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 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 #include "bonding_priv.h"
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
91 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
92 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
93 static int num_peer_notif
= 1;
97 static int use_carrier
= 1;
100 static char *primary_reselect
;
101 static char *lacp_rate
;
102 static int min_links
;
103 static char *ad_select
;
104 static char *xmit_hash_policy
;
105 static int arp_interval
;
106 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
107 static char *arp_validate
;
108 static char *arp_all_targets
;
109 static char *fail_over_mac
;
110 static int all_slaves_active
;
111 static struct bond_params bonding_defaults
;
112 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
113 static int packets_per_slave
= 1;
114 static int lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
116 module_param(max_bonds
, int, 0);
117 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
118 module_param(tx_queues
, int, 0);
119 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp
, num_peer_notif
, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp
, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na
, num_peer_notif
, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na
, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126 module_param(miimon
, int, 0);
127 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
128 module_param(updelay
, int, 0);
129 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
130 module_param(downdelay
, int, 0);
131 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
133 module_param(use_carrier
, int, 0);
134 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136 module_param(mode
, charp
, 0);
137 MODULE_PARM_DESC(mode
, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141 module_param(primary
, charp
, 0);
142 MODULE_PARM_DESC(primary
, "Primary network device to use");
143 module_param(primary_reselect
, charp
, 0);
144 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
149 "2 for only on active slave "
151 module_param(lacp_rate
, charp
, 0);
152 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154 module_param(ad_select
, charp
, 0);
155 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
158 module_param(min_links
, int, 0);
159 MODULE_PARM_DESC(min_links
, "Minimum number of available links before turning on carrier");
161 module_param(xmit_hash_policy
, charp
, 0);
162 MODULE_PARM_DESC(xmit_hash_policy
, "balance-xor and 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166 module_param(arp_interval
, int, 0);
167 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
168 module_param_array(arp_ip_target
, charp
, NULL
, 0);
169 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
170 module_param(arp_validate
, charp
, 0);
171 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174 module_param(arp_all_targets
, charp
, 0);
175 MODULE_PARM_DESC(arp_all_targets
, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac
, charp
, 0);
177 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180 module_param(all_slaves_active
, int, 0);
181 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184 module_param(resend_igmp
, int, 0);
185 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on "
187 module_param(packets_per_slave
, int, 0);
188 MODULE_PARM_DESC(packets_per_slave
, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191 module_param(lp_interval
, uint
, 0);
192 MODULE_PARM_DESC(lp_interval
, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
196 /*----------------------------- Global variables ----------------------------*/
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
202 int bond_net_id __read_mostly
;
204 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
205 static int arp_ip_count
;
206 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
207 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
208 static int lacp_fast
;
210 /*-------------------------- Forward declarations ---------------------------*/
212 static int bond_init(struct net_device
*bond_dev
);
213 static void bond_uninit(struct net_device
*bond_dev
);
214 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
215 struct rtnl_link_stats64
*stats
);
216 static void bond_slave_arr_handler(struct work_struct
*work
);
218 /*---------------------------- General routines -----------------------------*/
220 const char *bond_mode_name(int mode
)
222 static const char *names
[] = {
223 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
224 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
225 [BOND_MODE_XOR
] = "load balancing (xor)",
226 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
227 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
228 [BOND_MODE_TLB
] = "transmit load balancing",
229 [BOND_MODE_ALB
] = "adaptive load balancing",
232 if (mode
< BOND_MODE_ROUNDROBIN
|| mode
> BOND_MODE_ALB
)
238 /*---------------------------------- VLAN -----------------------------------*/
241 * bond_dev_queue_xmit - Prepare skb for xmit.
243 * @bond: bond device that got this skb for tx.
244 * @skb: hw accel VLAN tagged skb to transmit
245 * @slave_dev: slave that is supposed to xmit this skbuff
247 void bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
248 struct net_device
*slave_dev
)
250 skb
->dev
= slave_dev
;
252 BUILD_BUG_ON(sizeof(skb
->queue_mapping
) !=
253 sizeof(qdisc_skb_cb(skb
)->slave_dev_queue_mapping
));
254 skb
->queue_mapping
= qdisc_skb_cb(skb
)->slave_dev_queue_mapping
;
256 if (unlikely(netpoll_tx_running(bond
->dev
)))
257 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
262 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
263 * We don't protect the slave list iteration with a lock because:
264 * a. This operation is performed in IOCTL context,
265 * b. The operation is protected by the RTNL semaphore in the 8021q code,
266 * c. Holding a lock with BH disabled while directly calling a base driver
267 * entry point is generally a BAD idea.
269 * The design of synchronization/protection for this operation in the 8021q
270 * module is good for one or more VLAN devices over a single physical device
271 * and cannot be extended for a teaming solution like bonding, so there is a
272 * potential race condition here where a net device from the vlan group might
273 * be referenced (either by a base driver or the 8021q code) while it is being
274 * removed from the system. However, it turns out we're not making matters
275 * worse, and if it works for regular VLAN usage it will work here too.
279 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
280 * @bond_dev: bonding net device that got called
281 * @vid: vlan id being added
283 static int bond_vlan_rx_add_vid(struct net_device
*bond_dev
,
284 __be16 proto
, u16 vid
)
286 struct bonding
*bond
= netdev_priv(bond_dev
);
287 struct slave
*slave
, *rollback_slave
;
288 struct list_head
*iter
;
291 bond_for_each_slave(bond
, slave
, iter
) {
292 res
= vlan_vid_add(slave
->dev
, proto
, vid
);
300 /* unwind to the slave that failed */
301 bond_for_each_slave(bond
, rollback_slave
, iter
) {
302 if (rollback_slave
== slave
)
305 vlan_vid_del(rollback_slave
->dev
, proto
, vid
);
312 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
313 * @bond_dev: bonding net device that got called
314 * @vid: vlan id being removed
316 static int bond_vlan_rx_kill_vid(struct net_device
*bond_dev
,
317 __be16 proto
, u16 vid
)
319 struct bonding
*bond
= netdev_priv(bond_dev
);
320 struct list_head
*iter
;
323 bond_for_each_slave(bond
, slave
, iter
)
324 vlan_vid_del(slave
->dev
, proto
, vid
);
326 if (bond_is_lb(bond
))
327 bond_alb_clear_vlan(bond
, vid
);
332 /*------------------------------- Link status -------------------------------*/
334 /* Set the carrier state for the master according to the state of its
335 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
336 * do special 802.3ad magic.
338 * Returns zero if carrier state does not change, nonzero if it does.
340 int bond_set_carrier(struct bonding
*bond
)
342 struct list_head
*iter
;
345 if (!bond_has_slaves(bond
))
348 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
349 return bond_3ad_set_carrier(bond
);
351 bond_for_each_slave(bond
, slave
, iter
) {
352 if (slave
->link
== BOND_LINK_UP
) {
353 if (!netif_carrier_ok(bond
->dev
)) {
354 netif_carrier_on(bond
->dev
);
362 if (netif_carrier_ok(bond
->dev
)) {
363 netif_carrier_off(bond
->dev
);
369 /* Get link speed and duplex from the slave's base driver
370 * using ethtool. If for some reason the call fails or the
371 * values are invalid, set speed and duplex to -1,
374 static void bond_update_speed_duplex(struct slave
*slave
)
376 struct net_device
*slave_dev
= slave
->dev
;
377 struct ethtool_cmd ecmd
;
381 slave
->speed
= SPEED_UNKNOWN
;
382 slave
->duplex
= DUPLEX_UNKNOWN
;
384 res
= __ethtool_get_settings(slave_dev
, &ecmd
);
388 slave_speed
= ethtool_cmd_speed(&ecmd
);
389 if (slave_speed
== 0 || slave_speed
== ((__u32
) -1))
392 switch (ecmd
.duplex
) {
400 slave
->speed
= slave_speed
;
401 slave
->duplex
= ecmd
.duplex
;
406 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
;
481 /* If reporting, report that either there's no dev->do_ioctl,
482 * or both SIOCGMIIREG and get_link failed (meaning that we
483 * cannot report link status). If not reporting, pretend
486 return reporting
? -1 : BMSR_LSTATUS
;
489 /*----------------------------- Multicast list ------------------------------*/
491 /* Push the promiscuity flag down to appropriate slaves */
492 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
494 struct list_head
*iter
;
497 if (bond_uses_primary(bond
)) {
498 struct slave
*curr_active
= rtnl_dereference(bond
->curr_active_slave
);
501 err
= dev_set_promiscuity(curr_active
->dev
, inc
);
505 bond_for_each_slave(bond
, slave
, iter
) {
506 err
= dev_set_promiscuity(slave
->dev
, inc
);
514 /* Push the allmulti flag down to all slaves */
515 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
517 struct list_head
*iter
;
520 if (bond_uses_primary(bond
)) {
521 struct slave
*curr_active
= rtnl_dereference(bond
->curr_active_slave
);
524 err
= dev_set_allmulti(curr_active
->dev
, inc
);
528 bond_for_each_slave(bond
, slave
, iter
) {
529 err
= dev_set_allmulti(slave
->dev
, inc
);
537 /* Retrieve the list of registered multicast addresses for the bonding
538 * device and retransmit an IGMP JOIN request to the current active
541 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
543 struct bonding
*bond
= container_of(work
, struct bonding
,
546 if (!rtnl_trylock()) {
547 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
550 call_netdevice_notifiers(NETDEV_RESEND_IGMP
, bond
->dev
);
552 if (bond
->igmp_retrans
> 1) {
553 bond
->igmp_retrans
--;
554 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
559 /* Flush bond's hardware addresses from slave */
560 static void bond_hw_addr_flush(struct net_device
*bond_dev
,
561 struct net_device
*slave_dev
)
563 struct bonding
*bond
= netdev_priv(bond_dev
);
565 dev_uc_unsync(slave_dev
, bond_dev
);
566 dev_mc_unsync(slave_dev
, bond_dev
);
568 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
569 /* del lacpdu mc addr from mc list */
570 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
572 dev_mc_del(slave_dev
, lacpdu_multicast
);
576 /*--------------------------- Active slave change ---------------------------*/
578 /* Update the hardware address list and promisc/allmulti for the new and
579 * old active slaves (if any). Modes that are not using primary keep all
580 * slaves up date at all times; only the modes that use primary need to call
581 * this function to swap these settings during a failover.
583 static void bond_hw_addr_swap(struct bonding
*bond
, struct slave
*new_active
,
584 struct slave
*old_active
)
587 if (bond
->dev
->flags
& IFF_PROMISC
)
588 dev_set_promiscuity(old_active
->dev
, -1);
590 if (bond
->dev
->flags
& IFF_ALLMULTI
)
591 dev_set_allmulti(old_active
->dev
, -1);
593 bond_hw_addr_flush(bond
->dev
, old_active
->dev
);
597 /* FIXME: Signal errors upstream. */
598 if (bond
->dev
->flags
& IFF_PROMISC
)
599 dev_set_promiscuity(new_active
->dev
, 1);
601 if (bond
->dev
->flags
& IFF_ALLMULTI
)
602 dev_set_allmulti(new_active
->dev
, 1);
604 netif_addr_lock_bh(bond
->dev
);
605 dev_uc_sync(new_active
->dev
, bond
->dev
);
606 dev_mc_sync(new_active
->dev
, bond
->dev
);
607 netif_addr_unlock_bh(bond
->dev
);
612 * bond_set_dev_addr - clone slave's address to bond
613 * @bond_dev: bond net device
614 * @slave_dev: slave net device
616 * Should be called with RTNL held.
618 static void bond_set_dev_addr(struct net_device
*bond_dev
,
619 struct net_device
*slave_dev
)
621 netdev_dbg(bond_dev
, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
622 bond_dev
, slave_dev
, slave_dev
->addr_len
);
623 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
624 bond_dev
->addr_assign_type
= NET_ADDR_STOLEN
;
625 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond_dev
);
628 static struct slave
*bond_get_old_active(struct bonding
*bond
,
629 struct slave
*new_active
)
632 struct list_head
*iter
;
634 bond_for_each_slave(bond
, slave
, iter
) {
635 if (slave
== new_active
)
638 if (ether_addr_equal(bond
->dev
->dev_addr
, slave
->dev
->dev_addr
))
645 /* bond_do_fail_over_mac
647 * Perform special MAC address swapping for fail_over_mac settings
651 static void bond_do_fail_over_mac(struct bonding
*bond
,
652 struct slave
*new_active
,
653 struct slave
*old_active
)
655 u8 tmp_mac
[ETH_ALEN
];
656 struct sockaddr saddr
;
659 switch (bond
->params
.fail_over_mac
) {
660 case BOND_FOM_ACTIVE
:
662 bond_set_dev_addr(bond
->dev
, new_active
->dev
);
664 case BOND_FOM_FOLLOW
:
665 /* if new_active && old_active, swap them
666 * if just old_active, do nothing (going to no active slave)
667 * if just new_active, set new_active to bond's MAC
673 old_active
= bond_get_old_active(bond
, new_active
);
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 netdev_err(bond
->dev
, "Error %d setting MAC of slave %s\n",
688 -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 netdev_err(bond
->dev
, "Error %d setting MAC of slave %s\n",
701 -rv
, new_active
->dev
->name
);
705 netdev_err(bond
->dev
, "bond_do_fail_over_mac impossible: bad policy %d\n",
706 bond
->params
.fail_over_mac
);
712 static struct slave
*bond_choose_primary_or_current(struct bonding
*bond
)
714 struct slave
*prim
= rtnl_dereference(bond
->primary_slave
);
715 struct slave
*curr
= rtnl_dereference(bond
->curr_active_slave
);
717 if (!prim
|| prim
->link
!= BOND_LINK_UP
) {
718 if (!curr
|| curr
->link
!= BOND_LINK_UP
)
723 if (bond
->force_primary
) {
724 bond
->force_primary
= false;
728 if (!curr
|| curr
->link
!= BOND_LINK_UP
)
731 /* At this point, prim and curr are both up */
732 switch (bond
->params
.primary_reselect
) {
733 case BOND_PRI_RESELECT_ALWAYS
:
735 case BOND_PRI_RESELECT_BETTER
:
736 if (prim
->speed
< curr
->speed
)
738 if (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)
741 case BOND_PRI_RESELECT_FAILURE
:
744 netdev_err(bond
->dev
, "impossible primary_reselect %d\n",
745 bond
->params
.primary_reselect
);
751 * bond_find_best_slave - select the best available slave to be the active one
752 * @bond: our bonding struct
754 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
756 struct slave
*slave
, *bestslave
= NULL
;
757 struct list_head
*iter
;
758 int mintime
= bond
->params
.updelay
;
760 slave
= bond_choose_primary_or_current(bond
);
764 bond_for_each_slave(bond
, slave
, iter
) {
765 if (slave
->link
== BOND_LINK_UP
)
767 if (slave
->link
== BOND_LINK_BACK
&& bond_slave_is_up(slave
) &&
768 slave
->delay
< mintime
) {
769 mintime
= slave
->delay
;
777 static bool bond_should_notify_peers(struct bonding
*bond
)
782 slave
= rcu_dereference(bond
->curr_active_slave
);
785 netdev_dbg(bond
->dev
, "bond_should_notify_peers: slave %s\n",
786 slave
? slave
->dev
->name
: "NULL");
788 if (!slave
|| !bond
->send_peer_notif
||
789 !netif_carrier_ok(bond
->dev
) ||
790 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
797 * change_active_interface - change the active slave into the specified one
798 * @bond: our bonding struct
799 * @new: the new slave to make the active one
801 * Set the new slave to the bond's settings and unset them on the old
803 * Setting include flags, mc-list, promiscuity, allmulti, etc.
805 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
806 * because it is apparently the best available slave we have, even though its
807 * updelay hasn't timed out yet.
809 * Caller must hold RTNL.
