2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
57 #include <asm/system.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <linux/preempt.h>
77 #include <net/route.h>
78 #include <net/net_namespace.h>
79 #include <net/netns/generic.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV 0
88 #define BOND_LINK_ARP_INTERV 0
90 static int max_bonds
= BOND_DEFAULT_MAX_BONDS
;
91 static int tx_queues
= BOND_DEFAULT_TX_QUEUES
;
92 static int num_grat_arp
= 1;
93 static int num_unsol_na
= 1;
94 static int miimon
= BOND_LINK_MON_INTERV
;
97 static int use_carrier
= 1;
100 static char *primary_reselect
;
101 static char *lacp_rate
;
102 static char *ad_select
;
103 static char *xmit_hash_policy
;
104 static int arp_interval
= BOND_LINK_ARP_INTERV
;
105 static char *arp_ip_target
[BOND_MAX_ARP_TARGETS
];
106 static char *arp_validate
;
107 static char *fail_over_mac
;
108 static int all_slaves_active
= 0;
109 static struct bond_params bonding_defaults
;
110 static int resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
112 module_param(max_bonds
, int, 0);
113 MODULE_PARM_DESC(max_bonds
, "Max number of bonded devices");
114 module_param(tx_queues
, int, 0);
115 MODULE_PARM_DESC(tx_queues
, "Max number of transmit queues (default = 16)");
116 module_param(num_grat_arp
, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp
, "Number of gratuitous ARP packets to send on failover event");
118 module_param(num_unsol_na
, int, 0644);
119 MODULE_PARM_DESC(num_unsol_na
, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
120 module_param(miimon
, int, 0);
121 MODULE_PARM_DESC(miimon
, "Link check interval in milliseconds");
122 module_param(updelay
, int, 0);
123 MODULE_PARM_DESC(updelay
, "Delay before considering link up, in milliseconds");
124 module_param(downdelay
, int, 0);
125 MODULE_PARM_DESC(downdelay
, "Delay before considering link down, "
127 module_param(use_carrier
, int, 0);
128 MODULE_PARM_DESC(use_carrier
, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
129 "0 for off, 1 for on (default)");
130 module_param(mode
, charp
, 0);
131 MODULE_PARM_DESC(mode
, "Mode of operation : 0 for balance-rr, "
132 "1 for active-backup, 2 for balance-xor, "
133 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
134 "6 for balance-alb");
135 module_param(primary
, charp
, 0);
136 MODULE_PARM_DESC(primary
, "Primary network device to use");
137 module_param(primary_reselect
, charp
, 0);
138 MODULE_PARM_DESC(primary_reselect
, "Reselect primary slave "
140 "0 for always (default), "
141 "1 for only if speed of primary is "
143 "2 for only on active slave "
145 module_param(lacp_rate
, charp
, 0);
146 MODULE_PARM_DESC(lacp_rate
, "LACPDU tx rate to request from 802.3ad partner "
148 module_param(ad_select
, charp
, 0);
149 MODULE_PARM_DESC(ad_select
, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
150 module_param(xmit_hash_policy
, charp
, 0);
151 MODULE_PARM_DESC(xmit_hash_policy
, "XOR hashing method: 0 for layer 2 (default)"
152 ", 1 for layer 3+4");
153 module_param(arp_interval
, int, 0);
154 MODULE_PARM_DESC(arp_interval
, "arp interval in milliseconds");
155 module_param_array(arp_ip_target
, charp
, NULL
, 0);
156 MODULE_PARM_DESC(arp_ip_target
, "arp targets in n.n.n.n form");
157 module_param(arp_validate
, charp
, 0);
158 MODULE_PARM_DESC(arp_validate
, "validate src/dst of ARP probes: none (default), active, backup or all");
159 module_param(fail_over_mac
, charp
, 0);
160 MODULE_PARM_DESC(fail_over_mac
, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
161 module_param(all_slaves_active
, int, 0);
162 MODULE_PARM_DESC(all_slaves_active
, "Keep all frames received on an interface"
163 "by setting active flag for all slaves. "
164 "0 for never (default), 1 for always.");
165 module_param(resend_igmp
, int, 0);
166 MODULE_PARM_DESC(resend_igmp
, "Number of IGMP membership reports to send on link failure");
168 /*----------------------------- Global variables ----------------------------*/
170 #ifdef CONFIG_NET_POLL_CONTROLLER
171 atomic_t netpoll_block_tx
= ATOMIC_INIT(0);
174 int bond_net_id __read_mostly
;
176 static __be32 arp_target
[BOND_MAX_ARP_TARGETS
];
177 static int arp_ip_count
;
178 static int bond_mode
= BOND_MODE_ROUNDROBIN
;
179 static int xmit_hashtype
= BOND_XMIT_POLICY_LAYER2
;
180 static int lacp_fast
;
182 const struct bond_parm_tbl bond_lacp_tbl
[] = {
183 { "slow", AD_LACP_SLOW
},
184 { "fast", AD_LACP_FAST
},
188 const struct bond_parm_tbl bond_mode_tbl
[] = {
189 { "balance-rr", BOND_MODE_ROUNDROBIN
},
190 { "active-backup", BOND_MODE_ACTIVEBACKUP
},
191 { "balance-xor", BOND_MODE_XOR
},
192 { "broadcast", BOND_MODE_BROADCAST
},
193 { "802.3ad", BOND_MODE_8023AD
},
194 { "balance-tlb", BOND_MODE_TLB
},
195 { "balance-alb", BOND_MODE_ALB
},
199 const struct bond_parm_tbl xmit_hashtype_tbl
[] = {
200 { "layer2", BOND_XMIT_POLICY_LAYER2
},
201 { "layer3+4", BOND_XMIT_POLICY_LAYER34
},
202 { "layer2+3", BOND_XMIT_POLICY_LAYER23
},
206 const struct bond_parm_tbl arp_validate_tbl
[] = {
207 { "none", BOND_ARP_VALIDATE_NONE
},
208 { "active", BOND_ARP_VALIDATE_ACTIVE
},
209 { "backup", BOND_ARP_VALIDATE_BACKUP
},
210 { "all", BOND_ARP_VALIDATE_ALL
},
214 const struct bond_parm_tbl fail_over_mac_tbl
[] = {
215 { "none", BOND_FOM_NONE
},
216 { "active", BOND_FOM_ACTIVE
},
217 { "follow", BOND_FOM_FOLLOW
},
221 const struct bond_parm_tbl pri_reselect_tbl
[] = {
222 { "always", BOND_PRI_RESELECT_ALWAYS
},
223 { "better", BOND_PRI_RESELECT_BETTER
},
224 { "failure", BOND_PRI_RESELECT_FAILURE
},
228 struct bond_parm_tbl ad_select_tbl
[] = {
229 { "stable", BOND_AD_STABLE
},
230 { "bandwidth", BOND_AD_BANDWIDTH
},
231 { "count", BOND_AD_COUNT
},
235 /*-------------------------- Forward declarations ---------------------------*/
237 static void bond_send_gratuitous_arp(struct bonding
*bond
);
238 static int bond_init(struct net_device
*bond_dev
);
239 static void bond_uninit(struct net_device
*bond_dev
);
241 /*---------------------------- General routines -----------------------------*/
243 const char *bond_mode_name(int mode
)
245 static const char *names
[] = {
246 [BOND_MODE_ROUNDROBIN
] = "load balancing (round-robin)",
247 [BOND_MODE_ACTIVEBACKUP
] = "fault-tolerance (active-backup)",
248 [BOND_MODE_XOR
] = "load balancing (xor)",
249 [BOND_MODE_BROADCAST
] = "fault-tolerance (broadcast)",
250 [BOND_MODE_8023AD
] = "IEEE 802.3ad Dynamic link aggregation",
251 [BOND_MODE_TLB
] = "transmit load balancing",
252 [BOND_MODE_ALB
] = "adaptive load balancing",
255 if (mode
< 0 || mode
> BOND_MODE_ALB
)
261 /*---------------------------------- VLAN -----------------------------------*/
264 * bond_add_vlan - add a new vlan id on bond
265 * @bond: bond that got the notification
266 * @vlan_id: the vlan id to add
268 * Returns -ENOMEM if allocation failed.
270 static int bond_add_vlan(struct bonding
*bond
, unsigned short vlan_id
)
272 struct vlan_entry
*vlan
;
274 pr_debug("bond: %s, vlan id %d\n",
275 (bond
? bond
->dev
->name
: "None"), vlan_id
);
277 vlan
= kzalloc(sizeof(struct vlan_entry
), GFP_KERNEL
);
281 INIT_LIST_HEAD(&vlan
->vlan_list
);
282 vlan
->vlan_id
= vlan_id
;
284 write_lock_bh(&bond
->lock
);
286 list_add_tail(&vlan
->vlan_list
, &bond
->vlan_list
);
288 write_unlock_bh(&bond
->lock
);
290 pr_debug("added VLAN ID %d on bond %s\n", vlan_id
, bond
->dev
->name
);
296 * bond_del_vlan - delete a vlan id from bond
297 * @bond: bond that got the notification
298 * @vlan_id: the vlan id to delete
300 * returns -ENODEV if @vlan_id was not found in @bond.
302 static int bond_del_vlan(struct bonding
*bond
, unsigned short vlan_id
)
304 struct vlan_entry
*vlan
;
307 pr_debug("bond: %s, vlan id %d\n", bond
->dev
->name
, vlan_id
);
310 write_lock_bh(&bond
->lock
);
312 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
313 if (vlan
->vlan_id
== vlan_id
) {
314 list_del(&vlan
->vlan_list
);
316 if (bond_is_lb(bond
))
317 bond_alb_clear_vlan(bond
, vlan_id
);
319 pr_debug("removed VLAN ID %d from bond %s\n",
320 vlan_id
, bond
->dev
->name
);
324 if (list_empty(&bond
->vlan_list
) &&
325 (bond
->slave_cnt
== 0)) {
326 /* Last VLAN removed and no slaves, so
327 * restore block on adding VLANs. This will
328 * be removed once new slaves that are not
329 * VLAN challenged will be added.
331 bond
->dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
339 pr_debug("couldn't find VLAN ID %d in bond %s\n",
340 vlan_id
, bond
->dev
->name
);
343 write_unlock_bh(&bond
->lock
);
344 unblock_netpoll_tx();
349 * bond_has_challenged_slaves
350 * @bond: the bond we're working on
352 * Searches the slave list. Returns 1 if a vlan challenged slave
353 * was found, 0 otherwise.
355 * Assumes bond->lock is held.
