 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
967 struct bonding
*bond
= netdev_priv(dev
);
968 struct list_head
*iter
;
972 bond_for_each_slave(bond
, slave
, iter
) {
973 err
= slave_enable_netpoll(slave
);
975 bond_netpoll_cleanup(dev
);
982 static inline int slave_enable_netpoll(struct slave
*slave
)
986 static inline void slave_disable_netpoll(struct slave
*slave
)
989 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
994 /*---------------------------------- IOCTL ----------------------------------*/
996 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
997 netdev_features_t features
)
999 struct bonding
*bond
= netdev_priv(dev
);
1000 struct list_head
*iter
;
1001 netdev_features_t mask
;
1002 struct slave
*slave
;
1004 if (!bond_has_slaves(bond
)) {
1005 /* Disable adding VLANs to empty bond. But why? --mq */
1006 features
|= NETIF_F_VLAN_CHALLENGED
;
1011 features
&= ~NETIF_F_ONE_FOR_ALL
;
1012 features
|= NETIF_F_ALL_FOR_ALL
;
1014 bond_for_each_slave(bond
, slave
, iter
) {
1015 features
= netdev_increment_features(features
,
1016 slave
->dev
->features
,
1019 features
= netdev_add_tso_features(features
, mask
);
1024 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1025 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1026 NETIF_F_HIGHDMA | NETIF_F_LRO)
1028 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1029 NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL)
1031 static void bond_compute_features(struct bonding
*bond
)
1033 unsigned int flags
, dst_release_flag
= IFF_XMIT_DST_RELEASE
;
1034 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1035 netdev_features_t enc_features
= BOND_ENC_FEATURES
;
1036 struct net_device
*bond_dev
= bond
->dev
;
1037 struct list_head
*iter
;
1038 struct slave
*slave
;
1039 unsigned short max_hard_header_len
= ETH_HLEN
;
1040 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1041 u16 gso_max_segs
= GSO_MAX_SEGS
;
1043 if (!bond_has_slaves(bond
))
1045 vlan_features
&= NETIF_F_ALL_FOR_ALL
;
1047 bond_for_each_slave(bond
, slave
, iter
) {
1048 vlan_features
= netdev_increment_features(vlan_features
,
1049 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1051 enc_features
= netdev_increment_features(enc_features
,
1052 slave
->dev
->hw_enc_features
,
1054 dst_release_flag
&= slave
->dev
->priv_flags
;
1055 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1056 max_hard_header_len
= slave
->dev
->hard_header_len
;
1058 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1059 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1063 bond_dev
->vlan_features
= vlan_features
;
1064 bond_dev
->hw_enc_features
= enc_features
;
1065 bond_dev
->hard_header_len
= max_hard_header_len
;
1066 bond_dev
->gso_max_segs
= gso_max_segs
;
1067 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1069 flags
= bond_dev
->priv_flags
& ~IFF_XMIT_DST_RELEASE
;
1070 bond_dev
->priv_flags
= flags
| dst_release_flag
;
1072 netdev_change_features(bond_dev
);
1075 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1076 struct net_device
*slave_dev
)
1078 bond_dev
->header_ops
= slave_dev
->header_ops
;
1080 bond_dev
->type
= slave_dev
->type
;
1081 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1082 bond_dev
->addr_len
= slave_dev
->addr_len
;
1084 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1085 slave_dev
->addr_len
);
1088 /* On bonding slaves other than the currently active slave, suppress
1089 * duplicates except for alb non-mcast/bcast.
1091 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1092 struct slave
*slave
,
1093 struct bonding
*bond
)
1095 if (bond_is_slave_inactive(slave
)) {
1096 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1097 skb
->pkt_type
!= PACKET_BROADCAST
&&
1098 skb
->pkt_type
!= PACKET_MULTICAST
)
1105 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1107 struct sk_buff
*skb
= *pskb
;
1108 struct slave
*slave
;
1109 struct bonding
*bond
;
1110 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1112 int ret
= RX_HANDLER_ANOTHER
;
1114 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1116 return RX_HANDLER_CONSUMED
;
1120 slave
= bond_slave_get_rcu(skb
->dev
);
1123 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1125 ret
= recv_probe(skb
, bond
, slave
);
1126 if (ret
== RX_HANDLER_CONSUMED
) {
1132 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1133 return RX_HANDLER_EXACT
;
1136 skb
->dev
= bond
->dev
;
1138 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1139 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1140 skb
->pkt_type
== PACKET_HOST
) {
1142 if (unlikely(skb_cow_head(skb
,
1143 skb
->data
- skb_mac_header(skb
)))) {
1145 return RX_HANDLER_CONSUMED
;
1147 ether_addr_copy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
);
1153 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1154 struct net_device
*slave_dev
,
1155 struct slave
*slave
)
1159 err
= netdev_master_upper_dev_link_private(slave_dev
, bond_dev
, slave
);
1162 slave_dev
->flags
|= IFF_SLAVE
;
1163 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1167 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1168 struct net_device
*slave_dev
)
1170 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1171 slave_dev
->flags
&= ~IFF_SLAVE
;
1172 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1175 static struct slave
*bond_alloc_slave(struct bonding
*bond
)
1177 struct slave
*slave
= NULL
;
1179 slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1183 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1184 SLAVE_AD_INFO(slave
) = kzalloc(sizeof(struct ad_slave_info
),
1186 if (!SLAVE_AD_INFO(slave
)) {
1194 static void bond_free_slave(struct slave
*slave
)
1196 struct bonding
*bond
= bond_get_bond_by_slave(slave
);
1198 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1199 kfree(SLAVE_AD_INFO(slave
));
1204 /* enslave device <slave> to bond device <master> */
1205 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1207 struct bonding
*bond
= netdev_priv(bond_dev
);
1208 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1209 struct slave
*new_slave
= NULL
, *prev_slave
;
1210 struct sockaddr addr
;
1214 if (!bond
->params
.use_carrier
&&
1215 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1216 slave_ops
->ndo_do_ioctl
== NULL
) {
1217 pr_warn("%s: Warning: no link monitoring support for %s\n",
1218 bond_dev
->name
, slave_dev
->name
);
1221 /* already enslaved */
1222 if (slave_dev
->flags
& IFF_SLAVE
) {
1223 pr_debug("Error: Device was already enslaved\n");
1227 if (bond_dev
== slave_dev
) {
1228 pr_err("%s: cannot enslave bond to itself.\n", bond_dev
->name
);
1232 /* vlan challenged mutual exclusion */
1233 /* no need to lock since we're protected by rtnl_lock */
1234 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1235 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1236 if (vlan_uses_dev(bond_dev
)) {
1237 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1238 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1241 pr_warn("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1242 bond_dev
->name
, slave_dev
->name
,
1243 slave_dev
->name
, bond_dev
->name
);
1246 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1250 * Old ifenslave binaries are no longer supported. These can
1251 * be identified with moderate accuracy by the state of the slave:
1252 * the current ifenslave will set the interface down prior to
1253 * enslaving it; the old ifenslave will not.
1255 if ((slave_dev
->flags
& IFF_UP
)) {
1256 pr_err("%s is up - this may be due to an out of date ifenslave\n",
1259 goto err_undo_flags
;
1262 /* set bonding device ether type by slave - bonding netdevices are
1263 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1264 * there is a need to override some of the type dependent attribs/funcs.
1266 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1267 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1269 if (!bond_has_slaves(bond
)) {
1270 if (bond_dev
->type
!= slave_dev
->type
) {
1271 pr_debug("%s: change device type from %d to %d\n",
1273 bond_dev
->type
, slave_dev
->type
);
1275 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1277 res
= notifier_to_errno(res
);
1279 pr_err("%s: refused to change device type\n",
1282 goto err_undo_flags
;
1285 /* Flush unicast and multicast addresses */
1286 dev_uc_flush(bond_dev
);
1287 dev_mc_flush(bond_dev
);
1289 if (slave_dev
->type
!= ARPHRD_ETHER
)
1290 bond_setup_by_slave(bond_dev
, slave_dev
);
1292 ether_setup(bond_dev
);
1293 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1296 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1299 } else if (bond_dev
->type
!= slave_dev
->type
) {
1300 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1301 slave_dev
->name
, slave_dev
->type
, bond_dev
->type
);
1303 goto err_undo_flags
;
1306 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1307 if (!bond_has_slaves(bond
)) {
1308 pr_warn("%s: Warning: The first slave device specified does not support setting the MAC address\n",
1310 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
) {
1311 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1312 pr_warn("%s: Setting fail_over_mac to active for active-backup mode\n",
1315 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1316 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",
1319 goto err_undo_flags
;
1323 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1325 /* If this is the first slave, then we need to set the master's hardware
1326 * address to be the same as the slave's. */
1327 if (!bond_has_slaves(bond
) &&
1328 bond
->dev
->addr_assign_type
== NET_ADDR_RANDOM
)
1329 bond_set_dev_addr(bond
->dev
, slave_dev
);
1331 new_slave
= bond_alloc_slave(bond
);
1334 goto err_undo_flags
;
1337 new_slave
->bond
= bond
;
1338 new_slave
->dev
= slave_dev
;
1340 * Set the new_slave's queue_id to be zero. Queue ID mapping
1341 * is set via sysfs or module option if desired.
1343 new_slave
->queue_id
= 0;
1345 /* Save slave's original mtu and then set it to match the bond */
1346 new_slave
->original_mtu
= slave_dev
->mtu
;
1347 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1349 pr_debug("Error %d calling dev_set_mtu\n", res
);
1354 * Save slave's original ("permanent") mac address for modes
1355 * that need it, and for restoring it upon release, and then
1356 * set it to the master's address
1358 ether_addr_copy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
);
1360 if (!bond
->params
.fail_over_mac
||
1361 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1363 * Set slave to master's mac address. The application already
1364 * set the master's mac address to that of the first slave
1366 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1367 addr
.sa_family
= slave_dev
->type
;
1368 res
= dev_set_mac_address(slave_dev
, &addr
);
1370 pr_debug("Error %d calling set_mac_address\n", res
);
1371 goto err_restore_mtu
;
1375 /* open the slave since the application closed it */
1376 res
= dev_open(slave_dev
);
1378 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1379 goto err_restore_mac
;
1382 slave_dev
->priv_flags
|= IFF_BONDING
;
1384 if (bond_is_lb(bond
)) {
1385 /* bond_alb_init_slave() must be called before all other stages since
1386 * it might fail and we do not want to have to undo everything
1388 res
= bond_alb_init_slave(bond
, new_slave
);
1393 /* If the mode uses primary, then the following is handled by
1394 * bond_change_active_slave().
1396 if (!bond_uses_primary(bond
)) {
1397 /* set promiscuity level to new slave */
1398 if (bond_dev
->flags
& IFF_PROMISC
) {
1399 res
= dev_set_promiscuity(slave_dev
, 1);
1404 /* set allmulti level to new slave */
1405 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1406 res
= dev_set_allmulti(slave_dev
, 1);
1411 netif_addr_lock_bh(bond_dev
);
1413 dev_mc_sync_multiple(slave_dev
, bond_dev
);
1414 dev_uc_sync_multiple(slave_dev
, bond_dev
);
1416 netif_addr_unlock_bh(bond_dev
);
1419 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1420 /* add lacpdu mc addr to mc list */
1421 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1423 dev_mc_add(slave_dev
, lacpdu_multicast
);
1426 res
= vlan_vids_add_by_dev(slave_dev
, bond_dev
);
1428 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1429 bond_dev
->name
, slave_dev
->name
);
1433 prev_slave
= bond_last_slave(bond
);
1435 new_slave
->delay
= 0;
1436 new_slave
->link_failure_count
= 0;
1438 bond_update_speed_duplex(new_slave
);
1440 new_slave
->last_rx
= jiffies
-
1441 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1442 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
1443 new_slave
->target_last_arp_rx
[i
] = new_slave
->last_rx
;
1445 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1446 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1448 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1450 * miimon is set but a bonded network driver
1451 * does not support ETHTOOL/MII and
1452 * arp_interval is not set. Note: if
1453 * use_carrier is enabled, we will never go
1454 * here (because netif_carrier is always
1455 * supported); thus, we don't need to change
1456 * the messages for netif_carrier.
