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
);
835 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
836 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
838 if (bond_is_lb(bond
))
839 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
841 if (bond_uses_primary(bond
)) {
842 netdev_info(bond
->dev
, "making interface %s the new active one\n",
843 new_active
->dev
->name
);
848 if (bond_uses_primary(bond
))
849 bond_hw_addr_swap(bond
, new_active
, old_active
);
851 if (bond_is_lb(bond
)) {
852 bond_alb_handle_active_change(bond
, new_active
);
854 bond_set_slave_inactive_flags(old_active
,
855 BOND_SLAVE_NOTIFY_NOW
);
857 bond_set_slave_active_flags(new_active
,
858 BOND_SLAVE_NOTIFY_NOW
);
860 rcu_assign_pointer(bond
->curr_active_slave
, new_active
);
863 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
) {
865 bond_set_slave_inactive_flags(old_active
,
866 BOND_SLAVE_NOTIFY_NOW
);
869 bool should_notify_peers
= false;
871 bond_set_slave_active_flags(new_active
,
872 BOND_SLAVE_NOTIFY_NOW
);
874 if (bond
->params
.fail_over_mac
)
875 bond_do_fail_over_mac(bond
, new_active
,
878 if (netif_running(bond
->dev
)) {
879 bond
->send_peer_notif
=
880 bond
->params
.num_peer_notif
;
881 should_notify_peers
=
882 bond_should_notify_peers(bond
);
885 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, bond
->dev
);
886 if (should_notify_peers
)
887 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
892 /* resend IGMP joins since active slave has changed or
893 * all were sent on curr_active_slave.
894 * resend only if bond is brought up with the affected
895 * bonding modes and the retransmission is enabled
897 if (netif_running(bond
->dev
) && (bond
->params
.resend_igmp
> 0) &&
898 ((bond_uses_primary(bond
) && new_active
) ||
899 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
)) {
900 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
901 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 1);
906 * bond_select_active_slave - select a new active slave, if needed
907 * @bond: our bonding struct
909 * This functions should be called when one of the following occurs:
910 * - The old curr_active_slave has been released or lost its link.
911 * - The primary_slave has got its link back.
912 * - A slave has got its link back and there's no old curr_active_slave.
914 * Caller must hold RTNL.
916 void bond_select_active_slave(struct bonding
*bond
)
918 struct slave
*best_slave
;
923 best_slave
= bond_find_best_slave(bond
);
924 if (best_slave
!= rtnl_dereference(bond
->curr_active_slave
)) {
925 bond_change_active_slave(bond
, best_slave
);
926 rv
= bond_set_carrier(bond
);
930 if (netif_carrier_ok(bond
->dev
)) {
931 netdev_info(bond
->dev
, "first active interface up!\n");
933 netdev_info(bond
->dev
, "now running without any active interface!\n");
938 #ifdef CONFIG_NET_POLL_CONTROLLER
939 static inline int slave_enable_netpoll(struct slave
*slave
)
944 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
949 err
= __netpoll_setup(np
, slave
->dev
);
958 static inline void slave_disable_netpoll(struct slave
*slave
)
960 struct netpoll
*np
= slave
->np
;
966 __netpoll_free_async(np
);
969 static void bond_poll_controller(struct net_device
*bond_dev
)
971 struct bonding
*bond
= netdev_priv(bond_dev
);
972 struct slave
*slave
= NULL
;
973 struct list_head
*iter
;
974 struct ad_info ad_info
;
975 struct netpoll_info
*ni
;
976 const struct net_device_ops
*ops
;
978 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
979 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
);
1003 rcu_read_unlock_bh();
1006 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1008 struct bonding
*bond
= netdev_priv(bond_dev
);
1009 struct list_head
*iter
;
1010 struct slave
*slave
;
1012 bond_for_each_slave(bond
, slave
, iter
)
1013 if (bond_slave_is_up(slave
))
1014 slave_disable_netpoll(slave
);
1017 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
1019 struct bonding
*bond
= netdev_priv(dev
);
1020 struct list_head
*iter
;
1021 struct slave
*slave
;
1024 bond_for_each_slave(bond
, slave
, iter
) {
1025 err
= slave_enable_netpoll(slave
);
1027 bond_netpoll_cleanup(dev
);
1034 static inline int slave_enable_netpoll(struct slave
*slave
)
1038 static inline void slave_disable_netpoll(struct slave
*slave
)
1041 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1046 /*---------------------------------- IOCTL ----------------------------------*/
1048 static netdev_features_t
bond_fix_features(struct net_device
*dev
,
1049 netdev_features_t features
)
1051 struct bonding
*bond
= netdev_priv(dev
);
1052 struct list_head
*iter
;
1053 netdev_features_t mask
;
1054 struct slave
*slave
;
1058 features
&= ~NETIF_F_ONE_FOR_ALL
;
1059 features
|= NETIF_F_ALL_FOR_ALL
;
1061 bond_for_each_slave(bond
, slave
, iter
) {
1062 features
= netdev_increment_features(features
,
1063 slave
->dev
->features
,
1066 features
= netdev_add_tso_features(features
, mask
);
1071 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1072 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1073 NETIF_F_HIGHDMA | NETIF_F_LRO)
1075 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1078 static void bond_compute_features(struct bonding
*bond
)
1080 unsigned int dst_release_flag
= IFF_XMIT_DST_RELEASE
|
1081 IFF_XMIT_DST_RELEASE_PERM
;
1082 netdev_features_t vlan_features
= BOND_VLAN_FEATURES
;
1083 netdev_features_t enc_features
= BOND_ENC_FEATURES
;
1084 struct net_device
*bond_dev
= bond
->dev
;
1085 struct list_head
*iter
;
1086 struct slave
*slave
;
1087 unsigned short max_hard_header_len
= ETH_HLEN
;
1088 unsigned int gso_max_size
= GSO_MAX_SIZE
;
1089 u16 gso_max_segs
= GSO_MAX_SEGS
;
1091 if (!bond_has_slaves(bond
))
1093 vlan_features
&= NETIF_F_ALL_FOR_ALL
;
1095 bond_for_each_slave(bond
, slave
, iter
) {
1096 vlan_features
= netdev_increment_features(vlan_features
,
1097 slave
->dev
->vlan_features
, BOND_VLAN_FEATURES
);
1099 enc_features
= netdev_increment_features(enc_features
,
1100 slave
->dev
->hw_enc_features
,
1102 dst_release_flag
&= slave
->dev
->priv_flags
;
1103 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1104 max_hard_header_len
= slave
->dev
->hard_header_len
;
1106 gso_max_size
= min(gso_max_size
, slave
->dev
->gso_max_size
);
1107 gso_max_segs
= min(gso_max_segs
, slave
->dev
->gso_max_segs
);
1111 bond_dev
->vlan_features
= vlan_features
;
1112 bond_dev
->hw_enc_features
= enc_features
| NETIF_F_GSO_ENCAP_ALL
;
1113 bond_dev
->hard_header_len
= max_hard_header_len
;
1114 bond_dev
->gso_max_segs
= gso_max_segs
;
1115 netif_set_gso_max_size(bond_dev
, gso_max_size
);
1117 bond_dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
1118 if ((bond_dev
->priv_flags
& IFF_XMIT_DST_RELEASE_PERM
) &&
1119 dst_release_flag
== (IFF_XMIT_DST_RELEASE
| IFF_XMIT_DST_RELEASE_PERM
))
1120 bond_dev
->priv_flags
|= IFF_XMIT_DST_RELEASE
;
1122 netdev_change_features(bond_dev
);
1125 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1126 struct net_device
*slave_dev
)
1128 bond_dev
->header_ops
= slave_dev
->header_ops
;
1130 bond_dev
->type
= slave_dev
->type
;
1131 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1132 bond_dev
->addr_len
= slave_dev
->addr_len
;
1134 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1135 slave_dev
->addr_len
);
1138 /* On bonding slaves other than the currently active slave, suppress
1139 * duplicates except for alb non-mcast/bcast.
1141 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1142 struct slave
*slave
,
1143 struct bonding
*bond
)
1145 if (bond_is_slave_inactive(slave
)) {
1146 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1147 skb
->pkt_type
!= PACKET_BROADCAST
&&
1148 skb
->pkt_type
!= PACKET_MULTICAST
)
1155 static rx_handler_result_t
bond_handle_frame(struct sk_buff
**pskb
)
1157 struct sk_buff
*skb
= *pskb
;
1158 struct slave
*slave
;
1159 struct bonding
*bond
;
1160 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
1162 int ret
= RX_HANDLER_ANOTHER
;
1164 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1166 return RX_HANDLER_CONSUMED
;
1170 slave
= bond_slave_get_rcu(skb
->dev
);
1173 recv_probe
= ACCESS_ONCE(bond
->recv_probe
);
1175 ret
= recv_probe(skb
, bond
, slave
);
1176 if (ret
== RX_HANDLER_CONSUMED
) {
1182 if (bond_should_deliver_exact_match(skb
, slave
, bond
)) {
1183 return RX_HANDLER_EXACT
;
1186 skb
->dev
= bond
->dev
;
1188 if (BOND_MODE(bond
) == BOND_MODE_ALB
&&
1189 bond
->dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1190 skb
->pkt_type
== PACKET_HOST
) {
1192 if (unlikely(skb_cow_head(skb
,
1193 skb
->data
- skb_mac_header(skb
)))) {
1195 return RX_HANDLER_CONSUMED
;
1197 ether_addr_copy(eth_hdr(skb
)->h_dest
, bond
->dev
->dev_addr
);
1203 static int bond_master_upper_dev_link(struct net_device
*bond_dev
,
1204 struct net_device
*slave_dev
,
1205 struct slave
*slave
)
1209 err
= netdev_master_upper_dev_link_private(slave_dev
, bond_dev
, slave
);
1212 slave_dev
->flags
|= IFF_SLAVE
;
1213 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1217 static void bond_upper_dev_unlink(struct net_device
*bond_dev
,
1218 struct net_device
*slave_dev
)
1220 netdev_upper_dev_unlink(slave_dev
, bond_dev
);
1221 slave_dev
->flags
&= ~IFF_SLAVE
;
1222 rtmsg_ifinfo(RTM_NEWLINK
, slave_dev
, IFF_SLAVE
, GFP_KERNEL
);
1225 static struct slave
*bond_alloc_slave(struct bonding
*bond
)
1227 struct slave
*slave
= NULL
;
1229 slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1233 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1234 SLAVE_AD_INFO(slave
) = kzalloc(sizeof(struct ad_slave_info
),
1236 if (!SLAVE_AD_INFO(slave
)) {
1244 static void bond_free_slave(struct slave
*slave
)
1246 struct bonding
*bond
= bond_get_bond_by_slave(slave
);
1248 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1249 kfree(SLAVE_AD_INFO(slave
));
1254 static void bond_fill_ifbond(struct bonding
*bond
, struct ifbond
*info
)
1256 info
->bond_mode
= BOND_MODE(bond
);
1257 info
->miimon
= bond
->params
.miimon
;
1258 info
->num_slaves
= bond
->slave_cnt
;
1261 static void bond_fill_ifslave(struct slave
*slave
, struct ifslave
*info
)
1263 strcpy(info
->slave_name
, slave
->dev
->name
);
1264 info
->link
= slave
->link
;
1265 info
->state
= bond_slave_state(slave
);
1266 info
->link_failure_count
= slave
->link_failure_count
;
1269 static void bond_netdev_notify(struct net_device
*dev
,
1270 struct netdev_bonding_info
*info
)
1273 netdev_bonding_info_change(dev
, info
);
1277 static void bond_netdev_notify_work(struct work_struct
*_work
)
1279 struct netdev_notify_work
*w
=
1280 container_of(_work
, struct netdev_notify_work
, work
.work
);
1282 bond_netdev_notify(w
->dev
, &w
->bonding_info
);
1287 void bond_queue_slave_event(struct slave
*slave
)
1289 struct bonding
*bond
= slave
->bond
;
1290 struct netdev_notify_work
*nnw
= kzalloc(sizeof(*nnw
), GFP_ATOMIC
);
1295 dev_hold(slave
->dev
);
1296 nnw
->dev
= slave
->dev
;
1297 bond_fill_ifslave(slave
, &nnw
->bonding_info
.slave
);
1298 bond_fill_ifbond(bond
, &nnw
->bonding_info
.master
);
1299 INIT_DELAYED_WORK(&nnw
->work
, bond_netdev_notify_work
);
1301 queue_delayed_work(slave
->bond
->wq
, &nnw
->work
, 0);
1304 /* enslave device <slave> to bond device <master> */
1305 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1307 struct bonding
*bond
= netdev_priv(bond_dev
);
1308 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1309 struct slave
*new_slave
= NULL
, *prev_slave
;
1310 struct sockaddr addr
;
1314 if (!bond
->params
.use_carrier
&&
1315 slave_dev
->ethtool_ops
->get_link
== NULL
&&
1316 slave_ops
->ndo_do_ioctl
== NULL
) {
1317 netdev_warn(bond_dev
, "no link monitoring support for %s\n",
1321 /* already enslaved */
1322 if (slave_dev
->flags
& IFF_SLAVE
) {
1323 netdev_dbg(bond_dev
, "Error: Device was already enslaved\n");
1327 if (bond_dev
== slave_dev
) {
1328 netdev_err(bond_dev
, "cannot enslave bond to itself.\n");
1332 /* vlan challenged mutual exclusion */
1333 /* no need to lock since we're protected by rtnl_lock */
1334 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1335 netdev_dbg(bond_dev
, "%s is NETIF_F_VLAN_CHALLENGED\n",
1337 if (vlan_uses_dev(bond_dev
)) {
1338 netdev_err(bond_dev
, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1339 slave_dev
->name
, bond_dev
->name
);
1342 netdev_warn(bond_dev
, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1343 slave_dev
->name
, slave_dev
->name
,
1347 netdev_dbg(bond_dev
, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1351 /* Old ifenslave binaries are no longer supported. These can
1352 * be identified with moderate accuracy by the state of the slave:
1353 * the current ifenslave will set the interface down prior to
1354 * enslaving it; the old ifenslave will not.