357 static int bond_has_challenged_slaves(struct bonding
*bond
)
362 bond_for_each_slave(bond
, slave
, i
) {
363 if (slave
->dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
364 pr_debug("found VLAN challenged slave - %s\n",
370 pr_debug("no VLAN challenged slaves found\n");
375 * bond_next_vlan - safely skip to the next item in the vlans list.
376 * @bond: the bond we're working on
377 * @curr: item we're advancing from
379 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
380 * or @curr->next otherwise (even if it is @curr itself again).
382 * Caller must hold bond->lock
384 struct vlan_entry
*bond_next_vlan(struct bonding
*bond
, struct vlan_entry
*curr
)
386 struct vlan_entry
*next
, *last
;
388 if (list_empty(&bond
->vlan_list
))
392 next
= list_entry(bond
->vlan_list
.next
,
393 struct vlan_entry
, vlan_list
);
395 last
= list_entry(bond
->vlan_list
.prev
,
396 struct vlan_entry
, vlan_list
);
398 next
= list_entry(bond
->vlan_list
.next
,
399 struct vlan_entry
, vlan_list
);
401 next
= list_entry(curr
->vlan_list
.next
,
402 struct vlan_entry
, vlan_list
);
410 * bond_dev_queue_xmit - Prepare skb for xmit.
412 * @bond: bond device that got this skb for tx.
413 * @skb: hw accel VLAN tagged skb to transmit
414 * @slave_dev: slave that is supposed to xmit this skbuff
416 int bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
,
417 struct net_device
*slave_dev
)
419 skb
->dev
= slave_dev
;
421 if (unlikely(netpoll_tx_running(slave_dev
)))
422 bond_netpoll_send_skb(bond_get_slave_by_dev(bond
, slave_dev
), skb
);
430 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
431 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
433 * a. This operation is performed in IOCTL context,
434 * b. The operation is protected by the RTNL semaphore in the 8021q code,
435 * c. Holding a lock with BH disabled while directly calling a base driver
436 * entry point is generally a BAD idea.
438 * The design of synchronization/protection for this operation in the 8021q
439 * module is good for one or more VLAN devices over a single physical device
440 * and cannot be extended for a teaming solution like bonding, so there is a
441 * potential race condition here where a net device from the vlan group might
442 * be referenced (either by a base driver or the 8021q code) while it is being
443 * removed from the system. However, it turns out we're not making matters
444 * worse, and if it works for regular VLAN usage it will work here too.
448 * bond_vlan_rx_register - Propagates registration to slaves
449 * @bond_dev: bonding net device that got called
450 * @grp: vlan group being registered
452 static void bond_vlan_rx_register(struct net_device
*bond_dev
,
453 struct vlan_group
*grp
)
455 struct bonding
*bond
= netdev_priv(bond_dev
);
459 write_lock_bh(&bond
->lock
);
461 write_unlock_bh(&bond
->lock
);
463 bond_for_each_slave(bond
, slave
, i
) {
464 struct net_device
*slave_dev
= slave
->dev
;
465 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
467 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
468 slave_ops
->ndo_vlan_rx_register
) {
469 slave_ops
->ndo_vlan_rx_register(slave_dev
, grp
);
475 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
476 * @bond_dev: bonding net device that got called
477 * @vid: vlan id being added
479 static void bond_vlan_rx_add_vid(struct net_device
*bond_dev
, uint16_t vid
)
481 struct bonding
*bond
= netdev_priv(bond_dev
);
485 bond_for_each_slave(bond
, slave
, i
) {
486 struct net_device
*slave_dev
= slave
->dev
;
487 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
489 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
490 slave_ops
->ndo_vlan_rx_add_vid
) {
491 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vid
);
495 res
= bond_add_vlan(bond
, vid
);
497 pr_err("%s: Error: Failed to add vlan id %d\n",
498 bond_dev
->name
, vid
);
503 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
504 * @bond_dev: bonding net device that got called
505 * @vid: vlan id being removed
507 static void bond_vlan_rx_kill_vid(struct net_device
*bond_dev
, uint16_t vid
)
509 struct bonding
*bond
= netdev_priv(bond_dev
);
511 struct net_device
*vlan_dev
;
514 bond_for_each_slave(bond
, slave
, i
) {
515 struct net_device
*slave_dev
= slave
->dev
;
516 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
518 if ((slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) &&
519 slave_ops
->ndo_vlan_rx_kill_vid
) {
520 /* Save and then restore vlan_dev in the grp array,
521 * since the slave's driver might clear it.
523 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vid
);
524 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vid
);
525 vlan_group_set_device(bond
->vlgrp
, vid
, vlan_dev
);
529 res
= bond_del_vlan(bond
, vid
);
531 pr_err("%s: Error: Failed to remove vlan id %d\n",
532 bond_dev
->name
, vid
);
536 static void bond_add_vlans_on_slave(struct bonding
*bond
, struct net_device
*slave_dev
)
538 struct vlan_entry
*vlan
;
539 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
544 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
545 slave_ops
->ndo_vlan_rx_register
)
546 slave_ops
->ndo_vlan_rx_register(slave_dev
, bond
->vlgrp
);
548 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
549 !(slave_ops
->ndo_vlan_rx_add_vid
))
552 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
)
553 slave_ops
->ndo_vlan_rx_add_vid(slave_dev
, vlan
->vlan_id
);
556 static void bond_del_vlans_from_slave(struct bonding
*bond
,
557 struct net_device
*slave_dev
)
559 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
560 struct vlan_entry
*vlan
;
561 struct net_device
*vlan_dev
;
566 if (!(slave_dev
->features
& NETIF_F_HW_VLAN_FILTER
) ||
567 !(slave_ops
->ndo_vlan_rx_kill_vid
))
570 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
573 /* Save and then restore vlan_dev in the grp array,
574 * since the slave's driver might clear it.
576 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
577 slave_ops
->ndo_vlan_rx_kill_vid(slave_dev
, vlan
->vlan_id
);
578 vlan_group_set_device(bond
->vlgrp
, vlan
->vlan_id
, vlan_dev
);
582 if ((slave_dev
->features
& NETIF_F_HW_VLAN_RX
) &&
583 slave_ops
->ndo_vlan_rx_register
)
584 slave_ops
->ndo_vlan_rx_register(slave_dev
, NULL
);
587 /*------------------------------- Link status -------------------------------*/
590 * Set the carrier state for the master according to the state of its
591 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
592 * do special 802.3ad magic.
594 * Returns zero if carrier state does not change, nonzero if it does.
596 static int bond_set_carrier(struct bonding
*bond
)
601 if (bond
->slave_cnt
== 0)
604 if (bond
->params
.mode
== BOND_MODE_8023AD
)
605 return bond_3ad_set_carrier(bond
);
607 bond_for_each_slave(bond
, slave
, i
) {
608 if (slave
->link
== BOND_LINK_UP
) {
609 if (!netif_carrier_ok(bond
->dev
)) {
610 netif_carrier_on(bond
->dev
);
618 if (netif_carrier_ok(bond
->dev
)) {
619 netif_carrier_off(bond
->dev
);
626 * Get link speed and duplex from the slave's base driver
627 * using ethtool. If for some reason the call fails or the
628 * values are invalid, fake speed and duplex to 100/Full
631 static int bond_update_speed_duplex(struct slave
*slave
)
633 struct net_device
*slave_dev
= slave
->dev
;
634 struct ethtool_cmd etool
;
637 /* Fake speed and duplex */
638 slave
->speed
= SPEED_100
;
639 slave
->duplex
= DUPLEX_FULL
;
641 if (!slave_dev
->ethtool_ops
|| !slave_dev
->ethtool_ops
->get_settings
)
644 res
= slave_dev
->ethtool_ops
->get_settings(slave_dev
, &etool
);
648 switch (etool
.speed
) {
658 switch (etool
.duplex
) {
666 slave
->speed
= etool
.speed
;
667 slave
->duplex
= etool
.duplex
;
673 * if <dev> supports MII link status reporting, check its link status.
675 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
676 * depending upon the setting of the use_carrier parameter.
678 * Return either BMSR_LSTATUS, meaning that the link is up (or we
679 * can't tell and just pretend it is), or 0, meaning that the link is
682 * If reporting is non-zero, instead of faking link up, return -1 if
683 * both ETHTOOL and MII ioctls fail (meaning the device does not
684 * support them). If use_carrier is set, return whatever it says.
685 * It'd be nice if there was a good way to tell if a driver supports
686 * netif_carrier, but there really isn't.