1458 pr_warn("%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",
1459 bond_dev
->name
, slave_dev
->name
);
1460 } else if (link_reporting
== -1) {
1461 /* unable get link status using mii/ethtool */
1462 pr_warn("%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",
1463 bond_dev
->name
, slave_dev
->name
);
1467 /* check for initial state */
1468 if (bond
->params
.miimon
) {
1469 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1470 if (bond
->params
.updelay
) {
1471 new_slave
->link
= BOND_LINK_BACK
;
1472 new_slave
->delay
= bond
->params
.updelay
;
1474 new_slave
->link
= BOND_LINK_UP
;
1477 new_slave
->link
= BOND_LINK_DOWN
;
1479 } else if (bond
->params
.arp_interval
) {
1480 new_slave
->link
= (netif_carrier_ok(slave_dev
) ?
1481 BOND_LINK_UP
: BOND_LINK_DOWN
);
1483 new_slave
->link
= BOND_LINK_UP
;
1486 if (new_slave
->link
!= BOND_LINK_DOWN
)
1487 new_slave
->last_link_up
= jiffies
;
1488 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1489 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1490 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1492 if (bond_uses_primary(bond
) && bond
->params
.primary
[0]) {
1493 /* if there is a primary slave, remember it */
1494 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1495 bond
->primary_slave
= new_slave
;
1496 bond
->force_primary
= true;
1500 switch (BOND_MODE(bond
)) {
1501 case BOND_MODE_ACTIVEBACKUP
:
1502 bond_set_slave_inactive_flags(new_slave
,
1503 BOND_SLAVE_NOTIFY_NOW
);
1505 case BOND_MODE_8023AD
:
1506 /* in 802.3ad mode, the internal mechanism
1507 * will activate the slaves in the selected
1510 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1511 /* if this is the first slave */
1513 SLAVE_AD_INFO(new_slave
)->id
= 1;
1514 /* Initialize AD with the number of times that the AD timer is called in 1 second
1515 * can be called only after the mac address of the bond is set
1517 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1519 SLAVE_AD_INFO(new_slave
)->id
=
1520 SLAVE_AD_INFO(prev_slave
)->id
+ 1;
1523 bond_3ad_bind_slave(new_slave
);
1527 bond_set_active_slave(new_slave
);
1528 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1531 pr_debug("This slave is always active in trunk mode\n");
1533 /* always active in trunk mode */
1534 bond_set_active_slave(new_slave
);
1536 /* In trunking mode there is little meaning to curr_active_slave
1537 * anyway (it holds no special properties of the bond device),
1538 * so we can change it without calling change_active_interface()
1540 if (!bond
->curr_active_slave
&& new_slave
->link
== BOND_LINK_UP
)
1541 rcu_assign_pointer(bond
->curr_active_slave
, new_slave
);
1544 } /* switch(bond_mode) */
1546 #ifdef CONFIG_NET_POLL_CONTROLLER
1547 slave_dev
->npinfo
= bond
->dev
->npinfo
;
1548 if (slave_dev
->npinfo
) {
1549 if (slave_enable_netpoll(new_slave
)) {
1550 pr_info("Error, %s: master_dev is using netpoll, but new slave device does not support netpoll\n",
1558 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1561 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1565 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
, new_slave
);
1567 pr_debug("Error %d calling bond_master_upper_dev_link\n", res
);
1568 goto err_unregister
;
1571 res
= bond_sysfs_slave_add(new_slave
);
1573 pr_debug("Error %d calling bond_sysfs_slave_add\n", res
);
1574 goto err_upper_unlink
;
1578 bond_compute_features(bond
);
1579 bond_set_carrier(bond
);
1581 if (bond_uses_primary(bond
)) {
1583 write_lock_bh(&bond
->curr_slave_lock
);
1584 bond_select_active_slave(bond
);
1585 write_unlock_bh(&bond
->curr_slave_lock
);
1586 unblock_netpoll_tx();
1589 pr_info("%s: Enslaving %s as %s interface with %s link\n",
1590 bond_dev
->name
, slave_dev
->name
,
1591 bond_is_active_slave(new_slave
) ? "an active" : "a backup",
1592 new_slave
->link
!= BOND_LINK_DOWN
? "an up" : "a down");
1594 /* enslave is successful */
1597 /* Undo stages on error */
1599 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1602 netdev_rx_handler_unregister(slave_dev
);
1605 if (!bond_uses_primary(bond
))
1606 bond_hw_addr_flush(bond_dev
, slave_dev
);
1608 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1609 if (bond
->primary_slave
== new_slave
)
1610 bond
->primary_slave
= NULL
;
1611 if (bond
->curr_active_slave
== new_slave
) {
1613 write_lock_bh(&bond
->curr_slave_lock
);
1614 bond_change_active_slave(bond
, NULL
);
1615 bond_select_active_slave(bond
);
1616 write_unlock_bh(&bond
->curr_slave_lock
);
1617 unblock_netpoll_tx();
1619 slave_disable_netpoll(new_slave
);
1622 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1623 dev_close(slave_dev
);
1626 if (!bond
->params
.fail_over_mac
||
1627 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1628 /* XXX TODO - fom follow mode needs to change master's
1629 * MAC if this slave's MAC is in use by the bond, or at
1630 * least print a warning.
1632 ether_addr_copy(addr
.sa_data
, new_slave
->perm_hwaddr
);
1633 addr
.sa_family
= slave_dev
->type
;
1634 dev_set_mac_address(slave_dev
, &addr
);
1638 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1641 bond_free_slave(new_slave
);
1644 /* Enslave of first slave has failed and we need to fix master's mac */
1645 if (!bond_has_slaves(bond
) &&
1646 ether_addr_equal_64bits(bond_dev
->dev_addr
, slave_dev
->dev_addr
))
1647 eth_hw_addr_random(bond_dev
);
1653 * Try to release the slave device <slave> from the bond device <master>
1654 * It is legal to access curr_active_slave without a lock because all the function
1655 * is write-locked. If "all" is true it means that the function is being called
1656 * while destroying a bond interface and all slaves are being released.
1658 * The rules for slave state should be:
1659 * for Active/Backup:
1660 * Active stays on all backups go down
1661 * for Bonded connections:
1662 * The first up interface should be left on and all others downed.
1664 static int __bond_release_one(struct net_device
*bond_dev
,
1665 struct net_device
*slave_dev
,
1668 struct bonding
*bond
= netdev_priv(bond_dev
);
1669 struct slave
*slave
, *oldcurrent
;
1670 struct sockaddr addr
;
1671 int old_flags
= bond_dev
->flags
;
1672 netdev_features_t old_features
= bond_dev
->features
;
1674 /* slave is not a slave or master is not master of this slave */
1675 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1676 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1677 pr_err("%s: Error: cannot release %s\n",
1678 bond_dev
->name
, slave_dev
->name
);
1684 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1686 /* not a slave of this bond */
1687 pr_info("%s: %s not enslaved\n",
1688 bond_dev
->name
, slave_dev
->name
);
1689 unblock_netpoll_tx();
1693 bond_sysfs_slave_del(slave
);
1695 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1696 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1697 * for this slave anymore.
1699 netdev_rx_handler_unregister(slave_dev
);
1700 write_lock_bh(&bond
->lock
);
1702 /* Inform AD package of unbinding of slave. */
1703 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1704 bond_3ad_unbind_slave(slave
);
1706 write_unlock_bh(&bond
->lock
);
1708 pr_info("%s: Releasing %s interface %s\n",
1710 bond_is_active_slave(slave
) ? "active" : "backup",
1713 oldcurrent
= bond
->curr_active_slave
;
1715 bond
->current_arp_slave
= NULL
;
1717 if (!all
&& (!bond
->params
.fail_over_mac
||
1718 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
)) {
1719 if (ether_addr_equal_64bits(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1720 bond_has_slaves(bond
))
1721 pr_warn("%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",
1722 bond_dev
->name
, slave_dev
->name
,
1724 bond_dev
->name
, slave_dev
->name
);
1727 if (bond
->primary_slave
== slave
)
1728 bond
->primary_slave
= NULL
;
1730 if (oldcurrent
== slave
) {
1731 write_lock_bh(&bond
->curr_slave_lock
);
1732 bond_change_active_slave(bond
, NULL
);
1733 write_unlock_bh(&bond
->curr_slave_lock
);
1736 if (bond_is_lb(bond
)) {
1737 /* Must be called only after the slave has been
1738 * detached from the list and the curr_active_slave
1739 * has been cleared (if our_slave == old_current),
1740 * but before a new active slave is selected.
1742 bond_alb_deinit_slave(bond
, slave
);
1746 RCU_INIT_POINTER(bond
->curr_active_slave
, NULL
);
1747 } else if (oldcurrent
== slave
) {
1749 * Note that we hold RTNL over this sequence, so there
1750 * is no concern that another slave add/remove event
1753 write_lock_bh(&bond
->curr_slave_lock
);
1755 bond_select_active_slave(bond
);
1757 write_unlock_bh(&bond
->curr_slave_lock
);
1760 if (!bond_has_slaves(bond
)) {
1761 bond_set_carrier(bond
);
1762 eth_hw_addr_random(bond_dev
);
1764 if (vlan_uses_dev(bond_dev
)) {
1765 pr_warn("%s: Warning: clearing HW address of %s while it still has VLANs\n",
1766 bond_dev
->name
, bond_dev
->name
);
1767 pr_warn("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs\n",
1772 unblock_netpoll_tx();
1776 if (!bond_has_slaves(bond
)) {
1777 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
1778 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
1781 bond_compute_features(bond
);
1782 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1783 (old_features
& NETIF_F_VLAN_CHALLENGED
))
1784 pr_info("%s: last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1785 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1787 /* must do this from outside any spinlocks */
1788 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1790 /* If the mode uses primary, then this cases was handled above by
1791 * bond_change_active_slave(..., NULL)
1793 if (!bond_uses_primary(bond
)) {
1794 /* unset promiscuity level from slave
1795 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1796 * of the IFF_PROMISC flag in the bond_dev, but we need the
1797 * value of that flag before that change, as that was the value
1798 * when this slave was attached, so we cache at the start of the
1799 * function and use it here. Same goes for ALLMULTI below
1801 if (old_flags
& IFF_PROMISC
)
1802 dev_set_promiscuity(slave_dev
, -1);
1804 /* unset allmulti level from slave */
1805 if (old_flags
& IFF_ALLMULTI
)
1806 dev_set_allmulti(slave_dev
, -1);
1808 bond_hw_addr_flush(bond_dev
, slave_dev
);
1811 slave_disable_netpoll(slave
);
1813 /* close slave before restoring its mac address */
1814 dev_close(slave_dev
);
1816 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
||
1817 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1818 /* restore original ("permanent") mac address */
1819 ether_addr_copy(addr
.sa_data
, slave
->perm_hwaddr
);
1820 addr
.sa_family
= slave_dev
->type
;
1821 dev_set_mac_address(slave_dev
, &addr
);
1824 dev_set_mtu(slave_dev
, slave
->original_mtu
);
1826 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1828 bond_free_slave(slave
);
1830 return 0; /* deletion OK */
1833 /* A wrapper used because of ndo_del_link */
1834 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1836 return __bond_release_one(bond_dev
, slave_dev
, false);
1840 * First release a slave and then destroy the bond if no more slaves are left.
1841 * Must be under rtnl_lock when this function is called.