811 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
813 struct slave
*old_active
;
817 old_active
= rtnl_dereference(bond
->curr_active_slave
);
819 if (old_active
== new_active
)
823 new_active
->last_link_up
= jiffies
;
825 if (new_active
->link
== BOND_LINK_BACK
) {
826 if (bond_uses_primary(bond
)) {
827 netdev_info(bond
->dev
, "making interface %s the new active one %d ms earlier\n",
828 new_active
->dev
->name
,
829 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
832 new_active
->delay
= 0;
833 bond_set_slave_link_state(new_active
, BOND_LINK_UP
,
834 BOND_SLAVE_NOTIFY_NOW
);
836 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
837 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
839 if (bond_is_lb(bond
))
840 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
842 if (bond_uses_primary(bond
)) {
843 netdev_info(bond
->dev
, "making interface %s the new active one\n",
844 new_active
->dev
->name
);
849 if (bond_uses_primary(bond
))
850 bond_hw_addr_swap(bond
, new_active
, old_active
);
852 if (bond_is_lb(bond
)) {
853 bond_alb_handle_active_change(bond
, new_active
);
855 bond_set_slave_inactive_flags(old_active
,
856 BOND_SLAVE_NOTIFY_NOW
);
858 bond_set_slave_active_flags(new_active
,
859 BOND_SLAVE_NOTIFY_NOW
);
861 rcu_assign_pointer(bond
->curr_active_slave
, new_active
);
864 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
) {
866 bond_set_slave_inactive_flags(old_active
,
867 BOND_SLAVE_NOTIFY_NOW
);
870 bool should_notify_peers
= false;
872 bond_set_slave_active_flags(new_active
,
873 BOND_SLAVE_NOTIFY_NOW
);
875 if (bond
->params
.fail_over_mac
)
876 bond_do_fail_over_mac(bond
, new_active
,
879 if (netif_running(bond
->dev
)) {
880 bond
->send_peer_notif
=
881 bond
->params
.num_peer_notif
;
882 should_notify_peers
=
883 bond_should_notify_peers(bond
);
886 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
887 if (should_notify_peers
)
888 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
893 /* resend IGMP joins since active slave has changed or
894 * all were sent on curr_active_slave.
895 * resend only if bond is brought up with the affected
896 * bonding modes and the retransmission is enabled
898 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
899 ((bond_uses_primary(bond
) && new_active
) ||
900 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
)) {
901 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
902 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
907 * bond_select_active_slave - select a new active slave, if needed
908 * @bond: our bonding struct
910 * This functions should be called when one of the following occurs:
911 * - The old curr_active_slave has been released or lost its link.
912 * - The primary_slave has got its link back.
913 * - A slave has got its link back and there's no old curr_active_slave.
915 * Caller must hold RTNL.
917 void bond_select_active_slave(struct bonding
*bond
)
919 struct slave
*best_slave
;
924 best_slave
= bond_find_best_slave(bond
);
925 if (best_slave
!= rtnl_dereference(bond
->curr_active_slave
)) {
926 bond_change_active_slave(bond
, best_slave
);
927 rv
= bond_set_carrier(bond
);
931 if (netif_carrier_ok(bond
->dev
)) {
932 netdev_info(bond
->dev
, "first active interface up!\n");
934 netdev_info(bond
->dev
, "now running without any active interface!\n");
939 #ifdef CONFIG_NET_POLL_CONTROLLER
940 static inline int slave_enable_netpoll(struct slave
*slave
)
945 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
950 err
= __netpoll_setup(np
, slave
->dev
);
959 static inline void slave_disable_netpoll(struct slave
*slave
)
961 struct netpoll
*np
= slave
->np
;
967 __netpoll_free_async(np
);
970 static void bond_poll_controller(struct net_device
*bond_dev
)
972 struct bonding
*bond
= netdev_priv(bond_dev
);
973 struct slave
*slave
= NULL
;
974 struct list_head
*iter
;
975 struct ad_info ad_info
;
976 struct netpoll_info
*ni
;
977 const struct net_device_ops
*ops
;
979 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
980 if (bond_3ad_get_active_agg_info(bond
, &ad_info
))
983 bond_for_each_slave_rcu(bond
, slave
, iter
) {
984 ops
= slave
->dev
->netdev_ops
;
985 if (!bond_slave_is_up(slave
) || !ops
->ndo_poll_controller
)
988 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
989 struct aggregator
*agg
=
990 SLAVE_AD_INFO(slave
)->port
.aggregator
;
993 agg
->aggregator_identifier
!= ad_info
.aggregator_id
)
997 ni
= rcu_dereference_bh(slave
->dev
->npinfo
);
998 if (down_trylock(&ni
->dev_lock
))
1000 ops
->ndo_poll_controller(slave
->dev
);
1005 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1007 struct bonding
*bond
= netdev_priv(bond_dev
);
1008 struct list_head
*iter
;
1009 struct slave
*slave
;
1011 bond_for_each_slave(bond
, slave
, iter
)
1012 if (bond_slave_is_up(slave
))
1013 slave_disable_netpoll(slave
);
1016 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
1018 struct bonding
*bond
= netdev_priv(dev
);
1019 struct list_head
*iter
;
1020 struct slave
*slave
;
1023 bond_for_each_slave(bond
, slave
, iter
) {
1024 err
= slave_enable_netpoll(slave
);
1026 bond_netpoll_cleanup(dev
);
1033 static inline int slave_enable_netpoll(struct slave
*slave
)
1037 static inline void slave_disable_netpoll(struct slave
*slave
)
1040 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1045 /*---------------------------------- IOCTL ----------------------------------*/
1047 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1048 netdev_features_t features
)
1050 struct bonding
*bond
= netdev_priv(dev
);
1051 struct list_head
*iter
;
1052 netdev_features_t mask
;
1053 struct slave
*slave
;
1057 features
&= ~NETIF_F_ONE_FOR_ALL
;
1058 features
|= NETIF_F_ALL_FOR_ALL
;
1060 bond_for_each_slave(bond
, slave
, iter
) {
1061 features
= netdev_increment_features(features
,
1062 slave
->dev
->features
,
1065 features
= netdev_add_tso_features(features
, mask
);
1070 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1071 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1072 NETIF_F_HIGHDMA | NETIF_F_LRO)
1074 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1075 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1077 static void bond_compute_features(struct bonding
*bond
)
1079 unsigned int dst_release_flag
= IFF_XMIT_DST_RELEASE
|
1080 IFF_XMIT_DST_RELEASE_PERM
;
1081 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1082 netdev_features_t enc_features
= BOND_ENC_FEATURES
;
1083 struct net_device
*bond_dev
= bond
->dev
;
1084 struct list_head
*iter
;
1085 struct slave
*slave
;
1086 unsigned short max_hard_header_len
= ETH_HLEN
;
1087 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1088 u16 gso_max_segs
= GSO_MAX_SEGS
;
1090 if (!bond_has_slaves(bond
))
1092 vlan_features
&= NETIF_F_ALL_FOR_ALL
;
1094 bond_for_each_slave(bond
, slave
, iter
) {
1095 vlan_features
= netdev_increment_features(vlan_features
,
1096 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1098 enc_features
= netdev_increment_features(enc_features
,
1099 slave
->dev
->hw_enc_features
,
1101 dst_release_flag
&= slave
->dev
->priv_flags
;
1102 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1103 max_hard_header_len
= slave
->dev
->hard_header_len
;
1105 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1106 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1110 bond_dev
->vlan_features
= vlan_features
;
1111 bond_dev
->hw_enc_features
= enc_features
| NETIF_F_GSO_ENCAP_ALL
;
1112 bond_dev
->hard_header_len
= max_hard_header_len
;
1113 bond_dev
->gso_max_segs
= gso_max_segs
;
1114 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1116 bond_dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
1117 if ((bond_dev
->priv_flags
& IFF_XMIT_DST_RELEASE_PERM
) &&
1118 dst_release_flag
== (IFF_XMIT_DST_RELEASE
| IFF_XMIT_DST_RELEASE_PERM
))
1119 bond_dev
->priv_flags
|= IFF_XMIT_DST_RELEASE
;
1121 netdev_change_features(bond_dev
);
1124 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1125 struct net_device
*slave_dev
)
1127 bond_dev
->header_ops
= slave_dev
->header_ops
;
1129 bond_dev
->type
= slave_dev
->type
;
1130 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1131 bond_dev
->addr_len
= slave_dev
->addr_len
;
1133 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1134 slave_dev
->addr_len
);
1137 /* On bonding slaves other than the currently active slave, suppress
1138 * duplicates except for alb non-mcast/bcast.
1140 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1141 struct slave
*slave
,
1142 struct bonding
*bond
)
1144 if (bond_is_slave_inactive(slave
)) {
1145 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1146 skb
->pkt_type
!= PACKET_BROADCAST
&&
1147 skb
->pkt_type
!= PACKET_MULTICAST
)
1154 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1156 struct sk_buff
*skb
= *pskb
;
1157 struct slave
*slave
;
1158 struct bonding
*bond
;
1159 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1161 int ret
= RX_HANDLER_ANOTHER
;
1163 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1165 return RX_HANDLER_CONSUMED
;
1169 slave
= bond_slave_get_rcu(skb
->dev
);
1172 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1174 ret
= recv_probe(skb
, bond
, slave
);
1175 if (ret
== RX_HANDLER_CONSUMED
) {
1181 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1182 return RX_HANDLER_EXACT
;
1185 skb
->dev
= bond
->dev
;
1187 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1188 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1189 skb
->pkt_type
== PACKET_HOST
) {
1191 if (unlikely(skb_cow_head(skb
,
1192 skb
->data
- skb_mac_header(skb
)))) {
1194 return RX_HANDLER_CONSUMED
;
1196 ether_addr_copy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
);
1202 static enum netdev_lag_tx_type
bond_lag_tx_type(struct bonding
*bond
)
1204 switch (BOND_MODE(bond
)) {
1205 case BOND_MODE_ROUNDROBIN
:
1206 return NETDEV_LAG_TX_TYPE_ROUNDROBIN
;
1207 case BOND_MODE_ACTIVEBACKUP
:
1208 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP
;
1209 case BOND_MODE_BROADCAST
:
1210 return NETDEV_LAG_TX_TYPE_BROADCAST
;
1212 case BOND_MODE_8023AD
:
1213 return NETDEV_LAG_TX_TYPE_HASH
;
1215 return NETDEV_LAG_TX_TYPE_UNKNOWN
;
1219 static int bond_master_upper_dev_link(struct bonding
*bond
, struct slave
*slave
)
1221 struct netdev_lag_upper_info lag_upper_info
;
1224 lag_upper_info
.tx_type
= bond_lag_tx_type(bond
);
1225 err
= netdev_master_upper_dev_link(slave
->dev
, bond
->dev
, slave
,
1229 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, IFF_SLAVE
, GFP_KERNEL
);
1233 static void bond_upper_dev_unlink(struct bonding
*bond
, struct slave
*slave
)
1235 netdev_upper_dev_unlink(slave
->dev
, bond
->dev
);
1236 slave
->dev
->flags
&= ~IFF_SLAVE
;
1237 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, IFF_SLAVE
, GFP_KERNEL
);
1240 static struct slave
*bond_alloc_slave(struct bonding
*bond
)
1242 struct slave
*slave
= NULL
;
1244 slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1248 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1249 SLAVE_AD_INFO(slave
) = kzalloc(sizeof(struct ad_slave_info
),
1251 if (!SLAVE_AD_INFO(slave
)) {
1259 static void bond_free_slave(struct slave
*slave
)
1261 struct bonding
*bond
= bond_get_bond_by_slave(slave
);
1263 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1264 kfree(SLAVE_AD_INFO(slave
));
1269 static void bond_fill_ifbond(struct bonding
*bond
, struct ifbond
*info
)
1271 info
->bond_mode
= BOND_MODE(bond
);
1272 info
->miimon
= bond
->params
.miimon
;
1273 info
->num_slaves
= bond
->slave_cnt
;
1276 static void bond_fill_ifslave(struct slave
*slave
, struct ifslave
*info
)
1278 strcpy(info
->slave_name
, slave
->dev
->name
);
1279 info
->link
= slave
->link
;
1280 info
->state
= bond_slave_state(slave
);
1281 info
->link_failure_count
= slave
->link_failure_count
;
1284 static void bond_netdev_notify(struct net_device
*dev
,
1285 struct netdev_bonding_info
*info
)
1288 netdev_bonding_info_change(dev
, info
);
1292 static void bond_netdev_notify_work(struct work_struct
*_work
)
1294 struct netdev_notify_work
*w
=
1295 container_of(_work
, struct netdev_notify_work
, work
.work
);
1297 bond_netdev_notify(w
->dev
, &w
->bonding_info
);
1302 void bond_queue_slave_event(struct slave
*slave
)
1304 struct bonding
*bond
= slave
->bond
;
1305 struct netdev_notify_work
*nnw
= kzalloc(sizeof(*nnw
), GFP_ATOMIC
);
1310 dev_hold(slave
->dev
);
1311 nnw
->dev
= slave
->dev
;
1312 bond_fill_ifslave(slave
, &nnw
->bonding_info
.slave
);
1313 bond_fill_ifbond(bond
, &nnw
->bonding_info
.master
);
1314 INIT_DELAYED_WORK(&nnw
->work
, bond_netdev_notify_work
);
1316 queue_delayed_work(slave
->bond
->wq
, &nnw
->work
, 0);
1319 void bond_lower_state_changed(struct slave
*slave
)
1321 struct netdev_lag_lower_state_info info
;
1323 info
.link_up
= slave
->link
== BOND_LINK_UP
||
1324 slave
->link
== BOND_LINK_FAIL
;
1325 info
.tx_enabled
= bond_is_active_slave(slave
);
1326 netdev_lower_state_changed(slave
->dev
, &info
);
1329 /* enslave device <slave> to bond device <master> */
1330 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1332 struct bonding
*bond
= netdev_priv(bond_dev
);
1333 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1334 struct slave
*new_slave
= NULL
, *prev_slave
;
1335 struct sockaddr addr
;
1339 if (!bond
->params
.use_carrier
&&
1340 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1341 slave_ops
->ndo_do_ioctl
== NULL
) {
1342 netdev_warn(bond_dev
, "no link monitoring support for %s\n",
1346 /* already enslaved */
1347 if (slave_dev
->flags
& IFF_SLAVE
) {
1348 netdev_dbg(bond_dev
, "Error: Device was already enslaved\n");
1352 if (bond_dev
== slave_dev
) {
1353 netdev_err(bond_dev
, "cannot enslave bond to itself.\n");
1357 /* vlan challenged mutual exclusion */
1358 /* no need to lock since we're protected by rtnl_lock */
1359 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1360 netdev_dbg(bond_dev
, "%s is NETIF_F_VLAN_CHALLENGED\n",
1362 if (vlan_uses_dev(bond_dev
)) {
1363 netdev_err(bond_dev
, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1364 slave_dev
->name
, bond_dev
->name
);
1367 netdev_warn(bond_dev
, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1368 slave_dev
->name
, slave_dev
->name
,
1372 netdev_dbg(bond_dev
, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1376 /* Old ifenslave binaries are no longer supported. These can
1377 * be identified with moderate accuracy by the state of the slave:
1378 * the current ifenslave will set the interface down prior to
1379 * enslaving it; the old ifenslave will not.
1381 if (slave_dev
->flags
& IFF_UP
) {
1382 netdev_err(bond_dev
, "%s is up - this may be due to an out of date ifenslave\n",
1385 goto err_undo_flags
;
1388 /* set bonding device ether type by slave - bonding netdevices are
1389 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1390 * there is a need to override some of the type dependent attribs/funcs.