688 static int bond_check_dev_link(struct bonding
*bond
,
689 struct net_device
*slave_dev
, int reporting
)
691 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
692 int (*ioctl
)(struct net_device
*, struct ifreq
*, int);
694 struct mii_ioctl_data
*mii
;
696 if (!reporting
&& !netif_running(slave_dev
))
699 if (bond
->params
.use_carrier
)
700 return netif_carrier_ok(slave_dev
) ? BMSR_LSTATUS
: 0;
702 /* Try to get link status using Ethtool first. */
703 if (slave_dev
->ethtool_ops
) {
704 if (slave_dev
->ethtool_ops
->get_link
) {
707 link
= slave_dev
->ethtool_ops
->get_link(slave_dev
);
709 return link
? BMSR_LSTATUS
: 0;
713 /* Ethtool can't be used, fallback to MII ioctls. */
714 ioctl
= slave_ops
->ndo_do_ioctl
;
716 /* TODO: set pointer to correct ioctl on a per team member */
717 /* bases to make this more efficient. that is, once */
718 /* we determine the correct ioctl, we will always */
719 /* call it and not the others for that team */
723 * We cannot assume that SIOCGMIIPHY will also read a
724 * register; not all network drivers (e.g., e100)
728 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
729 strncpy(ifr
.ifr_name
, slave_dev
->name
, IFNAMSIZ
);
731 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIPHY
) == 0) {
732 mii
->reg_num
= MII_BMSR
;
733 if (IOCTL(slave_dev
, &ifr
, SIOCGMIIREG
) == 0)
734 return mii
->val_out
& BMSR_LSTATUS
;
739 * If reporting, report that either there's no dev->do_ioctl,
740 * or both SIOCGMIIREG and get_link failed (meaning that we
741 * cannot report link status). If not reporting, pretend
744 return reporting
? -1 : BMSR_LSTATUS
;
747 /*----------------------------- Multicast list ------------------------------*/
750 * Push the promiscuity flag down to appropriate slaves
752 static int bond_set_promiscuity(struct bonding
*bond
, int inc
)
755 if (USES_PRIMARY(bond
->params
.mode
)) {
756 /* write lock already acquired */
757 if (bond
->curr_active_slave
) {
758 err
= dev_set_promiscuity(bond
->curr_active_slave
->dev
,
764 bond_for_each_slave(bond
, slave
, i
) {
765 err
= dev_set_promiscuity(slave
->dev
, inc
);
774 * Push the allmulti flag down to all slaves
776 static int bond_set_allmulti(struct bonding
*bond
, int inc
)
779 if (USES_PRIMARY(bond
->params
.mode
)) {
780 /* write lock already acquired */
781 if (bond
->curr_active_slave
) {
782 err
= dev_set_allmulti(bond
->curr_active_slave
->dev
,
788 bond_for_each_slave(bond
, slave
, i
) {
789 err
= dev_set_allmulti(slave
->dev
, inc
);
798 * Add a Multicast address to slaves
801 static void bond_mc_add(struct bonding
*bond
, void *addr
)
803 if (USES_PRIMARY(bond
->params
.mode
)) {
804 /* write lock already acquired */
805 if (bond
->curr_active_slave
)
806 dev_mc_add(bond
->curr_active_slave
->dev
, addr
);
811 bond_for_each_slave(bond
, slave
, i
)
812 dev_mc_add(slave
->dev
, addr
);
817 * Remove a multicast address from slave
820 static void bond_mc_del(struct bonding
*bond
, void *addr
)
822 if (USES_PRIMARY(bond
->params
.mode
)) {
823 /* write lock already acquired */
824 if (bond
->curr_active_slave
)
825 dev_mc_del(bond
->curr_active_slave
->dev
, addr
);
829 bond_for_each_slave(bond
, slave
, i
) {
830 dev_mc_del(slave
->dev
, addr
);
836 static void __bond_resend_igmp_join_requests(struct net_device
*dev
)
838 struct in_device
*in_dev
;
841 in_dev
= __in_dev_get_rcu(dev
);
843 ip_mc_rejoin_groups(in_dev
);
848 * Retrieve the list of registered multicast addresses for the bonding
849 * device and retransmit an IGMP JOIN request to the current active
852 static void bond_resend_igmp_join_requests(struct bonding
*bond
)
854 struct net_device
*vlan_dev
;
855 struct vlan_entry
*vlan
;
857 read_lock(&bond
->lock
);
859 /* rejoin all groups on bond device */
860 __bond_resend_igmp_join_requests(bond
->dev
);
862 /* rejoin all groups on vlan devices */
864 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
865 vlan_dev
= vlan_group_get_device(bond
->vlgrp
,
868 __bond_resend_igmp_join_requests(vlan_dev
);
872 if (--bond
->igmp_retrans
> 0)
873 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, HZ
/5);
875 read_unlock(&bond
->lock
);
878 static void bond_resend_igmp_join_requests_delayed(struct work_struct
*work
)
880 struct bonding
*bond
= container_of(work
, struct bonding
,
882 bond_resend_igmp_join_requests(bond
);
886 * flush all members of flush->mc_list from device dev->mc_list
888 static void bond_mc_list_flush(struct net_device
*bond_dev
,
889 struct net_device
*slave_dev
)
891 struct bonding
*bond
= netdev_priv(bond_dev
);
892 struct netdev_hw_addr
*ha
;
894 netdev_for_each_mc_addr(ha
, bond_dev
)
895 dev_mc_del(slave_dev
, ha
->addr
);
897 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
898 /* del lacpdu mc addr from mc list */
899 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
901 dev_mc_del(slave_dev
, lacpdu_multicast
);
905 /*--------------------------- Active slave change ---------------------------*/
908 * Update the mc list and multicast-related flags for the new and
909 * old active slaves (if any) according to the multicast mode, and
910 * promiscuous flags unconditionally.
912 static void bond_mc_swap(struct bonding
*bond
, struct slave
*new_active
,
913 struct slave
*old_active
)
915 struct netdev_hw_addr
*ha
;
917 if (!USES_PRIMARY(bond
->params
.mode
))
918 /* nothing to do - mc list is already up-to-date on
924 if (bond
->dev
->flags
& IFF_PROMISC
)
925 dev_set_promiscuity(old_active
->dev
, -1);
927 if (bond
->dev
->flags
& IFF_ALLMULTI
)
928 dev_set_allmulti(old_active
->dev
, -1);
930 netdev_for_each_mc_addr(ha
, bond
->dev
)
931 dev_mc_del(old_active
->dev
, ha
->addr
);
935 /* FIXME: Signal errors upstream. */
936 if (bond
->dev
->flags
& IFF_PROMISC
)
937 dev_set_promiscuity(new_active
->dev
, 1);
939 if (bond
->dev
->flags
& IFF_ALLMULTI
)
940 dev_set_allmulti(new_active
->dev
, 1);
942 netdev_for_each_mc_addr(ha
, bond
->dev
)
943 dev_mc_add(new_active
->dev
, ha
->addr
);
948 * bond_do_fail_over_mac
950 * Perform special MAC address swapping for fail_over_mac settings
952 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
954 static void bond_do_fail_over_mac(struct bonding
*bond
,
955 struct slave
*new_active
,
956 struct slave
*old_active
)
957 __releases(&bond
->curr_slave_lock
)
958 __releases(&bond
->lock
)
959 __acquires(&bond
->lock
)
960 __acquires(&bond
->curr_slave_lock
)
962 u8 tmp_mac
[ETH_ALEN
];
963 struct sockaddr saddr
;
966 switch (bond
->params
.fail_over_mac
) {
967 case BOND_FOM_ACTIVE
:
969 memcpy(bond
->dev
->dev_addr
, new_active
->dev
->dev_addr
,
970 new_active
->dev
->addr_len
);
972 case BOND_FOM_FOLLOW
:
974 * if new_active && old_active, swap them
975 * if just old_active, do nothing (going to no active slave)
976 * if just new_active, set new_active to bond's MAC
981 write_unlock_bh(&bond
->curr_slave_lock
);
982 read_unlock(&bond
->lock
);
985 memcpy(tmp_mac
, new_active
->dev
->dev_addr
, ETH_ALEN
);
986 memcpy(saddr
.sa_data
, old_active
->dev
->dev_addr
,
988 saddr
.sa_family
= new_active
->dev
->type
;
990 memcpy(saddr
.sa_data
, bond
->dev
->dev_addr
, ETH_ALEN
);
991 saddr
.sa_family
= bond
->dev
->type
;
994 rv
= dev_set_mac_address(new_active
->dev
, &saddr
);
996 pr_err("%s: Error %d setting MAC of slave %s\n",
997 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1004 memcpy(saddr
.sa_data
, tmp_mac
, ETH_ALEN
);
1005 saddr
.sa_family
= old_active
->dev
->type
;
1007 rv
= dev_set_mac_address(old_active
->dev
, &saddr
);
1009 pr_err("%s: Error %d setting MAC of slave %s\n",
1010 bond
->dev
->name
, -rv
, new_active
->dev
->name
);
1012 read_lock(&bond
->lock
);
1013 write_lock_bh(&bond
->curr_slave_lock
);
1016 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1017 bond
->dev
->name
, bond
->params
.fail_over_mac
);
1023 static bool bond_should_change_active(struct bonding
*bond
)
1025 struct slave
*prim
= bond
->primary_slave
;
1026 struct slave
*curr
= bond
->curr_active_slave
;
1028 if (!prim
|| !curr
|| curr
->link
!= BOND_LINK_UP
)
1030 if (bond
->force_primary
) {
1031 bond
->force_primary
= false;
1034 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_BETTER
&&
1035 (prim
->speed
< curr
->speed
||
1036 (prim
->speed
== curr
->speed
&& prim
->duplex
<= curr
->duplex
)))
1038 if (bond
->params
.primary_reselect
== BOND_PRI_RESELECT_FAILURE
)
1044 * find_best_interface - select the best available slave to be the active one
1045 * @bond: our bonding struct
1047 * Warning: Caller must hold curr_slave_lock for writing.
1049 static struct slave
*bond_find_best_slave(struct bonding
*bond
)
1051 struct slave
*new_active
, *old_active
;
1052 struct slave
*bestslave
= NULL
;
1053 int mintime
= bond
->params
.updelay
;
1056 new_active
= bond
->curr_active_slave
;
1058 if (!new_active
) { /* there were no active slaves left */
1059 if (bond
->slave_cnt
> 0) /* found one slave */
1060 new_active
= bond
->first_slave
;
1062 return NULL
; /* still no slave, return NULL */
1065 if ((bond
->primary_slave
) &&
1066 bond
->primary_slave
->link
== BOND_LINK_UP
&&
1067 bond_should_change_active(bond
)) {
1068 new_active
= bond
->primary_slave
;
1071 /* remember where to stop iterating over the slaves */
1072 old_active
= new_active
;
1074 bond_for_each_slave_from(bond
, new_active
, i
, old_active
) {
1075 if (new_active
->link
== BOND_LINK_UP
) {
1077 } else if (new_active
->link
== BOND_LINK_BACK
&&
1078 IS_UP(new_active
->dev
)) {
1079 /* link up, but waiting for stabilization */
1080 if (new_active
->delay
< mintime
) {
1081 mintime
= new_active
->delay
;
1082 bestslave
= new_active
;
1091 * change_active_interface - change the active slave into the specified one
1092 * @bond: our bonding struct
1093 * @new: the new slave to make the active one
1095 * Set the new slave to the bond's settings and unset them on the old
1096 * curr_active_slave.
1097 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1099 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1100 * because it is apparently the best available slave we have, even though its
1101 * updelay hasn't timed out yet.
1103 * If new_active is not NULL, caller must hold bond->lock for read and
1104 * curr_slave_lock for write_bh.