1843 static int bond_release_and_destroy(struct net_device
*bond_dev
,
1844 struct net_device
*slave_dev
)
1846 struct bonding
*bond
= netdev_priv(bond_dev
);
1849 ret
= bond_release(bond_dev
, slave_dev
);
1850 if (ret
== 0 && !bond_has_slaves(bond
)) {
1851 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1852 pr_info("%s: Destroying bond %s\n",
1853 bond_dev
->name
, bond_dev
->name
);
1854 unregister_netdevice(bond_dev
);
1859 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1861 struct bonding
*bond
= netdev_priv(bond_dev
);
1863 info
->bond_mode
= BOND_MODE(bond
);
1864 info
->miimon
= bond
->params
.miimon
;
1866 info
->num_slaves
= bond
->slave_cnt
;
1871 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
1873 struct bonding
*bond
= netdev_priv(bond_dev
);
1874 struct list_head
*iter
;
1875 int i
= 0, res
= -ENODEV
;
1876 struct slave
*slave
;
1878 bond_for_each_slave(bond
, slave
, iter
) {
1879 if (i
++ == (int)info
->slave_id
) {
1881 strcpy(info
->slave_name
, slave
->dev
->name
);
1882 info
->link
= slave
->link
;
1883 info
->state
= bond_slave_state(slave
);
1884 info
->link_failure_count
= slave
->link_failure_count
;
1892 /*-------------------------------- Monitoring -------------------------------*/
1895 static int bond_miimon_inspect(struct bonding
*bond
)
1897 int link_state
, commit
= 0;
1898 struct list_head
*iter
;
1899 struct slave
*slave
;
1900 bool ignore_updelay
;
1902 ignore_updelay
= !bond
->curr_active_slave
? true : false;
1904 bond_for_each_slave_rcu(bond
, slave
, iter
) {
1905 slave
->new_link
= BOND_LINK_NOCHANGE
;
1907 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
1909 switch (slave
->link
) {
1914 slave
->link
= BOND_LINK_FAIL
;
1915 slave
->delay
= bond
->params
.downdelay
;
1917 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms\n",
1920 BOND_MODE_ACTIVEBACKUP
) ?
1921 (bond_is_active_slave(slave
) ?
1922 "active " : "backup ") : "",
1924 bond
->params
.downdelay
* bond
->params
.miimon
);
1927 case BOND_LINK_FAIL
:
1930 * recovered before downdelay expired
1932 slave
->link
= BOND_LINK_UP
;
1933 slave
->last_link_up
= jiffies
;
1934 pr_info("%s: link status up again after %d ms for interface %s\n",
1936 (bond
->params
.downdelay
- slave
->delay
) *
1937 bond
->params
.miimon
,
1942 if (slave
->delay
<= 0) {
1943 slave
->new_link
= BOND_LINK_DOWN
;
1951 case BOND_LINK_DOWN
:
1955 slave
->link
= BOND_LINK_BACK
;
1956 slave
->delay
= bond
->params
.updelay
;
1959 pr_info("%s: link status up for interface %s, enabling it in %d ms\n",
1960 bond
->dev
->name
, slave
->dev
->name
,
1961 ignore_updelay
? 0 :
1962 bond
->params
.updelay
*
1963 bond
->params
.miimon
);
1966 case BOND_LINK_BACK
:
1968 slave
->link
= BOND_LINK_DOWN
;
1969 pr_info("%s: link status down again after %d ms for interface %s\n",
1971 (bond
->params
.updelay
- slave
->delay
) *
1972 bond
->params
.miimon
,
1981 if (slave
->delay
<= 0) {
1982 slave
->new_link
= BOND_LINK_UP
;
1984 ignore_updelay
= false;
1996 static void bond_miimon_commit(struct bonding
*bond
)
1998 struct list_head
*iter
;
1999 struct slave
*slave
;
2001 bond_for_each_slave(bond
, slave
, iter
) {
2002 switch (slave
->new_link
) {
2003 case BOND_LINK_NOCHANGE
:
2007 slave
->link
= BOND_LINK_UP
;
2008 slave
->last_link_up
= jiffies
;
2010 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
2011 /* prevent it from being the active one */
2012 bond_set_backup_slave(slave
);
2013 } else if (BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
2014 /* make it immediately active */
2015 bond_set_active_slave(slave
);
2016 } else if (slave
!= bond
->primary_slave
) {
2017 /* prevent it from being the active one */
2018 bond_set_backup_slave(slave
);
2021 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex\n",
2022 bond
->dev
->name
, slave
->dev
->name
,
2023 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
2024 slave
->duplex
? "full" : "half");
2026 /* notify ad that the link status has changed */
2027 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
2028 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2030 if (bond_is_lb(bond
))
2031 bond_alb_handle_link_change(bond
, slave
,
2034 if (!bond
->curr_active_slave
||
2035 (slave
== bond
->primary_slave
))
2040 case BOND_LINK_DOWN
:
2041 if (slave
->link_failure_count
< UINT_MAX
)
2042 slave
->link_failure_count
++;
2044 slave
->link
= BOND_LINK_DOWN
;
2046 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
2047 BOND_MODE(bond
) == BOND_MODE_8023AD
)
2048 bond_set_slave_inactive_flags(slave
,
2049 BOND_SLAVE_NOTIFY_NOW
);
2051 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2052 bond
->dev
->name
, slave
->dev
->name
);
2054 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
2055 bond_3ad_handle_link_change(slave
,
2058 if (bond_is_lb(bond
))
2059 bond_alb_handle_link_change(bond
, slave
,
2062 if (slave
== bond
->curr_active_slave
)
2068 pr_err("%s: invalid new link %d on slave %s\n",
2069 bond
->dev
->name
, slave
->new_link
,
2071 slave
->new_link
= BOND_LINK_NOCHANGE
;
2079 write_lock_bh(&bond
->curr_slave_lock
);
2080 bond_select_active_slave(bond
);
2081 write_unlock_bh(&bond
->curr_slave_lock
);
2082 unblock_netpoll_tx();
2085 bond_set_carrier(bond
);
2091 * Really a wrapper that splits the mii monitor into two phases: an
2092 * inspection, then (if inspection indicates something needs to be done)
2093 * an acquisition of appropriate locks followed by a commit phase to
2094 * implement whatever link state changes are indicated.
2096 static void bond_mii_monitor(struct work_struct
*work
)
2098 struct bonding
*bond
= container_of(work
, struct bonding
,
2100 bool should_notify_peers
= false;
2101 unsigned long delay
;
2103 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2105 if (!bond_has_slaves(bond
))
2110 should_notify_peers
= bond_should_notify_peers(bond
);
2112 if (bond_miimon_inspect(bond
)) {
2115 /* Race avoidance with bond_close cancel of workqueue */
2116 if (!rtnl_trylock()) {
2118 should_notify_peers
= false;
2122 bond_miimon_commit(bond
);
2124 rtnl_unlock(); /* might sleep, hold no other locks */
2129 if (bond
->params
.miimon
)
2130 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2132 if (should_notify_peers
) {
2133 if (!rtnl_trylock())
2135 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2140 static bool bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2142 struct net_device
*upper
;
2143 struct list_head
*iter
;
2146 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2150 netdev_for_each_all_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2151 if (ip
== bond_confirm_addr(upper
, 0, ip
)) {
2162 * We go to the (large) trouble of VLAN tagging ARP frames because
2163 * switches in VLAN mode (especially if ports are configured as
2164 * "native" to a VLAN) might not pass non-tagged frames.
2166 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
,
2167 __be32 dest_ip
, __be32 src_ip
,
2168 struct bond_vlan_tag
*tags
)
2170 struct sk_buff
*skb
;
2173 pr_debug("arp %d on slave %s: dst %pI4 src %pI4\n",
2174 arp_op
, slave_dev
->name
, &dest_ip
, &src_ip
);
2176 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2177 NULL
, slave_dev
->dev_addr
, NULL
);
2180 net_err_ratelimited("ARP packet allocation failed\n");
2184 /* Go through all the tags backwards and add them to the packet */
2185 for (i
= BOND_MAX_VLAN_ENCAP
- 1; i
> 0; i
--) {
2186 if (!tags
[i
].vlan_id
)
2189 pr_debug("inner tag: proto %X vid %X\n",
2190 ntohs(tags
[i
].vlan_proto
), tags
[i
].vlan_id
);
2191 skb
= __vlan_put_tag(skb
, tags
[i
].vlan_proto
,
2194 net_err_ratelimited("failed to insert inner VLAN tag\n");
2198 /* Set the outer tag */
2199 if (tags
[0].vlan_id
) {
2200 pr_debug("outer tag: proto %X vid %X\n",
2201 ntohs(tags
[0].vlan_proto
), tags
[0].vlan_id
);
2202 skb
= vlan_put_tag(skb
, tags
[0].vlan_proto
, tags
[0].vlan_id
);
2204 net_err_ratelimited("failed to insert outer VLAN tag\n");
2211 /* Validate the device path between the @start_dev and the @end_dev.
2212 * The path is valid if the @end_dev is reachable through device
2214 * When the path is validated, collect any vlan information in the
2217 bool bond_verify_device_path(struct net_device
*start_dev
,
2218 struct net_device
*end_dev
,
2219 struct bond_vlan_tag
*tags
)
2221 struct net_device
*upper
;
2222 struct list_head
*iter
;
2225 if (start_dev
== end_dev
)
2228 netdev_for_each_upper_dev_rcu(start_dev
, upper
, iter
) {
2229 if (bond_verify_device_path(upper
, end_dev
, tags
)) {
2230 if (is_vlan_dev(upper
)) {
2231 idx
= vlan_get_encap_level(upper
);
2232 if (idx
>= BOND_MAX_VLAN_ENCAP
)
2235 tags
[idx
].vlan_proto
=
2236 vlan_dev_vlan_proto(upper
);
2237 tags
[idx
].vlan_id
= vlan_dev_vlan_id(upper
);
2246 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2249 struct bond_vlan_tag tags
[BOND_MAX_VLAN_ENCAP
];
2250 __be32
*targets
= bond
->params
.arp_targets
, addr
;
2254 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
&& targets
[i
]; i
++) {
2255 pr_debug("basa: target %pI4\n", &targets
[i
]);
2256 memset(tags
, 0, sizeof(tags
));
2258 /* Find out through which dev should the packet go */
2259 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2262 /* there's no route to target - try to send arp
2263 * probe to generate any traffic (arp_validate=0)
2265 if (bond
->params
.arp_validate
)
2266 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2269 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2274 /* bond device itself */
2275 if (rt
->dst
.dev
== bond
->dev
)
2279 ret
= bond_verify_device_path(bond
->dev
, rt
->dst
.dev
, tags
);
2285 /* Not our device - skip */
2286 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2287 bond
->dev
->name
, &targets
[i
],
2288 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2294 addr
= bond_confirm_addr(rt
->dst
.dev
, targets
[i
], 0);
2296 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2301 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2305 if (!sip
|| !bond_has_this_ip(bond
, tip
)) {
2306 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip
, &tip
);
2310 i
= bond_get_targets_ip(bond
->params
.arp_targets
, sip
);
2312 pr_debug("bva: sip %pI4 not found in targets\n", &sip
);
2315 slave
->last_rx
= jiffies
;
2316 slave
->target_last_arp_rx
[i
] = jiffies
;
2319 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2320 struct slave
*slave
)
2322 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2323 struct slave
*curr_active_slave
;
2324 unsigned char *arp_ptr
;
2326 int alen
, is_arp
= skb
->protocol
== __cpu_to_be16(ETH_P_ARP
);
2328 if (!slave_do_arp_validate(bond
, slave
)) {
2329 if ((slave_do_arp_validate_only(bond
) && is_arp
) ||
2330 !slave_do_arp_validate_only(bond
))
2331 slave
->last_rx
= jiffies
;
2332 return RX_HANDLER_ANOTHER
;
2333 } else if (!is_arp
) {
2334 return RX_HANDLER_ANOTHER
;
2337 alen
= arp_hdr_len(bond
->dev
);
2339 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2340 bond
->dev
->name
, skb
->dev
->name
);
2342 if (alen
> skb_headlen(skb
)) {
2343 arp
= kmalloc(alen
, GFP_ATOMIC
);
2346 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2350 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2351 skb
->pkt_type
== PACKET_OTHERHOST
||
2352 skb
->pkt_type
== PACKET_LOOPBACK
||
2353 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2354 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2358 arp_ptr
= (unsigned char *)(arp
+ 1);
2359 arp_ptr
+= bond
->dev
->addr_len
;
2360 memcpy(&sip
, arp_ptr
, 4);
2361 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2362 memcpy(&tip
, arp_ptr
, 4);
2364 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2365 bond
->dev
->name
, slave
->dev
->name
, bond_slave_state(slave
),
2366 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2369 curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2372 * Backup slaves won't see the ARP reply, but do come through
2373 * here for each ARP probe (so we swap the sip/tip to validate
2374 * the probe). In a "redundant switch, common router" type of
2375 * configuration, the ARP probe will (hopefully) travel from
2376 * the active, through one switch, the router, then the other
2377 * switch before reaching the backup.