1392 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1393 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1395 if (!bond_has_slaves(bond
)) {
1396 if (bond_dev
->type
!= slave_dev
->type
) {
1397 netdev_dbg(bond_dev
, "change device type from %d to %d\n",
1398 bond_dev
->type
, slave_dev
->type
);
1400 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1402 res
= notifier_to_errno(res
);
1404 netdev_err(bond_dev
, "refused to change device type\n");
1406 goto err_undo_flags
;
1409 /* Flush unicast and multicast addresses */
1410 dev_uc_flush(bond_dev
);
1411 dev_mc_flush(bond_dev
);
1413 if (slave_dev
->type
!= ARPHRD_ETHER
)
1414 bond_setup_by_slave(bond_dev
, slave_dev
);
1416 ether_setup(bond_dev
);
1417 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1420 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1423 } else if (bond_dev
->type
!= slave_dev
->type
) {
1424 netdev_err(bond_dev
, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1425 slave_dev
->name
, slave_dev
->type
, bond_dev
->type
);
1427 goto err_undo_flags
;
1430 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1431 netdev_warn(bond_dev
, "The slave device specified does not support setting the MAC address\n");
1432 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
&&
1433 bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1434 if (!bond_has_slaves(bond
)) {
1435 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1436 netdev_warn(bond_dev
, "Setting fail_over_mac to active for active-backup mode\n");
1438 netdev_err(bond_dev
, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1440 goto err_undo_flags
;
1445 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1447 /* If this is the first slave, then we need to set the master's hardware
1448 * address to be the same as the slave's.
1450 if (!bond_has_slaves(bond
) &&
1451 bond
->dev
->addr_assign_type
== NET_ADDR_RANDOM
)
1452 bond_set_dev_addr(bond
->dev
, slave_dev
);
1454 new_slave
= bond_alloc_slave(bond
);
1457 goto err_undo_flags
;
1460 new_slave
->bond
= bond
;
1461 new_slave
->dev
= slave_dev
;
1462 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1463 * is set via sysfs or module option if desired.
1465 new_slave
->queue_id
= 0;
1467 /* Save slave's original mtu and then set it to match the bond */
1468 new_slave
->original_mtu
= slave_dev
->mtu
;
1469 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1471 netdev_dbg(bond_dev
, "Error %d calling dev_set_mtu\n", res
);
1475 /* Save slave's original ("permanent") mac address for modes
1476 * that need it, and for restoring it upon release, and then
1477 * set it to the master's address
1479 ether_addr_copy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
);
1481 if (!bond
->params
.fail_over_mac
||
1482 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1483 /* Set slave to master's mac address. The application already
1484 * set the master's mac address to that of the first slave
1486 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1487 addr
.sa_family
= slave_dev
->type
;
1488 res
= dev_set_mac_address(slave_dev
, &addr
);
1490 netdev_dbg(bond_dev
, "Error %d calling set_mac_address\n", res
);
1491 goto err_restore_mtu
;
1495 /* set slave flag before open to prevent IPv6 addrconf */
1496 slave_dev
->flags
|= IFF_SLAVE
;
1498 /* open the slave since the application closed it */
1499 res
= dev_open(slave_dev
);
1501 netdev_dbg(bond_dev
, "Opening slave %s failed\n", slave_dev
->name
);
1502 goto err_restore_mac
;
1505 slave_dev
->priv_flags
|= IFF_BONDING
;
1506 /* initialize slave stats */
1507 dev_get_stats(new_slave
->dev
, &new_slave
->slave_stats
);
1509 if (bond_is_lb(bond
)) {
1510 /* bond_alb_init_slave() must be called before all other stages since
1511 * it might fail and we do not want to have to undo everything
1513 res
= bond_alb_init_slave(bond
, new_slave
);
1518 /* If the mode uses primary, then the following is handled by
1519 * bond_change_active_slave().
1521 if (!bond_uses_primary(bond
)) {
1522 /* set promiscuity level to new slave */
1523 if (bond_dev
->flags
& IFF_PROMISC
) {
1524 res
= dev_set_promiscuity(slave_dev
, 1);
1529 /* set allmulti level to new slave */
1530 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1531 res
= dev_set_allmulti(slave_dev
, 1);
1536 netif_addr_lock_bh(bond_dev
);
1538 dev_mc_sync_multiple(slave_dev
, bond_dev
);
1539 dev_uc_sync_multiple(slave_dev
, bond_dev
);
1541 netif_addr_unlock_bh(bond_dev
);
1544 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1545 /* add lacpdu mc addr to mc list */
1546 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1548 dev_mc_add(slave_dev
, lacpdu_multicast
);
1551 res
= vlan_vids_add_by_dev(slave_dev
, bond_dev
);
1553 netdev_err(bond_dev
, "Couldn't add bond vlan ids to %s\n",
1558 prev_slave
= bond_last_slave(bond
);
1560 new_slave
->delay
= 0;
1561 new_slave
->link_failure_count
= 0;
1563 bond_update_speed_duplex(new_slave
);
1565 new_slave
->last_rx
= jiffies
-
1566 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1567 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
1568 new_slave
->target_last_arp_rx
[i
] = new_slave
->last_rx
;
1570 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1571 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1573 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1574 /* miimon is set but a bonded network driver
1575 * does not support ETHTOOL/MII and
1576 * arp_interval is not set. Note: if
1577 * use_carrier is enabled, we will never go
1578 * here (because netif_carrier is always
1579 * supported); thus, we don't need to change
1580 * the messages for netif_carrier.
1582 netdev_warn(bond_dev
, "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",
1584 } else if (link_reporting
== -1) {
1585 /* unable get link status using mii/ethtool */
1586 netdev_warn(bond_dev
, "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",
1591 /* check for initial state */
1592 if (bond
->params
.miimon
) {
1593 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1594 if (bond
->params
.updelay
) {
1595 bond_set_slave_link_state(new_slave
,
1597 BOND_SLAVE_NOTIFY_NOW
);
1598 new_slave
->delay
= bond
->params
.updelay
;
1600 bond_set_slave_link_state(new_slave
,
1602 BOND_SLAVE_NOTIFY_NOW
);
1605 bond_set_slave_link_state(new_slave
, BOND_LINK_DOWN
,
1606 BOND_SLAVE_NOTIFY_NOW
);
1608 } else if (bond
->params
.arp_interval
) {
1609 bond_set_slave_link_state(new_slave
,
1610 (netif_carrier_ok(slave_dev
) ?
1611 BOND_LINK_UP
: BOND_LINK_DOWN
),
1612 BOND_SLAVE_NOTIFY_NOW
);
1614 bond_set_slave_link_state(new_slave
, BOND_LINK_UP
,
1615 BOND_SLAVE_NOTIFY_NOW
);
1618 if (new_slave
->link
!= BOND_LINK_DOWN
)
1619 new_slave
->last_link_up
= jiffies
;
1620 netdev_dbg(bond_dev
, "Initial state of slave_dev is BOND_LINK_%s\n",
1621 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1622 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1624 if (bond_uses_primary(bond
) && bond
->params
.primary
[0]) {
1625 /* if there is a primary slave, remember it */
1626 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1627 rcu_assign_pointer(bond
->primary_slave
, new_slave
);
1628 bond
->force_primary
= true;
1632 switch (BOND_MODE(bond
)) {
1633 case BOND_MODE_ACTIVEBACKUP
:
1634 bond_set_slave_inactive_flags(new_slave
,
1635 BOND_SLAVE_NOTIFY_NOW
);
1637 case BOND_MODE_8023AD
:
1638 /* in 802.3ad mode, the internal mechanism
1639 * will activate the slaves in the selected
1642 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1643 /* if this is the first slave */
1645 SLAVE_AD_INFO(new_slave
)->id
= 1;
1646 /* Initialize AD with the number of times that the AD timer is called in 1 second
1647 * can be called only after the mac address of the bond is set
1649 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1651 SLAVE_AD_INFO(new_slave
)->id
=
1652 SLAVE_AD_INFO(prev_slave
)->id
+ 1;
1655 bond_3ad_bind_slave(new_slave
);
1659 bond_set_active_slave(new_slave
);
1660 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1663 netdev_dbg(bond_dev
, "This slave is always active in trunk mode\n");
1665 /* always active in trunk mode */
1666 bond_set_active_slave(new_slave
);
1668 /* In trunking mode there is little meaning to curr_active_slave
1669 * anyway (it holds no special properties of the bond device),
1670 * so we can change it without calling change_active_interface()
1672 if (!rcu_access_pointer(bond
->curr_active_slave
) &&
1673 new_slave
->link
== BOND_LINK_UP
)
1674 rcu_assign_pointer(bond
->curr_active_slave
, new_slave
);
1677 } /* switch(bond_mode) */
1679 #ifdef CONFIG_NET_POLL_CONTROLLER
1680 slave_dev
->npinfo
= bond
->dev
->npinfo
;
1681 if (slave_dev
->npinfo
) {
1682 if (slave_enable_netpoll(new_slave
)) {
1683 netdev_info(bond_dev
, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1690 if (!(bond_dev
->features
& NETIF_F_LRO
))
1691 dev_disable_lro(slave_dev
);
1693 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1696 netdev_dbg(bond_dev
, "Error %d calling netdev_rx_handler_register\n", res
);
1700 res
= bond_master_upper_dev_link(bond
, new_slave
);
1702 netdev_dbg(bond_dev
, "Error %d calling bond_master_upper_dev_link\n", res
);
1703 goto err_unregister
;
1706 res
= bond_sysfs_slave_add(new_slave
);
1708 netdev_dbg(bond_dev
, "Error %d calling bond_sysfs_slave_add\n", res
);
1709 goto err_upper_unlink
;
1713 bond_compute_features(bond
);
1714 bond_set_carrier(bond
);
1716 if (bond_uses_primary(bond
)) {
1718 bond_select_active_slave(bond
);
1719 unblock_netpoll_tx();
1722 if (bond_mode_uses_xmit_hash(bond
))
1723 bond_update_slave_arr(bond
, NULL
);
1725 netdev_info(bond_dev
, "Enslaving %s as %s interface with %s link\n",
1727 bond_is_active_slave(new_slave
) ? "an active" : "a backup",
1728 new_slave
->link
!= BOND_LINK_DOWN
? "an up" : "a down");
1730 /* enslave is successful */
1731 bond_queue_slave_event(new_slave
);
1734 /* Undo stages on error */
1736 bond_upper_dev_unlink(bond
, new_slave
);
1739 netdev_rx_handler_unregister(slave_dev
);
1742 if (!bond_uses_primary(bond
))
1743 bond_hw_addr_flush(bond_dev
, slave_dev
);
1745 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1746 if (rcu_access_pointer(bond
->primary_slave
) == new_slave
)
1747 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
1748 if (rcu_access_pointer(bond
->curr_active_slave
) == new_slave
) {
1750 bond_change_active_slave(bond
, NULL
);
1751 bond_select_active_slave(bond
);
1752 unblock_netpoll_tx();
1754 /* either primary_slave or curr_active_slave might've changed */
1756 slave_disable_netpoll(new_slave
);
1759 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1760 dev_close(slave_dev
);
1763 slave_dev
->flags
&= ~IFF_SLAVE
;
1764 if (!bond
->params
.fail_over_mac
||
1765 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1766 /* XXX TODO - fom follow mode needs to change master's
1767 * MAC if this slave's MAC is in use by the bond, or at
1768 * least print a warning.
1770 ether_addr_copy(addr
.sa_data
, new_slave
->perm_hwaddr
);
1771 addr
.sa_family
= slave_dev
->type
;
1772 dev_set_mac_address(slave_dev
, &addr
);
1776 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1779 bond_free_slave(new_slave
);
1782 /* Enslave of first slave has failed and we need to fix master's mac */
1783 if (!bond_has_slaves(bond
)) {
1784 if (ether_addr_equal_64bits(bond_dev
->dev_addr
,
1785 slave_dev
->dev_addr
))
1786 eth_hw_addr_random(bond_dev
);
1787 if (bond_dev
->type
!= ARPHRD_ETHER
) {
1788 dev_close(bond_dev
);
1789 ether_setup(bond_dev
);
1790 bond_dev
->flags
|= IFF_MASTER
;
1791 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1798 /* Try to release the slave device <slave> from the bond device <master>
1799 * It is legal to access curr_active_slave without a lock because all the function
1800 * is RTNL-locked. If "all" is true it means that the function is being called
1801 * while destroying a bond interface and all slaves are being released.
1803 * The rules for slave state should be:
1804 * for Active/Backup:
1805 * Active stays on all backups go down
1806 * for Bonded connections:
1807 * The first up interface should be left on and all others downed.
1809 static int __bond_release_one(struct net_device
*bond_dev
,
1810 struct net_device
*slave_dev
,
1813 struct bonding
*bond
= netdev_priv(bond_dev
);
1814 struct slave
*slave
, *oldcurrent
;
1815 struct sockaddr addr
;
1816 int old_flags
= bond_dev
->flags
;
1817 netdev_features_t old_features
= bond_dev
->features
;
1819 /* slave is not a slave or master is not master of this slave */
1820 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1821 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1822 netdev_dbg(bond_dev
, "cannot release %s\n",
1829 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1831 /* not a slave of this bond */
1832 netdev_info(bond_dev
, "%s not enslaved\n",
1834 unblock_netpoll_tx();
1838 bond_set_slave_inactive_flags(slave
, BOND_SLAVE_NOTIFY_NOW
);
1840 bond_sysfs_slave_del(slave
);
1842 /* recompute stats just before removing the slave */
1843 bond_get_stats(bond
->dev
, &bond
->bond_stats
);
1845 bond_upper_dev_unlink(bond
, slave
);
1846 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1847 * for this slave anymore.
1849 netdev_rx_handler_unregister(slave_dev
);
1851 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1852 bond_3ad_unbind_slave(slave
);
1854 if (bond_mode_uses_xmit_hash(bond
))
1855 bond_update_slave_arr(bond
, slave
);
1857 netdev_info(bond_dev
, "Releasing %s interface %s\n",
1858 bond_is_active_slave(slave
) ? "active" : "backup",
1861 oldcurrent
= rcu_access_pointer(bond
->curr_active_slave
);
1863 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
1865 if (!all
&& (!bond
->params
.fail_over_mac
||
1866 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
)) {
1867 if (ether_addr_equal_64bits(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1868 bond_has_slaves(bond
))
1869 netdev_warn(bond_dev
, "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",
1870 slave_dev
->name
, slave
->perm_hwaddr
,
1871 bond_dev
->name
, slave_dev
->name
);
1874 if (rtnl_dereference(bond
->primary_slave
) == slave
)
1875 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
1877 if (oldcurrent
== slave
)
1878 bond_change_active_slave(bond
, NULL
);
1880 if (bond_is_lb(bond
)) {
1881 /* Must be called only after the slave has been
1882 * detached from the list and the curr_active_slave
1883 * has been cleared (if our_slave == old_current),
1884 * but before a new active slave is selected.