1106 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
)
1108 struct slave
*old_active
= bond
->curr_active_slave
;
1110 if (old_active
== new_active
)
1114 new_active
->jiffies
= jiffies
;
1116 if (new_active
->link
== BOND_LINK_BACK
) {
1117 if (USES_PRIMARY(bond
->params
.mode
)) {
1118 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1119 bond
->dev
->name
, new_active
->dev
->name
,
1120 (bond
->params
.updelay
- new_active
->delay
) * bond
->params
.miimon
);
1123 new_active
->delay
= 0;
1124 new_active
->link
= BOND_LINK_UP
;
1126 if (bond
->params
.mode
== BOND_MODE_8023AD
)
1127 bond_3ad_handle_link_change(new_active
, BOND_LINK_UP
);
1129 if (bond_is_lb(bond
))
1130 bond_alb_handle_link_change(bond
, new_active
, BOND_LINK_UP
);
1132 if (USES_PRIMARY(bond
->params
.mode
)) {
1133 pr_info("%s: making interface %s the new active one.\n",
1134 bond
->dev
->name
, new_active
->dev
->name
);
1139 if (USES_PRIMARY(bond
->params
.mode
))
1140 bond_mc_swap(bond
, new_active
, old_active
);
1142 if (bond_is_lb(bond
)) {
1143 bond_alb_handle_active_change(bond
, new_active
);
1145 bond_set_slave_inactive_flags(old_active
);
1147 bond_set_slave_active_flags(new_active
);
1149 bond
->curr_active_slave
= new_active
;
1152 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
) {
1154 bond_set_slave_inactive_flags(old_active
);
1157 bond_set_slave_active_flags(new_active
);
1159 if (bond
->params
.fail_over_mac
)
1160 bond_do_fail_over_mac(bond
, new_active
,
1163 if (netif_running(bond
->dev
)) {
1164 bond
->send_grat_arp
= bond
->params
.num_grat_arp
;
1165 bond_send_gratuitous_arp(bond
);
1167 bond
->send_unsol_na
= bond
->params
.num_unsol_na
;
1168 bond_send_unsolicited_na(bond
);
1171 write_unlock_bh(&bond
->curr_slave_lock
);
1172 read_unlock(&bond
->lock
);
1174 netdev_bonding_change(bond
->dev
, NETDEV_BONDING_FAILOVER
);
1176 read_lock(&bond
->lock
);
1177 write_lock_bh(&bond
->curr_slave_lock
);
1181 /* resend IGMP joins since active slave has changed or
1182 * all were sent on curr_active_slave */
1183 if (((USES_PRIMARY(bond
->params
.mode
) && new_active
) ||
1184 bond
->params
.mode
== BOND_MODE_ROUNDROBIN
) &&
1185 netif_running(bond
->dev
)) {
1186 bond
->igmp_retrans
= bond
->params
.resend_igmp
;
1187 queue_delayed_work(bond
->wq
, &bond
->mcast_work
, 0);
1192 * bond_select_active_slave - select a new active slave, if needed
1193 * @bond: our bonding struct
1195 * This functions should be called when one of the following occurs:
1196 * - The old curr_active_slave has been released or lost its link.
1197 * - The primary_slave has got its link back.
1198 * - A slave has got its link back and there's no old curr_active_slave.
1200 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1202 void bond_select_active_slave(struct bonding
*bond
)
1204 struct slave
*best_slave
;
1207 best_slave
= bond_find_best_slave(bond
);
1208 if (best_slave
!= bond
->curr_active_slave
) {
1209 bond_change_active_slave(bond
, best_slave
);
1210 rv
= bond_set_carrier(bond
);
1214 if (netif_carrier_ok(bond
->dev
)) {
1215 pr_info("%s: first active interface up!\n",
1218 pr_info("%s: now running without any active interface !\n",
1224 /*--------------------------- slave list handling ---------------------------*/
1227 * This function attaches the slave to the end of list.
1229 * bond->lock held for writing by caller.
1231 static void bond_attach_slave(struct bonding
*bond
, struct slave
*new_slave
)
1233 if (bond
->first_slave
== NULL
) { /* attaching the first slave */
1234 new_slave
->next
= new_slave
;
1235 new_slave
->prev
= new_slave
;
1236 bond
->first_slave
= new_slave
;
1238 new_slave
->next
= bond
->first_slave
;
1239 new_slave
->prev
= bond
->first_slave
->prev
;
1240 new_slave
->next
->prev
= new_slave
;
1241 new_slave
->prev
->next
= new_slave
;
1248 * This function detaches the slave from the list.
1249 * WARNING: no check is made to verify if the slave effectively
1250 * belongs to <bond>.
1251 * Nothing is freed on return, structures are just unchained.
1252 * If any slave pointer in bond was pointing to <slave>,
1253 * it should be changed by the calling function.
1255 * bond->lock held for writing by caller.
1257 static void bond_detach_slave(struct bonding
*bond
, struct slave
*slave
)
1260 slave
->next
->prev
= slave
->prev
;
1263 slave
->prev
->next
= slave
->next
;
1265 if (bond
->first_slave
== slave
) { /* slave is the first slave */
1266 if (bond
->slave_cnt
> 1) { /* there are more slave */
1267 bond
->first_slave
= slave
->next
;
1269 bond
->first_slave
= NULL
; /* slave was the last one */
1278 #ifdef CONFIG_NET_POLL_CONTROLLER
1279 static inline int slave_enable_netpoll(struct slave
*slave
)
1284 np
= kzalloc(sizeof(*np
), GFP_KERNEL
);
1289 np
->dev
= slave
->dev
;
1290 err
= __netpoll_setup(np
);
1299 static inline void slave_disable_netpoll(struct slave
*slave
)
1301 struct netpoll
*np
= slave
->np
;
1307 synchronize_rcu_bh();
1308 __netpoll_cleanup(np
);
1311 static inline bool slave_dev_support_netpoll(struct net_device
*slave_dev
)
1313 if (slave_dev
->priv_flags
& IFF_DISABLE_NETPOLL
)
1315 if (!slave_dev
->netdev_ops
->ndo_poll_controller
)
1320 static void bond_poll_controller(struct net_device
*bond_dev
)
1324 static void __bond_netpoll_cleanup(struct bonding
*bond
)
1326 struct slave
*slave
;
1329 bond_for_each_slave(bond
, slave
, i
)
1330 if (IS_UP(slave
->dev
))
1331 slave_disable_netpoll(slave
);
1333 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1335 struct bonding
*bond
= netdev_priv(bond_dev
);
1337 read_lock(&bond
->lock
);
1338 __bond_netpoll_cleanup(bond
);
1339 read_unlock(&bond
->lock
);
1342 static int bond_netpoll_setup(struct net_device
*dev
, struct netpoll_info
*ni
)
1344 struct bonding
*bond
= netdev_priv(dev
);
1345 struct slave
*slave
;
1348 read_lock(&bond
->lock
);
1349 bond_for_each_slave(bond
, slave
, i
) {
1350 if (!IS_UP(slave
->dev
))
1352 err
= slave_enable_netpoll(slave
);
1354 __bond_netpoll_cleanup(bond
);
1358 read_unlock(&bond
->lock
);
1362 static struct netpoll_info
*bond_netpoll_info(struct bonding
*bond
)
1364 return bond
->dev
->npinfo
;
1368 static inline int slave_enable_netpoll(struct slave
*slave
)
1372 static inline void slave_disable_netpoll(struct slave
*slave
)
1375 static void bond_netpoll_cleanup(struct net_device
*bond_dev
)
1380 /*---------------------------------- IOCTL ----------------------------------*/
1382 static int bond_sethwaddr(struct net_device
*bond_dev
,
1383 struct net_device
*slave_dev
)
1385 pr_debug("bond_dev=%p\n", bond_dev
);
1386 pr_debug("slave_dev=%p\n", slave_dev
);
1387 pr_debug("slave_dev->addr_len=%d\n", slave_dev
->addr_len
);
1388 memcpy(bond_dev
->dev_addr
, slave_dev
->dev_addr
, slave_dev
->addr_len
);
1392 #define BOND_VLAN_FEATURES \
1393 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1394 NETIF_F_HW_VLAN_FILTER)
1397 * Compute the common dev->feature set available to all slaves. Some
1398 * feature bits are managed elsewhere, so preserve those feature bits
1399 * on the master device.
1401 static int bond_compute_features(struct bonding
*bond
)
1403 struct slave
*slave
;
1404 struct net_device
*bond_dev
= bond
->dev
;
1405 u32 features
= bond_dev
->features
;
1406 u32 vlan_features
= 0;
1407 unsigned short max_hard_header_len
= max((u16
)ETH_HLEN
,
1408 bond_dev
->hard_header_len
);
1411 features
&= ~(NETIF_F_ALL_CSUM
| BOND_VLAN_FEATURES
);
1412 features
|= NETIF_F_GSO_MASK
| NETIF_F_NO_CSUM
;
1414 if (!bond
->first_slave
)
1417 features
&= ~NETIF_F_ONE_FOR_ALL
;
1419 vlan_features
= bond
->first_slave
->dev
->vlan_features
;
1420 bond_for_each_slave(bond
, slave
, i
) {
1421 features
= netdev_increment_features(features
,
1422 slave
->dev
->features
,
1423 NETIF_F_ONE_FOR_ALL
);
1424 vlan_features
= netdev_increment_features(vlan_features
,
1425 slave
->dev
->vlan_features
,
1426 NETIF_F_ONE_FOR_ALL
);
1427 if (slave
->dev
->hard_header_len
> max_hard_header_len
)
1428 max_hard_header_len
= slave
->dev
->hard_header_len
;
1432 features
|= (bond_dev
->features
& BOND_VLAN_FEATURES
);
1433 bond_dev
->features
= netdev_fix_features(bond_dev
, features
);
1434 bond_dev
->vlan_features
= netdev_fix_features(bond_dev
, vlan_features
);
1435 bond_dev
->hard_header_len
= max_hard_header_len
;
1440 static void bond_setup_by_slave(struct net_device
*bond_dev
,
1441 struct net_device
*slave_dev
)
1443 struct bonding
*bond
= netdev_priv(bond_dev
);
1445 bond_dev
->header_ops
= slave_dev
->header_ops
;
1447 bond_dev
->type
= slave_dev
->type
;
1448 bond_dev
->hard_header_len
= slave_dev
->hard_header_len
;
1449 bond_dev
->addr_len
= slave_dev
->addr_len
;
1451 memcpy(bond_dev
->broadcast
, slave_dev
->broadcast
,
1452 slave_dev
->addr_len
);
1453 bond
->setup_by_slave
= 1;
1456 /* On bonding slaves other than the currently active slave, suppress
1457 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
1458 * ARP on active-backup slaves with arp_validate enabled.