1356 if ((slave_dev
->flags
& IFF_UP
)) {
1357 netdev_err(bond_dev
, "%s is up - this may be due to an out of date ifenslave\n",
1360 goto err_undo_flags
;
1363 /* set bonding device ether type by slave - bonding netdevices are
1364 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1365 * there is a need to override some of the type dependent attribs/funcs.
1367 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1368 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1370 if (!bond_has_slaves(bond
)) {
1371 if (bond_dev
->type
!= slave_dev
->type
) {
1372 netdev_dbg(bond_dev
, "change device type from %d to %d\n",
1373 bond_dev
->type
, slave_dev
->type
);
1375 res
= call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE
,
1377 res
= notifier_to_errno(res
);
1379 netdev_err(bond_dev
, "refused to change device type\n");
1381 goto err_undo_flags
;
1384 /* Flush unicast and multicast addresses */
1385 dev_uc_flush(bond_dev
);
1386 dev_mc_flush(bond_dev
);
1388 if (slave_dev
->type
!= ARPHRD_ETHER
)
1389 bond_setup_by_slave(bond_dev
, slave_dev
);
1391 ether_setup(bond_dev
);
1392 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1395 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE
,
1398 } else if (bond_dev
->type
!= slave_dev
->type
) {
1399 netdev_err(bond_dev
, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1400 slave_dev
->name
, slave_dev
->type
, bond_dev
->type
);
1402 goto err_undo_flags
;
1405 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1406 netdev_warn(bond_dev
, "The slave device specified does not support setting the MAC address\n");
1407 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
&&
1408 bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1409 if (!bond_has_slaves(bond
)) {
1410 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1411 netdev_warn(bond_dev
, "Setting fail_over_mac to active for active-backup mode\n");
1413 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");
1415 goto err_undo_flags
;
1420 call_netdevice_notifiers(NETDEV_JOIN
, slave_dev
);
1422 /* If this is the first slave, then we need to set the master's hardware
1423 * address to be the same as the slave's.
1425 if (!bond_has_slaves(bond
) &&
1426 bond
->dev
->addr_assign_type
== NET_ADDR_RANDOM
)
1427 bond_set_dev_addr(bond
->dev
, slave_dev
);
1429 new_slave
= bond_alloc_slave(bond
);
1432 goto err_undo_flags
;
1435 new_slave
->bond
= bond
;
1436 new_slave
->dev
= slave_dev
;
1437 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1438 * is set via sysfs or module option if desired.
1440 new_slave
->queue_id
= 0;
1442 /* Save slave's original mtu and then set it to match the bond */
1443 new_slave
->original_mtu
= slave_dev
->mtu
;
1444 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1446 netdev_dbg(bond_dev
, "Error %d calling dev_set_mtu\n", res
);
1450 /* Save slave's original ("permanent") mac address for modes
1451 * that need it, and for restoring it upon release, and then
1452 * set it to the master's address
1454 ether_addr_copy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
);
1456 if (!bond
->params
.fail_over_mac
||
1457 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1458 /* Set slave to master's mac address. The application already
1459 * set the master's mac address to that of the first slave
1461 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1462 addr
.sa_family
= slave_dev
->type
;
1463 res
= dev_set_mac_address(slave_dev
, &addr
);
1465 netdev_dbg(bond_dev
, "Error %d calling set_mac_address\n", res
);
1466 goto err_restore_mtu
;
1470 /* open the slave since the application closed it */
1471 res
= dev_open(slave_dev
);
1473 netdev_dbg(bond_dev
, "Opening slave %s failed\n", slave_dev
->name
);
1474 goto err_restore_mac
;
1477 slave_dev
->priv_flags
|= IFF_BONDING
;
1478 /* initialize slave stats */
1479 dev_get_stats(new_slave
->dev
, &new_slave
->slave_stats
);
1481 if (bond_is_lb(bond
)) {
1482 /* bond_alb_init_slave() must be called before all other stages since
1483 * it might fail and we do not want to have to undo everything
1485 res
= bond_alb_init_slave(bond
, new_slave
);
1490 /* If the mode uses primary, then the following is handled by
1491 * bond_change_active_slave().
1493 if (!bond_uses_primary(bond
)) {
1494 /* set promiscuity level to new slave */
1495 if (bond_dev
->flags
& IFF_PROMISC
) {
1496 res
= dev_set_promiscuity(slave_dev
, 1);
1501 /* set allmulti level to new slave */
1502 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1503 res
= dev_set_allmulti(slave_dev
, 1);
1508 netif_addr_lock_bh(bond_dev
);
1510 dev_mc_sync_multiple(slave_dev
, bond_dev
);
1511 dev_uc_sync_multiple(slave_dev
, bond_dev
);
1513 netif_addr_unlock_bh(bond_dev
);
1516 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
1517 /* add lacpdu mc addr to mc list */
1518 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1520 dev_mc_add(slave_dev
, lacpdu_multicast
);
1523 res
= vlan_vids_add_by_dev(slave_dev
, bond_dev
);
1525 netdev_err(bond_dev
, "Couldn't add bond vlan ids to %s\n",
1530 prev_slave
= bond_last_slave(bond
);
1532 new_slave
->delay
= 0;
1533 new_slave
->link_failure_count
= 0;
1535 bond_update_speed_duplex(new_slave
);
1537 new_slave
->last_rx
= jiffies
-
1538 (msecs_to_jiffies(bond
->params
.arp_interval
) + 1);
1539 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
1540 new_slave
->target_last_arp_rx
[i
] = new_slave
->last_rx
;
1542 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1543 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1545 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1546 /* miimon is set but a bonded network driver
1547 * does not support ETHTOOL/MII and
1548 * arp_interval is not set. Note: if
1549 * use_carrier is enabled, we will never go
1550 * here (because netif_carrier is always
1551 * supported); thus, we don't need to change
1552 * the messages for netif_carrier.
1554 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",
1556 } else if (link_reporting
== -1) {
1557 /* unable get link status using mii/ethtool */
1558 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",
1563 /* check for initial state */
1564 if (bond
->params
.miimon
) {
1565 if (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
) {
1566 if (bond
->params
.updelay
) {
1567 bond_set_slave_link_state(new_slave
,
1569 new_slave
->delay
= bond
->params
.updelay
;
1571 bond_set_slave_link_state(new_slave
,
1575 bond_set_slave_link_state(new_slave
, BOND_LINK_DOWN
);
1577 } else if (bond
->params
.arp_interval
) {
1578 bond_set_slave_link_state(new_slave
,
1579 (netif_carrier_ok(slave_dev
) ?
1580 BOND_LINK_UP
: BOND_LINK_DOWN
));
1582 bond_set_slave_link_state(new_slave
, BOND_LINK_UP
);
1585 if (new_slave
->link
!= BOND_LINK_DOWN
)
1586 new_slave
->last_link_up
= jiffies
;
1587 netdev_dbg(bond_dev
, "Initial state of slave_dev is BOND_LINK_%s\n",
1588 new_slave
->link
== BOND_LINK_DOWN
? "DOWN" :
1589 (new_slave
->link
== BOND_LINK_UP
? "UP" : "BACK"));
1591 if (bond_uses_primary(bond
) && bond
->params
.primary
[0]) {
1592 /* if there is a primary slave, remember it */
1593 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1594 rcu_assign_pointer(bond
->primary_slave
, new_slave
);
1595 bond
->force_primary
= true;
1599 switch (BOND_MODE(bond
)) {
1600 case BOND_MODE_ACTIVEBACKUP
:
1601 bond_set_slave_inactive_flags(new_slave
,
1602 BOND_SLAVE_NOTIFY_NOW
);
1604 case BOND_MODE_8023AD
:
1605 /* in 802.3ad mode, the internal mechanism
1606 * will activate the slaves in the selected
1609 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1610 /* if this is the first slave */
1612 SLAVE_AD_INFO(new_slave
)->id
= 1;
1613 /* Initialize AD with the number of times that the AD timer is called in 1 second
1614 * can be called only after the mac address of the bond is set
1616 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
);
1618 SLAVE_AD_INFO(new_slave
)->id
=
1619 SLAVE_AD_INFO(prev_slave
)->id
+ 1;
1622 bond_3ad_bind_slave(new_slave
);
1626 bond_set_active_slave(new_slave
);
1627 bond_set_slave_inactive_flags(new_slave
, BOND_SLAVE_NOTIFY_NOW
);
1630 netdev_dbg(bond_dev
, "This slave is always active in trunk mode\n");
1632 /* always active in trunk mode */
1633 bond_set_active_slave(new_slave
);
1635 /* In trunking mode there is little meaning to curr_active_slave
1636 * anyway (it holds no special properties of the bond device),
1637 * so we can change it without calling change_active_interface()
1639 if (!rcu_access_pointer(bond
->curr_active_slave
) &&
1640 new_slave
->link
== BOND_LINK_UP
)
1641 rcu_assign_pointer(bond
->curr_active_slave
, new_slave
);
1644 } /* switch(bond_mode) */
1646 #ifdef CONFIG_NET_POLL_CONTROLLER
1647 slave_dev
->npinfo
= bond
->dev
->npinfo
;
1648 if (slave_dev
->npinfo
) {
1649 if (slave_enable_netpoll(new_slave
)) {
1650 netdev_info(bond_dev
, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1657 if (!(bond_dev
->features
& NETIF_F_LRO
))
1658 dev_disable_lro(slave_dev
);
1660 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
,
1663 netdev_dbg(bond_dev
, "Error %d calling netdev_rx_handler_register\n", res
);
1667 res
= bond_master_upper_dev_link(bond_dev
, slave_dev
, new_slave
);
1669 netdev_dbg(bond_dev
, "Error %d calling bond_master_upper_dev_link\n", res
);
1670 goto err_unregister
;
1673 res
= bond_sysfs_slave_add(new_slave
);
1675 netdev_dbg(bond_dev
, "Error %d calling bond_sysfs_slave_add\n", res
);
1676 goto err_upper_unlink
;
1680 bond_compute_features(bond
);
1681 bond_set_carrier(bond
);
1683 if (bond_uses_primary(bond
)) {
1685 bond_select_active_slave(bond
);
1686 unblock_netpoll_tx();
1689 if (bond_mode_uses_xmit_hash(bond
))
1690 bond_update_slave_arr(bond
, NULL
);
1692 netdev_info(bond_dev
, "Enslaving %s as %s interface with %s link\n",
1694 bond_is_active_slave(new_slave
) ? "an active" : "a backup",
1695 new_slave
->link
!= BOND_LINK_DOWN
? "an up" : "a down");
1697 /* enslave is successful */
1698 bond_queue_slave_event(new_slave
);
1701 /* Undo stages on error */
1703 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1706 netdev_rx_handler_unregister(slave_dev
);
1709 if (!bond_uses_primary(bond
))
1710 bond_hw_addr_flush(bond_dev
, slave_dev
);
1712 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1713 if (rcu_access_pointer(bond
->primary_slave
) == new_slave
)
1714 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
1715 if (rcu_access_pointer(bond
->curr_active_slave
) == new_slave
) {
1717 bond_change_active_slave(bond
, NULL
);
1718 bond_select_active_slave(bond
);
1719 unblock_netpoll_tx();
1721 /* either primary_slave or curr_active_slave might've changed */
1723 slave_disable_netpoll(new_slave
);
1726 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1727 dev_close(slave_dev
);
1730 if (!bond
->params
.fail_over_mac
||
1731 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1732 /* XXX TODO - fom follow mode needs to change master's
1733 * MAC if this slave's MAC is in use by the bond, or at
1734 * least print a warning.
1736 ether_addr_copy(addr
.sa_data
, new_slave
->perm_hwaddr
);
1737 addr
.sa_family
= slave_dev
->type
;
1738 dev_set_mac_address(slave_dev
, &addr
);
1742 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1745 bond_free_slave(new_slave
);
1748 /* Enslave of first slave has failed and we need to fix master's mac */
1749 if (!bond_has_slaves(bond
)) {
1750 if (ether_addr_equal_64bits(bond_dev
->dev_addr
,
1751 slave_dev
->dev_addr
))
1752 eth_hw_addr_random(bond_dev
);
1753 if (bond_dev
->type
!= ARPHRD_ETHER
) {
1754 ether_setup(bond_dev
);
1755 bond_dev
->flags
|= IFF_MASTER
;
1756 bond_dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1763 /* Try to release the slave device <slave> from the bond device <master>
1764 * It is legal to access curr_active_slave without a lock because all the function
1765 * is RTNL-locked. If "all" is true it means that the function is being called
1766 * while destroying a bond interface and all slaves are being released.
1768 * The rules for slave state should be:
1769 * for Active/Backup:
1770 * Active stays on all backups go down
1771 * for Bonded connections:
1772 * The first up interface should be left on and all others downed.
1774 static int __bond_release_one(struct net_device
*bond_dev
,
1775 struct net_device
*slave_dev
,
1778 struct bonding
*bond
= netdev_priv(bond_dev
);
1779 struct slave
*slave
, *oldcurrent
;
1780 struct sockaddr addr
;
1781 int old_flags
= bond_dev
->flags
;
1782 netdev_features_t old_features
= bond_dev
->features
;
1784 /* slave is not a slave or master is not master of this slave */
1785 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1786 !netdev_has_upper_dev(slave_dev
, bond_dev
)) {
1787 netdev_dbg(bond_dev
, "cannot release %s\n",
1794 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1796 /* not a slave of this bond */
1797 netdev_info(bond_dev
, "%s not enslaved\n",
1799 unblock_netpoll_tx();
1803 bond_sysfs_slave_del(slave
);
1805 /* recompute stats just before removing the slave */
1806 bond_get_stats(bond
->dev
, &bond
->bond_stats
);
1808 bond_upper_dev_unlink(bond_dev
, slave_dev
);
1809 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1810 * for this slave anymore.