1886 bond_alb_deinit_slave(bond
, slave
);
1890 RCU_INIT_POINTER(bond
->curr_active_slave
, NULL
);
1891 } else if (oldcurrent
== slave
) {
1892 /* Note that we hold RTNL over this sequence, so there
1893 * is no concern that another slave add/remove event
1896 bond_select_active_slave(bond
);
1899 if (!bond_has_slaves(bond
)) {
1900 bond_set_carrier(bond
);
1901 eth_hw_addr_random(bond_dev
);
1904 unblock_netpoll_tx();
1908 if (!bond_has_slaves(bond
)) {
1909 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
1910 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
1913 bond_compute_features(bond
);
1914 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1915 (old_features
& NETIF_F_VLAN_CHALLENGED
))
1916 netdev_info(bond_dev
, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1917 slave_dev
->name
, bond_dev
->name
);
1919 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1921 /* If the mode uses primary, then this case was handled above by
1922 * bond_change_active_slave(..., NULL)
1924 if (!bond_uses_primary(bond
)) {
1925 /* unset promiscuity level from slave
1926 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1927 * of the IFF_PROMISC flag in the bond_dev, but we need the
1928 * value of that flag before that change, as that was the value
1929 * when this slave was attached, so we cache at the start of the
1930 * function and use it here. Same goes for ALLMULTI below
1932 if (old_flags
& IFF_PROMISC
)
1933 dev_set_promiscuity(slave_dev
, -1);
1935 /* unset allmulti level from slave */
1936 if (old_flags
& IFF_ALLMULTI
)
1937 dev_set_allmulti(slave_dev
, -1);
1939 bond_hw_addr_flush(bond_dev
, slave_dev
);
1942 slave_disable_netpoll(slave
);
1944 /* close slave before restoring its mac address */
1945 dev_close(slave_dev
);
1947 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
||
1948 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1949 /* restore original ("permanent") mac address */
1950 ether_addr_copy(addr
.sa_data
, slave
->perm_hwaddr
);
1951 addr
.sa_family
= slave_dev
->type
;
1952 dev_set_mac_address(slave_dev
, &addr
);
1955 dev_set_mtu(slave_dev
, slave
->original_mtu
);
1957 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1959 bond_free_slave(slave
);
1964 /* A wrapper used because of ndo_del_link */
1965 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1967 return __bond_release_one(bond_dev
, slave_dev
, false);
1970 /* First release a slave and then destroy the bond if no more slaves are left.
1971 * Must be under rtnl_lock when this function is called.
1973 static int bond_release_and_destroy(struct net_device
*bond_dev
,
1974 struct net_device
*slave_dev
)
1976 struct bonding
*bond
= netdev_priv(bond_dev
);
1979 ret
= bond_release(bond_dev
, slave_dev
);
1980 if (ret
== 0 && !bond_has_slaves(bond
)) {
1981 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1982 netdev_info(bond_dev
, "Destroying bond %s\n",
1984 bond_remove_proc_entry(bond
);
1985 unregister_netdevice(bond_dev
);
1990 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1992 struct bonding
*bond
= netdev_priv(bond_dev
);
1993 bond_fill_ifbond(bond
, info
);
1997 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
1999 struct bonding
*bond
= netdev_priv(bond_dev
);
2000 struct list_head
*iter
;
2001 int i
= 0, res
= -ENODEV
;
2002 struct slave
*slave
;
2004 bond_for_each_slave(bond
, slave
, iter
) {
2005 if (i
++ == (int)info
->slave_id
) {
2007 bond_fill_ifslave(slave
, info
);
2015 /*-------------------------------- Monitoring -------------------------------*/
2017 /* called with rcu_read_lock() */
2018 static int bond_miimon_inspect(struct bonding
*bond
)
2020 int link_state
, commit
= 0;
2021 struct list_head
*iter
;
2022 struct slave
*slave
;
2023 bool ignore_updelay
;
2025 ignore_updelay
= !rcu_dereference(bond
->curr_active_slave
);
2027 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2028 slave
->new_link
= BOND_LINK_NOCHANGE
;
2030 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2032 switch (slave
->link
) {
2037 bond_set_slave_link_state(slave
, BOND_LINK_FAIL
,
2038 BOND_SLAVE_NOTIFY_LATER
);
2039 slave
->delay
= bond
->params
.downdelay
;
2041 netdev_info(bond
->dev
, "link status down for %sinterface %s, disabling it in %d ms\n",
2043 BOND_MODE_ACTIVEBACKUP
) ?
2044 (bond_is_active_slave(slave
) ?
2045 "active " : "backup ") : "",
2047 bond
->params
.downdelay
* bond
->params
.miimon
);
2050 case BOND_LINK_FAIL
:
2052 /* recovered before downdelay expired */
2053 bond_set_slave_link_state(slave
, BOND_LINK_UP
,
2054 BOND_SLAVE_NOTIFY_LATER
);
2055 slave
->last_link_up
= jiffies
;
2056 netdev_info(bond
->dev
, "link status up again after %d ms for interface %s\n",
2057 (bond
->params
.downdelay
- slave
->delay
) *
2058 bond
->params
.miimon
,
2063 if (slave
->delay
<= 0) {
2064 slave
->new_link
= BOND_LINK_DOWN
;
2072 case BOND_LINK_DOWN
:
2076 bond_set_slave_link_state(slave
, BOND_LINK_BACK
,
2077 BOND_SLAVE_NOTIFY_LATER
);
2078 slave
->delay
= bond
->params
.updelay
;
2081 netdev_info(bond
->dev
, "link status up for interface %s, enabling it in %d ms\n",
2083 ignore_updelay
? 0 :
2084 bond
->params
.updelay
*
2085 bond
->params
.miimon
);
2088 case BOND_LINK_BACK
:
2090 bond_set_slave_link_state(slave
,
2092 BOND_SLAVE_NOTIFY_LATER
);
2093 netdev_info(bond
->dev
, "link status down again after %d ms for interface %s\n",
2094 (bond
->params
.updelay
- slave
->delay
) *
2095 bond
->params
.miimon
,
2104 if (slave
->delay
<= 0) {
2105 slave
->new_link
= BOND_LINK_UP
;
2107 ignore_updelay
= false;
2119 static void bond_miimon_commit(struct bonding
*bond
)
2121 struct list_head
*iter
;
2122 struct slave
*slave
, *primary
;
2124 bond_for_each_slave(bond
, slave
, iter
) {
2125 switch (slave
->new_link
) {
2126 case BOND_LINK_NOCHANGE
:
2130 bond_set_slave_link_state(slave
, BOND_LINK_UP
,
2131 BOND_SLAVE_NOTIFY_NOW
);
2132 slave
->last_link_up
= jiffies
;
2134 primary
= rtnl_dereference(bond
->primary_slave
);
2135 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
2136 /* prevent it from being the active one */
2137 bond_set_backup_slave(slave
);
2138 } else if (BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
2139 /* make it immediately active */
2140 bond_set_active_slave(slave
);
2141 } else if (slave
!= primary
) {
2142 /* prevent it from being the active one */
2143 bond_set_backup_slave(slave
);
2146 netdev_info(bond
->dev
, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2148 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
2149 slave
->duplex
? "full" : "half");
2151 /* notify ad that the link status has changed */
2152 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
2153 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2155 if (bond_is_lb(bond
))
2156 bond_alb_handle_link_change(bond
, slave
,
2159 if (BOND_MODE(bond
) == BOND_MODE_XOR
)
2160 bond_update_slave_arr(bond
, NULL
);
2162 if (!bond
->curr_active_slave
|| slave
== primary
)
2167 case BOND_LINK_DOWN
:
2168 if (slave
->link_failure_count
< UINT_MAX
)
2169 slave
->link_failure_count
++;
2171 bond_set_slave_link_state(slave
, BOND_LINK_DOWN
,
2172 BOND_SLAVE_NOTIFY_NOW
);
2174 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
2175 BOND_MODE(bond
) == BOND_MODE_8023AD
)
2176 bond_set_slave_inactive_flags(slave
,
2177 BOND_SLAVE_NOTIFY_NOW
);
2179 netdev_info(bond
->dev
, "link status definitely down for interface %s, disabling it\n",
2182 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
2183 bond_3ad_handle_link_change(slave
,
2186 if (bond_is_lb(bond
))
2187 bond_alb_handle_link_change(bond
, slave
,
2190 if (BOND_MODE(bond
) == BOND_MODE_XOR
)
2191 bond_update_slave_arr(bond
, NULL
);
2193 if (slave
== rcu_access_pointer(bond
->curr_active_slave
))
2199 netdev_err(bond
->dev
, "invalid new link %d on slave %s\n",
2200 slave
->new_link
, slave
->dev
->name
);
2201 slave
->new_link
= BOND_LINK_NOCHANGE
;
2208 bond_select_active_slave(bond
);
2209 unblock_netpoll_tx();
2212 bond_set_carrier(bond
);
2217 * Really a wrapper that splits the mii monitor into two phases: an
2218 * inspection, then (if inspection indicates something needs to be done)
2219 * an acquisition of appropriate locks followed by a commit phase to
2220 * implement whatever link state changes are indicated.
2222 static void bond_mii_monitor(struct work_struct
*work
)
2224 struct bonding
*bond
= container_of(work
, struct bonding
,
2226 bool should_notify_peers
= false;
2227 unsigned long delay
;
2229 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2231 if (!bond_has_slaves(bond
))
2236 should_notify_peers
= bond_should_notify_peers(bond
);
2238 if (bond_miimon_inspect(bond
)) {
2241 /* Race avoidance with bond_close cancel of workqueue */
2242 if (!rtnl_trylock()) {
2244 should_notify_peers
= false;
2248 bond_miimon_commit(bond
);
2250 rtnl_unlock(); /* might sleep, hold no other locks */
2255 if (bond
->params
.miimon
)
2256 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2258 if (should_notify_peers
) {
2259 if (!rtnl_trylock())
2261 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2266 static bool bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2268 struct net_device
*upper
;
2269 struct list_head
*iter
;
2272 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2276 netdev_for_each_all_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2277 if (ip
== bond_confirm_addr(upper
, 0, ip
)) {
2287 /* We go to the (large) trouble of VLAN tagging ARP frames because
2288 * switches in VLAN mode (especially if ports are configured as
2289 * "native" to a VLAN) might not pass non-tagged frames.
2291 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
,
2292 __be32 dest_ip
, __be32 src_ip
,
2293 struct bond_vlan_tag
*tags
)
2295 struct sk_buff
*skb
;
2296 struct bond_vlan_tag
*outer_tag
= tags
;
2298 netdev_dbg(slave_dev
, "arp %d on slave %s: dst %pI4 src %pI4\n",
2299 arp_op
, slave_dev
->name
, &dest_ip
, &src_ip
);
2301 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2302 NULL
, slave_dev
->dev_addr
, NULL
);
2305 net_err_ratelimited("ARP packet allocation failed\n");
2309 if (!tags
|| tags
->vlan_proto
== VLAN_N_VID
)
2314 /* Go through all the tags backwards and add them to the packet */
2315 while (tags
->vlan_proto
!= VLAN_N_VID
) {
2316 if (!tags
->vlan_id
) {
2321 netdev_dbg(slave_dev
, "inner tag: proto %X vid %X\n",
2322 ntohs(outer_tag
->vlan_proto
), tags
->vlan_id
);
2323 skb
= vlan_insert_tag_set_proto(skb
, tags
->vlan_proto
,
2326 net_err_ratelimited("failed to insert inner VLAN tag\n");
2332 /* Set the outer tag */
2333 if (outer_tag
->vlan_id
) {
2334 netdev_dbg(slave_dev
, "outer tag: proto %X vid %X\n",
2335 ntohs(outer_tag
->vlan_proto
), outer_tag
->vlan_id
);
2336 __vlan_hwaccel_put_tag(skb
, outer_tag
->vlan_proto
,
2337 outer_tag
->vlan_id
);
2344 /* Validate the device path between the @start_dev and the @end_dev.
2345 * The path is valid if the @end_dev is reachable through device
2347 * When the path is validated, collect any vlan information in the
2350 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
2351 struct net_device
*end_dev
,
2354 struct bond_vlan_tag
*tags
;
2355 struct net_device
*upper
;
2356 struct list_head
*iter
;
2358 if (start_dev
== end_dev
) {
2359 tags
= kzalloc(sizeof(*tags
) * (level
+ 1), GFP_ATOMIC
);
2361 return ERR_PTR(-ENOMEM
);
2362 tags
[level
].vlan_proto
= VLAN_N_VID
;
2366 netdev_for_each_upper_dev_rcu(start_dev
, upper
, iter
) {
2367 tags
= bond_verify_device_path(upper
, end_dev
, level
+ 1);
2368 if (IS_ERR_OR_NULL(tags
)) {
2373 if (is_vlan_dev(upper
)) {
2374 tags
[level
].vlan_proto
= vlan_dev_vlan_proto(upper
);
2375 tags
[level
].vlan_id
= vlan_dev_vlan_id(upper
);
2384 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2387 struct bond_vlan_tag
*tags
;
2388 __be32
*targets
= bond
->params
.arp_targets
, addr
;
2391 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
&& targets
[i
]; i
++) {
2392 netdev_dbg(bond
->dev
, "basa: target %pI4\n", &targets
[i
]);
2395 /* Find out through which dev should the packet go */
2396 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2399 /* there's no route to target - try to send arp
2400 * probe to generate any traffic (arp_validate=0)
2402 if (bond
->params
.arp_validate
)
2403 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2406 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2411 /* bond device itself */
2412 if (rt
->dst
.dev
== bond
->dev
)
2416 tags
= bond_verify_device_path(bond
->dev
, rt
->dst
.dev
, 0);
2419 if (!IS_ERR_OR_NULL(tags
))
2422 /* Not our device - skip */
2423 netdev_dbg(bond
->dev
, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2424 &targets
[i
], rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2430 addr
= bond_confirm_addr(rt
->dst
.dev
, targets
[i
], 0);
2432 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2438 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2442 if (!sip
|| !bond_has_this_ip(bond
, tip
)) {
2443 netdev_dbg(bond
->dev
, "bva: sip %pI4 tip %pI4 not found\n",
2448 i
= bond_get_targets_ip(bond
->params
.arp_targets
, sip
);
2450 netdev_dbg(bond
->dev
, "bva: sip %pI4 not found in targets\n",
2454 slave
->last_rx
= jiffies
;
2455 slave
->target_last_arp_rx
[i
] = jiffies
;
2458 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2459 struct slave
*slave
)
2461 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2462 struct slave
*curr_active_slave
;
2463 unsigned char *arp_ptr
;
2465 int alen
, is_arp
= skb
->protocol
== __cpu_to_be16(ETH_P_ARP
);
2467 if (!slave_do_arp_validate(bond
, slave
)) {
2468 if ((slave_do_arp_validate_only(bond
) && is_arp
) ||
2469 !slave_do_arp_validate_only(bond
))
2470 slave
->last_rx
= jiffies
;
2471 return RX_HANDLER_ANOTHER
;
2472 } else if (!is_arp
) {
2473 return RX_HANDLER_ANOTHER
;
2476 alen
= arp_hdr_len(bond
->dev
);
2478 netdev_dbg(bond
->dev
, "bond_arp_rcv: skb->dev %s\n",
2481 if (alen
> skb_headlen(skb
)) {
2482 arp
= kmalloc(alen
, GFP_ATOMIC
);
2485 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2489 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2490 skb
->pkt_type
== PACKET_OTHERHOST
||
2491 skb
->pkt_type
== PACKET_LOOPBACK
||
2492 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2493 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2497 arp_ptr
= (unsigned char *)(arp
+ 1);
2498 arp_ptr
+= bond
->dev
->addr_len
;
2499 memcpy(&sip
, arp_ptr
, 4);
2500 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2501 memcpy(&tip
, arp_ptr
, 4);
2503 netdev_dbg(bond
->dev
, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2504 slave
->dev
->name
, bond_slave_state(slave
),
2505 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2508 curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2510 /* Backup slaves won't see the ARP reply, but do come through
2511 * here for each ARP probe (so we swap the sip/tip to validate
2512 * the probe). In a "redundant switch, common router" type of
2513 * configuration, the ARP probe will (hopefully) travel from
2514 * the active, through one switch, the router, then the other
2515 * switch before reaching the backup.
2517 * We 'trust' the arp requests if there is an active slave and
2518 * it received valid arp reply(s) after it became active. This
2519 * is done to avoid endless looping when we can't reach the
2520 * arp_ip_target and fool ourselves with our own arp requests.