1460 static bool bond_should_deliver_exact_match(struct sk_buff
*skb
,
1461 struct net_device
*slave_dev
,
1462 struct net_device
*bond_dev
)
1464 if (slave_dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
1465 if (slave_dev
->priv_flags
& IFF_SLAVE_NEEDARP
&&
1466 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
1469 if (bond_dev
->priv_flags
& IFF_MASTER_ALB
&&
1470 skb
->pkt_type
!= PACKET_BROADCAST
&&
1471 skb
->pkt_type
!= PACKET_MULTICAST
)
1474 if (bond_dev
->priv_flags
& IFF_MASTER_8023AD
&&
1475 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
1483 static struct sk_buff
*bond_handle_frame(struct sk_buff
*skb
)
1485 struct net_device
*slave_dev
;
1486 struct net_device
*bond_dev
;
1488 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1491 slave_dev
= skb
->dev
;
1492 bond_dev
= ACCESS_ONCE(slave_dev
->master
);
1493 if (unlikely(!bond_dev
))
1496 if (bond_dev
->priv_flags
& IFF_MASTER_ARPMON
)
1497 slave_dev
->last_rx
= jiffies
;
1499 if (bond_should_deliver_exact_match(skb
, slave_dev
, bond_dev
)) {
1500 skb
->deliver_no_wcard
= 1;
1504 skb
->dev
= bond_dev
;
1506 if (bond_dev
->priv_flags
& IFF_MASTER_ALB
&&
1507 bond_dev
->priv_flags
& IFF_BRIDGE_PORT
&&
1508 skb
->pkt_type
== PACKET_HOST
) {
1510 if (unlikely(skb_cow_head(skb
,
1511 skb
->data
- skb_mac_header(skb
)))) {
1515 memcpy(eth_hdr(skb
)->h_dest
, bond_dev
->dev_addr
, ETH_ALEN
);
1521 /* enslave device <slave> to bond device <master> */
1522 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1524 struct bonding
*bond
= netdev_priv(bond_dev
);
1525 const struct net_device_ops
*slave_ops
= slave_dev
->netdev_ops
;
1526 struct slave
*new_slave
= NULL
;
1527 struct netdev_hw_addr
*ha
;
1528 struct sockaddr addr
;
1530 int old_features
= bond_dev
->features
;
1533 if (!bond
->params
.use_carrier
&& slave_dev
->ethtool_ops
== NULL
&&
1534 slave_ops
->ndo_do_ioctl
== NULL
) {
1535 pr_warning("%s: Warning: no link monitoring support for %s\n",
1536 bond_dev
->name
, slave_dev
->name
);
1539 /* bond must be initialized by bond_open() before enslaving */
1540 if (!(bond_dev
->flags
& IFF_UP
)) {
1541 pr_warning("%s: master_dev is not up in bond_enslave\n",
1545 /* already enslaved */
1546 if (slave_dev
->flags
& IFF_SLAVE
) {
1547 pr_debug("Error, Device was already enslaved\n");
1551 /* vlan challenged mutual exclusion */
1552 /* no need to lock since we're protected by rtnl_lock */
1553 if (slave_dev
->features
& NETIF_F_VLAN_CHALLENGED
) {
1554 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1556 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1557 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
1560 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1561 bond_dev
->name
, slave_dev
->name
,
1562 slave_dev
->name
, bond_dev
->name
);
1563 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1566 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev
->name
);
1567 if (bond
->slave_cnt
== 0) {
1568 /* First slave, and it is not VLAN challenged,
1569 * so remove the block of adding VLANs over the bond.
1571 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
1576 * Old ifenslave binaries are no longer supported. These can
1577 * be identified with moderate accuracy by the state of the slave:
1578 * the current ifenslave will set the interface down prior to
1579 * enslaving it; the old ifenslave will not.
1581 if ((slave_dev
->flags
& IFF_UP
)) {
1582 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1585 goto err_undo_flags
;
1588 /* set bonding device ether type by slave - bonding netdevices are
1589 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1590 * there is a need to override some of the type dependent attribs/funcs.
1592 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1593 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1595 if (bond
->slave_cnt
== 0) {
1596 if (bond_dev
->type
!= slave_dev
->type
) {
1597 pr_debug("%s: change device type from %d to %d\n",
1599 bond_dev
->type
, slave_dev
->type
);
1601 res
= netdev_bonding_change(bond_dev
,
1602 NETDEV_PRE_TYPE_CHANGE
);
1603 res
= notifier_to_errno(res
);
1605 pr_err("%s: refused to change device type\n",
1608 goto err_undo_flags
;
1611 /* Flush unicast and multicast addresses */
1612 dev_uc_flush(bond_dev
);
1613 dev_mc_flush(bond_dev
);
1615 if (slave_dev
->type
!= ARPHRD_ETHER
)
1616 bond_setup_by_slave(bond_dev
, slave_dev
);
1618 ether_setup(bond_dev
);
1620 netdev_bonding_change(bond_dev
,
1621 NETDEV_POST_TYPE_CHANGE
);
1623 } else if (bond_dev
->type
!= slave_dev
->type
) {
1624 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1626 slave_dev
->type
, bond_dev
->type
);
1628 goto err_undo_flags
;
1631 if (slave_ops
->ndo_set_mac_address
== NULL
) {
1632 if (bond
->slave_cnt
== 0) {
1633 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1635 bond
->params
.fail_over_mac
= BOND_FOM_ACTIVE
;
1636 } else if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
1637 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1640 goto err_undo_flags
;
1644 /* If this is the first slave, then we need to set the master's hardware
1645 * address to be the same as the slave's. */
1646 if (is_zero_ether_addr(bond
->dev
->dev_addr
))
1647 memcpy(bond
->dev
->dev_addr
, slave_dev
->dev_addr
,
1648 slave_dev
->addr_len
);
1651 new_slave
= kzalloc(sizeof(struct slave
), GFP_KERNEL
);
1654 goto err_undo_flags
;
1658 * Set the new_slave's queue_id to be zero. Queue ID mapping
1659 * is set via sysfs or module option if desired.
1661 new_slave
->queue_id
= 0;
1663 /* Save slave's original mtu and then set it to match the bond */
1664 new_slave
->original_mtu
= slave_dev
->mtu
;
1665 res
= dev_set_mtu(slave_dev
, bond
->dev
->mtu
);
1667 pr_debug("Error %d calling dev_set_mtu\n", res
);
1672 * Save slave's original ("permanent") mac address for modes
1673 * that need it, and for restoring it upon release, and then
1674 * set it to the master's address
1676 memcpy(new_slave
->perm_hwaddr
, slave_dev
->dev_addr
, ETH_ALEN
);
1678 if (!bond
->params
.fail_over_mac
) {
1680 * Set slave to master's mac address. The application already
1681 * set the master's mac address to that of the first slave
1683 memcpy(addr
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
1684 addr
.sa_family
= slave_dev
->type
;
1685 res
= dev_set_mac_address(slave_dev
, &addr
);
1687 pr_debug("Error %d calling set_mac_address\n", res
);
1688 goto err_restore_mtu
;
1692 res
= netdev_set_bond_master(slave_dev
, bond_dev
);
1694 pr_debug("Error %d calling netdev_set_bond_master\n", res
);
1695 goto err_restore_mac
;
1697 res
= netdev_rx_handler_register(slave_dev
, bond_handle_frame
, NULL
);
1699 pr_debug("Error %d calling netdev_rx_handler_register\n", res
);
1700 goto err_unset_master
;
1703 /* open the slave since the application closed it */
1704 res
= dev_open(slave_dev
);
1706 pr_debug("Opening slave %s failed\n", slave_dev
->name
);
1707 goto err_unreg_rxhandler
;
1710 new_slave
->dev
= slave_dev
;
1711 slave_dev
->priv_flags
|= IFF_BONDING
;
1713 if (bond_is_lb(bond
)) {
1714 /* bond_alb_init_slave() must be called before all other stages since
1715 * it might fail and we do not want to have to undo everything
1717 res
= bond_alb_init_slave(bond
, new_slave
);
1722 /* If the mode USES_PRIMARY, then the new slave gets the
1723 * master's promisc (and mc) settings only if it becomes the
1724 * curr_active_slave, and that is taken care of later when calling
1725 * bond_change_active()
1727 if (!USES_PRIMARY(bond
->params
.mode
)) {
1728 /* set promiscuity level to new slave */
1729 if (bond_dev
->flags
& IFF_PROMISC
) {
1730 res
= dev_set_promiscuity(slave_dev
, 1);
1735 /* set allmulti level to new slave */
1736 if (bond_dev
->flags
& IFF_ALLMULTI
) {
1737 res
= dev_set_allmulti(slave_dev
, 1);
1742 netif_addr_lock_bh(bond_dev
);
1743 /* upload master's mc_list to new slave */
1744 netdev_for_each_mc_addr(ha
, bond_dev
)
1745 dev_mc_add(slave_dev
, ha
->addr
);
1746 netif_addr_unlock_bh(bond_dev
);
1749 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1750 /* add lacpdu mc addr to mc list */
1751 u8 lacpdu_multicast
[ETH_ALEN
] = MULTICAST_LACPDU_ADDR
;
1753 dev_mc_add(slave_dev
, lacpdu_multicast
);
1756 bond_add_vlans_on_slave(bond
, slave_dev
);
1758 write_lock_bh(&bond
->lock
);
1760 bond_attach_slave(bond
, new_slave
);
1762 new_slave
->delay
= 0;
1763 new_slave
->link_failure_count
= 0;
1765 bond_compute_features(bond
);
1767 write_unlock_bh(&bond
->lock
);
1769 read_lock(&bond
->lock
);
1771 new_slave
->last_arp_rx
= jiffies
;
1773 if (bond
->params
.miimon
&& !bond
->params
.use_carrier
) {
1774 link_reporting
= bond_check_dev_link(bond
, slave_dev
, 1);
1776 if ((link_reporting
== -1) && !bond
->params
.arp_interval
) {
1778 * miimon is set but a bonded network driver
1779 * does not support ETHTOOL/MII and
1780 * arp_interval is not set. Note: if
1781 * use_carrier is enabled, we will never go
1782 * here (because netif_carrier is always
1783 * supported); thus, we don't need to change
1784 * the messages for netif_carrier.