1812 netdev_rx_handler_unregister(slave_dev
);
1814 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
1815 bond_3ad_unbind_slave(slave
);
1817 if (bond_mode_uses_xmit_hash(bond
))
1818 bond_update_slave_arr(bond
, slave
);
1820 netdev_info(bond_dev
, "Releasing %s interface %s\n",
1821 bond_is_active_slave(slave
) ? "active" : "backup",
1824 oldcurrent
= rcu_access_pointer(bond
->curr_active_slave
);
1826 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
1828 if (!all
&& (!bond
->params
.fail_over_mac
||
1829 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
)) {
1830 if (ether_addr_equal_64bits(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1831 bond_has_slaves(bond
))
1832 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",
1833 slave_dev
->name
, slave
->perm_hwaddr
,
1834 bond_dev
->name
, slave_dev
->name
);
1837 if (rtnl_dereference(bond
->primary_slave
) == slave
)
1838 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
1840 if (oldcurrent
== slave
)
1841 bond_change_active_slave(bond
, NULL
);
1843 if (bond_is_lb(bond
)) {
1844 /* Must be called only after the slave has been
1845 * detached from the list and the curr_active_slave
1846 * has been cleared (if our_slave == old_current),
1847 * but before a new active slave is selected.
1849 bond_alb_deinit_slave(bond
, slave
);
1853 RCU_INIT_POINTER(bond
->curr_active_slave
, NULL
);
1854 } else if (oldcurrent
== slave
) {
1855 /* Note that we hold RTNL over this sequence, so there
1856 * is no concern that another slave add/remove event
1859 bond_select_active_slave(bond
);
1862 if (!bond_has_slaves(bond
)) {
1863 bond_set_carrier(bond
);
1864 eth_hw_addr_random(bond_dev
);
1867 unblock_netpoll_tx();
1871 if (!bond_has_slaves(bond
)) {
1872 call_netdevice_notifiers(NETDEV_CHANGEADDR
, bond
->dev
);
1873 call_netdevice_notifiers(NETDEV_RELEASE
, bond
->dev
);
1876 bond_compute_features(bond
);
1877 if (!(bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
1878 (old_features
& NETIF_F_VLAN_CHALLENGED
))
1879 netdev_info(bond_dev
, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1880 slave_dev
->name
, bond_dev
->name
);
1882 vlan_vids_del_by_dev(slave_dev
, bond_dev
);
1884 /* If the mode uses primary, then this case was handled above by
1885 * bond_change_active_slave(..., NULL)
1887 if (!bond_uses_primary(bond
)) {
1888 /* unset promiscuity level from slave
1889 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1890 * of the IFF_PROMISC flag in the bond_dev, but we need the
1891 * value of that flag before that change, as that was the value
1892 * when this slave was attached, so we cache at the start of the
1893 * function and use it here. Same goes for ALLMULTI below
1895 if (old_flags
& IFF_PROMISC
)
1896 dev_set_promiscuity(slave_dev
, -1);
1898 /* unset allmulti level from slave */
1899 if (old_flags
& IFF_ALLMULTI
)
1900 dev_set_allmulti(slave_dev
, -1);
1902 bond_hw_addr_flush(bond_dev
, slave_dev
);
1905 slave_disable_netpoll(slave
);
1907 /* close slave before restoring its mac address */
1908 dev_close(slave_dev
);
1910 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
||
1911 BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
1912 /* restore original ("permanent") mac address */
1913 ether_addr_copy(addr
.sa_data
, slave
->perm_hwaddr
);
1914 addr
.sa_family
= slave_dev
->type
;
1915 dev_set_mac_address(slave_dev
, &addr
);
1918 dev_set_mtu(slave_dev
, slave
->original_mtu
);
1920 slave_dev
->priv_flags
&= ~IFF_BONDING
;
1922 bond_free_slave(slave
);
1927 /* A wrapper used because of ndo_del_link */
1928 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1930 return __bond_release_one(bond_dev
, slave_dev
, false);
1933 /* First release a slave and then destroy the bond if no more slaves are left.
1934 * Must be under rtnl_lock when this function is called.
1936 static int bond_release_and_destroy(struct net_device
*bond_dev
,
1937 struct net_device
*slave_dev
)
1939 struct bonding
*bond
= netdev_priv(bond_dev
);
1942 ret
= bond_release(bond_dev
, slave_dev
);
1943 if (ret
== 0 && !bond_has_slaves(bond
)) {
1944 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
1945 netdev_info(bond_dev
, "Destroying bond %s\n",
1947 bond_remove_proc_entry(bond
);
1948 unregister_netdevice(bond_dev
);
1953 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
1955 struct bonding
*bond
= netdev_priv(bond_dev
);
1956 bond_fill_ifbond(bond
, info
);
1960 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
1962 struct bonding
*bond
= netdev_priv(bond_dev
);
1963 struct list_head
*iter
;
1964 int i
= 0, res
= -ENODEV
;
1965 struct slave
*slave
;
1967 bond_for_each_slave(bond
, slave
, iter
) {
1968 if (i
++ == (int)info
->slave_id
) {
1970 bond_fill_ifslave(slave
, info
);
1978 /*-------------------------------- Monitoring -------------------------------*/
1980 /* called with rcu_read_lock() */
1981 static int bond_miimon_inspect(struct bonding
*bond
)
1983 int link_state
, commit
= 0;
1984 struct list_head
*iter
;
1985 struct slave
*slave
;
1986 bool ignore_updelay
;
1988 ignore_updelay
= !rcu_dereference(bond
->curr_active_slave
);
1990 bond_for_each_slave_rcu(bond
, slave
, iter
) {
1991 slave
->new_link
= BOND_LINK_NOCHANGE
;
1993 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
1995 switch (slave
->link
) {
2000 bond_set_slave_link_state(slave
, BOND_LINK_FAIL
);
2001 slave
->delay
= bond
->params
.downdelay
;
2003 netdev_info(bond
->dev
, "link status down for %sinterface %s, disabling it in %d ms\n",
2005 BOND_MODE_ACTIVEBACKUP
) ?
2006 (bond_is_active_slave(slave
) ?
2007 "active " : "backup ") : "",
2009 bond
->params
.downdelay
* bond
->params
.miimon
);
2012 case BOND_LINK_FAIL
:
2014 /* recovered before downdelay expired */
2015 bond_set_slave_link_state(slave
, BOND_LINK_UP
);
2016 slave
->last_link_up
= jiffies
;
2017 netdev_info(bond
->dev
, "link status up again after %d ms for interface %s\n",
2018 (bond
->params
.downdelay
- slave
->delay
) *
2019 bond
->params
.miimon
,
2024 if (slave
->delay
<= 0) {
2025 slave
->new_link
= BOND_LINK_DOWN
;
2033 case BOND_LINK_DOWN
:
2037 bond_set_slave_link_state(slave
, BOND_LINK_BACK
);
2038 slave
->delay
= bond
->params
.updelay
;
2041 netdev_info(bond
->dev
, "link status up for interface %s, enabling it in %d ms\n",
2043 ignore_updelay
? 0 :
2044 bond
->params
.updelay
*
2045 bond
->params
.miimon
);
2048 case BOND_LINK_BACK
:
2050 bond_set_slave_link_state(slave
,
2052 netdev_info(bond
->dev
, "link status down again after %d ms for interface %s\n",
2053 (bond
->params
.updelay
- slave
->delay
) *
2054 bond
->params
.miimon
,
2063 if (slave
->delay
<= 0) {
2064 slave
->new_link
= BOND_LINK_UP
;
2066 ignore_updelay
= false;
2078 static void bond_miimon_commit(struct bonding
*bond
)
2080 struct list_head
*iter
;
2081 struct slave
*slave
, *primary
;
2083 bond_for_each_slave(bond
, slave
, iter
) {
2084 switch (slave
->new_link
) {
2085 case BOND_LINK_NOCHANGE
:
2089 bond_set_slave_link_state(slave
, BOND_LINK_UP
);
2090 slave
->last_link_up
= jiffies
;
2092 primary
= rtnl_dereference(bond
->primary_slave
);
2093 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
2094 /* prevent it from being the active one */
2095 bond_set_backup_slave(slave
);
2096 } else if (BOND_MODE(bond
) != BOND_MODE_ACTIVEBACKUP
) {
2097 /* make it immediately active */
2098 bond_set_active_slave(slave
);
2099 } else if (slave
!= primary
) {
2100 /* prevent it from being the active one */
2101 bond_set_backup_slave(slave
);
2104 netdev_info(bond
->dev
, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2106 slave
->speed
== SPEED_UNKNOWN
? 0 : slave
->speed
,
2107 slave
->duplex
? "full" : "half");
2109 /* notify ad that the link status has changed */
2110 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
2111 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2113 if (bond_is_lb(bond
))
2114 bond_alb_handle_link_change(bond
, slave
,
2117 if (BOND_MODE(bond
) == BOND_MODE_XOR
)
2118 bond_update_slave_arr(bond
, NULL
);
2120 if (!bond
->curr_active_slave
|| slave
== primary
)
2125 case BOND_LINK_DOWN
:
2126 if (slave
->link_failure_count
< UINT_MAX
)
2127 slave
->link_failure_count
++;
2129 bond_set_slave_link_state(slave
, BOND_LINK_DOWN
);
2131 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
2132 BOND_MODE(bond
) == BOND_MODE_8023AD
)
2133 bond_set_slave_inactive_flags(slave
,
2134 BOND_SLAVE_NOTIFY_NOW
);
2136 netdev_info(bond
->dev
, "link status definitely down for interface %s, disabling it\n",
2139 if (BOND_MODE(bond
) == BOND_MODE_8023AD
)
2140 bond_3ad_handle_link_change(slave
,
2143 if (bond_is_lb(bond
))
2144 bond_alb_handle_link_change(bond
, slave
,
2147 if (BOND_MODE(bond
) == BOND_MODE_XOR
)
2148 bond_update_slave_arr(bond
, NULL
);
2150 if (slave
== rcu_access_pointer(bond
->curr_active_slave
))
2156 netdev_err(bond
->dev
, "invalid new link %d on slave %s\n",
2157 slave
->new_link
, slave
->dev
->name
);
2158 slave
->new_link
= BOND_LINK_NOCHANGE
;
2165 bond_select_active_slave(bond
);
2166 unblock_netpoll_tx();
2169 bond_set_carrier(bond
);
2174 * Really a wrapper that splits the mii monitor into two phases: an
2175 * inspection, then (if inspection indicates something needs to be done)
2176 * an acquisition of appropriate locks followed by a commit phase to
2177 * implement whatever link state changes are indicated.
2179 static void bond_mii_monitor(struct work_struct
*work
)
2181 struct bonding
*bond
= container_of(work
, struct bonding
,
2183 bool should_notify_peers
= false;
2184 unsigned long delay
;
2186 delay
= msecs_to_jiffies(bond
->params
.miimon
);
2188 if (!bond_has_slaves(bond
))
2193 should_notify_peers
= bond_should_notify_peers(bond
);
2195 if (bond_miimon_inspect(bond
)) {
2198 /* Race avoidance with bond_close cancel of workqueue */
2199 if (!rtnl_trylock()) {
2201 should_notify_peers
= false;
2205 bond_miimon_commit(bond
);
2207 rtnl_unlock(); /* might sleep, hold no other locks */
2212 if (bond
->params
.miimon
)
2213 queue_delayed_work(bond
->wq
, &bond
->mii_work
, delay
);
2215 if (should_notify_peers
) {
2216 if (!rtnl_trylock())
2218 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
, bond
->dev
);
2223 static bool bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2225 struct net_device
*upper
;
2226 struct list_head
*iter
;
2229 if (ip
== bond_confirm_addr(bond
->dev
, 0, ip
))
2233 netdev_for_each_all_upper_dev_rcu(bond
->dev
, upper
, iter
) {
2234 if (ip
== bond_confirm_addr(upper
, 0, ip
)) {
2244 /* We go to the (large) trouble of VLAN tagging ARP frames because
2245 * switches in VLAN mode (especially if ports are configured as
2246 * "native" to a VLAN) might not pass non-tagged frames.
2248 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
,
2249 __be32 dest_ip
, __be32 src_ip
,
2250 struct bond_vlan_tag
*tags
)
2252 struct sk_buff
*skb
;
2253 struct bond_vlan_tag
*outer_tag
= tags
;
2255 netdev_dbg(slave_dev
, "arp %d on slave %s: dst %pI4 src %pI4\n",
2256 arp_op
, slave_dev
->name
, &dest_ip
, &src_ip
);
2258 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2259 NULL
, slave_dev
->dev_addr
, NULL
);
2262 net_err_ratelimited("ARP packet allocation failed\n");
2266 if (!tags
|| tags
->vlan_proto
== VLAN_N_VID
)
2271 /* Go through all the tags backwards and add them to the packet */
2272 while (tags
->vlan_proto
!= VLAN_N_VID
) {
2273 if (!tags
->vlan_id
) {
2278 netdev_dbg(slave_dev
, "inner tag: proto %X vid %X\n",
2279 ntohs(outer_tag
->vlan_proto
), tags
->vlan_id
);
2280 skb
= vlan_insert_tag_set_proto(skb
, tags
->vlan_proto
,
2283 net_err_ratelimited("failed to insert inner VLAN tag\n");
2289 /* Set the outer tag */
2290 if (outer_tag
->vlan_id
) {
2291 netdev_dbg(slave_dev
, "outer tag: proto %X vid %X\n",
2292 ntohs(outer_tag
->vlan_proto
), outer_tag
->vlan_id
);
2293 __vlan_hwaccel_put_tag(skb
, outer_tag
->vlan_proto
,
2294 outer_tag
->vlan_id
);
2301 /* Validate the device path between the @start_dev and the @end_dev.