2379 * We 'trust' the arp requests if there is an active slave and
2380 * it received valid arp reply(s) after it became active. This
2381 * is done to avoid endless looping when we can't reach the
2382 * arp_ip_target and fool ourselves with our own arp requests.
2385 if (bond_is_active_slave(slave
))
2386 bond_validate_arp(bond
, slave
, sip
, tip
);
2387 else if (curr_active_slave
&&
2388 time_after(slave_last_rx(bond
, curr_active_slave
),
2389 curr_active_slave
->last_link_up
))
2390 bond_validate_arp(bond
, slave
, tip
, sip
);
2393 if (arp
!= (struct arphdr
*)skb
->data
)
2395 return RX_HANDLER_ANOTHER
;
2398 /* function to verify if we're in the arp_interval timeslice, returns true if
2399 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2400 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2402 static bool bond_time_in_interval(struct bonding
*bond
, unsigned long last_act
,
2405 int delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2407 return time_in_range(jiffies
,
2408 last_act
- delta_in_ticks
,
2409 last_act
+ mod
* delta_in_ticks
+ delta_in_ticks
/2);
2413 * this function is called regularly to monitor each slave's link
2414 * ensuring that traffic is being sent and received when arp monitoring
2415 * is used in load-balancing mode. if the adapter has been dormant, then an
2416 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2417 * arp monitoring in active backup mode.
2419 static void bond_loadbalance_arp_mon(struct work_struct
*work
)
2421 struct bonding
*bond
= container_of(work
, struct bonding
,
2423 struct slave
*slave
, *oldcurrent
;
2424 struct list_head
*iter
;
2425 int do_failover
= 0, slave_state_changed
= 0;
2427 if (!bond_has_slaves(bond
))
2432 oldcurrent
= ACCESS_ONCE(bond
->curr_active_slave
);
2433 /* see if any of the previous devices are up now (i.e. they have
2434 * xmt and rcv traffic). the curr_active_slave does not come into
2435 * the picture unless it is null. also, slave->last_link_up is not
2436 * needed here because we send an arp on each slave and give a slave
2437 * as long as it needs to get the tx/rx within the delta.
2438 * TODO: what about up/down delay in arp mode? it wasn't here before
2441 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2442 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2444 if (slave
->link
!= BOND_LINK_UP
) {
2445 if (bond_time_in_interval(bond
, trans_start
, 1) &&
2446 bond_time_in_interval(bond
, slave
->last_rx
, 1)) {
2448 slave
->link
= BOND_LINK_UP
;
2449 slave_state_changed
= 1;
2451 /* primary_slave has no meaning in round-robin
2452 * mode. the window of a slave being up and
2453 * curr_active_slave being null after enslaving
2457 pr_info("%s: link status definitely up for interface %s\n",
2462 pr_info("%s: interface %s is now up\n",
2468 /* slave->link == BOND_LINK_UP */
2470 /* not all switches will respond to an arp request
2471 * when the source ip is 0, so don't take the link down
2472 * if we don't know our ip yet
2474 if (!bond_time_in_interval(bond
, trans_start
, 2) ||
2475 !bond_time_in_interval(bond
, slave
->last_rx
, 2)) {
2477 slave
->link
= BOND_LINK_DOWN
;
2478 slave_state_changed
= 1;
2480 if (slave
->link_failure_count
< UINT_MAX
)
2481 slave
->link_failure_count
++;
2483 pr_info("%s: interface %s is now down\n",
2484 bond
->dev
->name
, slave
->dev
->name
);
2486 if (slave
== oldcurrent
)
2491 /* note: if switch is in round-robin mode, all links
2492 * must tx arp to ensure all links rx an arp - otherwise
2493 * links may oscillate or not come up at all; if switch is
2494 * in something like xor mode, there is nothing we can
2495 * do - all replies will be rx'ed on same link causing slaves
2496 * to be unstable during low/no traffic periods
2498 if (bond_slave_is_up(slave
))
2499 bond_arp_send_all(bond
, slave
);
2504 if (do_failover
|| slave_state_changed
) {
2505 if (!rtnl_trylock())
2508 if (slave_state_changed
) {
2509 bond_slave_state_change(bond
);
2510 } else if (do_failover
) {
2511 /* the bond_select_active_slave must hold RTNL
2512 * and curr_slave_lock for write.
2515 write_lock_bh(&bond
->curr_slave_lock
);
2517 bond_select_active_slave(bond
);
2519 write_unlock_bh(&bond
->curr_slave_lock
);
2520 unblock_netpoll_tx();
2526 if (bond
->params
.arp_interval
)
2527 queue_delayed_work(bond
->wq
, &bond
->arp_work
,
2528 msecs_to_jiffies(bond
->params
.arp_interval
));
2532 * Called to inspect slaves for active-backup mode ARP monitor link state
2533 * changes. Sets new_link in slaves to specify what action should take
2534 * place for the slave. Returns 0 if no changes are found, >0 if changes
2535 * to link states must be committed.
2537 * Called with rcu_read_lock hold.
2539 static int bond_ab_arp_inspect(struct bonding
*bond
)
2541 unsigned long trans_start
, last_rx
;
2542 struct list_head
*iter
;
2543 struct slave
*slave
;
2546 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2547 slave
->new_link
= BOND_LINK_NOCHANGE
;
2548 last_rx
= slave_last_rx(bond
, slave
);
2550 if (slave
->link
!= BOND_LINK_UP
) {
2551 if (bond_time_in_interval(bond
, last_rx
, 1)) {
2552 slave
->new_link
= BOND_LINK_UP
;
2559 * Give slaves 2*delta after being enslaved or made
2560 * active. This avoids bouncing, as the last receive
2561 * times need a full ARP monitor cycle to be updated.
2563 if (bond_time_in_interval(bond
, slave
->last_link_up
, 2))
2567 * Backup slave is down if:
2568 * - No current_arp_slave AND
2569 * - more than 3*delta since last receive AND
2570 * - the bond has an IP address
2572 * Note: a non-null current_arp_slave indicates
2573 * the curr_active_slave went down and we are
2574 * searching for a new one; under this condition
2575 * we only take the curr_active_slave down - this
2576 * gives each slave a chance to tx/rx traffic
2577 * before being taken out
2579 if (!bond_is_active_slave(slave
) &&
2580 !bond
->current_arp_slave
&&
2581 !bond_time_in_interval(bond
, last_rx
, 3)) {
2582 slave
->new_link
= BOND_LINK_DOWN
;
2587 * Active slave is down if:
2588 * - more than 2*delta since transmitting OR
2589 * - (more than 2*delta since receive AND
2590 * the bond has an IP address)
2592 trans_start
= dev_trans_start(slave
->dev
);
2593 if (bond_is_active_slave(slave
) &&
2594 (!bond_time_in_interval(bond
, trans_start
, 2) ||
2595 !bond_time_in_interval(bond
, last_rx
, 2))) {
2596 slave
->new_link
= BOND_LINK_DOWN
;
2605 * Called to commit link state changes noted by inspection step of
2606 * active-backup mode ARP monitor.
2608 * Called with RTNL hold.
2610 static void bond_ab_arp_commit(struct bonding
*bond
)
2612 unsigned long trans_start
;
2613 struct list_head
*iter
;
2614 struct slave
*slave
;
2616 bond_for_each_slave(bond
, slave
, iter
) {
2617 switch (slave
->new_link
) {
2618 case BOND_LINK_NOCHANGE
:
2622 trans_start
= dev_trans_start(slave
->dev
);
2623 if (bond
->curr_active_slave
!= slave
||
2624 (!bond
->curr_active_slave
&&
2625 bond_time_in_interval(bond
, trans_start
, 1))) {
2626 slave
->link
= BOND_LINK_UP
;
2627 if (bond
->current_arp_slave
) {
2628 bond_set_slave_inactive_flags(
2629 bond
->current_arp_slave
,
2630 BOND_SLAVE_NOTIFY_NOW
);
2631 bond
->current_arp_slave
= NULL
;
2634 pr_info("%s: link status definitely up for interface %s\n",
2635 bond
->dev
->name
, slave
->dev
->name
);
2637 if (!bond
->curr_active_slave
||
2638 (slave
== bond
->primary_slave
))
2645 case BOND_LINK_DOWN
:
2646 if (slave
->link_failure_count
< UINT_MAX
)
2647 slave
->link_failure_count
++;
2649 slave
->link
= BOND_LINK_DOWN
;
2650 bond_set_slave_inactive_flags(slave
,
2651 BOND_SLAVE_NOTIFY_NOW
);
2653 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2654 bond
->dev
->name
, slave
->dev
->name
);
2656 if (slave
== bond
->curr_active_slave
) {
2657 bond
->current_arp_slave
= NULL
;
2664 pr_err("%s: impossible: new_link %d on slave %s\n",
2665 bond
->dev
->name
, slave
->new_link
,
2673 write_lock_bh(&bond
->curr_slave_lock
);
2674 bond_select_active_slave(bond
);
2675 write_unlock_bh(&bond
->curr_slave_lock
);
2676 unblock_netpoll_tx();
2679 bond_set_carrier(bond
);
2683 * Send ARP probes for active-backup mode ARP monitor.
2685 * Called with rcu_read_lock hold.