2523 if (bond_is_active_slave(slave
))
2524 bond_validate_arp(bond
, slave
, sip
, tip
);
2525 else if (curr_active_slave
&&
2526 time_after(slave_last_rx(bond
, curr_active_slave
),
2527 curr_active_slave
->last_link_up
))
2528 bond_validate_arp(bond
, slave
, tip
, sip
);
2531 if (arp
!= (struct arphdr
*)skb
->data
)
2533 return RX_HANDLER_ANOTHER
;
2536 /* function to verify if we're in the arp_interval timeslice, returns true if
2537 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2538 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2540 static bool bond_time_in_interval(struct bonding
*bond
, unsigned long last_act
,
2543 int delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2545 return time_in_range(jiffies
,
2546 last_act
- delta_in_ticks
,
2547 last_act
+ mod
* delta_in_ticks
+ delta_in_ticks
/2);
2550 /* This function is called regularly to monitor each slave's link
2551 * ensuring that traffic is being sent and received when arp monitoring
2552 * is used in load-balancing mode. if the adapter has been dormant, then an
2553 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2554 * arp monitoring in active backup mode.
2556 static void bond_loadbalance_arp_mon(struct work_struct
*work
)
2558 struct bonding
*bond
= container_of(work
, struct bonding
,
2560 struct slave
*slave
, *oldcurrent
;
2561 struct list_head
*iter
;
2562 int do_failover
= 0, slave_state_changed
= 0;
2564 if (!bond_has_slaves(bond
))
2569 oldcurrent
= rcu_dereference(bond
->curr_active_slave
);
2570 /* see if any of the previous devices are up now (i.e. they have
2571 * xmt and rcv traffic). the curr_active_slave does not come into
2572 * the picture unless it is null. also, slave->last_link_up is not
2573 * needed here because we send an arp on each slave and give a slave
2574 * as long as it needs to get the tx/rx within the delta.
2575 * TODO: what about up/down delay in arp mode? it wasn't here before
2578 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2579 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2581 if (slave
->link
!= BOND_LINK_UP
) {
2582 if (bond_time_in_interval(bond
, trans_start
, 1) &&
2583 bond_time_in_interval(bond
, slave
->last_rx
, 1)) {
2585 slave
->link
= BOND_LINK_UP
;
2586 slave_state_changed
= 1;
2588 /* primary_slave has no meaning in round-robin
2589 * mode. the window of a slave being up and
2590 * curr_active_slave being null after enslaving
2594 netdev_info(bond
->dev
, "link status definitely up for interface %s\n",
2598 netdev_info(bond
->dev
, "interface %s is now up\n",
2603 /* slave->link == BOND_LINK_UP */
2605 /* not all switches will respond to an arp request
2606 * when the source ip is 0, so don't take the link down
2607 * if we don't know our ip yet
2609 if (!bond_time_in_interval(bond
, trans_start
, 2) ||
2610 !bond_time_in_interval(bond
, slave
->last_rx
, 2)) {
2612 slave
->link
= BOND_LINK_DOWN
;
2613 slave_state_changed
= 1;
2615 if (slave
->link_failure_count
< UINT_MAX
)
2616 slave
->link_failure_count
++;
2618 netdev_info(bond
->dev
, "interface %s is now down\n",
2621 if (slave
== oldcurrent
)
2626 /* note: if switch is in round-robin mode, all links
2627 * must tx arp to ensure all links rx an arp - otherwise
2628 * links may oscillate or not come up at all; if switch is
2629 * in something like xor mode, there is nothing we can
2630 * do - all replies will be rx'ed on same link causing slaves
2631 * to be unstable during low/no traffic periods
2633 if (bond_slave_is_up(slave
))
2634 bond_arp_send_all(bond
, slave
);
2639 if (do_failover
|| slave_state_changed
) {
2640 if (!rtnl_trylock())
2643 if (slave_state_changed
) {
2644 bond_slave_state_change(bond
);
2645 if (BOND_MODE(bond
) == BOND_MODE_XOR
)
2646 bond_update_slave_arr(bond
, NULL
);
2650 bond_select_active_slave(bond
);
2651 unblock_netpoll_tx();
2657 if (bond
->params
.arp_interval
)
2658 queue_delayed_work(bond
->wq
, &bond
->arp_work
,
2659 msecs_to_jiffies(bond
->params
.arp_interval
));
2662 /* Called to inspect slaves for active-backup mode ARP monitor link state
2663 * changes. Sets new_link in slaves to specify what action should take
2664 * place for the slave. Returns 0 if no changes are found, >0 if changes
2665 * to link states must be committed.
2667 * Called with rcu_read_lock held.
2669 static int bond_ab_arp_inspect(struct bonding
*bond
)
2671 unsigned long trans_start
, last_rx
;
2672 struct list_head
*iter
;
2673 struct slave
*slave
;
2676 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2677 slave
->new_link
= BOND_LINK_NOCHANGE
;
2678 last_rx
= slave_last_rx(bond
, slave
);
2680 if (slave
->link
!= BOND_LINK_UP
) {
2681 if (bond_time_in_interval(bond
, last_rx
, 1)) {
2682 slave
->new_link
= BOND_LINK_UP
;
2688 /* Give slaves 2*delta after being enslaved or made
2689 * active. This avoids bouncing, as the last receive
2690 * times need a full ARP monitor cycle to be updated.
2692 if (bond_time_in_interval(bond
, slave
->last_link_up
, 2))
2695 /* Backup slave is down if:
2696 * - No current_arp_slave AND
2697 * - more than 3*delta since last receive AND
2698 * - the bond has an IP address
2700 * Note: a non-null current_arp_slave indicates
2701 * the curr_active_slave went down and we are
2702 * searching for a new one; under this condition
2703 * we only take the curr_active_slave down - this
2704 * gives each slave a chance to tx/rx traffic
2705 * before being taken out
2707 if (!bond_is_active_slave(slave
) &&
2708 !rcu_access_pointer(bond
->current_arp_slave
) &&
2709 !bond_time_in_interval(bond
, last_rx
, 3)) {
2710 slave
->new_link
= BOND_LINK_DOWN
;
2714 /* Active slave is down if:
2715 * - more than 2*delta since transmitting OR
2716 * - (more than 2*delta since receive AND
2717 * the bond has an IP address)
2719 trans_start
= dev_trans_start(slave
->dev
);
2720 if (bond_is_active_slave(slave
) &&
2721 (!bond_time_in_interval(bond
, trans_start
, 2) ||
2722 !bond_time_in_interval(bond
, last_rx
, 2))) {
2723 slave
->new_link
= BOND_LINK_DOWN
;
2731 /* Called to commit link state changes noted by inspection step of
2732 * active-backup mode ARP monitor.
2734 * Called with RTNL hold.
2736 static void bond_ab_arp_commit(struct bonding
*bond
)
2738 unsigned long trans_start
;
2739 struct list_head
*iter
;
2740 struct slave
*slave
;
2742 bond_for_each_slave(bond
, slave
, iter
) {
2743 switch (slave
->new_link
) {
2744 case BOND_LINK_NOCHANGE
:
2748 trans_start
= dev_trans_start(slave
->dev
);
2749 if (rtnl_dereference(bond
->curr_active_slave
) != slave
||
2750 (!rtnl_dereference(bond
->curr_active_slave
) &&
2751 bond_time_in_interval(bond
, trans_start
, 1))) {
2752 struct slave
*current_arp_slave
;
2754 current_arp_slave
= rtnl_dereference(bond
->current_arp_slave
);
2755 bond_set_slave_link_state(slave
, BOND_LINK_UP
,
2756 BOND_SLAVE_NOTIFY_NOW
);
2757 if (current_arp_slave
) {
2758 bond_set_slave_inactive_flags(
2760 BOND_SLAVE_NOTIFY_NOW
);
2761 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
2764 netdev_info(bond
->dev
, "link status definitely up for interface %s\n",
2767 if (!rtnl_dereference(bond
->curr_active_slave
) ||
2768 slave
== rtnl_dereference(bond
->primary_slave
))
2775 case BOND_LINK_DOWN
:
2776 if (slave
->link_failure_count
< UINT_MAX
)
2777 slave
->link_failure_count
++;
2779 bond_set_slave_link_state(slave
, BOND_LINK_DOWN
,
2780 BOND_SLAVE_NOTIFY_NOW
);
2781 bond_set_slave_inactive_flags(slave
,
2782 BOND_SLAVE_NOTIFY_NOW
);
2784 netdev_info(bond
->dev
, "link status definitely down for interface %s, disabling it\n",
2787 if (slave
== rtnl_dereference(bond
->curr_active_slave
)) {
2788 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
2795 netdev_err(bond
->dev
, "impossible: new_link %d on slave %s\n",
2796 slave
->new_link
, slave
->dev
->name
);
2802 bond_select_active_slave(bond
);
2803 unblock_netpoll_tx();
2806 bond_set_carrier(bond
);
2809 /* Send ARP probes for active-backup mode ARP monitor.
2811 * Called with rcu_read_lock held.
2813 static bool bond_ab_arp_probe(struct bonding
*bond
)
2815 struct slave
*slave
, *before
= NULL
, *new_slave
= NULL
,
2816 *curr_arp_slave
= rcu_dereference(bond
->current_arp_slave
),
2817 *curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2818 struct list_head
*iter
;
2820 bool should_notify_rtnl
= BOND_SLAVE_NOTIFY_LATER
;
2822 if (curr_arp_slave
&& curr_active_slave
)
2823 netdev_info(bond
->dev
, "PROBE: c_arp %s && cas %s BAD\n",
2824 curr_arp_slave
->dev
->name
,
2825 curr_active_slave
->dev
->name
);
2827 if (curr_active_slave
) {
2828 bond_arp_send_all(bond
, curr_active_slave
);
2829 return should_notify_rtnl
;
2832 /* if we don't have a curr_active_slave, search for the next available
2833 * backup slave from the current_arp_slave and make it the candidate
2834 * for becoming the curr_active_slave
2837 if (!curr_arp_slave
) {
2838 curr_arp_slave
= bond_first_slave_rcu(bond
);
2839 if (!curr_arp_slave
)
2840 return should_notify_rtnl
;
2843 bond_set_slave_inactive_flags(curr_arp_slave
, BOND_SLAVE_NOTIFY_LATER
);
2845 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2846 if (!found
&& !before
&& bond_slave_is_up(slave
))
2849 if (found
&& !new_slave
&& bond_slave_is_up(slave
))
2851 /* if the link state is up at this point, we
2852 * mark it down - this can happen if we have
2853 * simultaneous link failures and
2854 * reselect_active_interface doesn't make this
2855 * one the current slave so it is still marked
2856 * up when it is actually down
2858 if (!bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
2859 bond_set_slave_link_state(slave
, BOND_LINK_DOWN
,
2860 BOND_SLAVE_NOTIFY_LATER
);
2861 if (slave
->link_failure_count
< UINT_MAX
)
2862 slave
->link_failure_count
++;
2864 bond_set_slave_inactive_flags(slave
,
2865 BOND_SLAVE_NOTIFY_LATER
);
2867 netdev_info(bond
->dev
, "backup interface %s is now down\n",
2870 if (slave
== curr_arp_slave
)
2874 if (!new_slave
&& before
)
2880 bond_set_slave_link_state(new_slave
, BOND_LINK_BACK
,
2881 BOND_SLAVE_NOTIFY_LATER
);
2882 bond_set_slave_active_flags(new_slave
, BOND_SLAVE_NOTIFY_LATER
);
2883 bond_arp_send_all(bond
, new_slave
);
2884 new_slave
->last_link_up
= jiffies
;
2885 rcu_assign_pointer(bond
->current_arp_slave
, new_slave
);
2888 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2889 if (slave
->should_notify
|| slave
->should_notify_link
) {
2890 should_notify_rtnl
= BOND_SLAVE_NOTIFY_NOW
;
2894 return should_notify_rtnl
;
2897 static void bond_activebackup_arp_mon(struct work_struct
*work
)
2899 struct bonding
*bond
= container_of(work
, struct bonding
,
2901 bool should_notify_peers
= false;
2902 bool should_notify_rtnl
= false;
2905 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2907 if (!bond_has_slaves(bond
))
2912 should_notify_peers
= bond_should_notify_peers(bond
);
2914 if (bond_ab_arp_inspect(bond
)) {
2917 /* Race avoidance with bond_close flush of workqueue */
2918 if (!rtnl_trylock()) {
2920 should_notify_peers
= false;
2924 bond_ab_arp_commit(bond
);
2930 should_notify_rtnl
= bond_ab_arp_probe(bond
);
2934 if (bond
->params
.arp_interval
)
2935 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2937 if (should_notify_peers
|| should_notify_rtnl
) {
2938 if (!rtnl_trylock())
2941 if (should_notify_peers
)
2942 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
2944 if (should_notify_rtnl
) {
2945 bond_slave_state_notify(bond
);
2946 bond_slave_link_notify(bond
);
2953 /*-------------------------- netdev event handling --------------------------*/
2955 /* Change device name */
2956 static int bond_event_changename(struct bonding
*bond
)
2958 bond_remove_proc_entry(bond
);
2959 bond_create_proc_entry(bond
);
2961 bond_debug_reregister(bond
);
2966 static int bond_master_netdev_event(unsigned long event
,
2967 struct net_device
*bond_dev
)
2969 struct bonding
*event_bond
= netdev_priv(bond_dev
);
2972 case NETDEV_CHANGENAME
:
2973 return bond_event_changename(event_bond
);
2974 case NETDEV_UNREGISTER
:
2975 bond_remove_proc_entry(event_bond
);
2977 case NETDEV_REGISTER
:
2978 bond_create_proc_entry(event_bond
);
2980 case NETDEV_NOTIFY_PEERS
:
2981 if (event_bond
->send_peer_notif
)
2982 event_bond
->send_peer_notif
--;
2991 static int bond_slave_netdev_event(unsigned long event
,
2992 struct net_device
*slave_dev
)
2994 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
), *primary
;
2995 struct bonding
*bond
;
2996 struct net_device
*bond_dev
;
2998 /* A netdev event can be generated while enslaving a device
2999 * before netdev_rx_handler_register is called in which case
3000 * slave will be NULL
3004 bond_dev
= slave
->bond
->dev
;
3006 primary
= rtnl_dereference(bond
->primary_slave
);
3009 case NETDEV_UNREGISTER
:
3010 if (bond_dev
->type
!= ARPHRD_ETHER
)
3011 bond_release_and_destroy(bond_dev
, slave_dev
);
3013 bond_release(bond_dev
, slave_dev
);
3017 bond_update_speed_duplex(slave
);
3018 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
3019 bond_3ad_adapter_speed_duplex_changed(slave
);
3022 /* Refresh slave-array if applicable!
3023 * If the setup does not use miimon or arpmon (mode-specific!),
3024 * then these events will not cause the slave-array to be
3025 * refreshed. This will cause xmit to use a slave that is not
3026 * usable. Avoid such situation by refeshing the array at these
3027 * events. If these (miimon/arpmon) parameters are configured
3028 * then array gets refreshed twice and that should be fine!
3030 if (bond_mode_uses_xmit_hash(bond
))
3031 bond_update_slave_arr(bond
, NULL
);
3033 case NETDEV_CHANGEMTU
:
3034 /* TODO: Should slaves be allowed to
3035 * independently alter their MTU? For
3036 * an active-backup bond, slaves need
3037 * not be the same type of device, so
3038 * MTUs may vary. For other modes,
3039 * slaves arguably should have the
3040 * same MTUs. To do this, we'd need to
3041 * take over the slave's change_mtu
3042 * function for the duration of their
3046 case NETDEV_CHANGENAME
:
3047 /* we don't care if we don't have primary set */
3048 if (!bond_uses_primary(bond
) ||
3049 !bond
->params
.primary
[0])
3052 if (slave
== primary
) {
3053 /* slave's name changed - he's no longer primary */
3054 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
3055 } else if (!strcmp(slave_dev
->name
, bond
->params
.primary
)) {
3056 /* we have a new primary slave */
3057 rcu_assign_pointer(bond
->primary_slave
, slave
);
3058 } else { /* we didn't change primary - exit */
3062 netdev_info(bond
->dev
, "Primary slave changed to %s, reselecting active slave\n",
3063 primary
? slave_dev
->name
: "none");
3066 bond_select_active_slave(bond
);
3067 unblock_netpoll_tx();
3069 case NETDEV_FEAT_CHANGE
:
3070 bond_compute_features(bond
);
3072 case NETDEV_RESEND_IGMP
:
3073 /* Propagate to master device */
3074 call_netdevice_notifiers(event
, slave
->bond
->dev
);
3083 /* bond_netdev_event: handle netdev notifier chain events.