2302 * The path is valid if the @end_dev is reachable through device
2304 * When the path is validated, collect any vlan information in the
2307 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
2308 struct net_device
*end_dev
,
2311 struct bond_vlan_tag
*tags
;
2312 struct net_device
*upper
;
2313 struct list_head
*iter
;
2315 if (start_dev
== end_dev
) {
2316 tags
= kzalloc(sizeof(*tags
) * (level
+ 1), GFP_ATOMIC
);
2318 return ERR_PTR(-ENOMEM
);
2319 tags
[level
].vlan_proto
= VLAN_N_VID
;
2323 netdev_for_each_upper_dev_rcu(start_dev
, upper
, iter
) {
2324 tags
= bond_verify_device_path(upper
, end_dev
, level
+ 1);
2325 if (IS_ERR_OR_NULL(tags
)) {
2330 if (is_vlan_dev(upper
)) {
2331 tags
[level
].vlan_proto
= vlan_dev_vlan_proto(upper
);
2332 tags
[level
].vlan_id
= vlan_dev_vlan_id(upper
);
2341 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2344 struct bond_vlan_tag
*tags
;
2345 __be32
*targets
= bond
->params
.arp_targets
, addr
;
2348 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
&& targets
[i
]; i
++) {
2349 netdev_dbg(bond
->dev
, "basa: target %pI4\n", &targets
[i
]);
2352 /* Find out through which dev should the packet go */
2353 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2356 /* there's no route to target - try to send arp
2357 * probe to generate any traffic (arp_validate=0)
2359 if (bond
->params
.arp_validate
)
2360 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2363 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2368 /* bond device itself */
2369 if (rt
->dst
.dev
== bond
->dev
)
2373 tags
= bond_verify_device_path(bond
->dev
, rt
->dst
.dev
, 0);
2376 if (!IS_ERR_OR_NULL(tags
))
2379 /* Not our device - skip */
2380 netdev_dbg(bond
->dev
, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2381 &targets
[i
], rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2387 addr
= bond_confirm_addr(rt
->dst
.dev
, targets
[i
], 0);
2389 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2395 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2399 if (!sip
|| !bond_has_this_ip(bond
, tip
)) {
2400 netdev_dbg(bond
->dev
, "bva: sip %pI4 tip %pI4 not found\n",
2405 i
= bond_get_targets_ip(bond
->params
.arp_targets
, sip
);
2407 netdev_dbg(bond
->dev
, "bva: sip %pI4 not found in targets\n",
2411 slave
->last_rx
= jiffies
;
2412 slave
->target_last_arp_rx
[i
] = jiffies
;
2415 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
,
2416 struct slave
*slave
)
2418 struct arphdr
*arp
= (struct arphdr
*)skb
->data
;
2419 struct slave
*curr_active_slave
;
2420 unsigned char *arp_ptr
;
2422 int alen
, is_arp
= skb
->protocol
== __cpu_to_be16(ETH_P_ARP
);
2424 if (!slave_do_arp_validate(bond
, slave
)) {
2425 if ((slave_do_arp_validate_only(bond
) && is_arp
) ||
2426 !slave_do_arp_validate_only(bond
))
2427 slave
->last_rx
= jiffies
;
2428 return RX_HANDLER_ANOTHER
;
2429 } else if (!is_arp
) {
2430 return RX_HANDLER_ANOTHER
;
2433 alen
= arp_hdr_len(bond
->dev
);
2435 netdev_dbg(bond
->dev
, "bond_arp_rcv: skb->dev %s\n",
2438 if (alen
> skb_headlen(skb
)) {
2439 arp
= kmalloc(alen
, GFP_ATOMIC
);
2442 if (skb_copy_bits(skb
, 0, arp
, alen
) < 0)
2446 if (arp
->ar_hln
!= bond
->dev
->addr_len
||
2447 skb
->pkt_type
== PACKET_OTHERHOST
||
2448 skb
->pkt_type
== PACKET_LOOPBACK
||
2449 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2450 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2454 arp_ptr
= (unsigned char *)(arp
+ 1);
2455 arp_ptr
+= bond
->dev
->addr_len
;
2456 memcpy(&sip
, arp_ptr
, 4);
2457 arp_ptr
+= 4 + bond
->dev
->addr_len
;
2458 memcpy(&tip
, arp_ptr
, 4);
2460 netdev_dbg(bond
->dev
, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2461 slave
->dev
->name
, bond_slave_state(slave
),
2462 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2465 curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2467 /* Backup slaves won't see the ARP reply, but do come through
2468 * here for each ARP probe (so we swap the sip/tip to validate
2469 * the probe). In a "redundant switch, common router" type of
2470 * configuration, the ARP probe will (hopefully) travel from
2471 * the active, through one switch, the router, then the other
2472 * switch before reaching the backup.
2474 * We 'trust' the arp requests if there is an active slave and
2475 * it received valid arp reply(s) after it became active. This
2476 * is done to avoid endless looping when we can't reach the
2477 * arp_ip_target and fool ourselves with our own arp requests.
2480 if (bond_is_active_slave(slave
))
2481 bond_validate_arp(bond
, slave
, sip
, tip
);
2482 else if (curr_active_slave
&&
2483 time_after(slave_last_rx(bond
, curr_active_slave
),
2484 curr_active_slave
->last_link_up
))
2485 bond_validate_arp(bond
, slave
, tip
, sip
);
2488 if (arp
!= (struct arphdr
*)skb
->data
)
2490 return RX_HANDLER_ANOTHER
;
2493 /* function to verify if we're in the arp_interval timeslice, returns true if
2494 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2495 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2497 static bool bond_time_in_interval(struct bonding
*bond
, unsigned long last_act
,
2500 int delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2502 return time_in_range(jiffies
,
2503 last_act
- delta_in_ticks
,
2504 last_act
+ mod
* delta_in_ticks
+ delta_in_ticks
/2);
2507 /* This function is called regularly to monitor each slave's link
2508 * ensuring that traffic is being sent and received when arp monitoring
2509 * is used in load-balancing mode. if the adapter has been dormant, then an
2510 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2511 * arp monitoring in active backup mode.
2513 static void bond_loadbalance_arp_mon(struct work_struct
*work
)
2515 struct bonding
*bond
= container_of(work
, struct bonding
,
2517 struct slave
*slave
, *oldcurrent
;
2518 struct list_head
*iter
;
2519 int do_failover
= 0, slave_state_changed
= 0;
2521 if (!bond_has_slaves(bond
))
2526 oldcurrent
= rcu_dereference(bond
->curr_active_slave
);
2527 /* see if any of the previous devices are up now (i.e. they have
2528 * xmt and rcv traffic). the curr_active_slave does not come into
2529 * the picture unless it is null. also, slave->last_link_up is not
2530 * needed here because we send an arp on each slave and give a slave
2531 * as long as it needs to get the tx/rx within the delta.
2532 * TODO: what about up/down delay in arp mode? it wasn't here before
2535 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2536 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2538 if (slave
->link
!= BOND_LINK_UP
) {
2539 if (bond_time_in_interval(bond
, trans_start
, 1) &&
2540 bond_time_in_interval(bond
, slave
->last_rx
, 1)) {
2542 slave
->link
= BOND_LINK_UP
;
2543 slave_state_changed
= 1;
2545 /* primary_slave has no meaning in round-robin
2546 * mode. the window of a slave being up and
2547 * curr_active_slave being null after enslaving
2551 netdev_info(bond
->dev
, "link status definitely up for interface %s\n",
2555 netdev_info(bond
->dev
, "interface %s is now up\n",
2560 /* slave->link == BOND_LINK_UP */
2562 /* not all switches will respond to an arp request
2563 * when the source ip is 0, so don't take the link down
2564 * if we don't know our ip yet
2566 if (!bond_time_in_interval(bond
, trans_start
, 2) ||
2567 !bond_time_in_interval(bond
, slave
->last_rx
, 2)) {
2569 slave
->link
= BOND_LINK_DOWN
;
2570 slave_state_changed
= 1;
2572 if (slave
->link_failure_count
< UINT_MAX
)
2573 slave
->link_failure_count
++;
2575 netdev_info(bond
->dev
, "interface %s is now down\n",
2578 if (slave
== oldcurrent
)
2583 /* note: if switch is in round-robin mode, all links
2584 * must tx arp to ensure all links rx an arp - otherwise
2585 * links may oscillate or not come up at all; if switch is
2586 * in something like xor mode, there is nothing we can
2587 * do - all replies will be rx'ed on same link causing slaves
2588 * to be unstable during low/no traffic periods
2590 if (bond_slave_is_up(slave
))
2591 bond_arp_send_all(bond
, slave
);
2596 if (do_failover
|| slave_state_changed
) {
2597 if (!rtnl_trylock())
2600 if (slave_state_changed
) {
2601 bond_slave_state_change(bond
);
2602 if (BOND_MODE(bond
) == BOND_MODE_XOR
)
2603 bond_update_slave_arr(bond
, NULL
);
2607 bond_select_active_slave(bond
);
2608 unblock_netpoll_tx();
2614 if (bond
->params
.arp_interval
)
2615 queue_delayed_work(bond
->wq
, &bond
->arp_work
,
2616 msecs_to_jiffies(bond
->params
.arp_interval
));
2619 /* Called to inspect slaves for active-backup mode ARP monitor link state
2620 * changes. Sets new_link in slaves to specify what action should take
2621 * place for the slave. Returns 0 if no changes are found, >0 if changes
2622 * to link states must be committed.
2624 * Called with rcu_read_lock held.
2626 static int bond_ab_arp_inspect(struct bonding
*bond
)
2628 unsigned long trans_start
, last_rx
;
2629 struct list_head
*iter
;
2630 struct slave
*slave
;
2633 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2634 slave
->new_link
= BOND_LINK_NOCHANGE
;
2635 last_rx
= slave_last_rx(bond
, slave
);
2637 if (slave
->link
!= BOND_LINK_UP
) {
2638 if (bond_time_in_interval(bond
, last_rx
, 1)) {
2639 slave
->new_link
= BOND_LINK_UP
;
2645 /* Give slaves 2*delta after being enslaved or made
2646 * active. This avoids bouncing, as the last receive
2647 * times need a full ARP monitor cycle to be updated.
2649 if (bond_time_in_interval(bond
, slave
->last_link_up
, 2))
2652 /* Backup slave is down if:
2653 * - No current_arp_slave AND
2654 * - more than 3*delta since last receive AND
2655 * - the bond has an IP address
2657 * Note: a non-null current_arp_slave indicates
2658 * the curr_active_slave went down and we are
2659 * searching for a new one; under this condition
2660 * we only take the curr_active_slave down - this
2661 * gives each slave a chance to tx/rx traffic
2662 * before being taken out
2664 if (!bond_is_active_slave(slave
) &&
2665 !rcu_access_pointer(bond
->current_arp_slave
) &&
2666 !bond_time_in_interval(bond
, last_rx
, 3)) {
2667 slave
->new_link
= BOND_LINK_DOWN
;
2671 /* Active slave is down if:
2672 * - more than 2*delta since transmitting OR
2673 * - (more than 2*delta since receive AND
2674 * the bond has an IP address)
2676 trans_start
= dev_trans_start(slave
->dev
);
2677 if (bond_is_active_slave(slave
) &&
2678 (!bond_time_in_interval(bond
, trans_start
, 2) ||
2679 !bond_time_in_interval(bond
, last_rx
, 2))) {
2680 slave
->new_link
= BOND_LINK_DOWN
;
2688 /* Called to commit link state changes noted by inspection step of
2689 * active-backup mode ARP monitor.
2691 * Called with RTNL hold.
2693 static void bond_ab_arp_commit(struct bonding
*bond
)
2695 unsigned long trans_start
;
2696 struct list_head
*iter
;
2697 struct slave
*slave
;
2699 bond_for_each_slave(bond
, slave
, iter
) {
2700 switch (slave
->new_link
) {
2701 case BOND_LINK_NOCHANGE
:
2705 trans_start
= dev_trans_start(slave
->dev
);
2706 if (rtnl_dereference(bond
->curr_active_slave
) != slave
||
2707 (!rtnl_dereference(bond
->curr_active_slave
) &&
2708 bond_time_in_interval(bond
, trans_start
, 1))) {
2709 struct slave
*current_arp_slave
;
2711 current_arp_slave
= rtnl_dereference(bond
->current_arp_slave
);
2712 bond_set_slave_link_state(slave
, BOND_LINK_UP
);
2713 if (current_arp_slave
) {
2714 bond_set_slave_inactive_flags(
2716 BOND_SLAVE_NOTIFY_NOW
);
2717 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
2720 netdev_info(bond
->dev
, "link status definitely up for interface %s\n",
2723 if (!rtnl_dereference(bond
->curr_active_slave
) ||
2724 slave
== rtnl_dereference(bond
->primary_slave
))
2731 case BOND_LINK_DOWN
:
2732 if (slave
->link_failure_count
< UINT_MAX
)
2733 slave
->link_failure_count
++;
2735 bond_set_slave_link_state(slave
, BOND_LINK_DOWN
);
2736 bond_set_slave_inactive_flags(slave
,
2737 BOND_SLAVE_NOTIFY_NOW
);
2739 netdev_info(bond
->dev
, "link status definitely down for interface %s, disabling it\n",
2742 if (slave
== rtnl_dereference(bond
->curr_active_slave
)) {
2743 RCU_INIT_POINTER(bond
->current_arp_slave
, NULL
);
2750 netdev_err(bond
->dev
, "impossible: new_link %d on slave %s\n",
2751 slave
->new_link
, slave
->dev
->name
);
2757 bond_select_active_slave(bond
);
2758 unblock_netpoll_tx();
2761 bond_set_carrier(bond
);
2764 /* Send ARP probes for active-backup mode ARP monitor.
2766 * Called with rcu_read_lock held.