1786 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1787 bond_dev
->name
, slave_dev
->name
);
1788 } else if (link_reporting
== -1) {
1789 /* unable get link status using mii/ethtool */
1790 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1791 bond_dev
->name
, slave_dev
->name
);
1795 /* check for initial state */
1796 if (!bond
->params
.miimon
||
1797 (bond_check_dev_link(bond
, slave_dev
, 0) == BMSR_LSTATUS
)) {
1798 if (bond
->params
.updelay
) {
1799 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1800 new_slave
->link
= BOND_LINK_BACK
;
1801 new_slave
->delay
= bond
->params
.updelay
;
1803 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1804 new_slave
->link
= BOND_LINK_UP
;
1806 new_slave
->jiffies
= jiffies
;
1808 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1809 new_slave
->link
= BOND_LINK_DOWN
;
1812 if (bond_update_speed_duplex(new_slave
) &&
1813 (new_slave
->link
!= BOND_LINK_DOWN
)) {
1814 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1815 bond_dev
->name
, new_slave
->dev
->name
);
1817 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1818 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1823 if (USES_PRIMARY(bond
->params
.mode
) && bond
->params
.primary
[0]) {
1824 /* if there is a primary slave, remember it */
1825 if (strcmp(bond
->params
.primary
, new_slave
->dev
->name
) == 0) {
1826 bond
->primary_slave
= new_slave
;
1827 bond
->force_primary
= true;
1831 write_lock_bh(&bond
->curr_slave_lock
);
1833 switch (bond
->params
.mode
) {
1834 case BOND_MODE_ACTIVEBACKUP
:
1835 bond_set_slave_inactive_flags(new_slave
);
1836 bond_select_active_slave(bond
);
1838 case BOND_MODE_8023AD
:
1839 /* in 802.3ad mode, the internal mechanism
1840 * will activate the slaves in the selected
1843 bond_set_slave_inactive_flags(new_slave
);
1844 /* if this is the first slave */
1845 if (bond
->slave_cnt
== 1) {
1846 SLAVE_AD_INFO(new_slave
).id
= 1;
1847 /* Initialize AD with the number of times that the AD timer is called in 1 second
1848 * can be called only after the mac address of the bond is set
1850 bond_3ad_initialize(bond
, 1000/AD_TIMER_INTERVAL
,
1851 bond
->params
.lacp_fast
);
1853 SLAVE_AD_INFO(new_slave
).id
=
1854 SLAVE_AD_INFO(new_slave
->prev
).id
+ 1;
1857 bond_3ad_bind_slave(new_slave
);
1861 new_slave
->state
= BOND_STATE_ACTIVE
;
1862 bond_set_slave_inactive_flags(new_slave
);
1863 bond_select_active_slave(bond
);
1866 pr_debug("This slave is always active in trunk mode\n");
1868 /* always active in trunk mode */
1869 new_slave
->state
= BOND_STATE_ACTIVE
;
1871 /* In trunking mode there is little meaning to curr_active_slave
1872 * anyway (it holds no special properties of the bond device),
1873 * so we can change it without calling change_active_interface()
1875 if (!bond
->curr_active_slave
)
1876 bond
->curr_active_slave
= new_slave
;
1879 } /* switch(bond_mode) */
1881 write_unlock_bh(&bond
->curr_slave_lock
);
1883 bond_set_carrier(bond
);
1885 #ifdef CONFIG_NET_POLL_CONTROLLER
1886 slave_dev
->npinfo
= bond_netpoll_info(bond
);
1887 if (slave_dev
->npinfo
) {
1888 if (slave_enable_netpoll(new_slave
)) {
1889 read_unlock(&bond
->lock
);
1890 pr_info("Error, %s: master_dev is using netpoll, "
1891 "but new slave device does not support netpoll.\n",
1899 read_unlock(&bond
->lock
);
1901 res
= bond_create_slave_symlinks(bond_dev
, slave_dev
);
1905 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1906 bond_dev
->name
, slave_dev
->name
,
1907 new_slave
->state
== BOND_STATE_ACTIVE
? "n active" : " backup",
1908 new_slave
->link
!= BOND_LINK_DOWN
? "n up" : " down");
1910 /* enslave is successful */
1913 /* Undo stages on error */
1915 dev_close(slave_dev
);
1917 err_unreg_rxhandler
:
1918 netdev_rx_handler_unregister(slave_dev
);
1921 netdev_set_bond_master(slave_dev
, NULL
);
1924 if (!bond
->params
.fail_over_mac
) {
1925 /* XXX TODO - fom follow mode needs to change master's
1926 * MAC if this slave's MAC is in use by the bond, or at
1927 * least print a warning.
1929 memcpy(addr
.sa_data
, new_slave
->perm_hwaddr
, ETH_ALEN
);
1930 addr
.sa_family
= slave_dev
->type
;
1931 dev_set_mac_address(slave_dev
, &addr
);
1935 dev_set_mtu(slave_dev
, new_slave
->original_mtu
);
1941 bond_dev
->features
= old_features
;
1947 * Try to release the slave device <slave> from the bond device <master>
1948 * It is legal to access curr_active_slave without a lock because all the function
1951 * The rules for slave state should be:
1952 * for Active/Backup:
1953 * Active stays on all backups go down
1954 * for Bonded connections:
1955 * The first up interface should be left on and all others downed.
1957 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
1959 struct bonding
*bond
= netdev_priv(bond_dev
);
1960 struct slave
*slave
, *oldcurrent
;
1961 struct sockaddr addr
;
1963 /* slave is not a slave or master is not master of this slave */
1964 if (!(slave_dev
->flags
& IFF_SLAVE
) ||
1965 (slave_dev
->master
!= bond_dev
)) {
1966 pr_err("%s: Error: cannot release %s.\n",
1967 bond_dev
->name
, slave_dev
->name
);
1972 netdev_bonding_change(bond_dev
, NETDEV_BONDING_DESLAVE
);
1973 write_lock_bh(&bond
->lock
);
1975 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
1977 /* not a slave of this bond */
1978 pr_info("%s: %s not enslaved\n",
1979 bond_dev
->name
, slave_dev
->name
);
1980 write_unlock_bh(&bond
->lock
);
1981 unblock_netpoll_tx();
1985 if (!bond
->params
.fail_over_mac
) {
1986 if (!compare_ether_addr(bond_dev
->dev_addr
, slave
->perm_hwaddr
) &&
1987 bond
->slave_cnt
> 1)
1988 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1989 bond_dev
->name
, slave_dev
->name
,
1991 bond_dev
->name
, slave_dev
->name
);
1994 /* Inform AD package of unbinding of slave. */
1995 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
1996 /* must be called before the slave is
1997 * detached from the list
1999 bond_3ad_unbind_slave(slave
);
2002 pr_info("%s: releasing %s interface %s\n",
2004 (slave
->state
== BOND_STATE_ACTIVE
) ? "active" : "backup",
2007 oldcurrent
= bond
->curr_active_slave
;
2009 bond
->current_arp_slave
= NULL
;
2011 /* release the slave from its bond */
2012 bond_detach_slave(bond
, slave
);
2014 bond_compute_features(bond
);
2016 if (bond
->primary_slave
== slave
)
2017 bond
->primary_slave
= NULL
;
2019 if (oldcurrent
== slave
)
2020 bond_change_active_slave(bond
, NULL
);
2022 if (bond_is_lb(bond
)) {
2023 /* Must be called only after the slave has been
2024 * detached from the list and the curr_active_slave
2025 * has been cleared (if our_slave == old_current),
2026 * but before a new active slave is selected.
2028 write_unlock_bh(&bond
->lock
);
2029 bond_alb_deinit_slave(bond
, slave
);
2030 write_lock_bh(&bond
->lock
);
2033 if (oldcurrent
== slave
) {
2035 * Note that we hold RTNL over this sequence, so there
2036 * is no concern that another slave add/remove event
2039 write_unlock_bh(&bond
->lock
);
2040 read_lock(&bond
->lock
);
2041 write_lock_bh(&bond
->curr_slave_lock
);
2043 bond_select_active_slave(bond
);
2045 write_unlock_bh(&bond
->curr_slave_lock
);
2046 read_unlock(&bond
->lock
);
2047 write_lock_bh(&bond
->lock
);
2050 if (bond
->slave_cnt
== 0) {
2051 bond_set_carrier(bond
);
2053 /* if the last slave was removed, zero the mac address
2054 * of the master so it will be set by the application
2055 * to the mac address of the first slave
2057 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2060 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2062 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2063 bond_dev
->name
, bond_dev
->name
);
2064 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2067 } else if ((bond_dev
->features
& NETIF_F_VLAN_CHALLENGED
) &&
2068 !bond_has_challenged_slaves(bond
)) {
2069 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2070 bond_dev
->name
, slave_dev
->name
, bond_dev
->name
);
2071 bond_dev
->features
&= ~NETIF_F_VLAN_CHALLENGED
;
2074 write_unlock_bh(&bond
->lock
);
2075 unblock_netpoll_tx();
2077 /* must do this from outside any spinlocks */
2078 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2080 bond_del_vlans_from_slave(bond
, slave_dev
);
2082 /* If the mode USES_PRIMARY, then we should only remove its
2083 * promisc and mc settings if it was the curr_active_slave, but that was
2084 * already taken care of above when we detached the slave
2086 if (!USES_PRIMARY(bond
->params
.mode
)) {
2087 /* unset promiscuity level from slave */
2088 if (bond_dev
->flags
& IFF_PROMISC
)
2089 dev_set_promiscuity(slave_dev
, -1);
2091 /* unset allmulti level from slave */
2092 if (bond_dev
->flags
& IFF_ALLMULTI
)
2093 dev_set_allmulti(slave_dev
, -1);
2095 /* flush master's mc_list from slave */
2096 netif_addr_lock_bh(bond_dev
);
2097 bond_mc_list_flush(bond_dev
, slave_dev
);
2098 netif_addr_unlock_bh(bond_dev
);
2101 netdev_rx_handler_unregister(slave_dev
);
2102 netdev_set_bond_master(slave_dev
, NULL
);
2104 slave_disable_netpoll(slave
);
2106 /* close slave before restoring its mac address */
2107 dev_close(slave_dev
);
2109 if (bond
->params
.fail_over_mac
!= BOND_FOM_ACTIVE
) {
2110 /* restore original ("permanent") mac address */
2111 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2112 addr
.sa_family
= slave_dev
->type
;
2113 dev_set_mac_address(slave_dev
, &addr
);
2116 dev_set_mtu(slave_dev
, slave
->original_mtu
);
2118 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2119 IFF_SLAVE_INACTIVE
| IFF_BONDING
|
2124 return 0; /* deletion OK */
2128 * First release a slave and than destroy the bond if no more slaves are left.
2129 * Must be under rtnl_lock when this function is called.
2131 static int bond_release_and_destroy(struct net_device
*bond_dev
,
2132 struct net_device
*slave_dev
)
2134 struct bonding
*bond
= netdev_priv(bond_dev
);
2137 ret
= bond_release(bond_dev
, slave_dev
);
2138 if ((ret
== 0) && (bond
->slave_cnt
== 0)) {
2139 bond_dev
->priv_flags
|= IFF_DISABLE_NETPOLL
;
2140 pr_info("%s: destroying bond %s.\n",
2141 bond_dev
->name
, bond_dev
->name
);
2142 unregister_netdevice(bond_dev
);
2148 * This function releases all slaves.
2150 static int bond_release_all(struct net_device
*bond_dev
)
2152 struct bonding
*bond
= netdev_priv(bond_dev
);
2153 struct slave
*slave
;
2154 struct net_device
*slave_dev
;
2155 struct sockaddr addr
;
2157 write_lock_bh(&bond
->lock
);
2159 netif_carrier_off(bond_dev
);
2161 if (bond
->slave_cnt
== 0)
2164 bond
->current_arp_slave
= NULL
;
2165 bond
->primary_slave
= NULL
;
2166 bond_change_active_slave(bond
, NULL
);
2168 while ((slave
= bond
->first_slave
) != NULL
) {
2169 /* Inform AD package of unbinding of slave
2170 * before slave is detached from the list.