3085 * This function receives events for the netdev chain. The caller (an
3086 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3087 * locks for us to safely manipulate the slave devices (RTNL lock,
3090 static int bond_netdev_event(struct notifier_block
*this,
3091 unsigned long event
, void *ptr
)
3093 struct net_device
*event_dev
= netdev_notifier_info_to_dev(ptr
);
3095 netdev_dbg(event_dev
, "event: %lx\n", event
);
3097 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3100 if (event_dev
->flags
& IFF_MASTER
) {
3101 netdev_dbg(event_dev
, "IFF_MASTER\n");
3102 return bond_master_netdev_event(event
, event_dev
);
3105 if (event_dev
->flags
& IFF_SLAVE
) {
3106 netdev_dbg(event_dev
, "IFF_SLAVE\n");
3107 return bond_slave_netdev_event(event
, event_dev
);
3113 static struct notifier_block bond_netdev_notifier
= {
3114 .notifier_call
= bond_netdev_event
,
3117 /*---------------------------- Hashing Policies -----------------------------*/
3119 /* L2 hash helper */
3120 static inline u32
bond_eth_hash(struct sk_buff
*skb
)
3122 struct ethhdr
*ep
, hdr_tmp
;
3124 ep
= skb_header_pointer(skb
, 0, sizeof(hdr_tmp
), &hdr_tmp
);
3126 return ep
->h_dest
[5] ^ ep
->h_source
[5] ^ ep
->h_proto
;
3130 /* Extract the appropriate headers based on bond's xmit policy */
3131 static bool bond_flow_dissect(struct bonding
*bond
, struct sk_buff
*skb
,
3132 struct flow_keys
*fk
)
3134 const struct ipv6hdr
*iph6
;
3135 const struct iphdr
*iph
;
3136 int noff
, proto
= -1;
3138 if (bond
->params
.xmit_policy
> BOND_XMIT_POLICY_LAYER23
)
3139 return skb_flow_dissect_flow_keys(skb
, fk
, 0);
3141 fk
->ports
.ports
= 0;
3142 noff
= skb_network_offset(skb
);
3143 if (skb
->protocol
== htons(ETH_P_IP
)) {
3144 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph
))))
3147 iph_to_flow_copy_v4addrs(fk
, iph
);
3148 noff
+= iph
->ihl
<< 2;
3149 if (!ip_is_fragment(iph
))
3150 proto
= iph
->protocol
;
3151 } else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
3152 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph6
))))
3154 iph6
= ipv6_hdr(skb
);
3155 iph_to_flow_copy_v6addrs(fk
, iph6
);
3156 noff
+= sizeof(*iph6
);
3157 proto
= iph6
->nexthdr
;
3161 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
&& proto
>= 0)
3162 fk
->ports
.ports
= skb_flow_get_ports(skb
, noff
, proto
);
3168 * bond_xmit_hash - generate a hash value based on the xmit policy
3169 * @bond: bonding device
3170 * @skb: buffer to use for headers
3172 * This function will extract the necessary headers from the skb buffer and use
3173 * them to generate a hash based on the xmit_policy set in the bonding device
3175 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
)
3177 struct flow_keys flow
;
3180 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_ENCAP34
&&
3184 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER2
||
3185 !bond_flow_dissect(bond
, skb
, &flow
))
3186 return bond_eth_hash(skb
);
3188 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER23
||
3189 bond
->params
.xmit_policy
== BOND_XMIT_POLICY_ENCAP23
)
3190 hash
= bond_eth_hash(skb
);
3192 hash
= (__force u32
)flow
.ports
.ports
;
3193 hash
^= (__force u32
)flow_get_u32_dst(&flow
) ^
3194 (__force u32
)flow_get_u32_src(&flow
);
3195 hash
^= (hash
>> 16);
3196 hash
^= (hash
>> 8);
3201 /*-------------------------- Device entry points ----------------------------*/
3203 static void bond_work_init_all(struct bonding
*bond
)
3205 INIT_DELAYED_WORK(&bond
->mcast_work
,
3206 bond_resend_igmp_join_requests_delayed
);
3207 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3208 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3209 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3210 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3212 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3213 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3214 INIT_DELAYED_WORK(&bond
->slave_arr_work
, bond_slave_arr_handler
);
3217 static void bond_work_cancel_all(struct bonding
*bond
)
3219 cancel_delayed_work_sync(&bond
->mii_work
);
3220 cancel_delayed_work_sync(&bond
->arp_work
);
3221 cancel_delayed_work_sync(&bond
->alb_work
);
3222 cancel_delayed_work_sync(&bond
->ad_work
);
3223 cancel_delayed_work_sync(&bond
->mcast_work
);
3224 cancel_delayed_work_sync(&bond
->slave_arr_work
);
3227 static int bond_open(struct net_device
*bond_dev
)
3229 struct bonding
*bond
= netdev_priv(bond_dev
);
3230 struct list_head
*iter
;
3231 struct slave
*slave
;
3233 /* reset slave->backup and slave->inactive */
3234 if (bond_has_slaves(bond
)) {
3235 bond_for_each_slave(bond
, slave
, iter
) {
3236 if (bond_uses_primary(bond
) &&
3237 slave
!= rcu_access_pointer(bond
->curr_active_slave
)) {
3238 bond_set_slave_inactive_flags(slave
,
3239 BOND_SLAVE_NOTIFY_NOW
);
3240 } else if (BOND_MODE(bond
) != BOND_MODE_8023AD
) {
3241 bond_set_slave_active_flags(slave
,
3242 BOND_SLAVE_NOTIFY_NOW
);
3247 bond_work_init_all(bond
);
3249 if (bond_is_lb(bond
)) {
3250 /* bond_alb_initialize must be called before the timer
3253 if (bond_alb_initialize(bond
, (BOND_MODE(bond
) == BOND_MODE_ALB
)))
3255 if (bond
->params
.tlb_dynamic_lb
)
3256 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3259 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3260 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3262 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3263 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3264 bond
->recv_probe
= bond_arp_rcv
;
3267 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
3268 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3269 /* register to receive LACPDUs */
3270 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3271 bond_3ad_initiate_agg_selection(bond
, 1);
3274 if (bond_mode_uses_xmit_hash(bond
))
3275 bond_update_slave_arr(bond
, NULL
);
3280 static int bond_close(struct net_device
*bond_dev
)
3282 struct bonding
*bond
= netdev_priv(bond_dev
);
3284 bond_work_cancel_all(bond
);
3285 bond
->send_peer_notif
= 0;
3286 if (bond_is_lb(bond
))
3287 bond_alb_deinitialize(bond
);
3288 bond
->recv_probe
= NULL
;
3293 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3294 struct rtnl_link_stats64
*stats
)
3296 struct bonding
*bond
= netdev_priv(bond_dev
);
3297 struct rtnl_link_stats64 temp
;
3298 struct list_head
*iter
;
3299 struct slave
*slave
;
3301 memcpy(stats
, &bond
->bond_stats
, sizeof(*stats
));
3303 bond_for_each_slave(bond
, slave
, iter
) {
3304 const struct rtnl_link_stats64
*sstats
=
3305 dev_get_stats(slave
->dev
, &temp
);
3306 struct rtnl_link_stats64
*pstats
= &slave
->slave_stats
;
3308 stats
->rx_packets
+= sstats
->rx_packets
- pstats
->rx_packets
;
3309 stats
->rx_bytes
+= sstats
->rx_bytes
- pstats
->rx_bytes
;
3310 stats
->rx_errors
+= sstats
->rx_errors
- pstats
->rx_errors
;
3311 stats
->rx_dropped
+= sstats
->rx_dropped
- pstats
->rx_dropped
;
3313 stats
->tx_packets
+= sstats
->tx_packets
- pstats
->tx_packets
;;
3314 stats
->tx_bytes
+= sstats
->tx_bytes
- pstats
->tx_bytes
;
3315 stats
->tx_errors
+= sstats
->tx_errors
- pstats
->tx_errors
;
3316 stats
->tx_dropped
+= sstats
->tx_dropped
- pstats
->tx_dropped
;
3318 stats
->multicast
+= sstats
->multicast
- pstats
->multicast
;
3319 stats
->collisions
+= sstats
->collisions
- pstats
->collisions
;
3321 stats
->rx_length_errors
+= sstats
->rx_length_errors
- pstats
->rx_length_errors
;
3322 stats
->rx_over_errors
+= sstats
->rx_over_errors
- pstats
->rx_over_errors
;
3323 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
- pstats
->rx_crc_errors
;
3324 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
- pstats
->rx_frame_errors
;
3325 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
- pstats
->rx_fifo_errors
;
3326 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
- pstats
->rx_missed_errors
;
3328 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
- pstats
->tx_aborted_errors
;
3329 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
- pstats
->tx_carrier_errors
;
3330 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
- pstats
->tx_fifo_errors
;
3331 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
- pstats
->tx_heartbeat_errors
;
3332 stats
->tx_window_errors
+= sstats
->tx_window_errors
- pstats
->tx_window_errors
;
3334 /* save off the slave stats for the next run */
3335 memcpy(pstats
, sstats
, sizeof(*sstats
));
3337 memcpy(&bond
->bond_stats
, stats
, sizeof(*stats
));
3342 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3344 struct bonding
*bond
= netdev_priv(bond_dev
);
3345 struct net_device
*slave_dev
= NULL
;
3346 struct ifbond k_binfo
;
3347 struct ifbond __user
*u_binfo
= NULL
;
3348 struct ifslave k_sinfo
;
3349 struct ifslave __user
*u_sinfo
= NULL
;
3350 struct mii_ioctl_data
*mii
= NULL
;
3351 struct bond_opt_value newval
;
3355 netdev_dbg(bond_dev
, "bond_ioctl: cmd=%d\n", cmd
);
3366 /* We do this again just in case we were called by SIOCGMIIREG
3367 * instead of SIOCGMIIPHY.
3373 if (mii
->reg_num
== 1) {
3375 if (netif_carrier_ok(bond
->dev
))
3376 mii
->val_out
= BMSR_LSTATUS
;
3380 case BOND_INFO_QUERY_OLD
:
3381 case SIOCBONDINFOQUERY
:
3382 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3384 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3387 res
= bond_info_query(bond_dev
, &k_binfo
);
3389 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3393 case BOND_SLAVE_INFO_QUERY_OLD
:
3394 case SIOCBONDSLAVEINFOQUERY
:
3395 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3397 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3400 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3402 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3410 net
= dev_net(bond_dev
);
3412 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3415 slave_dev
= __dev_get_by_name(net
, ifr
->ifr_slave
);
3417 netdev_dbg(bond_dev
, "slave_dev=%p:\n", slave_dev
);
3422 netdev_dbg(bond_dev
, "slave_dev->name=%s:\n", slave_dev
->name
);
3424 case BOND_ENSLAVE_OLD
:
3425 case SIOCBONDENSLAVE
:
3426 res
= bond_enslave(bond_dev
, slave_dev
);
3428 case BOND_RELEASE_OLD
:
3429 case SIOCBONDRELEASE
:
3430 res
= bond_release(bond_dev
, slave_dev
);
3432 case BOND_SETHWADDR_OLD
:
3433 case SIOCBONDSETHWADDR
:
3434 bond_set_dev_addr(bond_dev
, slave_dev
);
3437 case BOND_CHANGE_ACTIVE_OLD
:
3438 case SIOCBONDCHANGEACTIVE
:
3439 bond_opt_initstr(&newval
, slave_dev
->name
);
3440 res
= __bond_opt_set(bond
, BOND_OPT_ACTIVE_SLAVE
, &newval
);
3449 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3451 struct bonding
*bond
= netdev_priv(bond_dev
);
3453 if (change
& IFF_PROMISC
)
3454 bond_set_promiscuity(bond
,
3455 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3457 if (change
& IFF_ALLMULTI
)
3458 bond_set_allmulti(bond
,
3459 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3462 static void bond_set_rx_mode(struct net_device
*bond_dev
)
3464 struct bonding
*bond
= netdev_priv(bond_dev
);
3465 struct list_head
*iter
;
3466 struct slave
*slave
;
3469 if (bond_uses_primary(bond
)) {
3470 slave
= rcu_dereference(bond
->curr_active_slave
);
3472 dev_uc_sync(slave
->dev
, bond_dev
);
3473 dev_mc_sync(slave
->dev
, bond_dev
);
3476 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3477 dev_uc_sync_multiple(slave
->dev
, bond_dev
);
3478 dev_mc_sync_multiple(slave
->dev
, bond_dev
);
3484 static int bond_neigh_init(struct neighbour
*n
)
3486 struct bonding
*bond
= netdev_priv(n
->dev
);
3487 const struct net_device_ops
*slave_ops
;
3488 struct neigh_parms parms
;
3489 struct slave
*slave
;
3492 slave
= bond_first_slave(bond
);
3495 slave_ops
= slave
->dev
->netdev_ops
;
3496 if (!slave_ops
->ndo_neigh_setup
)
3499 parms
.neigh_setup
= NULL
;
3500 parms
.neigh_cleanup
= NULL
;
3501 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3505 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3506 * after the last slave has been detached. Assumes that all slaves
3507 * utilize the same neigh_cleanup (true at this writing as only user
3510 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3512 if (!parms
.neigh_setup
)
3515 return parms
.neigh_setup(n
);
3518 /* The bonding ndo_neigh_setup is called at init time beofre any
3519 * slave exists. So we must declare proxy setup function which will
3520 * be used at run time to resolve the actual slave neigh param setup.
3522 * It's also called by master devices (such as vlans) to setup their
3523 * underlying devices. In that case - do nothing, we're already set up from
3526 static int bond_neigh_setup(struct net_device
*dev
,
3527 struct neigh_parms
*parms
)
3529 /* modify only our neigh_parms */
3530 if (parms
->dev
== dev
)
3531 parms
->neigh_setup
= bond_neigh_init
;
3536 /* Change the MTU of all of a master's slaves to match the master */
3537 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3539 struct bonding
*bond
= netdev_priv(bond_dev
);
3540 struct slave
*slave
, *rollback_slave
;
3541 struct list_head
*iter
;
3544 netdev_dbg(bond_dev
, "bond=%p, new_mtu=%d\n", bond
, new_mtu
);
3546 bond_for_each_slave(bond
, slave
, iter
) {
3547 netdev_dbg(bond_dev
, "s %p c_m %p\n",
3548 slave
, slave
->dev
->netdev_ops
->ndo_change_mtu
);
3550 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3553 /* If we failed to set the slave's mtu to the new value
3554 * we must abort the operation even in ACTIVE_BACKUP
3555 * mode, because if we allow the backup slaves to have
3556 * different mtu values than the active slave we'll
3557 * need to change their mtu when doing a failover. That
3558 * means changing their mtu from timer context, which
3559 * is probably not a good idea.
3561 netdev_dbg(bond_dev
, "err %d %s\n", res
,
3567 bond_dev
->mtu
= new_mtu
;
3572 /* unwind from head to the slave that failed */
3573 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3576 if (rollback_slave
== slave
)
3579 tmp_res
= dev_set_mtu(rollback_slave
->dev
, bond_dev
->mtu
);
3581 netdev_dbg(bond_dev
, "unwind err %d dev %s\n",
3582 tmp_res
, rollback_slave
->dev
->name
);
3589 /* Change HW address
3591 * Note that many devices must be down to change the HW address, and
3592 * downing the master releases all slaves. We can make bonds full of
3593 * bonding devices to test this, however.