2768 static bool bond_ab_arp_probe(struct bonding
*bond
)
2770 struct slave
*slave
, *before
= NULL
, *new_slave
= NULL
,
2771 *curr_arp_slave
= rcu_dereference(bond
->current_arp_slave
),
2772 *curr_active_slave
= rcu_dereference(bond
->curr_active_slave
);
2773 struct list_head
*iter
;
2775 bool should_notify_rtnl
= BOND_SLAVE_NOTIFY_LATER
;
2777 if (curr_arp_slave
&& curr_active_slave
)
2778 netdev_info(bond
->dev
, "PROBE: c_arp %s && cas %s BAD\n",
2779 curr_arp_slave
->dev
->name
,
2780 curr_active_slave
->dev
->name
);
2782 if (curr_active_slave
) {
2783 bond_arp_send_all(bond
, curr_active_slave
);
2784 return should_notify_rtnl
;
2787 /* if we don't have a curr_active_slave, search for the next available
2788 * backup slave from the current_arp_slave and make it the candidate
2789 * for becoming the curr_active_slave
2792 if (!curr_arp_slave
) {
2793 curr_arp_slave
= bond_first_slave_rcu(bond
);
2794 if (!curr_arp_slave
)
2795 return should_notify_rtnl
;
2798 bond_set_slave_inactive_flags(curr_arp_slave
, BOND_SLAVE_NOTIFY_LATER
);
2800 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2801 if (!found
&& !before
&& bond_slave_is_up(slave
))
2804 if (found
&& !new_slave
&& bond_slave_is_up(slave
))
2806 /* if the link state is up at this point, we
2807 * mark it down - this can happen if we have
2808 * simultaneous link failures and
2809 * reselect_active_interface doesn't make this
2810 * one the current slave so it is still marked
2811 * up when it is actually down
2813 if (!bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
2814 bond_set_slave_link_state(slave
, BOND_LINK_DOWN
);
2815 if (slave
->link_failure_count
< UINT_MAX
)
2816 slave
->link_failure_count
++;
2818 bond_set_slave_inactive_flags(slave
,
2819 BOND_SLAVE_NOTIFY_LATER
);
2821 netdev_info(bond
->dev
, "backup interface %s is now down\n",
2824 if (slave
== curr_arp_slave
)
2828 if (!new_slave
&& before
)
2834 bond_set_slave_link_state(new_slave
, BOND_LINK_BACK
);
2835 bond_set_slave_active_flags(new_slave
, BOND_SLAVE_NOTIFY_LATER
);
2836 bond_arp_send_all(bond
, new_slave
);
2837 new_slave
->last_link_up
= jiffies
;
2838 rcu_assign_pointer(bond
->current_arp_slave
, new_slave
);
2841 bond_for_each_slave_rcu(bond
, slave
, iter
) {
2842 if (slave
->should_notify
) {
2843 should_notify_rtnl
= BOND_SLAVE_NOTIFY_NOW
;
2847 return should_notify_rtnl
;
2850 static void bond_activebackup_arp_mon(struct work_struct
*work
)
2852 struct bonding
*bond
= container_of(work
, struct bonding
,
2854 bool should_notify_peers
= false;
2855 bool should_notify_rtnl
= false;
2858 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2860 if (!bond_has_slaves(bond
))
2865 should_notify_peers
= bond_should_notify_peers(bond
);
2867 if (bond_ab_arp_inspect(bond
)) {
2870 /* Race avoidance with bond_close flush of workqueue */
2871 if (!rtnl_trylock()) {
2873 should_notify_peers
= false;
2877 bond_ab_arp_commit(bond
);
2883 should_notify_rtnl
= bond_ab_arp_probe(bond
);
2887 if (bond
->params
.arp_interval
)
2888 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
2890 if (should_notify_peers
|| should_notify_rtnl
) {
2891 if (!rtnl_trylock())
2894 if (should_notify_peers
)
2895 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS
,
2897 if (should_notify_rtnl
)
2898 bond_slave_state_notify(bond
);
2904 /*-------------------------- netdev event handling --------------------------*/
2906 /* Change device name */
2907 static int bond_event_changename(struct bonding
*bond
)
2909 bond_remove_proc_entry(bond
);
2910 bond_create_proc_entry(bond
);
2912 bond_debug_reregister(bond
);
2917 static int bond_master_netdev_event(unsigned long event
,
2918 struct net_device
*bond_dev
)
2920 struct bonding
*event_bond
= netdev_priv(bond_dev
);
2923 case NETDEV_CHANGENAME
:
2924 return bond_event_changename(event_bond
);
2925 case NETDEV_UNREGISTER
:
2926 bond_remove_proc_entry(event_bond
);
2928 case NETDEV_REGISTER
:
2929 bond_create_proc_entry(event_bond
);
2931 case NETDEV_NOTIFY_PEERS
:
2932 if (event_bond
->send_peer_notif
)
2933 event_bond
->send_peer_notif
--;
2942 static int bond_slave_netdev_event(unsigned long event
,
2943 struct net_device
*slave_dev
)
2945 struct slave
*slave
= bond_slave_get_rtnl(slave_dev
), *primary
;
2946 struct bonding
*bond
;
2947 struct net_device
*bond_dev
;
2951 /* A netdev event can be generated while enslaving a device
2952 * before netdev_rx_handler_register is called in which case
2953 * slave will be NULL
2957 bond_dev
= slave
->bond
->dev
;
2959 primary
= rtnl_dereference(bond
->primary_slave
);
2962 case NETDEV_UNREGISTER
:
2963 if (bond_dev
->type
!= ARPHRD_ETHER
)
2964 bond_release_and_destroy(bond_dev
, slave_dev
);
2966 bond_release(bond_dev
, slave_dev
);
2970 old_speed
= slave
->speed
;
2971 old_duplex
= slave
->duplex
;
2973 bond_update_speed_duplex(slave
);
2975 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
2976 if (old_speed
!= slave
->speed
)
2977 bond_3ad_adapter_speed_changed(slave
);
2978 if (old_duplex
!= slave
->duplex
)
2979 bond_3ad_adapter_duplex_changed(slave
);
2983 /* Refresh slave-array if applicable!
2984 * If the setup does not use miimon or arpmon (mode-specific!),
2985 * then these events will not cause the slave-array to be
2986 * refreshed. This will cause xmit to use a slave that is not
2987 * usable. Avoid such situation by refeshing the array at these
2988 * events. If these (miimon/arpmon) parameters are configured
2989 * then array gets refreshed twice and that should be fine!
2991 if (bond_mode_uses_xmit_hash(bond
))
2992 bond_update_slave_arr(bond
, NULL
);
2994 case NETDEV_CHANGEMTU
:
2995 /* TODO: Should slaves be allowed to
2996 * independently alter their MTU? For
2997 * an active-backup bond, slaves need
2998 * not be the same type of device, so
2999 * MTUs may vary. For other modes,
3000 * slaves arguably should have the
3001 * same MTUs. To do this, we'd need to
3002 * take over the slave's change_mtu
3003 * function for the duration of their
3007 case NETDEV_CHANGENAME
:
3008 /* we don't care if we don't have primary set */
3009 if (!bond_uses_primary(bond
) ||
3010 !bond
->params
.primary
[0])
3013 if (slave
== primary
) {
3014 /* slave's name changed - he's no longer primary */
3015 RCU_INIT_POINTER(bond
->primary_slave
, NULL
);
3016 } else if (!strcmp(slave_dev
->name
, bond
->params
.primary
)) {
3017 /* we have a new primary slave */
3018 rcu_assign_pointer(bond
->primary_slave
, slave
);
3019 } else { /* we didn't change primary - exit */
3023 netdev_info(bond
->dev
, "Primary slave changed to %s, reselecting active slave\n",
3024 primary
? slave_dev
->name
: "none");
3027 bond_select_active_slave(bond
);
3028 unblock_netpoll_tx();
3030 case NETDEV_FEAT_CHANGE
:
3031 bond_compute_features(bond
);
3033 case NETDEV_RESEND_IGMP
:
3034 /* Propagate to master device */
3035 call_netdevice_notifiers(event
, slave
->bond
->dev
);
3044 /* bond_netdev_event: handle netdev notifier chain events.
3046 * This function receives events for the netdev chain. The caller (an
3047 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3048 * locks for us to safely manipulate the slave devices (RTNL lock,
3051 static int bond_netdev_event(struct notifier_block
*this,
3052 unsigned long event
, void *ptr
)
3054 struct net_device
*event_dev
= netdev_notifier_info_to_dev(ptr
);
3056 netdev_dbg(event_dev
, "event: %lx\n", event
);
3058 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3061 if (event_dev
->flags
& IFF_MASTER
) {
3062 netdev_dbg(event_dev
, "IFF_MASTER\n");
3063 return bond_master_netdev_event(event
, event_dev
);
3066 if (event_dev
->flags
& IFF_SLAVE
) {
3067 netdev_dbg(event_dev
, "IFF_SLAVE\n");
3068 return bond_slave_netdev_event(event
, event_dev
);
3074 static struct notifier_block bond_netdev_notifier
= {
3075 .notifier_call
= bond_netdev_event
,
3078 /*---------------------------- Hashing Policies -----------------------------*/
3080 /* L2 hash helper */
3081 static inline u32
bond_eth_hash(struct sk_buff
*skb
)
3083 struct ethhdr
*ep
, hdr_tmp
;
3085 ep
= skb_header_pointer(skb
, 0, sizeof(hdr_tmp
), &hdr_tmp
);
3087 return ep
->h_dest
[5] ^ ep
->h_source
[5] ^ ep
->h_proto
;
3091 /* Extract the appropriate headers based on bond's xmit policy */
3092 static bool bond_flow_dissect(struct bonding
*bond
, struct sk_buff
*skb
,
3093 struct flow_keys
*fk
)
3095 const struct ipv6hdr
*iph6
;
3096 const struct iphdr
*iph
;
3097 int noff
, proto
= -1;
3099 if (bond
->params
.xmit_policy
> BOND_XMIT_POLICY_LAYER23
)
3100 return skb_flow_dissect_flow_keys(skb
, fk
);
3102 fk
->ports
.ports
= 0;
3103 noff
= skb_network_offset(skb
);
3104 if (skb
->protocol
== htons(ETH_P_IP
)) {
3105 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph
))))
3108 iph_to_flow_copy_v4addrs(fk
, iph
);
3109 noff
+= iph
->ihl
<< 2;
3110 if (!ip_is_fragment(iph
))
3111 proto
= iph
->protocol
;
3112 } else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
3113 if (unlikely(!pskb_may_pull(skb
, noff
+ sizeof(*iph6
))))
3115 iph6
= ipv6_hdr(skb
);
3116 iph_to_flow_copy_v6addrs(fk
, iph6
);
3117 noff
+= sizeof(*iph6
);
3118 proto
= iph6
->nexthdr
;
3122 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER34
&& proto
>= 0)
3123 fk
->ports
.ports
= skb_flow_get_ports(skb
, noff
, proto
);
3129 * bond_xmit_hash - generate a hash value based on the xmit policy
3130 * @bond: bonding device
3131 * @skb: buffer to use for headers
3133 * This function will extract the necessary headers from the skb buffer and use
3134 * them to generate a hash based on the xmit_policy set in the bonding device
3136 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
)
3138 struct flow_keys flow
;
3141 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER2
||
3142 !bond_flow_dissect(bond
, skb
, &flow
))
3143 return bond_eth_hash(skb
);
3145 if (bond
->params
.xmit_policy
== BOND_XMIT_POLICY_LAYER23
||
3146 bond
->params
.xmit_policy
== BOND_XMIT_POLICY_ENCAP23
)
3147 hash
= bond_eth_hash(skb
);
3149 hash
= (__force u32
)flow
.ports
.ports
;
3150 hash
^= (__force u32
)flow_get_u32_dst(&flow
) ^
3151 (__force u32
)flow_get_u32_src(&flow
);
3152 hash
^= (hash
>> 16);
3153 hash
^= (hash
>> 8);
3158 /*-------------------------- Device entry points ----------------------------*/
3160 static void bond_work_init_all(struct bonding
*bond
)
3162 INIT_DELAYED_WORK(&bond
->mcast_work
,
3163 bond_resend_igmp_join_requests_delayed
);
3164 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3165 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3166 if (BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3167 INIT_DELAYED_WORK(&bond
->arp_work
, bond_activebackup_arp_mon
);
3169 INIT_DELAYED_WORK(&bond
->arp_work
, bond_loadbalance_arp_mon
);
3170 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3171 INIT_DELAYED_WORK(&bond
->slave_arr_work
, bond_slave_arr_handler
);
3174 static void bond_work_cancel_all(struct bonding
*bond
)
3176 cancel_delayed_work_sync(&bond
->mii_work
);
3177 cancel_delayed_work_sync(&bond
->arp_work
);
3178 cancel_delayed_work_sync(&bond
->alb_work
);
3179 cancel_delayed_work_sync(&bond
->ad_work
);
3180 cancel_delayed_work_sync(&bond
->mcast_work
);
3181 cancel_delayed_work_sync(&bond
->slave_arr_work
);
3184 static int bond_open(struct net_device
*bond_dev
)
3186 struct bonding
*bond
= netdev_priv(bond_dev
);
3187 struct list_head
*iter
;
3188 struct slave
*slave
;
3190 /* reset slave->backup and slave->inactive */
3191 if (bond_has_slaves(bond
)) {
3192 bond_for_each_slave(bond
, slave
, iter
) {
3193 if (bond_uses_primary(bond
) &&
3194 slave
!= rcu_access_pointer(bond
->curr_active_slave
)) {
3195 bond_set_slave_inactive_flags(slave
,
3196 BOND_SLAVE_NOTIFY_NOW
);
3197 } else if (BOND_MODE(bond
) != BOND_MODE_8023AD
) {
3198 bond_set_slave_active_flags(slave
,
3199 BOND_SLAVE_NOTIFY_NOW
);
3204 bond_work_init_all(bond
);
3206 if (bond_is_lb(bond
)) {
3207 /* bond_alb_initialize must be called before the timer
3210 if (bond_alb_initialize(bond
, (BOND_MODE(bond
) == BOND_MODE_ALB
)))
3212 if (bond
->params
.tlb_dynamic_lb
)
3213 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3216 if (bond
->params
.miimon
) /* link check interval, in milliseconds. */
3217 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3219 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3220 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3221 bond
->recv_probe
= bond_arp_rcv
;
3224 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
3225 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3226 /* register to receive LACPDUs */
3227 bond
->recv_probe
= bond_3ad_lacpdu_recv
;
3228 bond_3ad_initiate_agg_selection(bond
, 1);
3231 if (bond_mode_uses_xmit_hash(bond
))
3232 bond_update_slave_arr(bond
, NULL
);
3237 static int bond_close(struct net_device
*bond_dev
)
3239 struct bonding
*bond
= netdev_priv(bond_dev
);
3241 bond_work_cancel_all(bond
);
3242 bond
->send_peer_notif
= 0;
3243 if (bond_is_lb(bond
))
3244 bond_alb_deinitialize(bond
);
3245 bond
->recv_probe
= NULL
;
3250 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3251 struct rtnl_link_stats64
*stats
)
3253 struct bonding
*bond
= netdev_priv(bond_dev
);
3254 struct rtnl_link_stats64 temp
;
3255 struct list_head
*iter
;
3256 struct slave
*slave
;
3258 memcpy(stats
, &bond
->bond_stats
, sizeof(*stats
));
3260 bond_for_each_slave(bond
, slave
, iter
) {
3261 const struct rtnl_link_stats64
*sstats
=
3262 dev_get_stats(slave
->dev
, &temp
);
3263 struct rtnl_link_stats64
*pstats
= &slave
->slave_stats
;
3265 stats
->rx_packets
+= sstats
->rx_packets
- pstats
->rx_packets
;
3266 stats
->rx_bytes
+= sstats
->rx_bytes
- pstats
->rx_bytes
;
3267 stats
->rx_errors
+= sstats
->rx_errors
- pstats
->rx_errors
;
3268 stats
->rx_dropped
+= sstats
->rx_dropped
- pstats
->rx_dropped
;
3270 stats
->tx_packets
+= sstats
->tx_packets
- pstats
->tx_packets
;;
3271 stats
->tx_bytes
+= sstats
->tx_bytes
- pstats
->tx_bytes
;
3272 stats
->tx_errors
+= sstats
->tx_errors
- pstats
->tx_errors
;
3273 stats
->tx_dropped
+= sstats
->tx_dropped
- pstats
->tx_dropped
;
3275 stats
->multicast
+= sstats
->multicast
- pstats
->multicast
;
3276 stats
->collisions
+= sstats
->collisions
- pstats
->collisions
;
3278 stats
->rx_length_errors
+= sstats
->rx_length_errors
- pstats
->rx_length_errors
;
3279 stats
->rx_over_errors
+= sstats
->rx_over_errors
- pstats
->rx_over_errors
;
3280 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
- pstats
->rx_crc_errors
;
3281 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
- pstats
->rx_frame_errors
;
3282 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
- pstats
->rx_fifo_errors
;
3283 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
- pstats
->rx_missed_errors
;
3285 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
- pstats
->tx_aborted_errors
;
3286 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
- pstats
->tx_carrier_errors
;
3287 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
- pstats
->tx_fifo_errors
;
3288 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
- pstats
->tx_heartbeat_errors
;
3289 stats
->tx_window_errors
+= sstats
->tx_window_errors
- pstats
->tx_window_errors
;
3291 /* save off the slave stats for the next run */
3292 memcpy(pstats
, sstats
, sizeof(*sstats
));
3294 memcpy(&bond
->bond_stats
, stats
, sizeof(*stats
));
3299 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3301 struct bonding
*bond
= netdev_priv(bond_dev
);
3302 struct net_device
*slave_dev
= NULL
;
3303 struct ifbond k_binfo
;
3304 struct ifbond __user
*u_binfo
= NULL
;
3305 struct ifslave k_sinfo
;
3306 struct ifslave __user
*u_sinfo
= NULL
;
3307 struct mii_ioctl_data
*mii
= NULL
;
3308 struct bond_opt_value newval
;
3312 netdev_dbg(bond_dev
, "bond_ioctl: cmd=%d\n", cmd
);
3323 /* We do this again just in case we were called by SIOCGMIIREG
3324 * instead of SIOCGMIIPHY.