2687 static bool bond_ab_arp_probe(struct bonding
*bond
)
2689 struct slave
*slave
, *before
= NULL
, *new_slave
= NULL
,
2690 *curr_arp_slave
= rcu_dereference(bond
->current_arp_slave
),
2691 *curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2692 struct list_head
*iter
;
2694 bool should_notify_rtnl
= BOND_SLAVE_NOTIFY_LATER
;
2696 if (curr_arp_slave
&& curr_active_slave
)
2697 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2698 curr_arp_slave
->dev
->name
,
2699 curr_active_slave
->dev
->name
);
2701 if (curr_active_slave
) {
2702 bond_arp_send_all(bond
, curr_active_slave
);
2703 return should_notify_rtnl
;
2706 /* if we don't have a curr_active_slave, search for the next available
2707 * backup slave from the current_arp_slave and make it the candidate
2708 * for becoming the curr_active_slave
2711 if (!curr_arp_slave
) {
2712 curr_arp_slave
= bond_first_slave_rcu(bond
);
2713 if (!curr_arp_slave
)
2714 return should_notify_rtnl
;
2717 bond_set_slave_inactive_flags(curr_arp_slave
, BOND_SLAVE_NOTIFY_LATER
);
2719 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2720 if (!found
&& !before
&& bond_slave_is_up(slave
))
2723 if (found
&& !new_slave
&& bond_slave_is_up(slave
))
2725 /* if the link state is up at this point, we
2726 * mark it down - this can happen if we have
2727 * simultaneous link failures and
2728 * reselect_active_interface doesn't make this
2729 * one the current slave so it is still marked
2730 * up when it is actually down
2732 if (!bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
2733 slave
->link
= BOND_LINK_DOWN
;
2734 if (slave
->link_failure_count
< UINT_MAX
)
2735 slave
->link_failure_count
++;
2737 bond_set_slave_inactive_flags(slave
,
2738 BOND_SLAVE_NOTIFY_LATER
);
2740 pr_info("%s: backup interface %s is now down\n",
2741 bond
->dev
->name
, slave
->dev
->name
);
2743 if (slave
== curr_arp_slave
)
2747 if (!new_slave
&& before
)
2753 new_slave
->link
= BOND_LINK_BACK
;
2754 bond_set_slave_active_flags(new_slave
, BOND_SLAVE_NOTIFY_LATER
);
2755 bond_arp_send_all(bond
, new_slave
);
2756 new_slave
->last_link_up
= jiffies
;
2757 rcu_assign_pointer(bond
->current_arp_slave
, new_slave
);
2760 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2761 if (slave
->should_notify
) {
2762 should_notify_rtnl
= BOND_SLAVE_NOTIFY_NOW
;
2766 return should_notify_rtnl
;
2769 static void bond_activebackup_arp_mon(struct work_struct
*work
)
2771 struct bonding
*bond
= container_of(work
, struct bonding
,
2773 bool should_notify_peers
= false;
2774 bool should_notify_rtnl
= false;
2777 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2779 if (!bond_has_slaves(bond
))
2784 should_notify_peers
= bond_should_notify_peers(bond
);
2786 if (bond_ab_arp_inspect(bond
)) {
2789 /* Race avoidance with bond_close flush of workqueue */
2790 if (!rtnl_trylock()) {
2792 should_notify_peers
= false;
2796 bond_ab_arp_commit(bond
);
2802 should_notify_rtnl
= bond_ab_arp_probe(bond
);
2806 if (bond
->params
.arp_interval
)
2807 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2809 if (should_notify_peers
|| should_notify_rtnl
) {
2810 if (!rtnl_trylock())
2813 if (should_notify_peers
)
2814 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
2816 if (should_notify_rtnl
)
2817 bond_slave_state_notify(bond
);
2823 /*-------------------------- netdev event handling --------------------------*/
2826 * Change device name
2828 static int bond_event_changename(struct bonding
*bond
)
2830 bond_remove_proc_entry(bond
);
2831 bond_create_proc_entry(bond
);
2833 bond_debug_reregister(bond
);
2838 static int bond_master_netdev_event(unsigned long event
,
2839 struct net_device
*bond_dev
)
2841 struct bonding
*event_bond
= netdev_priv(bond_dev
);
2844 case NETDEV_CHANGENAME
:
2845 return bond_event_changename(event_bond
);
2846 case NETDEV_UNREGISTER
:
2847 bond_remove_proc_entry(event_bond
);
2849 case NETDEV_REGISTER
:
2850 bond_create_proc_entry(event_bond
);
2852 case NETDEV_NOTIFY_PEERS
:
2853 if (event_bond
->send_peer_notif
)
2854 event_bond
->send_peer_notif
--;
2863 static int bond_slave_netdev_event(unsigned long event
,
2864 struct net_device
*slave_dev
)
2866 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
);
2867 struct bonding
*bond
;
2868 struct net_device
*bond_dev
;
2872 /* A netdev event can be generated while enslaving a device
2873 * before netdev_rx_handler_register is called in which case
2874 * slave will be NULL
2878 bond_dev
= slave
->bond
->dev
;
2882 case NETDEV_UNREGISTER
:
2883 if (bond_dev
->type
!= ARPHRD_ETHER
)
2884 bond_release_and_destroy(bond_dev
, slave_dev
);
2886 bond_release(bond_dev
, slave_dev
);
2890 old_speed
= slave
->speed
;
2891 old_duplex
= slave
->duplex
;
2893 bond_update_speed_duplex(slave
);
2895 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
2896 if (old_speed
!= slave
->speed
)
2897 bond_3ad_adapter_speed_changed(slave
);
2898 if (old_duplex
!= slave
->duplex
)
2899 bond_3ad_adapter_duplex_changed(slave
);
2904 * ... Or is it this?
2907 case NETDEV_CHANGEMTU
:
2909 * TODO: Should slaves be allowed to
2910 * independently alter their MTU? For
2911 * an active-backup bond, slaves need
2912 * not be the same type of device, so
2913 * MTUs may vary. For other modes,
2914 * slaves arguably should have the
2915 * same MTUs. To do this, we'd need to
2916 * take over the slave's change_mtu
2917 * function for the duration of their
2921 case NETDEV_CHANGENAME
:
2922 /* we don't care if we don't have primary set */
2923 if (!bond_uses_primary(bond
) ||
2924 !bond
->params
.primary
[0])
2927 if (slave
== bond
->primary_slave
) {
2928 /* slave's name changed - he's no longer primary */
2929 bond
->primary_slave
= NULL
;
2930 } else if (!strcmp(slave_dev
->name
, bond
->params
.primary
)) {
2931 /* we have a new primary slave */
2932 bond
->primary_slave
= slave
;
2933 } else { /* we didn't change primary - exit */
2937 pr_info("%s: Primary slave changed to %s, reselecting active slave\n",
2939 bond
->primary_slave
? slave_dev
->name
: "none");
2942 write_lock_bh(&bond
->curr_slave_lock
);
2943 bond_select_active_slave(bond
);
2944 write_unlock_bh(&bond
->curr_slave_lock
);
2945 unblock_netpoll_tx();
2947 case NETDEV_FEAT_CHANGE
:
2948 bond_compute_features(bond
);
2950 case NETDEV_RESEND_IGMP
:
2951 /* Propagate to master device */
2952 call_netdevice_notifiers(event
, slave
->bond
->dev
);
2962 * bond_netdev_event: handle netdev notifier chain events.
2964 * This function receives events for the netdev chain. The caller (an
2965 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2966 * locks for us to safely manipulate the slave devices (RTNL lock,
2969 static int bond_netdev_event(struct notifier_block
*this,
2970 unsigned long event
, void *ptr
)
2972 struct net_device
*event_dev
= netdev_notifier_info_to_dev(ptr
);
2974 pr_debug("event_dev: %s, event: %lx\n",
2975 event_dev
? event_dev
->name
: "None", event
);
2977 if (!(event_dev
->priv_flags
& IFF_BONDING
))
2980 if (event_dev
->flags
& IFF_MASTER
) {
2981 pr_debug("IFF_MASTER\n");
2982 return bond_master_netdev_event(event
, event_dev
);
2985 if (event_dev
->flags
& IFF_SLAVE
) {
2986 pr_debug("IFF_SLAVE\n");
2987 return bond_slave_netdev_event(event
, event_dev
);
2993 static struct notifier_block bond_netdev_notifier
= {
2994 .notifier_call
= bond_netdev_event
,
2997 /*---------------------------- Hashing Policies -----------------------------*/
2999 /* L2 hash helper */
3000 static inline u32
bond_eth_hash(struct sk_buff
*skb
)
3002 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3004 if (skb_headlen(skb
) >= offsetof(struct ethhdr
, h_proto
))
3005 return data
->h_dest
[5] ^ data
->h_source
[5];
3010 /* Extract the appropriate headers based on bond's xmit policy */
3011 static bool bond_flow_dissect(struct bonding
*bond
, struct sk_buff
*skb
,
3012 struct flow_keys
*fk
)
3014 const struct ipv6hdr
*iph6
;
3015 const struct iphdr
*iph
;
3016 int noff
, proto
= -1;
3018 if (bond
->params
.xmit_policy
> BOND_XMIT_POLICY_LAYER23
)
3019 return skb_flow_dissect(skb
, fk
);
3022 noff
= skb_network_offset(skb
);
3023 if (skb
->protocol
== htons(ETH_P_IP
)) {
3024 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph
))))
3027 fk
->src
= iph
->saddr
;
3028 fk
->dst
= iph
->daddr
;
3029 noff
+= iph
->ihl
<< 2;
3030 if (!ip_is_fragment(iph
))
3031 proto
= iph
->protocol
;
3032 } else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
3033 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph6
))))
3035 iph6
= ipv6_hdr(skb
);
3036 fk
->src
= (__force __be32
)ipv6_addr_hash(&iph6
->saddr
);
3037 fk
->dst
= (__force __be32
)ipv6_addr_hash(&iph6
->daddr
);
3038 noff
+= sizeof(*iph6
);
3039 proto
= iph6
->nexthdr
;
3043 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
&& proto
>= 0)
3044 fk
->ports
= skb_flow_get_ports(skb
, noff
, proto
);
3050 * bond_xmit_hash - generate a hash value based on the xmit policy
3051 * @bond: bonding device
3052 * @skb: buffer to use for headers
3054 * This function will extract the necessary headers from the skb buffer and use
3055 * them to generate a hash based on the xmit_policy set in the bonding device
3057 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
)
3059 struct flow_keys flow
;
3062 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER2
||
3063 !bond_flow_dissect(bond
, skb
, &flow
))
3064 return bond_eth_hash(skb
);
3066 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER23
||
3067 bond
->params
.xmit_policy
== BOND_XMIT_POLICY_ENCAP23
)
3068 hash
= bond_eth_hash(skb
);
3070 hash
= (__force u32
)flow
.ports
;
3071 hash
^= (__force u32
)flow
.dst
^ (__force u32
)flow
.src
;
3072 hash
^= (hash
>> 16);
3073 hash
^= (hash
>> 8);
3078 /*-------------------------- Device entry points ----------------------------*/
3080 static void bond_work_init_all(struct bonding
*bond
)
3082 INIT_DELAYED_WORK(&bond
->mcast_work
,
3083 bond_resend_igmp_join_requests_delayed
);
3084 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3085 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3086 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3087 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3089 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3090 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3093 static void bond_work_cancel_all(struct bonding
*bond
)
3095 cancel_delayed_work_sync(&bond
->mii_work
);
3096 cancel_delayed_work_sync(&bond
->arp_work
);
3097 cancel_delayed_work_sync(&bond
->alb_work
);
3098 cancel_delayed_work_sync(&bond
->ad_work
);
3099 cancel_delayed_work_sync(&bond
->mcast_work
);
3102 static int bond_open(struct net_device
*bond_dev
)
3104 struct bonding
*bond
= netdev_priv(bond_dev
);
3105 struct list_head
*iter
;
3106 struct slave
*slave
;
3108 /* reset slave->backup and slave->inactive */
3109 read_lock(&bond
->lock
);
3110 if (bond_has_slaves(bond
)) {
3111 read_lock(&bond
->curr_slave_lock
);
3112 bond_for_each_slave(bond
, slave
, iter
) {
3113 if (bond_uses_primary(bond
)
3114 && (slave
!= bond
->curr_active_slave
)) {
3115 bond_set_slave_inactive_flags(slave
,
3116 BOND_SLAVE_NOTIFY_NOW
);
3118 bond_set_slave_active_flags(slave
,
3119 BOND_SLAVE_NOTIFY_NOW
);
3122 read_unlock(&bond
->curr_slave_lock
);
3124 read_unlock(&bond
->lock
);
3126 bond_work_init_all(bond
);
3128 if (bond_is_lb(bond
)) {
3129 /* bond_alb_initialize must be called before the timer
3132 if (bond_alb_initialize(bond
, (BOND_MODE(bond
) == BOND_MODE_ALB
)))
3134 if (bond
->params
.tlb_dynamic_lb
)
3135 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3138 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3139 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3141 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3142 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3143 bond
->recv_probe
= bond_arp_rcv
;
3146 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
3147 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3148 /* register to receive LACPDUs */
3149 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3150 bond_3ad_initiate_agg_selection(bond
, 1);
3156 static int bond_close(struct net_device
*bond_dev
)
3158 struct bonding
*bond
= netdev_priv(bond_dev
);
3160 bond_work_cancel_all(bond
);
3161 bond
->send_peer_notif
= 0;
3162 if (bond_is_lb(bond
))
3163 bond_alb_deinitialize(bond
);
3164 bond
->recv_probe
= NULL
;
3169 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3170 struct rtnl_link_stats64
*stats
)
3172 struct bonding
*bond
= netdev_priv(bond_dev
);
3173 struct rtnl_link_stats64 temp
;
3174 struct list_head
*iter
;
3175 struct slave
*slave
;
3177 memset(stats
, 0, sizeof(*stats
));
3179 read_lock_bh(&bond
->lock
);
3180 bond_for_each_slave(bond
, slave
, iter
) {
3181 const struct rtnl_link_stats64
*sstats
=
3182 dev_get_stats(slave
->dev
, &temp
);
3184 stats
->rx_packets
+= sstats
->rx_packets
;
3185 stats
->rx_bytes
+= sstats
->rx_bytes
;
3186 stats
->rx_errors
+= sstats
->rx_errors
;
3187 stats
->rx_dropped
+= sstats
->rx_dropped
;
3189 stats
->tx_packets
+= sstats
->tx_packets
;
3190 stats
->tx_bytes
+= sstats
->tx_bytes
;
3191 stats
->tx_errors
+= sstats
->tx_errors
;
3192 stats
->tx_dropped
+= sstats
->tx_dropped
;
3194 stats
->multicast
+= sstats
->multicast
;
3195 stats
->collisions
+= sstats
->collisions
;
3197 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3198 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3199 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3200 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3201 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3202 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3204 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3205 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3206 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3207 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3208 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3210 read_unlock_bh(&bond
->lock
);
3215 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3217 struct bonding
*bond
= netdev_priv(bond_dev
);
3218 struct net_device
*slave_dev
= NULL
;
3219 struct ifbond k_binfo
;
3220 struct ifbond __user
*u_binfo
= NULL
;
3221 struct ifslave k_sinfo
;
3222 struct ifslave __user
*u_sinfo
= NULL
;
3223 struct mii_ioctl_data
*mii
= NULL
;
3224 struct bond_opt_value newval
;
3228 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3240 * We do this again just in case we were called by SIOCGMIIREG
3241 * instead of SIOCGMIIPHY.