3595 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3597 struct bonding
*bond
= netdev_priv(bond_dev
);
3598 struct slave
*slave
, *rollback_slave
;
3599 struct sockaddr
*sa
= addr
, tmp_sa
;
3600 struct list_head
*iter
;
3603 if (BOND_MODE(bond
) == BOND_MODE_ALB
)
3604 return bond_alb_set_mac_address(bond_dev
, addr
);
3607 netdev_dbg(bond_dev
, "bond=%p\n", bond
);
3609 /* If fail_over_mac is enabled, do nothing and return success.
3610 * Returning an error causes ifenslave to fail.
3612 if (bond
->params
.fail_over_mac
&&
3613 BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3616 if (!is_valid_ether_addr(sa
->sa_data
))
3617 return -EADDRNOTAVAIL
;
3619 bond_for_each_slave(bond
, slave
, iter
) {
3620 netdev_dbg(bond_dev
, "slave %p %s\n", slave
, slave
->dev
->name
);
3621 res
= dev_set_mac_address(slave
->dev
, addr
);
3623 /* TODO: consider downing the slave
3625 * User should expect communications
3626 * breakage anyway until ARP finish
3629 netdev_dbg(bond_dev
, "err %d %s\n", res
, slave
->dev
->name
);
3635 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3639 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3640 tmp_sa
.sa_family
= bond_dev
->type
;
3642 /* unwind from head to the slave that failed */
3643 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3646 if (rollback_slave
== slave
)
3649 tmp_res
= dev_set_mac_address(rollback_slave
->dev
, &tmp_sa
);
3651 netdev_dbg(bond_dev
, "unwind err %d dev %s\n",
3652 tmp_res
, rollback_slave
->dev
->name
);
3660 * bond_xmit_slave_id - transmit skb through slave with slave_id
3661 * @bond: bonding device that is transmitting
3662 * @skb: buffer to transmit
3663 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3665 * This function tries to transmit through slave with slave_id but in case
3666 * it fails, it tries to find the first available slave for transmission.
3667 * The skb is consumed in all cases, thus the function is void.
3669 static void bond_xmit_slave_id(struct bonding
*bond
, struct sk_buff
*skb
, int slave_id
)
3671 struct list_head
*iter
;
3672 struct slave
*slave
;
3675 /* Here we start from the slave with slave_id */
3676 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3678 if (bond_slave_can_tx(slave
)) {
3679 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3685 /* Here we start from the first slave up to slave_id */
3687 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3690 if (bond_slave_can_tx(slave
)) {
3691 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3695 /* no slave that can tx has been found */
3696 bond_tx_drop(bond
->dev
, skb
);
3700 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3701 * @bond: bonding device to use
3703 * Based on the value of the bonding device's packets_per_slave parameter
3704 * this function generates a slave id, which is usually used as the next
3705 * slave to transmit through.
3707 static u32
bond_rr_gen_slave_id(struct bonding
*bond
)
3710 struct reciprocal_value reciprocal_packets_per_slave
;
3711 int packets_per_slave
= bond
->params
.packets_per_slave
;
3713 switch (packets_per_slave
) {
3715 slave_id
= prandom_u32();
3718 slave_id
= bond
->rr_tx_counter
;
3721 reciprocal_packets_per_slave
=
3722 bond
->params
.reciprocal_packets_per_slave
;
3723 slave_id
= reciprocal_divide(bond
->rr_tx_counter
,
3724 reciprocal_packets_per_slave
);
3727 bond
->rr_tx_counter
++;
3732 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3734 struct bonding
*bond
= netdev_priv(bond_dev
);
3735 struct iphdr
*iph
= ip_hdr(skb
);
3736 struct slave
*slave
;
3739 /* Start with the curr_active_slave that joined the bond as the
3740 * default for sending IGMP traffic. For failover purposes one
3741 * needs to maintain some consistency for the interface that will
3742 * send the join/membership reports. The curr_active_slave found
3743 * will send all of this type of traffic.
3745 if (iph
->protocol
== IPPROTO_IGMP
&& skb
->protocol
== htons(ETH_P_IP
)) {
3746 slave
= rcu_dereference(bond
->curr_active_slave
);
3748 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3750 bond_xmit_slave_id(bond
, skb
, 0);
3752 int slave_cnt
= ACCESS_ONCE(bond
->slave_cnt
);
3754 if (likely(slave_cnt
)) {
3755 slave_id
= bond_rr_gen_slave_id(bond
);
3756 bond_xmit_slave_id(bond
, skb
, slave_id
% slave_cnt
);
3758 bond_tx_drop(bond_dev
, skb
);
3762 return NETDEV_TX_OK
;
3765 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3766 * the bond has a usable interface.
3768 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3770 struct bonding
*bond
= netdev_priv(bond_dev
);
3771 struct slave
*slave
;
3773 slave
= rcu_dereference(bond
->curr_active_slave
);
3775 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3777 bond_tx_drop(bond_dev
, skb
);
3779 return NETDEV_TX_OK
;
3782 /* Use this to update slave_array when (a) it's not appropriate to update
3783 * slave_array right away (note that update_slave_array() may sleep)
3784 * and / or (b) RTNL is not held.
3786 void bond_slave_arr_work_rearm(struct bonding
*bond
, unsigned long delay
)
3788 queue_delayed_work(bond
->wq
, &bond
->slave_arr_work
, delay
);
3791 /* Slave array work handler. Holds only RTNL */
3792 static void bond_slave_arr_handler(struct work_struct
*work
)
3794 struct bonding
*bond
= container_of(work
, struct bonding
,
3795 slave_arr_work
.work
);
3798 if (!rtnl_trylock())
3801 ret
= bond_update_slave_arr(bond
, NULL
);
3804 pr_warn_ratelimited("Failed to update slave array from WT\n");
3810 bond_slave_arr_work_rearm(bond
, 1);
3813 /* Build the usable slaves array in control path for modes that use xmit-hash
3814 * to determine the slave interface -
3815 * (a) BOND_MODE_8023AD
3817 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3819 * The caller is expected to hold RTNL only and NO other lock!
3821 int bond_update_slave_arr(struct bonding
*bond
, struct slave
*skipslave
)
3823 struct slave
*slave
;
3824 struct list_head
*iter
;
3825 struct bond_up_slave
*new_arr
, *old_arr
;
3829 #ifdef CONFIG_LOCKDEP
3830 WARN_ON(lockdep_is_held(&bond
->mode_lock
));
3833 new_arr
= kzalloc(offsetof(struct bond_up_slave
, arr
[bond
->slave_cnt
]),
3837 pr_err("Failed to build slave-array.\n");
3840 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
3841 struct ad_info ad_info
;
3843 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3844 pr_debug("bond_3ad_get_active_agg_info failed\n");
3845 kfree_rcu(new_arr
, rcu
);
3846 /* No active aggragator means it's not safe to use
3847 * the previous array.
3849 old_arr
= rtnl_dereference(bond
->slave_arr
);
3851 RCU_INIT_POINTER(bond
->slave_arr
, NULL
);
3852 kfree_rcu(old_arr
, rcu
);
3856 agg_id
= ad_info
.aggregator_id
;
3858 bond_for_each_slave(bond
, slave
, iter
) {
3859 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
3860 struct aggregator
*agg
;
3862 agg
= SLAVE_AD_INFO(slave
)->port
.aggregator
;
3863 if (!agg
|| agg
->aggregator_identifier
!= agg_id
)
3866 if (!bond_slave_can_tx(slave
))
3868 if (skipslave
== slave
)
3870 new_arr
->arr
[new_arr
->count
++] = slave
;
3873 old_arr
= rtnl_dereference(bond
->slave_arr
);
3874 rcu_assign_pointer(bond
->slave_arr
, new_arr
);
3876 kfree_rcu(old_arr
, rcu
);
3878 if (ret
!= 0 && skipslave
) {
3881 /* Rare situation where caller has asked to skip a specific
3882 * slave but allocation failed (most likely!). BTW this is
3883 * only possible when the call is initiated from
3884 * __bond_release_one(). In this situation; overwrite the
3885 * skipslave entry in the array with the last entry from the
3886 * array to avoid a situation where the xmit path may choose
3887 * this to-be-skipped slave to send a packet out.
3889 old_arr
= rtnl_dereference(bond
->slave_arr
);
3890 for (idx
= 0; idx
< old_arr
->count
; idx
++) {
3891 if (skipslave
== old_arr
->arr
[idx
]) {
3893 old_arr
->arr
[old_arr
->count
-1];
3902 /* Use this Xmit function for 3AD as well as XOR modes. The current
3903 * usable slave array is formed in the control path. The xmit function
3904 * just calculates hash and sends the packet out.
3906 static int bond_3ad_xor_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3908 struct bonding
*bond
= netdev_priv(dev
);
3909 struct slave
*slave
;
3910 struct bond_up_slave
*slaves
;
3913 slaves
= rcu_dereference(bond
->slave_arr
);
3914 count
= slaves
? ACCESS_ONCE(slaves
->count
) : 0;
3915 if (likely(count
)) {
3916 slave
= slaves
->arr
[bond_xmit_hash(bond
, skb
) % count
];
3917 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3919 bond_tx_drop(dev
, skb
);
3922 return NETDEV_TX_OK
;
3925 /* in broadcast mode, we send everything to all usable interfaces. */
3926 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3928 struct bonding
*bond
= netdev_priv(bond_dev
);
3929 struct slave
*slave
= NULL
;
3930 struct list_head
*iter
;
3932 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3933 if (bond_is_last_slave(bond
, slave
))
3935 if (bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
3936 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
3939 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3940 bond_dev
->name
, __func__
);
3943 bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3946 if (slave
&& bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
)
3947 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3949 bond_tx_drop(bond_dev
, skb
);
3951 return NETDEV_TX_OK
;
3954 /*------------------------- Device initialization ---------------------------*/
3956 /* Lookup the slave that corresponds to a qid */
3957 static inline int bond_slave_override(struct bonding
*bond
,
3958 struct sk_buff
*skb
)
3960 struct slave
*slave
= NULL
;
3961 struct list_head
*iter
;
3963 if (!skb
->queue_mapping
)
3966 /* Find out if any slaves have the same mapping as this skb. */
3967 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3968 if (slave
->queue_id
== skb
->queue_mapping
) {
3969 if (bond_slave_is_up(slave
) &&
3970 slave
->link
== BOND_LINK_UP
) {
3971 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3974 /* If the slave isn't UP, use default transmit policy. */
3983 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
3984 void *accel_priv
, select_queue_fallback_t fallback
)
3986 /* This helper function exists to help dev_pick_tx get the correct
3987 * destination queue. Using a helper function skips a call to
3988 * skb_tx_hash and will put the skbs in the queue we expect on their
3989 * way down to the bonding driver.
3991 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
3993 /* Save the original txq to restore before passing to the driver */
3994 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
3996 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
3998 txq
-= dev
->real_num_tx_queues
;
3999 } while (txq
>= dev
->real_num_tx_queues
);
4004 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4006 struct bonding
*bond
= netdev_priv(dev
);
4008 if (bond_should_override_tx_queue(bond
) &&
4009 !bond_slave_override(bond
, skb
))
4010 return NETDEV_TX_OK
;
4012 switch (BOND_MODE(bond
)) {
4013 case BOND_MODE_ROUNDROBIN
:
4014 return bond_xmit_roundrobin(skb
, dev
);
4015 case BOND_MODE_ACTIVEBACKUP
:
4016 return bond_xmit_activebackup(skb
, dev
);
4017 case BOND_MODE_8023AD
:
4019 return bond_3ad_xor_xmit(skb
, dev
);
4020 case BOND_MODE_BROADCAST
:
4021 return bond_xmit_broadcast(skb
, dev
);
4023 return bond_alb_xmit(skb
, dev
);
4025 return bond_tlb_xmit(skb
, dev
);
4027 /* Should never happen, mode already checked */
4028 netdev_err(dev
, "Unknown bonding mode %d\n", BOND_MODE(bond
));
4030 bond_tx_drop(dev
, skb
);
4031 return NETDEV_TX_OK
;
4035 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4037 struct bonding
*bond
= netdev_priv(dev
);
4038 netdev_tx_t ret
= NETDEV_TX_OK
;
4040 /* If we risk deadlock from transmitting this in the
4041 * netpoll path, tell netpoll to queue the frame for later tx
4043 if (unlikely(is_netpoll_tx_blocked(dev
)))
4044 return NETDEV_TX_BUSY
;
4047 if (bond_has_slaves(bond
))
4048 ret
= __bond_start_xmit(skb
, dev
);
4050 bond_tx_drop(dev
, skb
);
4056 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
4057 struct ethtool_cmd
*ecmd
)
4059 struct bonding
*bond
= netdev_priv(bond_dev
);
4060 unsigned long speed
= 0;
4061 struct list_head
*iter
;
4062 struct slave
*slave
;
4064 ecmd
->duplex
= DUPLEX_UNKNOWN
;
4065 ecmd
->port
= PORT_OTHER
;
4067 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4068 * do not need to check mode. Though link speed might not represent
4069 * the true receive or transmit bandwidth (not all modes are symmetric)
4070 * this is an accurate maximum.
4072 bond_for_each_slave(bond
, slave
, iter
) {
4073 if (bond_slave_can_tx(slave
)) {
4074 if (slave
->speed
!= SPEED_UNKNOWN
)
4075 speed
+= slave
->speed
;
4076 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
4077 slave
->duplex
!= DUPLEX_UNKNOWN
)
4078 ecmd
->duplex
= slave
->duplex
;
4081 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
4086 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4087 struct ethtool_drvinfo
*drvinfo
)
4089 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
4090 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
4091 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
4095 static const struct ethtool_ops bond_ethtool_ops
= {
4096 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4097 .get_settings
= bond_ethtool_get_settings
,
4098 .get_link
= ethtool_op_get_link
,
4101 static const struct net_device_ops bond_netdev_ops
= {
4102 .ndo_init
= bond_init
,
4103 .ndo_uninit
= bond_uninit
,
4104 .ndo_open
= bond_open
,
4105 .ndo_stop
= bond_close
,
4106 .ndo_start_xmit
= bond_start_xmit
,
4107 .ndo_select_queue
= bond_select_queue
,
4108 .ndo_get_stats64
= bond_get_stats
,
4109 .ndo_do_ioctl
= bond_do_ioctl
,
4110 .ndo_change_rx_flags
= bond_change_rx_flags
,
4111 .ndo_set_rx_mode
= bond_set_rx_mode
,
4112 .ndo_change_mtu
= bond_change_mtu
,
4113 .ndo_set_mac_address
= bond_set_mac_address
,
4114 .ndo_neigh_setup
= bond_neigh_setup
,
4115 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4116 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4117 #ifdef CONFIG_NET_POLL_CONTROLLER
4118 .ndo_netpoll_setup
= bond_netpoll_setup
,
4119 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4120 .ndo_poll_controller
= bond_poll_controller
,
4122 .ndo_add_slave
= bond_enslave
,
4123 .ndo_del_slave
= bond_release
,
4124 .ndo_fix_features
= bond_fix_features
,
4125 .ndo_bridge_setlink
= switchdev_port_bridge_setlink
,
4126 .ndo_bridge_getlink
= switchdev_port_bridge_getlink
,
4127 .ndo_bridge_dellink
= switchdev_port_bridge_dellink
,
4128 .ndo_fdb_add
= switchdev_port_fdb_add
,
4129 .ndo_fdb_del
= switchdev_port_fdb_del
,
4130 .ndo_fdb_dump
= switchdev_port_fdb_dump
,
4131 .ndo_features_check
= passthru_features_check
,
4134 static const struct device_type bond_type
= {
4138 static void bond_destructor(struct net_device
*bond_dev
)
4140 struct bonding
*bond
= netdev_priv(bond_dev
);
4142 destroy_workqueue(bond
->wq
);
4143 free_netdev(bond_dev
);
4146 void bond_setup(struct net_device
*bond_dev
)
4148 struct bonding
*bond
= netdev_priv(bond_dev
);
4150 spin_lock_init(&bond
->mode_lock
);
4151 bond
->params
= bonding_defaults
;
4153 /* Initialize pointers */
4154 bond
->dev
= bond_dev
;
4156 /* Initialize the device entry points */
4157 ether_setup(bond_dev
);
4158 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4159 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4161 bond_dev
->destructor
= bond_destructor
;
4163 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
4165 /* Initialize the device options */
4166 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4167 bond_dev
->priv_flags
|= IFF_BONDING
| IFF_UNICAST_FLT
| IFF_NO_QUEUE
;
4168 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
4170 /* don't acquire bond device's netif_tx_lock when transmitting */
4171 bond_dev
->features
|= NETIF_F_LLTX
;
4173 /* By default, we declare the bond to be fully
4174 * VLAN hardware accelerated capable. Special
4175 * care is taken in the various xmit functions
4176 * when there are slaves that are not hw accel
4180 /* Don't allow bond devices to change network namespaces. */
4181 bond_dev
->features
|= NETIF_F_NETNS_LOCAL
;
4183 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4184 NETIF_F_HW_VLAN_CTAG_TX
|
4185 NETIF_F_HW_VLAN_CTAG_RX
|
4186 NETIF_F_HW_VLAN_CTAG_FILTER
;
4188 bond_dev
->hw_features
|= NETIF_F_GSO_ENCAP_ALL
;
4189 bond_dev
->features
|= bond_dev
->hw_features
;
4192 /* Destroy a bonding device.