3330 if (mii
->reg_num
== 1) {
3332 if (netif_carrier_ok(bond
->dev
))
3333 mii
->val_out
= BMSR_LSTATUS
;
3337 case BOND_INFO_QUERY_OLD
:
3338 case SIOCBONDINFOQUERY
:
3339 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3341 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3344 res
= bond_info_query(bond_dev
, &k_binfo
);
3346 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3350 case BOND_SLAVE_INFO_QUERY_OLD
:
3351 case SIOCBONDSLAVEINFOQUERY
:
3352 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3354 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3357 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3359 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3367 net
= dev_net(bond_dev
);
3369 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
3372 slave_dev
= __dev_get_by_name(net
, ifr
->ifr_slave
);
3374 netdev_dbg(bond_dev
, "slave_dev=%p:\n", slave_dev
);
3379 netdev_dbg(bond_dev
, "slave_dev->name=%s:\n", slave_dev
->name
);
3381 case BOND_ENSLAVE_OLD
:
3382 case SIOCBONDENSLAVE
:
3383 res
= bond_enslave(bond_dev
, slave_dev
);
3385 case BOND_RELEASE_OLD
:
3386 case SIOCBONDRELEASE
:
3387 res
= bond_release(bond_dev
, slave_dev
);
3389 case BOND_SETHWADDR_OLD
:
3390 case SIOCBONDSETHWADDR
:
3391 bond_set_dev_addr(bond_dev
, slave_dev
);
3394 case BOND_CHANGE_ACTIVE_OLD
:
3395 case SIOCBONDCHANGEACTIVE
:
3396 bond_opt_initstr(&newval
, slave_dev
->name
);
3397 res
= __bond_opt_set(bond
, BOND_OPT_ACTIVE_SLAVE
, &newval
);
3406 static void bond_change_rx_flags(struct net_device
*bond_dev
, int change
)
3408 struct bonding
*bond
= netdev_priv(bond_dev
);
3410 if (change
& IFF_PROMISC
)
3411 bond_set_promiscuity(bond
,
3412 bond_dev
->flags
& IFF_PROMISC
? 1 : -1);
3414 if (change
& IFF_ALLMULTI
)
3415 bond_set_allmulti(bond
,
3416 bond_dev
->flags
& IFF_ALLMULTI
? 1 : -1);
3419 static void bond_set_rx_mode(struct net_device
*bond_dev
)
3421 struct bonding
*bond
= netdev_priv(bond_dev
);
3422 struct list_head
*iter
;
3423 struct slave
*slave
;
3426 if (bond_uses_primary(bond
)) {
3427 slave
= rcu_dereference(bond
->curr_active_slave
);
3429 dev_uc_sync(slave
->dev
, bond_dev
);
3430 dev_mc_sync(slave
->dev
, bond_dev
);
3433 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3434 dev_uc_sync_multiple(slave
->dev
, bond_dev
);
3435 dev_mc_sync_multiple(slave
->dev
, bond_dev
);
3441 static int bond_neigh_init(struct neighbour
*n
)
3443 struct bonding
*bond
= netdev_priv(n
->dev
);
3444 const struct net_device_ops
*slave_ops
;
3445 struct neigh_parms parms
;
3446 struct slave
*slave
;
3449 slave
= bond_first_slave(bond
);
3452 slave_ops
= slave
->dev
->netdev_ops
;
3453 if (!slave_ops
->ndo_neigh_setup
)
3456 parms
.neigh_setup
= NULL
;
3457 parms
.neigh_cleanup
= NULL
;
3458 ret
= slave_ops
->ndo_neigh_setup(slave
->dev
, &parms
);
3462 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3463 * after the last slave has been detached. Assumes that all slaves
3464 * utilize the same neigh_cleanup (true at this writing as only user
3467 n
->parms
->neigh_cleanup
= parms
.neigh_cleanup
;
3469 if (!parms
.neigh_setup
)
3472 return parms
.neigh_setup(n
);
3475 /* The bonding ndo_neigh_setup is called at init time beofre any
3476 * slave exists. So we must declare proxy setup function which will
3477 * be used at run time to resolve the actual slave neigh param setup.
3479 * It's also called by master devices (such as vlans) to setup their
3480 * underlying devices. In that case - do nothing, we're already set up from
3483 static int bond_neigh_setup(struct net_device
*dev
,
3484 struct neigh_parms
*parms
)
3486 /* modify only our neigh_parms */
3487 if (parms
->dev
== dev
)
3488 parms
->neigh_setup
= bond_neigh_init
;
3493 /* Change the MTU of all of a master's slaves to match the master */
3494 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3496 struct bonding
*bond
= netdev_priv(bond_dev
);
3497 struct slave
*slave
, *rollback_slave
;
3498 struct list_head
*iter
;
3501 netdev_dbg(bond_dev
, "bond=%p, new_mtu=%d\n", bond
, new_mtu
);
3503 bond_for_each_slave(bond
, slave
, iter
) {
3504 netdev_dbg(bond_dev
, "s %p c_m %p\n",
3505 slave
, slave
->dev
->netdev_ops
->ndo_change_mtu
);
3507 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3510 /* If we failed to set the slave's mtu to the new value
3511 * we must abort the operation even in ACTIVE_BACKUP
3512 * mode, because if we allow the backup slaves to have
3513 * different mtu values than the active slave we'll
3514 * need to change their mtu when doing a failover. That
3515 * means changing their mtu from timer context, which
3516 * is probably not a good idea.
3518 netdev_dbg(bond_dev
, "err %d %s\n", res
,
3524 bond_dev
->mtu
= new_mtu
;
3529 /* unwind from head to the slave that failed */
3530 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3533 if (rollback_slave
== slave
)
3536 tmp_res
= dev_set_mtu(rollback_slave
->dev
, bond_dev
->mtu
);
3538 netdev_dbg(bond_dev
, "unwind err %d dev %s\n",
3539 tmp_res
, rollback_slave
->dev
->name
);
3546 /* Change HW address
3548 * Note that many devices must be down to change the HW address, and
3549 * downing the master releases all slaves. We can make bonds full of
3550 * bonding devices to test this, however.
3552 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
3554 struct bonding
*bond
= netdev_priv(bond_dev
);
3555 struct slave
*slave
, *rollback_slave
;
3556 struct sockaddr
*sa
= addr
, tmp_sa
;
3557 struct list_head
*iter
;
3560 if (BOND_MODE(bond
) == BOND_MODE_ALB
)
3561 return bond_alb_set_mac_address(bond_dev
, addr
);
3564 netdev_dbg(bond_dev
, "bond=%p\n", bond
);
3566 /* If fail_over_mac is enabled, do nothing and return success.
3567 * Returning an error causes ifenslave to fail.
3569 if (bond
->params
.fail_over_mac
&&
3570 BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
)
3573 if (!is_valid_ether_addr(sa
->sa_data
))
3574 return -EADDRNOTAVAIL
;
3576 bond_for_each_slave(bond
, slave
, iter
) {
3577 netdev_dbg(bond_dev
, "slave %p %s\n", slave
, slave
->dev
->name
);
3578 res
= dev_set_mac_address(slave
->dev
, addr
);
3580 /* TODO: consider downing the slave
3582 * User should expect communications
3583 * breakage anyway until ARP finish
3586 netdev_dbg(bond_dev
, "err %d %s\n", res
, slave
->dev
->name
);
3592 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
3596 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
3597 tmp_sa
.sa_family
= bond_dev
->type
;
3599 /* unwind from head to the slave that failed */
3600 bond_for_each_slave(bond
, rollback_slave
, iter
) {
3603 if (rollback_slave
== slave
)
3606 tmp_res
= dev_set_mac_address(rollback_slave
->dev
, &tmp_sa
);
3608 netdev_dbg(bond_dev
, "unwind err %d dev %s\n",
3609 tmp_res
, rollback_slave
->dev
->name
);
3617 * bond_xmit_slave_id - transmit skb through slave with slave_id
3618 * @bond: bonding device that is transmitting
3619 * @skb: buffer to transmit
3620 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3622 * This function tries to transmit through slave with slave_id but in case
3623 * it fails, it tries to find the first available slave for transmission.
3624 * The skb is consumed in all cases, thus the function is void.
3626 static void bond_xmit_slave_id(struct bonding
*bond
, struct sk_buff
*skb
, int slave_id
)
3628 struct list_head
*iter
;
3629 struct slave
*slave
;
3632 /* Here we start from the slave with slave_id */
3633 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3635 if (bond_slave_can_tx(slave
)) {
3636 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3642 /* Here we start from the first slave up to slave_id */
3644 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3647 if (bond_slave_can_tx(slave
)) {
3648 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3652 /* no slave that can tx has been found */
3653 bond_tx_drop(bond
->dev
, skb
);
3657 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3658 * @bond: bonding device to use
3660 * Based on the value of the bonding device's packets_per_slave parameter
3661 * this function generates a slave id, which is usually used as the next
3662 * slave to transmit through.
3664 static u32
bond_rr_gen_slave_id(struct bonding
*bond
)
3667 struct reciprocal_value reciprocal_packets_per_slave
;
3668 int packets_per_slave
= bond
->params
.packets_per_slave
;
3670 switch (packets_per_slave
) {
3672 slave_id
= prandom_u32();
3675 slave_id
= bond
->rr_tx_counter
;
3678 reciprocal_packets_per_slave
=
3679 bond
->params
.reciprocal_packets_per_slave
;
3680 slave_id
= reciprocal_divide(bond
->rr_tx_counter
,
3681 reciprocal_packets_per_slave
);
3684 bond
->rr_tx_counter
++;
3689 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3691 struct bonding
*bond
= netdev_priv(bond_dev
);
3692 struct iphdr
*iph
= ip_hdr(skb
);
3693 struct slave
*slave
;
3696 /* Start with the curr_active_slave that joined the bond as the
3697 * default for sending IGMP traffic. For failover purposes one
3698 * needs to maintain some consistency for the interface that will
3699 * send the join/membership reports. The curr_active_slave found
3700 * will send all of this type of traffic.
3702 if (iph
->protocol
== IPPROTO_IGMP
&& skb
->protocol
== htons(ETH_P_IP
)) {
3703 slave
= rcu_dereference(bond
->curr_active_slave
);
3705 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3707 bond_xmit_slave_id(bond
, skb
, 0);
3709 int slave_cnt
= ACCESS_ONCE(bond
->slave_cnt
);
3711 if (likely(slave_cnt
)) {
3712 slave_id
= bond_rr_gen_slave_id(bond
);
3713 bond_xmit_slave_id(bond
, skb
, slave_id
% slave_cnt
);
3715 bond_tx_drop(bond_dev
, skb
);
3719 return NETDEV_TX_OK
;
3722 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3723 * the bond has a usable interface.
3725 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3727 struct bonding
*bond
= netdev_priv(bond_dev
);
3728 struct slave
*slave
;
3730 slave
= rcu_dereference(bond
->curr_active_slave
);
3732 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3734 bond_tx_drop(bond_dev
, skb
);
3736 return NETDEV_TX_OK
;
3739 /* Use this to update slave_array when (a) it's not appropriate to update
3740 * slave_array right away (note that update_slave_array() may sleep)
3741 * and / or (b) RTNL is not held.