2172 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2173 bond_3ad_unbind_slave(slave
);
2175 slave_dev
= slave
->dev
;
2176 bond_detach_slave(bond
, slave
);
2178 /* now that the slave is detached, unlock and perform
2179 * all the undo steps that should not be called from
2182 write_unlock_bh(&bond
->lock
);
2184 if (bond_is_lb(bond
)) {
2185 /* must be called only after the slave
2186 * has been detached from the list
2188 bond_alb_deinit_slave(bond
, slave
);
2191 bond_compute_features(bond
);
2193 bond_destroy_slave_symlinks(bond_dev
, slave_dev
);
2194 bond_del_vlans_from_slave(bond
, slave_dev
);
2196 /* If the mode USES_PRIMARY, then we should only remove its
2197 * promisc and mc settings if it was the curr_active_slave, but that was
2198 * already taken care of above when we detached the slave
2200 if (!USES_PRIMARY(bond
->params
.mode
)) {
2201 /* unset promiscuity level from slave */
2202 if (bond_dev
->flags
& IFF_PROMISC
)
2203 dev_set_promiscuity(slave_dev
, -1);
2205 /* unset allmulti level from slave */
2206 if (bond_dev
->flags
& IFF_ALLMULTI
)
2207 dev_set_allmulti(slave_dev
, -1);
2209 /* flush master's mc_list from slave */
2210 netif_addr_lock_bh(bond_dev
);
2211 bond_mc_list_flush(bond_dev
, slave_dev
);
2212 netif_addr_unlock_bh(bond_dev
);
2215 netdev_rx_handler_unregister(slave_dev
);
2216 netdev_set_bond_master(slave_dev
, NULL
);
2218 slave_disable_netpoll(slave
);
2220 /* close slave before restoring its mac address */
2221 dev_close(slave_dev
);
2223 if (!bond
->params
.fail_over_mac
) {
2224 /* restore original ("permanent") mac address*/
2225 memcpy(addr
.sa_data
, slave
->perm_hwaddr
, ETH_ALEN
);
2226 addr
.sa_family
= slave_dev
->type
;
2227 dev_set_mac_address(slave_dev
, &addr
);
2230 slave_dev
->priv_flags
&= ~(IFF_MASTER_8023AD
| IFF_MASTER_ALB
|
2231 IFF_SLAVE_INACTIVE
);
2235 /* re-acquire the lock before getting the next slave */
2236 write_lock_bh(&bond
->lock
);
2239 /* zero the mac address of the master so it will be
2240 * set by the application to the mac address of the
2243 memset(bond_dev
->dev_addr
, 0, bond_dev
->addr_len
);
2246 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
2248 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2249 bond_dev
->name
, bond_dev
->name
);
2250 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2254 pr_info("%s: released all slaves\n", bond_dev
->name
);
2257 write_unlock_bh(&bond
->lock
);
2262 * This function changes the active slave to slave <slave_dev>.
2263 * It returns -EINVAL in the following cases.
2264 * - <slave_dev> is not found in the list.
2265 * - There is not active slave now.
2266 * - <slave_dev> is already active.
2267 * - The link state of <slave_dev> is not BOND_LINK_UP.
2268 * - <slave_dev> is not running.
2269 * In these cases, this function does nothing.
2270 * In the other cases, current_slave pointer is changed and 0 is returned.
2272 static int bond_ioctl_change_active(struct net_device
*bond_dev
, struct net_device
*slave_dev
)
2274 struct bonding
*bond
= netdev_priv(bond_dev
);
2275 struct slave
*old_active
= NULL
;
2276 struct slave
*new_active
= NULL
;
2279 if (!USES_PRIMARY(bond
->params
.mode
))
2282 /* Verify that master_dev is indeed the master of slave_dev */
2283 if (!(slave_dev
->flags
& IFF_SLAVE
) || (slave_dev
->master
!= bond_dev
))
2286 read_lock(&bond
->lock
);
2288 read_lock(&bond
->curr_slave_lock
);
2289 old_active
= bond
->curr_active_slave
;
2290 read_unlock(&bond
->curr_slave_lock
);
2292 new_active
= bond_get_slave_by_dev(bond
, slave_dev
);
2295 * Changing to the current active: do nothing; return success.
2297 if (new_active
&& (new_active
== old_active
)) {
2298 read_unlock(&bond
->lock
);
2304 (new_active
->link
== BOND_LINK_UP
) &&
2305 IS_UP(new_active
->dev
)) {
2307 write_lock_bh(&bond
->curr_slave_lock
);
2308 bond_change_active_slave(bond
, new_active
);
2309 write_unlock_bh(&bond
->curr_slave_lock
);
2310 unblock_netpoll_tx();
2314 read_unlock(&bond
->lock
);
2319 static int bond_info_query(struct net_device
*bond_dev
, struct ifbond
*info
)
2321 struct bonding
*bond
= netdev_priv(bond_dev
);
2323 info
->bond_mode
= bond
->params
.mode
;
2324 info
->miimon
= bond
->params
.miimon
;
2326 read_lock(&bond
->lock
);
2327 info
->num_slaves
= bond
->slave_cnt
;
2328 read_unlock(&bond
->lock
);
2333 static int bond_slave_info_query(struct net_device
*bond_dev
, struct ifslave
*info
)
2335 struct bonding
*bond
= netdev_priv(bond_dev
);
2336 struct slave
*slave
;
2337 int i
, res
= -ENODEV
;
2339 read_lock(&bond
->lock
);
2341 bond_for_each_slave(bond
, slave
, i
) {
2342 if (i
== (int)info
->slave_id
) {
2344 strcpy(info
->slave_name
, slave
->dev
->name
);
2345 info
->link
= slave
->link
;
2346 info
->state
= slave
->state
;
2347 info
->link_failure_count
= slave
->link_failure_count
;
2352 read_unlock(&bond
->lock
);
2357 /*-------------------------------- Monitoring -------------------------------*/
2360 static int bond_miimon_inspect(struct bonding
*bond
)
2362 struct slave
*slave
;
2363 int i
, link_state
, commit
= 0;
2364 bool ignore_updelay
;
2366 ignore_updelay
= !bond
->curr_active_slave
? true : false;
2368 bond_for_each_slave(bond
, slave
, i
) {
2369 slave
->new_link
= BOND_LINK_NOCHANGE
;
2371 link_state
= bond_check_dev_link(bond
, slave
->dev
, 0);
2373 switch (slave
->link
) {
2378 slave
->link
= BOND_LINK_FAIL
;
2379 slave
->delay
= bond
->params
.downdelay
;
2381 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2383 (bond
->params
.mode
==
2384 BOND_MODE_ACTIVEBACKUP
) ?
2385 ((slave
->state
== BOND_STATE_ACTIVE
) ?
2386 "active " : "backup ") : "",
2388 bond
->params
.downdelay
* bond
->params
.miimon
);
2391 case BOND_LINK_FAIL
:
2394 * recovered before downdelay expired
2396 slave
->link
= BOND_LINK_UP
;
2397 slave
->jiffies
= jiffies
;
2398 pr_info("%s: link status up again after %d ms for interface %s.\n",
2400 (bond
->params
.downdelay
- slave
->delay
) *
2401 bond
->params
.miimon
,
2406 if (slave
->delay
<= 0) {
2407 slave
->new_link
= BOND_LINK_DOWN
;
2415 case BOND_LINK_DOWN
:
2419 slave
->link
= BOND_LINK_BACK
;
2420 slave
->delay
= bond
->params
.updelay
;
2423 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2424 bond
->dev
->name
, slave
->dev
->name
,
2425 ignore_updelay
? 0 :
2426 bond
->params
.updelay
*
2427 bond
->params
.miimon
);
2430 case BOND_LINK_BACK
:
2432 slave
->link
= BOND_LINK_DOWN
;
2433 pr_info("%s: link status down again after %d ms for interface %s.\n",
2435 (bond
->params
.updelay
- slave
->delay
) *
2436 bond
->params
.miimon
,
2445 if (slave
->delay
<= 0) {
2446 slave
->new_link
= BOND_LINK_UP
;
2448 ignore_updelay
= false;
2460 static void bond_miimon_commit(struct bonding
*bond
)
2462 struct slave
*slave
;
2465 bond_for_each_slave(bond
, slave
, i
) {
2466 switch (slave
->new_link
) {
2467 case BOND_LINK_NOCHANGE
:
2471 slave
->link
= BOND_LINK_UP
;
2472 slave
->jiffies
= jiffies
;
2474 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
2475 /* prevent it from being the active one */
2476 slave
->state
= BOND_STATE_BACKUP
;
2477 } else if (bond
->params
.mode
!= BOND_MODE_ACTIVEBACKUP
) {
2478 /* make it immediately active */
2479 slave
->state
= BOND_STATE_ACTIVE
;
2480 } else if (slave
!= bond
->primary_slave
) {
2481 /* prevent it from being the active one */
2482 slave
->state
= BOND_STATE_BACKUP
;
2485 bond_update_speed_duplex(slave
);
2487 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2488 bond
->dev
->name
, slave
->dev
->name
,
2489 slave
->speed
, slave
->duplex
? "full" : "half");
2491 /* notify ad that the link status has changed */
2492 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2493 bond_3ad_handle_link_change(slave
, BOND_LINK_UP
);
2495 if (bond_is_lb(bond
))
2496 bond_alb_handle_link_change(bond
, slave
,
2499 if (!bond
->curr_active_slave
||
2500 (slave
== bond
->primary_slave
))
2505 case BOND_LINK_DOWN
:
2506 if (slave
->link_failure_count
< UINT_MAX
)
2507 slave
->link_failure_count
++;
2509 slave
->link
= BOND_LINK_DOWN
;
2511 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
||
2512 bond
->params
.mode
== BOND_MODE_8023AD
)
2513 bond_set_slave_inactive_flags(slave
);
2515 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2516 bond
->dev
->name
, slave
->dev
->name
);
2518 if (bond
->params
.mode
== BOND_MODE_8023AD
)
2519 bond_3ad_handle_link_change(slave
,
2522 if (bond_is_lb(bond
))
2523 bond_alb_handle_link_change(bond
, slave
,
2526 if (slave
== bond
->curr_active_slave
)
2532 pr_err("%s: invalid new link %d on slave %s\n",
2533 bond
->dev
->name
, slave
->new_link
,
2535 slave
->new_link
= BOND_LINK_NOCHANGE
;
2543 write_lock_bh(&bond
->curr_slave_lock
);
2544 bond_select_active_slave(bond
);
2545 write_unlock_bh(&bond
->curr_slave_lock
);
2546 unblock_netpoll_tx();
2549 bond_set_carrier(bond
);
2555 * Really a wrapper that splits the mii monitor into two phases: an
2556 * inspection, then (if inspection indicates something needs to be done)
2557 * an acquisition of appropriate locks followed by a commit phase to
2558 * implement whatever link state changes are indicated.