4193 * Must be under rtnl_lock when this function is called.
4195 static void bond_uninit(struct net_device
*bond_dev
)
4197 struct bonding
*bond
= netdev_priv(bond_dev
);
4198 struct list_head
*iter
;
4199 struct slave
*slave
;
4200 struct bond_up_slave
*arr
;
4202 bond_netpoll_cleanup(bond_dev
);
4204 /* Release the bonded slaves */
4205 bond_for_each_slave(bond
, slave
, iter
)
4206 __bond_release_one(bond_dev
, slave
->dev
, true);
4207 netdev_info(bond_dev
, "Released all slaves\n");
4209 arr
= rtnl_dereference(bond
->slave_arr
);
4211 RCU_INIT_POINTER(bond
->slave_arr
, NULL
);
4212 kfree_rcu(arr
, rcu
);
4215 list_del(&bond
->bond_list
);
4217 bond_debug_unregister(bond
);
4220 /*------------------------- Module initialization ---------------------------*/
4222 static int bond_check_params(struct bond_params
*params
)
4224 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
4225 struct bond_opt_value newval
;
4226 const struct bond_opt_value
*valptr
;
4227 int arp_all_targets_value
;
4228 u16 ad_actor_sys_prio
= 0;
4229 u16 ad_user_port_key
= 0;
4231 /* Convert string parameters. */
4233 bond_opt_initstr(&newval
, mode
);
4234 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_MODE
), &newval
);
4236 pr_err("Error: Invalid bonding mode \"%s\"\n", mode
);
4239 bond_mode
= valptr
->value
;
4242 if (xmit_hash_policy
) {
4243 if ((bond_mode
!= BOND_MODE_XOR
) &&
4244 (bond_mode
!= BOND_MODE_8023AD
) &&
4245 (bond_mode
!= BOND_MODE_TLB
)) {
4246 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4247 bond_mode_name(bond_mode
));
4249 bond_opt_initstr(&newval
, xmit_hash_policy
);
4250 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH
),
4253 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4257 xmit_hashtype
= valptr
->value
;
4262 if (bond_mode
!= BOND_MODE_8023AD
) {
4263 pr_info("lacp_rate param is irrelevant in mode %s\n",
4264 bond_mode_name(bond_mode
));
4266 bond_opt_initstr(&newval
, lacp_rate
);
4267 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE
),
4270 pr_err("Error: Invalid lacp rate \"%s\"\n",
4274 lacp_fast
= valptr
->value
;
4279 bond_opt_initstr(&newval
, ad_select
);
4280 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT
),
4283 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select
);
4286 params
->ad_select
= valptr
->value
;
4287 if (bond_mode
!= BOND_MODE_8023AD
)
4288 pr_warn("ad_select param only affects 802.3ad mode\n");
4290 params
->ad_select
= BOND_AD_STABLE
;
4293 if (max_bonds
< 0) {
4294 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4295 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4296 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4300 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4306 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4311 if (downdelay
< 0) {
4312 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4313 downdelay
, INT_MAX
);
4317 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4318 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4323 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4324 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4329 /* reset values for 802.3ad/TLB/ALB */
4330 if (!bond_mode_uses_arp(bond_mode
)) {
4332 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");
4333 pr_warn("Forcing miimon to 100msec\n");
4334 miimon
= BOND_DEFAULT_MIIMON
;
4338 if (tx_queues
< 1 || tx_queues
> 255) {
4339 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4340 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4341 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4344 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4345 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4347 all_slaves_active
= 0;
4350 if (resend_igmp
< 0 || resend_igmp
> 255) {
4351 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4352 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4353 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4356 bond_opt_initval(&newval
, packets_per_slave
);
4357 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE
), &newval
)) {
4358 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4359 packets_per_slave
, USHRT_MAX
);
4360 packets_per_slave
= 1;
4363 if (bond_mode
== BOND_MODE_ALB
) {
4364 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",
4369 if (updelay
|| downdelay
) {
4370 /* just warn the user the up/down delay will have
4371 * no effect since miimon is zero...
4373 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",
4374 updelay
, downdelay
);
4377 /* don't allow arp monitoring */
4379 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4380 miimon
, arp_interval
);
4384 if ((updelay
% miimon
) != 0) {
4385 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4386 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4391 if ((downdelay
% miimon
) != 0) {
4392 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4394 (downdelay
/ miimon
) * miimon
);
4397 downdelay
/= miimon
;
4400 if (arp_interval
< 0) {
4401 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4402 arp_interval
, INT_MAX
);
4406 for (arp_ip_count
= 0, i
= 0;
4407 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4410 /* not a complete check, but good enough to catch mistakes */
4411 if (!in4_pton(arp_ip_target
[i
], -1, (u8
*)&ip
, -1, NULL
) ||
4412 !bond_is_ip_target_ok(ip
)) {
4413 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4417 if (bond_get_targets_ip(arp_target
, ip
) == -1)
4418 arp_target
[arp_ip_count
++] = ip
;
4420 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4425 if (arp_interval
&& !arp_ip_count
) {
4426 /* don't allow arping if no arp_ip_target given... */
4427 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4433 if (!arp_interval
) {
4434 pr_err("arp_validate requires arp_interval\n");
4438 bond_opt_initstr(&newval
, arp_validate
);
4439 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE
),
4442 pr_err("Error: invalid arp_validate \"%s\"\n",
4446 arp_validate_value
= valptr
->value
;
4448 arp_validate_value
= 0;
4451 arp_all_targets_value
= 0;
4452 if (arp_all_targets
) {
4453 bond_opt_initstr(&newval
, arp_all_targets
);
4454 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS
),
4457 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4459 arp_all_targets_value
= 0;
4461 arp_all_targets_value
= valptr
->value
;
4466 pr_info("MII link monitoring set to %d ms\n", miimon
);
4467 } else if (arp_interval
) {
4468 valptr
= bond_opt_get_val(BOND_OPT_ARP_VALIDATE
,
4469 arp_validate_value
);
4470 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4471 arp_interval
, valptr
->string
, arp_ip_count
);
4473 for (i
= 0; i
< arp_ip_count
; i
++)
4474 pr_cont(" %s", arp_ip_target
[i
]);
4478 } else if (max_bonds
) {
4479 /* miimon and arp_interval not set, we need one so things
4480 * work as expected, see bonding.txt for details
4482 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");
4485 if (primary
&& !bond_mode_uses_primary(bond_mode
)) {
4486 /* currently, using a primary only makes sense
4487 * in active backup, TLB or ALB modes
4489 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4490 primary
, bond_mode_name(bond_mode
));
4494 if (primary
&& primary_reselect
) {
4495 bond_opt_initstr(&newval
, primary_reselect
);
4496 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT
),
4499 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4503 primary_reselect_value
= valptr
->value
;
4505 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4508 if (fail_over_mac
) {
4509 bond_opt_initstr(&newval
, fail_over_mac
);
4510 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC
),
4513 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4517 fail_over_mac_value
= valptr
->value
;
4518 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4519 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4521 fail_over_mac_value
= BOND_FOM_NONE
;
4524 bond_opt_initstr(&newval
, "default");
4525 valptr
= bond_opt_parse(
4526 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO
),
4529 pr_err("Error: No ad_actor_sys_prio default value");
4532 ad_actor_sys_prio
= valptr
->value
;
4534 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY
),
4537 pr_err("Error: No ad_user_port_key default value");
4540 ad_user_port_key
= valptr
->value
;
4542 if (lp_interval
== 0) {
4543 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4544 INT_MAX
, BOND_ALB_DEFAULT_LP_INTERVAL
);
4545 lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
4548 /* fill params struct with the proper values */
4549 params
->mode
= bond_mode
;
4550 params
->xmit_policy
= xmit_hashtype
;
4551 params
->miimon
= miimon
;
4552 params
->num_peer_notif
= num_peer_notif
;
4553 params
->arp_interval
= arp_interval
;
4554 params
->arp_validate
= arp_validate_value
;
4555 params
->arp_all_targets
= arp_all_targets_value
;
4556 params
->updelay
= updelay
;
4557 params
->downdelay
= downdelay
;
4558 params
->use_carrier
= use_carrier
;
4559 params
->lacp_fast
= lacp_fast
;
4560 params
->primary
[0] = 0;
4561 params
->primary_reselect
= primary_reselect_value
;
4562 params
->fail_over_mac
= fail_over_mac_value
;
4563 params
->tx_queues
= tx_queues
;
4564 params
->all_slaves_active
= all_slaves_active
;
4565 params
->resend_igmp
= resend_igmp
;
4566 params
->min_links
= min_links
;
4567 params
->lp_interval
= lp_interval
;
4568 params
->packets_per_slave
= packets_per_slave
;
4569 params
->tlb_dynamic_lb
= 1; /* Default value */
4570 params
->ad_actor_sys_prio
= ad_actor_sys_prio
;
4571 eth_zero_addr(params
->ad_actor_system
);
4572 params
->ad_user_port_key
= ad_user_port_key
;
4573 if (packets_per_slave
> 0) {
4574 params
->reciprocal_packets_per_slave
=
4575 reciprocal_value(packets_per_slave
);
4577 /* reciprocal_packets_per_slave is unused if
4578 * packets_per_slave is 0 or 1, just initialize it
4580 params
->reciprocal_packets_per_slave
=
4581 (struct reciprocal_value
) { 0 };
4585 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4586 params
->primary
[IFNAMSIZ
- 1] = 0;
4589 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4594 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4595 static struct lock_class_key bonding_netdev_addr_lock_key
;
4596 static struct lock_class_key bonding_tx_busylock_key
;
4598 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4599 struct netdev_queue
*txq
,
4602 lockdep_set_class(&txq
->_xmit_lock
,
4603 &bonding_netdev_xmit_lock_key
);
4606 static void bond_set_lockdep_class(struct net_device
*dev
)
4608 lockdep_set_class(&dev
->addr_list_lock
,
4609 &bonding_netdev_addr_lock_key
);
4610 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4611 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4614 /* Called from registration process */
4615 static int bond_init(struct net_device
*bond_dev
)
4617 struct bonding
*bond
= netdev_priv(bond_dev
);
4618 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4620 netdev_dbg(bond_dev
, "Begin bond_init\n");
4622 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4626 bond_set_lockdep_class(bond_dev
);
4628 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4630 bond_prepare_sysfs_group(bond
);
4632 bond_debug_register(bond
);
4634 /* Ensure valid dev_addr */
4635 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4636 bond_dev
->addr_assign_type
== NET_ADDR_PERM
)
4637 eth_hw_addr_random(bond_dev
);
4642 unsigned int bond_get_num_tx_queues(void)
4647 /* Create a new bond based on the specified name and bonding parameters.
4648 * If name is NULL, obtain a suitable "bond%d" name for us.
4649 * Caller must NOT hold rtnl_lock; we need to release it here before we
4650 * set up our sysfs entries.
4652 int bond_create(struct net
*net
, const char *name
)
4654 struct net_device
*bond_dev
;
4655 struct bonding
*bond
;
4656 struct alb_bond_info
*bond_info
;
4661 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4662 name
? name
: "bond%d", NET_NAME_UNKNOWN
,
4663 bond_setup
, tx_queues
);
4665 pr_err("%s: eek! can't alloc netdev!\n", name
);
4671 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4672 * It is set to 0 by default which is wrong.
4674 bond
= netdev_priv(bond_dev
);
4675 bond_info
= &(BOND_ALB_INFO(bond
));
4676 bond_info
->rx_hashtbl_used_head
= RLB_NULL_INDEX
;
4678 dev_net_set(bond_dev
, net
);
4679 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4681 res
= register_netdevice(bond_dev
);
4683 netif_carrier_off(bond_dev
);
4687 bond_destructor(bond_dev
);
4691 static int __net_init
bond_net_init(struct net
*net
)
4693 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4696 INIT_LIST_HEAD(&bn
->dev_list
);
4698 bond_create_proc_dir(bn
);
4699 bond_create_sysfs(bn
);
4704 static void __net_exit
bond_net_exit(struct net
*net
)
4706 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4707 struct bonding
*bond
, *tmp_bond
;
4710 bond_destroy_sysfs(bn
);
4712 /* Kill off any bonds created after unregistering bond rtnl ops */
4714 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4715 unregister_netdevice_queue(bond
->dev
, &list
);
4716 unregister_netdevice_many(&list
);
4719 bond_destroy_proc_dir(bn
);
4722 static struct pernet_operations bond_net_ops
= {
4723 .init
= bond_net_init
,
4724 .exit
= bond_net_exit
,
4726 .size
= sizeof(struct bond_net
),
4729 static int __init
bonding_init(void)
4734 pr_info("%s", bond_version
);
4736 res
= bond_check_params(&bonding_defaults
);
4740 res
= register_pernet_subsys(&bond_net_ops
);
4744 res
= bond_netlink_init();
4748 bond_create_debugfs();
4750 for (i
= 0; i
< max_bonds
; i
++) {
4751 res
= bond_create(&init_net
, NULL
);
4756 register_netdevice_notifier(&bond_netdev_notifier
);
4760 bond_destroy_debugfs();
4761 bond_netlink_fini();
4763 unregister_pernet_subsys(&bond_net_ops
);
4768 static void __exit
bonding_exit(void)
4770 unregister_netdevice_notifier(&bond_netdev_notifier
);
4772 bond_destroy_debugfs();
4774 bond_netlink_fini();
4775 unregister_pernet_subsys(&bond_net_ops
);
4777 #ifdef CONFIG_NET_POLL_CONTROLLER
4778 /* Make sure we don't have an imbalance on our netpoll blocking */
4779 WARN_ON(atomic_read(&netpoll_block_tx
));
4783 module_init(bonding_init
);
4784 module_exit(bonding_exit
);
4785 MODULE_LICENSE("GPL");
4786 MODULE_VERSION(DRV_VERSION
);
4787 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4788 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");