3743 void bond_slave_arr_work_rearm(struct bonding
*bond
, unsigned long delay
)
3745 queue_delayed_work(bond
->wq
, &bond
->slave_arr_work
, delay
);
3748 /* Slave array work handler. Holds only RTNL */
3749 static void bond_slave_arr_handler(struct work_struct
*work
)
3751 struct bonding
*bond
= container_of(work
, struct bonding
,
3752 slave_arr_work
.work
);
3755 if (!rtnl_trylock())
3758 ret
= bond_update_slave_arr(bond
, NULL
);
3761 pr_warn_ratelimited("Failed to update slave array from WT\n");
3767 bond_slave_arr_work_rearm(bond
, 1);
3770 /* Build the usable slaves array in control path for modes that use xmit-hash
3771 * to determine the slave interface -
3772 * (a) BOND_MODE_8023AD
3774 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3776 * The caller is expected to hold RTNL only and NO other lock!
3778 int bond_update_slave_arr(struct bonding
*bond
, struct slave
*skipslave
)
3780 struct slave
*slave
;
3781 struct list_head
*iter
;
3782 struct bond_up_slave
*new_arr
, *old_arr
;
3787 #ifdef CONFIG_LOCKDEP
3788 WARN_ON(lockdep_is_held(&bond
->mode_lock
));
3791 new_arr
= kzalloc(offsetof(struct bond_up_slave
, arr
[bond
->slave_cnt
]),
3795 pr_err("Failed to build slave-array.\n");
3798 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
3799 struct ad_info ad_info
;
3801 if (bond_3ad_get_active_agg_info(bond
, &ad_info
)) {
3802 pr_debug("bond_3ad_get_active_agg_info failed\n");
3803 kfree_rcu(new_arr
, rcu
);
3804 /* No active aggragator means it's not safe to use
3805 * the previous array.
3807 old_arr
= rtnl_dereference(bond
->slave_arr
);
3809 RCU_INIT_POINTER(bond
->slave_arr
, NULL
);
3810 kfree_rcu(old_arr
, rcu
);
3814 slaves_in_agg
= ad_info
.ports
;
3815 agg_id
= ad_info
.aggregator_id
;
3817 bond_for_each_slave(bond
, slave
, iter
) {
3818 if (BOND_MODE(bond
) == BOND_MODE_8023AD
) {
3819 struct aggregator
*agg
;
3821 agg
= SLAVE_AD_INFO(slave
)->port
.aggregator
;
3822 if (!agg
|| agg
->aggregator_identifier
!= agg_id
)
3825 if (!bond_slave_can_tx(slave
))
3827 if (skipslave
== slave
)
3829 new_arr
->arr
[new_arr
->count
++] = slave
;
3832 old_arr
= rtnl_dereference(bond
->slave_arr
);
3833 rcu_assign_pointer(bond
->slave_arr
, new_arr
);
3835 kfree_rcu(old_arr
, rcu
);
3837 if (ret
!= 0 && skipslave
) {
3840 /* Rare situation where caller has asked to skip a specific
3841 * slave but allocation failed (most likely!). BTW this is
3842 * only possible when the call is initiated from
3843 * __bond_release_one(). In this situation; overwrite the
3844 * skipslave entry in the array with the last entry from the
3845 * array to avoid a situation where the xmit path may choose
3846 * this to-be-skipped slave to send a packet out.
3848 old_arr
= rtnl_dereference(bond
->slave_arr
);
3849 for (idx
= 0; idx
< old_arr
->count
; idx
++) {
3850 if (skipslave
== old_arr
->arr
[idx
]) {
3852 old_arr
->arr
[old_arr
->count
-1];
3861 /* Use this Xmit function for 3AD as well as XOR modes. The current
3862 * usable slave array is formed in the control path. The xmit function
3863 * just calculates hash and sends the packet out.
3865 static int bond_3ad_xor_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3867 struct bonding
*bond
= netdev_priv(dev
);
3868 struct slave
*slave
;
3869 struct bond_up_slave
*slaves
;
3872 slaves
= rcu_dereference(bond
->slave_arr
);
3873 count
= slaves
? ACCESS_ONCE(slaves
->count
) : 0;
3874 if (likely(count
)) {
3875 slave
= slaves
->arr
[bond_xmit_hash(bond
, skb
) % count
];
3876 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3878 bond_tx_drop(dev
, skb
);
3881 return NETDEV_TX_OK
;
3884 /* in broadcast mode, we send everything to all usable interfaces. */
3885 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
3887 struct bonding
*bond
= netdev_priv(bond_dev
);
3888 struct slave
*slave
= NULL
;
3889 struct list_head
*iter
;
3891 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3892 if (bond_is_last_slave(bond
, slave
))
3894 if (bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
) {
3895 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
3898 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3899 bond_dev
->name
, __func__
);
3902 bond_dev_queue_xmit(bond
, skb2
, slave
->dev
);
3905 if (slave
&& bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
)
3906 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3908 bond_tx_drop(bond_dev
, skb
);
3910 return NETDEV_TX_OK
;
3913 /*------------------------- Device initialization ---------------------------*/
3915 /* Lookup the slave that corresponds to a qid */
3916 static inline int bond_slave_override(struct bonding
*bond
,
3917 struct sk_buff
*skb
)
3919 struct slave
*slave
= NULL
;
3920 struct list_head
*iter
;
3922 if (!skb
->queue_mapping
)
3925 /* Find out if any slaves have the same mapping as this skb. */
3926 bond_for_each_slave_rcu(bond
, slave
, iter
) {
3927 if (slave
->queue_id
== skb
->queue_mapping
) {
3928 if (bond_slave_is_up(slave
) &&
3929 slave
->link
== BOND_LINK_UP
) {
3930 bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
3933 /* If the slave isn't UP, use default transmit policy. */
3942 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
3943 void *accel_priv
, select_queue_fallback_t fallback
)
3945 /* This helper function exists to help dev_pick_tx get the correct
3946 * destination queue. Using a helper function skips a call to
3947 * skb_tx_hash and will put the skbs in the queue we expect on their
3948 * way down to the bonding driver.
3950 u16 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) : 0;
3952 /* Save the original txq to restore before passing to the driver */
3953 qdisc_skb_cb(skb
)->slave_dev_queue_mapping
= skb
->queue_mapping
;
3955 if (unlikely(txq
>= dev
->real_num_tx_queues
)) {
3957 txq
-= dev
->real_num_tx_queues
;
3958 } while (txq
>= dev
->real_num_tx_queues
);
3963 static netdev_tx_t
__bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3965 struct bonding
*bond
= netdev_priv(dev
);
3967 if (bond_should_override_tx_queue(bond
) &&
3968 !bond_slave_override(bond
, skb
))
3969 return NETDEV_TX_OK
;
3971 switch (BOND_MODE(bond
)) {
3972 case BOND_MODE_ROUNDROBIN
:
3973 return bond_xmit_roundrobin(skb
, dev
);
3974 case BOND_MODE_ACTIVEBACKUP
:
3975 return bond_xmit_activebackup(skb
, dev
);
3976 case BOND_MODE_8023AD
:
3978 return bond_3ad_xor_xmit(skb
, dev
);
3979 case BOND_MODE_BROADCAST
:
3980 return bond_xmit_broadcast(skb
, dev
);
3982 return bond_alb_xmit(skb
, dev
);
3984 return bond_tlb_xmit(skb
, dev
);
3986 /* Should never happen, mode already checked */
3987 netdev_err(dev
, "Unknown bonding mode %d\n", BOND_MODE(bond
));
3989 bond_tx_drop(dev
, skb
);
3990 return NETDEV_TX_OK
;
3994 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
3996 struct bonding
*bond
= netdev_priv(dev
);
3997 netdev_tx_t ret
= NETDEV_TX_OK
;
3999 /* If we risk deadlock from transmitting this in the
4000 * netpoll path, tell netpoll to queue the frame for later tx
4002 if (unlikely(is_netpoll_tx_blocked(dev
)))
4003 return NETDEV_TX_BUSY
;
4006 if (bond_has_slaves(bond
))
4007 ret
= __bond_start_xmit(skb
, dev
);
4009 bond_tx_drop(dev
, skb
);
4015 static int bond_ethtool_get_settings(struct net_device
*bond_dev
,
4016 struct ethtool_cmd
*ecmd
)
4018 struct bonding
*bond
= netdev_priv(bond_dev
);
4019 unsigned long speed
= 0;
4020 struct list_head
*iter
;
4021 struct slave
*slave
;
4023 ecmd
->duplex
= DUPLEX_UNKNOWN
;
4024 ecmd
->port
= PORT_OTHER
;
4026 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4027 * do not need to check mode. Though link speed might not represent
4028 * the true receive or transmit bandwidth (not all modes are symmetric)
4029 * this is an accurate maximum.
4031 bond_for_each_slave(bond
, slave
, iter
) {
4032 if (bond_slave_can_tx(slave
)) {
4033 if (slave
->speed
!= SPEED_UNKNOWN
)
4034 speed
+= slave
->speed
;
4035 if (ecmd
->duplex
== DUPLEX_UNKNOWN
&&
4036 slave
->duplex
!= DUPLEX_UNKNOWN
)
4037 ecmd
->duplex
= slave
->duplex
;
4040 ethtool_cmd_speed_set(ecmd
, speed
? : SPEED_UNKNOWN
);
4045 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4046 struct ethtool_drvinfo
*drvinfo
)
4048 strlcpy(drvinfo
->driver
, DRV_NAME
, sizeof(drvinfo
->driver
));
4049 strlcpy(drvinfo
->version
, DRV_VERSION
, sizeof(drvinfo
->version
));
4050 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
), "%d",
4054 static const struct ethtool_ops bond_ethtool_ops
= {
4055 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4056 .get_settings
= bond_ethtool_get_settings
,
4057 .get_link
= ethtool_op_get_link
,
4060 static const struct net_device_ops bond_netdev_ops
= {
4061 .ndo_init
= bond_init
,
4062 .ndo_uninit
= bond_uninit
,
4063 .ndo_open
= bond_open
,
4064 .ndo_stop
= bond_close
,
4065 .ndo_start_xmit
= bond_start_xmit
,
4066 .ndo_select_queue
= bond_select_queue
,
4067 .ndo_get_stats64
= bond_get_stats
,
4068 .ndo_do_ioctl
= bond_do_ioctl
,
4069 .ndo_change_rx_flags
= bond_change_rx_flags
,
4070 .ndo_set_rx_mode
= bond_set_rx_mode
,
4071 .ndo_change_mtu
= bond_change_mtu
,
4072 .ndo_set_mac_address
= bond_set_mac_address
,
4073 .ndo_neigh_setup
= bond_neigh_setup
,
4074 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4075 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4076 #ifdef CONFIG_NET_POLL_CONTROLLER
4077 .ndo_netpoll_setup
= bond_netpoll_setup
,
4078 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4079 .ndo_poll_controller
= bond_poll_controller
,
4081 .ndo_add_slave
= bond_enslave
,
4082 .ndo_del_slave
= bond_release
,
4083 .ndo_fix_features
= bond_fix_features
,
4084 .ndo_bridge_setlink
= switchdev_port_bridge_setlink
,
4085 .ndo_bridge_getlink
= switchdev_port_bridge_getlink
,
4086 .ndo_bridge_dellink
= switchdev_port_bridge_dellink
,
4087 .ndo_fdb_add
= switchdev_port_fdb_add
,
4088 .ndo_fdb_del
= switchdev_port_fdb_del
,
4089 .ndo_fdb_dump
= switchdev_port_fdb_dump
,
4090 .ndo_features_check
= passthru_features_check
,
4093 static const struct device_type bond_type
= {
4097 static void bond_destructor(struct net_device
*bond_dev
)
4099 struct bonding
*bond
= netdev_priv(bond_dev
);
4101 destroy_workqueue(bond
->wq
);
4102 free_netdev(bond_dev
);
4105 void bond_setup(struct net_device
*bond_dev
)
4107 struct bonding
*bond
= netdev_priv(bond_dev
);
4109 spin_lock_init(&bond
->mode_lock
);
4110 bond
->params
= bonding_defaults
;
4112 /* Initialize pointers */
4113 bond
->dev
= bond_dev
;
4115 /* Initialize the device entry points */
4116 ether_setup(bond_dev
);
4117 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4118 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4120 bond_dev
->destructor
= bond_destructor
;
4122 SET_NETDEV_DEVTYPE(bond_dev
, &bond_type
);
4124 /* Initialize the device options */
4125 bond_dev
->tx_queue_len
= 0;
4126 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4127 bond_dev
->priv_flags
|= IFF_BONDING
| IFF_UNICAST_FLT
;
4128 bond_dev
->priv_flags
&= ~(IFF_XMIT_DST_RELEASE
| IFF_TX_SKB_SHARING
);
4130 /* don't acquire bond device's netif_tx_lock when transmitting */
4131 bond_dev
->features
|= NETIF_F_LLTX
;
4133 /* By default, we declare the bond to be fully
4134 * VLAN hardware accelerated capable. Special
4135 * care is taken in the various xmit functions
4136 * when there are slaves that are not hw accel
4140 /* Don't allow bond devices to change network namespaces. */
4141 bond_dev
->features
|= NETIF_F_NETNS_LOCAL
;
4143 bond_dev
->hw_features
= BOND_VLAN_FEATURES
|
4144 NETIF_F_HW_VLAN_CTAG_TX
|
4145 NETIF_F_HW_VLAN_CTAG_RX
|
4146 NETIF_F_HW_VLAN_CTAG_FILTER
;
4148 bond_dev
->hw_features
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_HW_CSUM
);
4149 bond_dev
->hw_features
|= NETIF_F_GSO_ENCAP_ALL
;
4150 bond_dev
->features
|= bond_dev
->hw_features
;
4153 /* Destroy a bonding device.
4154 * Must be under rtnl_lock when this function is called.