3248 if (mii
->reg_num
== 1) {
3250 read_lock(&bond
->lock
);
3251 read_lock(&bond
->curr_slave_lock
);
3252 if (netif_carrier_ok(bond
->dev
))
3253 mii
->val_out
= BMSR_LSTATUS
;
3255 read_unlock(&bond
->curr_slave_lock
);
3256 read_unlock(&bond
->lock
);
3260 case BOND_INFO_QUERY_OLD
:
3261 case SIOCBONDINFOQUERY
:
3262 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3264 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3267 res
= bond_info_query(bond_dev
, &k_binfo
);
3269 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3273 case BOND_SLAVE_INFO_QUERY_OLD
:
3274 case SIOCBONDSLAVEINFOQUERY
:
3275 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3277 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3280 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3282 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3291 net
= dev_net(bond_dev
);
3293 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3296 slave_dev
= __dev_get_by_name(net
, ifr
->ifr_slave
);
3298 pr_debug("slave_dev=%p:\n", slave_dev
);
3303 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3305 case BOND_ENSLAVE_OLD
:
3306 case SIOCBONDENSLAVE
:
3307 res
= bond_enslave(bond_dev
, slave_dev
);
3309 case BOND_RELEASE_OLD
:
3310 case SIOCBONDRELEASE
:
3311 res
= bond_release(bond_dev
, slave_dev
);
3313 case BOND_SETHWADDR_OLD
:
3314 case SIOCBONDSETHWADDR
:
3315 bond_set_dev_addr(bond_dev
, slave_dev
);
3318 case BOND_CHANGE_ACTIVE_OLD
:
3319 case SIOCBONDCHANGEACTIVE
:
3320 bond_opt_initstr(&newval
, slave_dev
->name
);
3321 res
= __bond_opt_set(bond
, BOND_OPT_ACTIVE_SLAVE
, &newval
);
3330 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3332 struct bonding
*bond
= netdev_priv(bond_dev
);
3334 if (change
& IFF_PROMISC
)
3335 bond_set_promiscuity(bond
,
3336 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3338 if (change
& IFF_ALLMULTI
)
3339 bond_set_allmulti(bond
,
3340 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3343 static void bond_set_rx_mode(struct net_device
*bond_dev
)
3345 struct bonding
*bond
= netdev_priv(bond_dev
);
3346 struct list_head
*iter
;
3347 struct slave
*slave
;
3351 if (bond_uses_primary(bond
)) {
3352 slave
= rcu_dereference(bond
->curr_active_slave
);
3354 dev_uc_sync(slave
->dev
, bond_dev
);
3355 dev_mc_sync(slave
->dev
, bond_dev
);
3358 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3359 dev_uc_sync_multiple(slave
->dev
, bond_dev
);
3360 dev_mc_sync_multiple(slave
->dev
, bond_dev
);
3366 static int bond_neigh_init(struct neighbour
*n
)
3368 struct bonding
*bond
= netdev_priv(n
->dev
);
3369 const struct net_device_ops
*slave_ops
;
3370 struct neigh_parms parms
;
3371 struct slave
*slave
;
3374 slave
= bond_first_slave(bond
);
3377 slave_ops
= slave
->dev
->netdev_ops
;
3378 if (!slave_ops
->ndo_neigh_setup
)
3381 parms
.neigh_setup
= NULL
;
3382 parms
.neigh_cleanup
= NULL
;
3383 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3388 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3389 * after the last slave has been detached. Assumes that all slaves
3390 * utilize the same neigh_cleanup (true at this writing as only user
3393 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3395 if (!parms
.neigh_setup
)
3398 return parms
.neigh_setup(n
);
3402 * The bonding ndo_neigh_setup is called at init time beofre any
3403 * slave exists. So we must declare proxy setup function which will
3404 * be used at run time to resolve the actual slave neigh param setup.
3406 * It's also called by master devices (such as vlans) to setup their
3407 * underlying devices. In that case - do nothing, we're already set up from
3410 static int bond_neigh_setup(struct net_device
*dev
,
3411 struct neigh_parms
*parms
)
3413 /* modify only our neigh_parms */
3414 if (parms
->dev
== dev
)
3415 parms
->neigh_setup
= bond_neigh_init
;
3421 * Change the MTU of all of a master's slaves to match the master
3423 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3425 struct bonding
*bond
= netdev_priv(bond_dev
);
3426 struct slave
*slave
, *rollback_slave
;
3427 struct list_head
*iter
;
3430 pr_debug("bond=%p, name=%s, new_mtu=%d\n",
3431 bond
, bond_dev
? bond_dev
->name
: "None", new_mtu
);
3433 /* Can't hold bond->lock with bh disabled here since
3434 * some base drivers panic. On the other hand we can't
3435 * hold bond->lock without bh disabled because we'll
3436 * deadlock. The only solution is to rely on the fact
3437 * that we're under rtnl_lock here, and the slaves
3438 * list won't change. This doesn't solve the problem
3439 * of setting the slave's MTU while it is
3440 * transmitting, but the assumption is that the base
3441 * driver can handle that.
3443 * TODO: figure out a way to safely iterate the slaves
3444 * list, but without holding a lock around the actual
3445 * call to the base driver.
3448 bond_for_each_slave(bond
, slave
, iter
) {
3449 pr_debug("s %p c_m %p\n",
3450 slave
, slave
->dev
->netdev_ops
->ndo_change_mtu
);
3452 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3455 /* If we failed to set the slave's mtu to the new value
3456 * we must abort the operation even in ACTIVE_BACKUP
3457 * mode, because if we allow the backup slaves to have
3458 * different mtu values than the active slave we'll
3459 * need to change their mtu when doing a failover. That
3460 * means changing their mtu from timer context, which
3461 * is probably not a good idea.
3463 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3468 bond_dev
->mtu
= new_mtu
;
3473 /* unwind from head to the slave that failed */
3474 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3477 if (rollback_slave
== slave
)
3480 tmp_res
= dev_set_mtu(rollback_slave
->dev
, bond_dev
->mtu
);
3482 pr_debug("unwind err %d dev %s\n",
3483 tmp_res
, rollback_slave
->dev
->name
);
3493 * Note that many devices must be down to change the HW address, and
3494 * downing the master releases all slaves. We can make bonds full of
3495 * bonding devices to test this, however.
3497 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3499 struct bonding
*bond
= netdev_priv(bond_dev
);
3500 struct slave
*slave
, *rollback_slave
;
3501 struct sockaddr
*sa
= addr
, tmp_sa
;
3502 struct list_head
*iter
;
3505 if (BOND_MODE(bond
) == BOND_MODE_ALB
)
3506 return bond_alb_set_mac_address(bond_dev
, addr
);
3509 pr_debug("bond=%p, name=%s\n",
3510 bond
, bond_dev
? bond_dev
->name
: "None");
3512 /* If fail_over_mac is enabled, do nothing and return success.
3513 * Returning an error causes ifenslave to fail.
3515 if (bond
->params
.fail_over_mac
&&
3516 BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3519 if (!is_valid_ether_addr(sa
->sa_data
))
3520 return -EADDRNOTAVAIL
;
3522 /* Can't hold bond->lock with bh disabled here since
3523 * some base drivers panic. On the other hand we can't
3524 * hold bond->lock without bh disabled because we'll
3525 * deadlock. The only solution is to rely on the fact
3526 * that we're under rtnl_lock here, and the slaves
3527 * list won't change. This doesn't solve the problem
3528 * of setting the slave's hw address while it is
3529 * transmitting, but the assumption is that the base
3530 * driver can handle that.
3532 * TODO: figure out a way to safely iterate the slaves
3533 * list, but without holding a lock around the actual
3534 * call to the base driver.
3537 bond_for_each_slave(bond
, slave
, iter
) {
3538 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
3539 res
= dev_set_mac_address(slave
->dev
, addr
);
3541 /* TODO: consider downing the slave
3543 * User should expect communications
3544 * breakage anyway until ARP finish
3547 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3553 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3557 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3558 tmp_sa
.sa_family
= bond_dev
->type
;
3560 /* unwind from head to the slave that failed */
3561 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3564 if (rollback_slave
== slave
)
3567 tmp_res
= dev_set_mac_address(rollback_slave
->dev
, &tmp_sa
);
3569 pr_debug("unwind err %d dev %s\n",
3570 tmp_res
, rollback_slave
->dev
->name
);
3578 * bond_xmit_slave_id - transmit skb through slave with slave_id
3579 * @bond: bonding device that is transmitting
3580 * @skb: buffer to transmit
3581 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3583 * This function tries to transmit through slave with slave_id but in case
3584 * it fails, it tries to find the first available slave for transmission.
3585 * The skb is consumed in all cases, thus the function is void.
3587 static void bond_xmit_slave_id(struct bonding
*bond
, struct sk_buff
*skb
, int slave_id
)
3589 struct list_head
*iter
;
3590 struct slave
*slave
;
3593 /* Here we start from the slave with slave_id */
3594 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3596 if (bond_slave_can_tx(slave
)) {
3597 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3603 /* Here we start from the first slave up to slave_id */
3605 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3608 if (bond_slave_can_tx(slave
)) {
3609 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3613 /* no slave that can tx has been found */
3614 dev_kfree_skb_any(skb
);
3618 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3619 * @bond: bonding device to use
3621 * Based on the value of the bonding device's packets_per_slave parameter
3622 * this function generates a slave id, which is usually used as the next
3623 * slave to transmit through.
3625 static u32
bond_rr_gen_slave_id(struct bonding
*bond
)
3628 struct reciprocal_value reciprocal_packets_per_slave
;
3629 int packets_per_slave
= bond
->params
.packets_per_slave
;
3631 switch (packets_per_slave
) {
3633 slave_id
= prandom_u32();
3636 slave_id
= bond
->rr_tx_counter
;
3639 reciprocal_packets_per_slave
=
3640 bond
->params
.reciprocal_packets_per_slave
;
3641 slave_id
= reciprocal_divide(bond
->rr_tx_counter
,
3642 reciprocal_packets_per_slave
);
3645 bond
->rr_tx_counter
++;
3650 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3652 struct bonding
*bond
= netdev_priv(bond_dev
);
3653 struct iphdr
*iph
= ip_hdr(skb
);
3654 struct slave
*slave
;
3657 /* Start with the curr_active_slave that joined the bond as the
3658 * default for sending IGMP traffic. For failover purposes one
3659 * needs to maintain some consistency for the interface that will
3660 * send the join/membership reports. The curr_active_slave found
3661 * will send all of this type of traffic.
3663 if (iph
->protocol
== IPPROTO_IGMP
&& skb
->protocol
== htons(ETH_P_IP
)) {
3664 slave
= rcu_dereference(bond
->curr_active_slave
);
3665 if (slave
&& bond_slave_can_tx(slave
))
3666 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3668 bond_xmit_slave_id(bond
, skb
, 0);
3670 slave_id
= bond_rr_gen_slave_id(bond
);
3671 bond_xmit_slave_id(bond
, skb
, slave_id
% bond
->slave_cnt
);
3674 return NETDEV_TX_OK
;
3678 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3679 * the bond has a usable interface.
3681 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3683 struct bonding
*bond
= netdev_priv(bond_dev
);
3684 struct slave
*slave
;
3686 slave
= rcu_dereference(bond
->curr_active_slave
);
3688 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3690 dev_kfree_skb_any(skb
);
3692 return NETDEV_TX_OK
;
3695 /* In bond_xmit_xor() , we determine the output device by using a pre-
3696 * determined xmit_hash_policy(), If the selected device is not enabled,
3697 * find the next active slave.
3699 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3701 struct bonding
*bond
= netdev_priv(bond_dev
);
3703 bond_xmit_slave_id(bond
, skb
, bond_xmit_hash(bond
, skb
) % bond
->slave_cnt
);
3705 return NETDEV_TX_OK
;
3708 /* in broadcast mode, we send everything to all usable interfaces. */
3709 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3711 struct bonding
*bond
= netdev_priv(bond_dev
);
3712 struct slave
*slave
= NULL
;
3713 struct list_head
*iter
;
3715 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3716 if (bond_is_last_slave(bond
, slave
))
3718 if (bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
3719 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
3722 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3723 bond_dev
->name
, __func__
);
3726 /* bond_dev_queue_xmit always returns 0 */
3727 bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3730 if (slave
&& bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
)
3731 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3733 dev_kfree_skb_any(skb
);
3735 return NETDEV_TX_OK
;
3738 /*------------------------- Device initialization ---------------------------*/
3741 * Lookup the slave that corresponds to a qid
3743 static inline int bond_slave_override(struct bonding
*bond
,
3744 struct sk_buff
*skb
)
3746 struct slave
*slave
= NULL
;
3747 struct list_head
*iter
;
3749 if (!skb
->queue_mapping
)
3752 /* Find out if any slaves have the same mapping as this skb. */
3753 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3754 if (slave
->queue_id
== skb
->queue_mapping
) {
3755 if (bond_slave_can_tx(slave
)) {
3756 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3759 /* If the slave isn't UP, use default transmit policy. */
3768 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
3769 void *accel_priv
, select_queue_fallback_t fallback
)
3772 * This helper function exists to help dev_pick_tx get the correct
3773 * destination queue. Using a helper function skips a call to
3774 * skb_tx_hash and will put the skbs in the queue we expect on their
3775 * way down to the bonding driver.
3777 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
3780 * Save the original txq to restore before passing to the driver
3782 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
3784 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
3786 txq
-= dev
->real_num_tx_queues
;
3787 } while (txq
>= dev
->real_num_tx_queues
);
3792 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3794 struct bonding
*bond
= netdev_priv(dev
);
3796 if (bond_should_override_tx_queue(bond
) &&
3797 !bond_slave_override(bond
, skb
))
3798 return NETDEV_TX_OK
;
3800 switch (BOND_MODE(bond
)) {
3801 case BOND_MODE_ROUNDROBIN
:
3802 return bond_xmit_roundrobin(skb
, dev
);
3803 case BOND_MODE_ACTIVEBACKUP
:
3804 return bond_xmit_activebackup(skb
, dev
);
3806 return bond_xmit_xor(skb
, dev
);
3807 case BOND_MODE_BROADCAST
:
3808 return bond_xmit_broadcast(skb
, dev
);
3809 case BOND_MODE_8023AD
:
3810 return bond_3ad_xmit_xor(skb
, dev
);
3812 return bond_alb_xmit(skb
, dev
);
3814 return bond_tlb_xmit(skb
, dev
);
3816 /* Should never happen, mode already checked */
3817 pr_err("%s: Error: Unknown bonding mode %d\n",
3818 dev
->name
, BOND_MODE(bond
));
3820 dev_kfree_skb_any(skb
);
3821 return NETDEV_TX_OK
;
3825 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3827 struct bonding
*bond
= netdev_priv(dev
);
3828 netdev_tx_t ret
= NETDEV_TX_OK
;
3831 * If we risk deadlock from transmitting this in the
3832 * netpoll path, tell netpoll to queue the frame for later tx
3834 if (unlikely(is_netpoll_tx_blocked(dev
)))
3835 return NETDEV_TX_BUSY
;
3838 if (bond_has_slaves(bond
))
3839 ret
= __bond_start_xmit(skb
, dev
);
3841 dev_kfree_skb_any(skb
);
3847 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
3848 struct ethtool_cmd
*ecmd
)
3850 struct bonding
*bond
= netdev_priv(bond_dev
);
3851 unsigned long speed
= 0;
3852 struct list_head
*iter
;
3853 struct slave
*slave
;
3855 ecmd
->duplex
= DUPLEX_UNKNOWN
;
3856 ecmd
->port
= PORT_OTHER
;
3858 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3859 * do not need to check mode. Though link speed might not represent
3860 * the true receive or transmit bandwidth (not all modes are symmetric)
3861 * this is an accurate maximum.
3863 read_lock(&bond
->lock
);
3864 bond_for_each_slave(bond
, slave
, iter
) {
3865 if (bond_slave_can_tx(slave
)) {
3866 if (slave
->speed
!= SPEED_UNKNOWN
)
3867 speed
+= slave
->speed
;
3868 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
3869 slave
->duplex
!= DUPLEX_UNKNOWN
)
3870 ecmd
->duplex
= slave
->duplex
;
3873 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
3874 read_unlock(&bond
->lock
);
3879 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
3880 struct ethtool_drvinfo
*drvinfo
)
3882 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
3883 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
3884 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
3888 static const struct ethtool_ops bond_ethtool_ops
= {
3889 .get_drvinfo
= bond_ethtool_get_drvinfo
,
3890 .get_settings
= bond_ethtool_get_settings
,
3891 .get_link
= ethtool_op_get_link
,
3894 static const struct net_device_ops bond_netdev_ops
= {
3895 .ndo_init
= bond_init
,
3896 .ndo_uninit
= bond_uninit
,
3897 .ndo_open
= bond_open
,
3898 .ndo_stop
= bond_close
,
3899 .ndo_start_xmit
= bond_start_xmit
,
3900 .ndo_select_queue
= bond_select_queue
,
3901 .ndo_get_stats64
= bond_get_stats
,
3902 .ndo_do_ioctl
= bond_do_ioctl
,
3903 .ndo_change_rx_flags
= bond_change_rx_flags
,
3904 .ndo_set_rx_mode
= bond_set_rx_mode
,
3905 .ndo_change_mtu
= bond_change_mtu
,
3906 .ndo_set_mac_address
= bond_set_mac_address
,
3907 .ndo_neigh_setup
= bond_neigh_setup
,
3908 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
3909 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
3910 #ifdef CONFIG_NET_POLL_CONTROLLER
3911 .ndo_netpoll_setup
= bond_netpoll_setup
,
3912 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
3913 .ndo_poll_controller
= bond_poll_controller
,
3915 .ndo_add_slave
= bond_enslave
,
3916 .ndo_del_slave
= bond_release
,
3917 .ndo_fix_features
= bond_fix_features
,
3920 static const struct device_type bond_type
= {
3924 static void bond_destructor(struct net_device
*bond_dev
)
3926 struct bonding
*bond
= netdev_priv(bond_dev
);
3928 destroy_workqueue(bond
->wq
);
3929 free_netdev(bond_dev
);
3932 void bond_setup(struct net_device
*bond_dev
)
3934 struct bonding
*bond
= netdev_priv(bond_dev
);
3936 /* initialize rwlocks */
3937 rwlock_init(&bond
->lock
);
3938 rwlock_init(&bond
->curr_slave_lock
);
3939 bond
->params
= bonding_defaults
;
3941 /* Initialize pointers */
3942 bond
->dev
= bond_dev
;
3944 /* Initialize the device entry points */
3945 ether_setup(bond_dev
);
3946 bond_dev
->netdev_ops
= &bond_netdev_ops
;
3947 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
3949 bond_dev
->destructor
= bond_destructor
;
3951 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
3953 /* Initialize the device options */
3954 bond_dev
->tx_queue_len
= 0;
3955 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
3956 bond_dev
->priv_flags
|= IFF_BONDING
| IFF_UNICAST_FLT
;
3957 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
3959 /* At first, we block adding VLANs. That's the only way to
3960 * prevent problems that occur when adding VLANs over an
3961 * empty bond. The block will be removed once non-challenged
3962 * slaves are enslaved.
3964 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
3966 /* don't acquire bond device's netif_tx_lock when
3968 bond_dev
->features
|= NETIF_F_LLTX
;
3970 /* By default, we declare the bond to be fully
3971 * VLAN hardware accelerated capable. Special
3972 * care is taken in the various xmit functions
3973 * when there are slaves that are not hw accel
3977 /* Don't allow bond devices to change network namespaces. */
3978 bond_dev
->features
|= NETIF_F_NETNS_LOCAL
;
3980 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
3981 NETIF_F_HW_VLAN_CTAG_TX
|
3982 NETIF_F_HW_VLAN_CTAG_RX
|
3983 NETIF_F_HW_VLAN_CTAG_FILTER
;
3985 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
3986 bond_dev
->hw_features
|= NETIF_F_GSO_UDP_TUNNEL
;
3987 bond_dev
->features
|= bond_dev
->hw_features
;
3991 * Destroy a bonding device.
3992 * Must be under rtnl_lock when this function is called.
3994 static void bond_uninit(struct net_device
*bond_dev
)
3996 struct bonding
*bond
= netdev_priv(bond_dev
);
3997 struct list_head
*iter
;
3998 struct slave
*slave
;
4000 bond_netpoll_cleanup(bond_dev
);
4002 /* Release the bonded slaves */
4003 bond_for_each_slave(bond
, slave
, iter
)
4004 __bond_release_one(bond_dev
, slave
->dev
, true);
4005 pr_info("%s: Released all slaves\n", bond_dev
->name
);
4007 list_del(&bond
->bond_list
);
4009 bond_debug_unregister(bond
);
4012 /*------------------------- Module initialization ---------------------------*/
4014 static int bond_check_params(struct bond_params
*params
)
4016 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
4017 struct bond_opt_value newval
;
4018 const struct bond_opt_value
*valptr
;
4019 int arp_all_targets_value
;
4022 * Convert string parameters.