4156 static void bond_uninit(struct net_device
*bond_dev
)
4158 struct bonding
*bond
= netdev_priv(bond_dev
);
4159 struct list_head
*iter
;
4160 struct slave
*slave
;
4161 struct bond_up_slave
*arr
;
4163 bond_netpoll_cleanup(bond_dev
);
4165 /* Release the bonded slaves */
4166 bond_for_each_slave(bond
, slave
, iter
)
4167 __bond_release_one(bond_dev
, slave
->dev
, true);
4168 netdev_info(bond_dev
, "Released all slaves\n");
4170 arr
= rtnl_dereference(bond
->slave_arr
);
4172 RCU_INIT_POINTER(bond
->slave_arr
, NULL
);
4173 kfree_rcu(arr
, rcu
);
4176 list_del(&bond
->bond_list
);
4178 bond_debug_unregister(bond
);
4181 /*------------------------- Module initialization ---------------------------*/
4183 static int bond_check_params(struct bond_params
*params
)
4185 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
, i
;
4186 struct bond_opt_value newval
;
4187 const struct bond_opt_value
*valptr
;
4188 int arp_all_targets_value
;
4189 u16 ad_actor_sys_prio
= 0;
4190 u16 ad_user_port_key
= 0;
4192 /* Convert string parameters. */
4194 bond_opt_initstr(&newval
, mode
);
4195 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_MODE
), &newval
);
4197 pr_err("Error: Invalid bonding mode \"%s\"\n", mode
);
4200 bond_mode
= valptr
->value
;
4203 if (xmit_hash_policy
) {
4204 if ((bond_mode
!= BOND_MODE_XOR
) &&
4205 (bond_mode
!= BOND_MODE_8023AD
) &&
4206 (bond_mode
!= BOND_MODE_TLB
)) {
4207 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4208 bond_mode_name(bond_mode
));
4210 bond_opt_initstr(&newval
, xmit_hash_policy
);
4211 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH
),
4214 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4218 xmit_hashtype
= valptr
->value
;
4223 if (bond_mode
!= BOND_MODE_8023AD
) {
4224 pr_info("lacp_rate param is irrelevant in mode %s\n",
4225 bond_mode_name(bond_mode
));
4227 bond_opt_initstr(&newval
, lacp_rate
);
4228 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE
),
4231 pr_err("Error: Invalid lacp rate \"%s\"\n",
4235 lacp_fast
= valptr
->value
;
4240 bond_opt_initstr(&newval
, ad_select
);
4241 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT
),
4244 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select
);
4247 params
->ad_select
= valptr
->value
;
4248 if (bond_mode
!= BOND_MODE_8023AD
)
4249 pr_warn("ad_select param only affects 802.3ad mode\n");
4251 params
->ad_select
= BOND_AD_STABLE
;
4254 if (max_bonds
< 0) {
4255 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4256 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4257 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4261 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4267 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4272 if (downdelay
< 0) {
4273 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4274 downdelay
, INT_MAX
);
4278 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4279 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4284 if (num_peer_notif
< 0 || num_peer_notif
> 255) {
4285 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4290 /* reset values for 802.3ad/TLB/ALB */
4291 if (!bond_mode_uses_arp(bond_mode
)) {
4293 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");
4294 pr_warn("Forcing miimon to 100msec\n");
4295 miimon
= BOND_DEFAULT_MIIMON
;
4299 if (tx_queues
< 1 || tx_queues
> 255) {
4300 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4301 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4302 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4305 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4306 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4308 all_slaves_active
= 0;
4311 if (resend_igmp
< 0 || resend_igmp
> 255) {
4312 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4313 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4314 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4317 bond_opt_initval(&newval
, packets_per_slave
);
4318 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE
), &newval
)) {
4319 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4320 packets_per_slave
, USHRT_MAX
);
4321 packets_per_slave
= 1;
4324 if (bond_mode
== BOND_MODE_ALB
) {
4325 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",
4330 if (updelay
|| downdelay
) {
4331 /* just warn the user the up/down delay will have
4332 * no effect since miimon is zero...
4334 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",
4335 updelay
, downdelay
);
4338 /* don't allow arp monitoring */
4340 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4341 miimon
, arp_interval
);
4345 if ((updelay
% miimon
) != 0) {
4346 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4347 updelay
, miimon
, (updelay
/ miimon
) * miimon
);
4352 if ((downdelay
% miimon
) != 0) {
4353 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4355 (downdelay
/ miimon
) * miimon
);
4358 downdelay
/= miimon
;
4361 if (arp_interval
< 0) {
4362 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4363 arp_interval
, INT_MAX
);
4367 for (arp_ip_count
= 0, i
= 0;
4368 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[i
]; i
++) {
4371 /* not a complete check, but good enough to catch mistakes */
4372 if (!in4_pton(arp_ip_target
[i
], -1, (u8
*)&ip
, -1, NULL
) ||
4373 !bond_is_ip_target_ok(ip
)) {
4374 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4378 if (bond_get_targets_ip(arp_target
, ip
) == -1)
4379 arp_target
[arp_ip_count
++] = ip
;
4381 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4386 if (arp_interval
&& !arp_ip_count
) {
4387 /* don't allow arping if no arp_ip_target given... */
4388 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4394 if (!arp_interval
) {
4395 pr_err("arp_validate requires arp_interval\n");
4399 bond_opt_initstr(&newval
, arp_validate
);
4400 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE
),
4403 pr_err("Error: invalid arp_validate \"%s\"\n",
4407 arp_validate_value
= valptr
->value
;
4409 arp_validate_value
= 0;
4412 arp_all_targets_value
= 0;
4413 if (arp_all_targets
) {
4414 bond_opt_initstr(&newval
, arp_all_targets
);
4415 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS
),
4418 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4420 arp_all_targets_value
= 0;
4422 arp_all_targets_value
= valptr
->value
;
4427 pr_info("MII link monitoring set to %d ms\n", miimon
);
4428 } else if (arp_interval
) {
4429 valptr
= bond_opt_get_val(BOND_OPT_ARP_VALIDATE
,
4430 arp_validate_value
);
4431 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4432 arp_interval
, valptr
->string
, arp_ip_count
);
4434 for (i
= 0; i
< arp_ip_count
; i
++)
4435 pr_cont(" %s", arp_ip_target
[i
]);
4439 } else if (max_bonds
) {
4440 /* miimon and arp_interval not set, we need one so things
4441 * work as expected, see bonding.txt for details
4443 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");
4446 if (primary
&& !bond_mode_uses_primary(bond_mode
)) {
4447 /* currently, using a primary only makes sense
4448 * in active backup, TLB or ALB modes
4450 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4451 primary
, bond_mode_name(bond_mode
));
4455 if (primary
&& primary_reselect
) {
4456 bond_opt_initstr(&newval
, primary_reselect
);
4457 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT
),
4460 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4464 primary_reselect_value
= valptr
->value
;
4466 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4469 if (fail_over_mac
) {
4470 bond_opt_initstr(&newval
, fail_over_mac
);
4471 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC
),
4474 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4478 fail_over_mac_value
= valptr
->value
;
4479 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4480 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4482 fail_over_mac_value
= BOND_FOM_NONE
;
4485 bond_opt_initstr(&newval
, "default");
4486 valptr
= bond_opt_parse(
4487 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO
),
4490 pr_err("Error: No ad_actor_sys_prio default value");
4493 ad_actor_sys_prio
= valptr
->value
;
4495 valptr
= bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY
),
4498 pr_err("Error: No ad_user_port_key default value");
4501 ad_user_port_key
= valptr
->value
;
4503 if (lp_interval
== 0) {
4504 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4505 INT_MAX
, BOND_ALB_DEFAULT_LP_INTERVAL
);
4506 lp_interval
= BOND_ALB_DEFAULT_LP_INTERVAL
;
4509 /* fill params struct with the proper values */
4510 params
->mode
= bond_mode
;
4511 params
->xmit_policy
= xmit_hashtype
;
4512 params
->miimon
= miimon
;
4513 params
->num_peer_notif
= num_peer_notif
;
4514 params
->arp_interval
= arp_interval
;
4515 params
->arp_validate
= arp_validate_value
;
4516 params
->arp_all_targets
= arp_all_targets_value
;
4517 params
->updelay
= updelay
;
4518 params
->downdelay
= downdelay
;
4519 params
->use_carrier
= use_carrier
;
4520 params
->lacp_fast
= lacp_fast
;
4521 params
->primary
[0] = 0;
4522 params
->primary_reselect
= primary_reselect_value
;
4523 params
->fail_over_mac
= fail_over_mac_value
;
4524 params
->tx_queues
= tx_queues
;
4525 params
->all_slaves_active
= all_slaves_active
;
4526 params
->resend_igmp
= resend_igmp
;
4527 params
->min_links
= min_links
;
4528 params
->lp_interval
= lp_interval
;
4529 params
->packets_per_slave
= packets_per_slave
;
4530 params
->tlb_dynamic_lb
= 1; /* Default value */
4531 params
->ad_actor_sys_prio
= ad_actor_sys_prio
;
4532 eth_zero_addr(params
->ad_actor_system
);
4533 params
->ad_user_port_key
= ad_user_port_key
;
4534 if (packets_per_slave
> 0) {
4535 params
->reciprocal_packets_per_slave
=
4536 reciprocal_value(packets_per_slave
);
4538 /* reciprocal_packets_per_slave is unused if
4539 * packets_per_slave is 0 or 1, just initialize it
4541 params
->reciprocal_packets_per_slave
=
4542 (struct reciprocal_value
) { 0 };
4546 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4547 params
->primary
[IFNAMSIZ
- 1] = 0;
4550 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4555 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4556 static struct lock_class_key bonding_netdev_addr_lock_key
;
4557 static struct lock_class_key bonding_tx_busylock_key
;
4559 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4560 struct netdev_queue
*txq
,
4563 lockdep_set_class(&txq
->_xmit_lock
,
4564 &bonding_netdev_xmit_lock_key
);
4567 static void bond_set_lockdep_class(struct net_device
*dev
)
4569 lockdep_set_class(&dev
->addr_list_lock
,
4570 &bonding_netdev_addr_lock_key
);
4571 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4572 dev
->qdisc_tx_busylock
= &bonding_tx_busylock_key
;
4575 /* Called from registration process */
4576 static int bond_init(struct net_device
*bond_dev
)
4578 struct bonding
*bond
= netdev_priv(bond_dev
);
4579 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4581 netdev_dbg(bond_dev
, "Begin bond_init\n");
4583 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4587 bond_set_lockdep_class(bond_dev
);
4589 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4591 bond_prepare_sysfs_group(bond
);
4593 bond_debug_register(bond
);
4595 /* Ensure valid dev_addr */
4596 if (is_zero_ether_addr(bond_dev
->dev_addr
) &&
4597 bond_dev
->addr_assign_type
== NET_ADDR_PERM
)
4598 eth_hw_addr_random(bond_dev
);
4603 unsigned int bond_get_num_tx_queues(void)
4608 /* Create a new bond based on the specified name and bonding parameters.
4609 * If name is NULL, obtain a suitable "bond%d" name for us.
4610 * Caller must NOT hold rtnl_lock; we need to release it here before we
4611 * set up our sysfs entries.
4613 int bond_create(struct net
*net
, const char *name
)
4615 struct net_device
*bond_dev
;
4616 struct bonding
*bond
;
4617 struct alb_bond_info
*bond_info
;
4622 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
),
4623 name
? name
: "bond%d", NET_NAME_UNKNOWN
,
4624 bond_setup
, tx_queues
);
4626 pr_err("%s: eek! can't alloc netdev!\n", name
);
4632 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4633 * It is set to 0 by default which is wrong.
4635 bond
= netdev_priv(bond_dev
);
4636 bond_info
= &(BOND_ALB_INFO(bond
));
4637 bond_info
->rx_hashtbl_used_head
= RLB_NULL_INDEX
;
4639 dev_net_set(bond_dev
, net
);
4640 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
4642 res
= register_netdevice(bond_dev
);
4644 netif_carrier_off(bond_dev
);
4648 bond_destructor(bond_dev
);
4652 static int __net_init
bond_net_init(struct net
*net
)
4654 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4657 INIT_LIST_HEAD(&bn
->dev_list
);
4659 bond_create_proc_dir(bn
);
4660 bond_create_sysfs(bn
);
4665 static void __net_exit
bond_net_exit(struct net
*net
)
4667 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
4668 struct bonding
*bond
, *tmp_bond
;
4671 bond_destroy_sysfs(bn
);
4673 /* Kill off any bonds created after unregistering bond rtnl ops */
4675 list_for_each_entry_safe(bond
, tmp_bond
, &bn
->dev_list
, bond_list
)
4676 unregister_netdevice_queue(bond
->dev
, &list
);
4677 unregister_netdevice_many(&list
);
4680 bond_destroy_proc_dir(bn
);
4683 static struct pernet_operations bond_net_ops
= {
4684 .init
= bond_net_init
,
4685 .exit
= bond_net_exit
,
4687 .size
= sizeof(struct bond_net
),
4690 static int __init
bonding_init(void)
4695 pr_info("%s", bond_version
);
4697 res
= bond_check_params(&bonding_defaults
);
4701 res
= register_pernet_subsys(&bond_net_ops
);
4705 res
= bond_netlink_init();
4709 bond_create_debugfs();
4711 for (i
= 0; i
< max_bonds
; i
++) {
4712 res
= bond_create(&init_net
, NULL
);
4717 register_netdevice_notifier(&bond_netdev_notifier
);
4721 bond_destroy_debugfs();
4722 bond_netlink_fini();
4724 unregister_pernet_subsys(&bond_net_ops
);
4729 static void __exit
bonding_exit(void)
4731 unregister_netdevice_notifier(&bond_netdev_notifier
);
4733 bond_destroy_debugfs();
4735 bond_netlink_fini();
4736 unregister_pernet_subsys(&bond_net_ops
);
4738 #ifdef CONFIG_NET_POLL_CONTROLLER
4739 /* Make sure we don't have an imbalance on our netpoll blocking */
4740 WARN_ON(atomic_read(&netpoll_block_tx
));
4744 module_init(bonding_init
);
4745 module_exit(bonding_exit
);
4746 MODULE_LICENSE("GPL");
4747 MODULE_VERSION(DRV_VERSION
);
4748 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
4749 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");