4025 bond_opt_initstr(&newval
, mode
);
4026 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_MODE
), &newval
);
4028 pr_err("Error: Invalid bonding mode \"%s\"\n", mode
);
4031 bond_mode
= valptr
->value
;
4034 if (xmit_hash_policy
) {
4035 if ((bond_mode
!= BOND_MODE_XOR
) &&
4036 (bond_mode
!= BOND_MODE_8023AD
) &&
4037 (bond_mode
!= BOND_MODE_TLB
)) {
4038 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4039 bond_mode_name(bond_mode
));
4041 bond_opt_initstr(&newval
, xmit_hash_policy
);
4042 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH
),
4045 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4049 xmit_hashtype
= valptr
->value
;
4054 if (bond_mode
!= BOND_MODE_8023AD
) {
4055 pr_info("lacp_rate param is irrelevant in mode %s\n",
4056 bond_mode_name(bond_mode
));
4058 bond_opt_initstr(&newval
, lacp_rate
);
4059 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE
),
4062 pr_err("Error: Invalid lacp rate \"%s\"\n",
4066 lacp_fast
= valptr
->value
;
4071 bond_opt_initstr(&newval
, lacp_rate
);
4072 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT
),
4075 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select
);
4078 params
->ad_select
= valptr
->value
;
4079 if (bond_mode
!= BOND_MODE_8023AD
)
4080 pr_warn("ad_select param only affects 802.3ad mode\n");
4082 params
->ad_select
= BOND_AD_STABLE
;
4085 if (max_bonds
< 0) {
4086 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4087 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4088 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4092 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4098 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4103 if (downdelay
< 0) {
4104 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4105 downdelay
, INT_MAX
);
4109 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4110 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4115 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4116 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4121 /* reset values for 802.3ad/TLB/ALB */
4122 if (!bond_mode_uses_arp(bond_mode
)) {
4124 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4125 pr_warn("Forcing miimon to 100msec\n");
4126 miimon
= BOND_DEFAULT_MIIMON
;
4130 if (tx_queues
< 1 || tx_queues
> 255) {
4131 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4132 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4133 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4136 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4137 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4139 all_slaves_active
= 0;
4142 if (resend_igmp
< 0 || resend_igmp
> 255) {
4143 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4144 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4145 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4148 bond_opt_initval(&newval
, packets_per_slave
);
4149 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE
), &newval
)) {
4150 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4151 packets_per_slave
, USHRT_MAX
);
4152 packets_per_slave
= 1;
4155 if (bond_mode
== BOND_MODE_ALB
) {
4156 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",
4161 if (updelay
|| downdelay
) {
4162 /* just warn the user the up/down delay will have
4163 * no effect since miimon is zero...
4165 pr_warn("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",
4166 updelay
, downdelay
);
4169 /* don't allow arp monitoring */
4171 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4172 miimon
, arp_interval
);
4176 if ((updelay
% miimon
) != 0) {
4177 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4178 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4183 if ((downdelay
% miimon
) != 0) {
4184 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4186 (downdelay
/ miimon
) * miimon
);
4189 downdelay
/= miimon
;
4192 if (arp_interval
< 0) {
4193 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4194 arp_interval
, INT_MAX
);
4198 for (arp_ip_count
= 0, i
= 0;
4199 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4200 /* not complete check, but should be good enough to
4203 if (!in4_pton(arp_ip_target
[i
], -1, (u8
*)&ip
, -1, NULL
) ||
4204 !bond_is_ip_target_ok(ip
)) {
4205 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4209 if (bond_get_targets_ip(arp_target
, ip
) == -1)
4210 arp_target
[arp_ip_count
++] = ip
;
4212 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4217 if (arp_interval
&& !arp_ip_count
) {
4218 /* don't allow arping if no arp_ip_target given... */
4219 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4225 if (!arp_interval
) {
4226 pr_err("arp_validate requires arp_interval\n");
4230 bond_opt_initstr(&newval
, arp_validate
);
4231 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE
),
4234 pr_err("Error: invalid arp_validate \"%s\"\n",
4238 arp_validate_value
= valptr
->value
;
4240 arp_validate_value
= 0;
4243 arp_all_targets_value
= 0;
4244 if (arp_all_targets
) {
4245 bond_opt_initstr(&newval
, arp_all_targets
);
4246 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS
),
4249 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4251 arp_all_targets_value
= 0;
4253 arp_all_targets_value
= valptr
->value
;
4258 pr_info("MII link monitoring set to %d ms\n", miimon
);
4259 } else if (arp_interval
) {
4260 valptr
= bond_opt_get_val(BOND_OPT_ARP_VALIDATE
,
4261 arp_validate_value
);
4262 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4263 arp_interval
, valptr
->string
, arp_ip_count
);
4265 for (i
= 0; i
< arp_ip_count
; i
++)
4266 pr_cont(" %s", arp_ip_target
[i
]);
4270 } else if (max_bonds
) {
4271 /* miimon and arp_interval not set, we need one so things
4272 * work as expected, see bonding.txt for details
4274 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");
4277 if (primary
&& !bond_mode_uses_primary(bond_mode
)) {
4278 /* currently, using a primary only makes sense
4279 * in active backup, TLB or ALB modes
4281 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4282 primary
, bond_mode_name(bond_mode
));
4286 if (primary
&& primary_reselect
) {
4287 bond_opt_initstr(&newval
, primary_reselect
);
4288 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT
),
4291 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4295 primary_reselect_value
= valptr
->value
;
4297 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4300 if (fail_over_mac
) {
4301 bond_opt_initstr(&newval
, fail_over_mac
);
4302 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC
),
4305 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4309 fail_over_mac_value
= valptr
->value
;
4310 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4311 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4313 fail_over_mac_value
= BOND_FOM_NONE
;
4316 if (lp_interval
== 0) {
4317 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4318 INT_MAX
, BOND_ALB_DEFAULT_LP_INTERVAL
);
4319 lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
4322 /* fill params struct with the proper values */
4323 params
->mode
= bond_mode
;
4324 params
->xmit_policy
= xmit_hashtype
;
4325 params
->miimon
= miimon
;
4326 params
->num_peer_notif
= num_peer_notif
;
4327 params
->arp_interval
= arp_interval
;
4328 params
->arp_validate
= arp_validate_value
;
4329 params
->arp_all_targets
= arp_all_targets_value
;
4330 params
->updelay
= updelay
;
4331 params
->downdelay
= downdelay
;
4332 params
->use_carrier
= use_carrier
;
4333 params
->lacp_fast
= lacp_fast
;
4334 params
->primary
[0] = 0;
4335 params
->primary_reselect
= primary_reselect_value
;
4336 params
->fail_over_mac
= fail_over_mac_value
;
4337 params
->tx_queues
= tx_queues
;
4338 params
->all_slaves_active
= all_slaves_active
;
4339 params
->resend_igmp
= resend_igmp
;
4340 params
->min_links
= min_links
;
4341 params
->lp_interval
= lp_interval
;
4342 params
->packets_per_slave
= packets_per_slave
;
4343 params
->tlb_dynamic_lb
= 1; /* Default value */
4344 if (packets_per_slave
> 0) {
4345 params
->reciprocal_packets_per_slave
=
4346 reciprocal_value(packets_per_slave
);
4348 /* reciprocal_packets_per_slave is unused if
4349 * packets_per_slave is 0 or 1, just initialize it
4351 params
->reciprocal_packets_per_slave
=
4352 (struct reciprocal_value
) { 0 };
4356 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4357 params
->primary
[IFNAMSIZ
- 1] = 0;
4360 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4365 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4366 static struct lock_class_key bonding_netdev_addr_lock_key
;
4367 static struct lock_class_key bonding_tx_busylock_key
;
4369 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4370 struct netdev_queue
*txq
,
4373 lockdep_set_class(&txq
->_xmit_lock
,
4374 &bonding_netdev_xmit_lock_key
);
4377 static void bond_set_lockdep_class(struct net_device
*dev
)
4379 lockdep_set_class(&dev
->addr_list_lock
,
4380 &bonding_netdev_addr_lock_key
);
4381 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4382 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4386 * Called from registration process
4388 static int bond_init(struct net_device
*bond_dev
)
4390 struct bonding
*bond
= netdev_priv(bond_dev
);
4391 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4392 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
4394 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4397 * Initialize locks that may be required during
4398 * en/deslave operations. All of the bond_open work
4399 * (of which this is part) should really be moved to
4400 * a phase prior to dev_open
4402 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
4403 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
4405 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4409 bond_set_lockdep_class(bond_dev
);
4411 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4413 bond_prepare_sysfs_group(bond
);
4415 bond_debug_register(bond
);
4417 /* Ensure valid dev_addr */
4418 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4419 bond_dev
->addr_assign_type
== NET_ADDR_PERM
)
4420 eth_hw_addr_random(bond_dev
);
4425 unsigned int bond_get_num_tx_queues(void)
4430 /* Create a new bond based on the specified name and bonding parameters.
4431 * If name is NULL, obtain a suitable "bond%d" name for us.
4432 * Caller must NOT hold rtnl_lock; we need to release it here before we
4433 * set up our sysfs entries.
4435 int bond_create(struct net
*net
, const char *name
)
4437 struct net_device
*bond_dev
;
4442 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4443 name
? name
: "bond%d",
4444 bond_setup
, tx_queues
);
4446 pr_err("%s: eek! can't alloc netdev!\n", name
);
4451 dev_net_set(bond_dev
, net
);
4452 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4454 res
= register_netdevice(bond_dev
);
4456 netif_carrier_off(bond_dev
);
4460 bond_destructor(bond_dev
);
4464 static int __net_init
bond_net_init(struct net
*net
)
4466 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4469 INIT_LIST_HEAD(&bn
->dev_list
);
4471 bond_create_proc_dir(bn
);
4472 bond_create_sysfs(bn
);
4477 static void __net_exit
bond_net_exit(struct net
*net
)
4479 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4480 struct bonding
*bond
, *tmp_bond
;
4483 bond_destroy_sysfs(bn
);
4484 bond_destroy_proc_dir(bn
);
4486 /* Kill off any bonds created after unregistering bond rtnl ops */
4488 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4489 unregister_netdevice_queue(bond
->dev
, &list
);
4490 unregister_netdevice_many(&list
);
4494 static struct pernet_operations bond_net_ops
= {
4495 .init
= bond_net_init
,
4496 .exit
= bond_net_exit
,
4498 .size
= sizeof(struct bond_net
),
4501 static int __init
bonding_init(void)
4506 pr_info("%s", bond_version
);
4508 res
= bond_check_params(&bonding_defaults
);
4512 res
= register_pernet_subsys(&bond_net_ops
);
4516 res
= bond_netlink_init();
4520 bond_create_debugfs();
4522 for (i
= 0; i
< max_bonds
; i
++) {
4523 res
= bond_create(&init_net
, NULL
);
4528 register_netdevice_notifier(&bond_netdev_notifier
);
4532 bond_destroy_debugfs();
4533 bond_netlink_fini();
4535 unregister_pernet_subsys(&bond_net_ops
);
4540 static void __exit
bonding_exit(void)
4542 unregister_netdevice_notifier(&bond_netdev_notifier
);
4544 bond_destroy_debugfs();
4546 bond_netlink_fini();
4547 unregister_pernet_subsys(&bond_net_ops
);
4549 #ifdef CONFIG_NET_POLL_CONTROLLER
4551 * Make sure we don't have an imbalance on our netpoll blocking
4553 WARN_ON(atomic_read(&netpoll_block_tx
));
4557 module_init(bonding_init
);
4558 module_exit(bonding_exit
);
4559 MODULE_LICENSE("GPL");
4560 MODULE_VERSION(DRV_VERSION
);
4561 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4562 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");