2560 void bond_mii_monitor(struct work_struct
*work
)
2562 struct bonding
*bond
= container_of(work
, struct bonding
,
2565 read_lock(&bond
->lock
);
2566 if (bond
->kill_timers
)
2569 if (bond
->slave_cnt
== 0)
2572 if (bond
->send_grat_arp
) {
2573 read_lock(&bond
->curr_slave_lock
);
2574 bond_send_gratuitous_arp(bond
);
2575 read_unlock(&bond
->curr_slave_lock
);
2578 if (bond
->send_unsol_na
) {
2579 read_lock(&bond
->curr_slave_lock
);
2580 bond_send_unsolicited_na(bond
);
2581 read_unlock(&bond
->curr_slave_lock
);
2584 if (bond_miimon_inspect(bond
)) {
2585 read_unlock(&bond
->lock
);
2587 read_lock(&bond
->lock
);
2589 bond_miimon_commit(bond
);
2591 read_unlock(&bond
->lock
);
2592 rtnl_unlock(); /* might sleep, hold no other locks */
2593 read_lock(&bond
->lock
);
2597 if (bond
->params
.miimon
)
2598 queue_delayed_work(bond
->wq
, &bond
->mii_work
,
2599 msecs_to_jiffies(bond
->params
.miimon
));
2601 read_unlock(&bond
->lock
);
2604 static __be32
bond_glean_dev_ip(struct net_device
*dev
)
2606 struct in_device
*idev
;
2607 struct in_ifaddr
*ifa
;
2614 idev
= __in_dev_get_rcu(dev
);
2618 ifa
= idev
->ifa_list
;
2622 addr
= ifa
->ifa_local
;
2628 static int bond_has_this_ip(struct bonding
*bond
, __be32 ip
)
2630 struct vlan_entry
*vlan
;
2632 if (ip
== bond
->master_ip
)
2635 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2636 if (ip
== vlan
->vlan_ip
)
2644 * We go to the (large) trouble of VLAN tagging ARP frames because
2645 * switches in VLAN mode (especially if ports are configured as
2646 * "native" to a VLAN) might not pass non-tagged frames.
2648 static void bond_arp_send(struct net_device
*slave_dev
, int arp_op
, __be32 dest_ip
, __be32 src_ip
, unsigned short vlan_id
)
2650 struct sk_buff
*skb
;
2652 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op
,
2653 slave_dev
->name
, dest_ip
, src_ip
, vlan_id
);
2655 skb
= arp_create(arp_op
, ETH_P_ARP
, dest_ip
, slave_dev
, src_ip
,
2656 NULL
, slave_dev
->dev_addr
, NULL
);
2659 pr_err("ARP packet allocation failed\n");
2663 skb
= vlan_put_tag(skb
, vlan_id
);
2665 pr_err("failed to insert VLAN tag\n");
2673 static void bond_arp_send_all(struct bonding
*bond
, struct slave
*slave
)
2676 __be32
*targets
= bond
->params
.arp_targets
;
2677 struct vlan_entry
*vlan
;
2678 struct net_device
*vlan_dev
;
2681 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
); i
++) {
2684 pr_debug("basa: target %x\n", targets
[i
]);
2686 pr_debug("basa: empty vlan: arp_send\n");
2687 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2688 bond
->master_ip
, 0);
2693 * If VLANs are configured, we do a route lookup to
2694 * determine which VLAN interface would be used, so we
2695 * can tag the ARP with the proper VLAN tag.
2697 rt
= ip_route_output(dev_net(bond
->dev
), targets
[i
], 0,
2700 if (net_ratelimit()) {
2701 pr_warning("%s: no route to arp_ip_target %pI4\n",
2702 bond
->dev
->name
, &targets
[i
]);
2708 * This target is not on a VLAN
2710 if (rt
->dst
.dev
== bond
->dev
) {
2712 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2713 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2714 bond
->master_ip
, 0);
2719 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2720 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2721 if (vlan_dev
== rt
->dst
.dev
) {
2722 vlan_id
= vlan
->vlan_id
;
2723 pr_debug("basa: vlan match on %s %d\n",
2724 vlan_dev
->name
, vlan_id
);
2731 bond_arp_send(slave
->dev
, ARPOP_REQUEST
, targets
[i
],
2732 vlan
->vlan_ip
, vlan_id
);
2736 if (net_ratelimit()) {
2737 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2738 bond
->dev
->name
, &targets
[i
],
2739 rt
->dst
.dev
? rt
->dst
.dev
->name
: "NULL");
2746 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2747 * for each VLAN above us.
2749 * Caller must hold curr_slave_lock for read or better
2751 static void bond_send_gratuitous_arp(struct bonding
*bond
)
2753 struct slave
*slave
= bond
->curr_active_slave
;
2754 struct vlan_entry
*vlan
;
2755 struct net_device
*vlan_dev
;
2757 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2758 bond
->dev
->name
, slave
? slave
->dev
->name
: "NULL");
2760 if (!slave
|| !bond
->send_grat_arp
||
2761 test_bit(__LINK_STATE_LINKWATCH_PENDING
, &slave
->dev
->state
))
2764 bond
->send_grat_arp
--;
2766 if (bond
->master_ip
) {
2767 bond_arp_send(slave
->dev
, ARPOP_REPLY
, bond
->master_ip
,
2768 bond
->master_ip
, 0);
2774 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
2775 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
2776 if (vlan
->vlan_ip
) {
2777 bond_arp_send(slave
->dev
, ARPOP_REPLY
, vlan
->vlan_ip
,
2778 vlan
->vlan_ip
, vlan
->vlan_id
);
2783 static void bond_validate_arp(struct bonding
*bond
, struct slave
*slave
, __be32 sip
, __be32 tip
)
2786 __be32
*targets
= bond
->params
.arp_targets
;
2788 for (i
= 0; (i
< BOND_MAX_ARP_TARGETS
) && targets
[i
]; i
++) {
2789 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2790 &sip
, &tip
, i
, &targets
[i
],
2791 bond_has_this_ip(bond
, tip
));
2792 if (sip
== targets
[i
]) {
2793 if (bond_has_this_ip(bond
, tip
))
2794 slave
->last_arp_rx
= jiffies
;
2800 static int bond_arp_rcv(struct sk_buff
*skb
, struct net_device
*dev
, struct packet_type
*pt
, struct net_device
*orig_dev
)
2803 struct slave
*slave
;
2804 struct bonding
*bond
;
2805 unsigned char *arp_ptr
;
2808 if (dev
->priv_flags
& IFF_802_1Q_VLAN
) {
2810 * When using VLANS and bonding, dev and oriv_dev may be
2811 * incorrect if the physical interface supports VLAN
2812 * acceleration. With this change ARP validation now
2813 * works for hosts only reachable on the VLAN interface.
2815 dev
= vlan_dev_real_dev(dev
);
2816 orig_dev
= dev_get_by_index_rcu(dev_net(skb
->dev
),skb
->skb_iif
);
2819 if (!(dev
->priv_flags
& IFF_BONDING
) || !(dev
->flags
& IFF_MASTER
))
2822 bond
= netdev_priv(dev
);
2823 read_lock(&bond
->lock
);
2825 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2826 bond
->dev
->name
, skb
->dev
? skb
->dev
->name
: "NULL",
2827 orig_dev
? orig_dev
->name
: "NULL");
2829 slave
= bond_get_slave_by_dev(bond
, orig_dev
);
2830 if (!slave
|| !slave_do_arp_validate(bond
, slave
))
2833 skb
= skb_share_check(skb
, GFP_ATOMIC
);
2837 if (!pskb_may_pull(skb
, arp_hdr_len(dev
)))
2841 if (arp
->ar_hln
!= dev
->addr_len
||
2842 skb
->pkt_type
== PACKET_OTHERHOST
||
2843 skb
->pkt_type
== PACKET_LOOPBACK
||
2844 arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
2845 arp
->ar_pro
!= htons(ETH_P_IP
) ||
2849 arp_ptr
= (unsigned char *)(arp
+ 1);
2850 arp_ptr
+= dev
->addr_len
;
2851 memcpy(&sip
, arp_ptr
, 4);
2852 arp_ptr
+= 4 + dev
->addr_len
;
2853 memcpy(&tip
, arp_ptr
, 4);
2855 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2856 bond
->dev
->name
, slave
->dev
->name
, slave
->state
,
2857 bond
->params
.arp_validate
, slave_do_arp_validate(bond
, slave
),
2861 * Backup slaves won't see the ARP reply, but do come through
2862 * here for each ARP probe (so we swap the sip/tip to validate
2863 * the probe). In a "redundant switch, common router" type of
2864 * configuration, the ARP probe will (hopefully) travel from
2865 * the active, through one switch, the router, then the other
2866 * switch before reaching the backup.
2868 if (slave
->state
== BOND_STATE_ACTIVE
)
2869 bond_validate_arp(bond
, slave
, sip
, tip
);
2871 bond_validate_arp(bond
, slave
, tip
, sip
);
2874 read_unlock(&bond
->lock
);
2877 return NET_RX_SUCCESS
;
2881 * this function is called regularly to monitor each slave's link
2882 * ensuring that traffic is being sent and received when arp monitoring
2883 * is used in load-balancing mode. if the adapter has been dormant, then an
2884 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2885 * arp monitoring in active backup mode.
2887 void bond_loadbalance_arp_mon(struct work_struct
*work
)
2889 struct bonding
*bond
= container_of(work
, struct bonding
,
2891 struct slave
*slave
, *oldcurrent
;
2892 int do_failover
= 0;
2896 read_lock(&bond
->lock
);
2898 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
2900 if (bond
->kill_timers
)
2903 if (bond
->slave_cnt
== 0)
2906 read_lock(&bond
->curr_slave_lock
);
2907 oldcurrent
= bond
->curr_active_slave
;
2908 read_unlock(&bond
->curr_slave_lock
);
2910 /* see if any of the previous devices are up now (i.e. they have
2911 * xmt and rcv traffic). the curr_active_slave does not come into
2912 * the picture unless it is null. also, slave->jiffies is not needed
2913 * here because we send an arp on each slave and give a slave as
2914 * long as it needs to get the tx/rx within the delta.
2915 * TODO: what about up/down delay in arp mode? it wasn't here before
2918 bond_for_each_slave(bond
, slave
, i
) {
2919 unsigned long trans_start
= dev_trans_start(slave
->dev
);
2921 if (slave
->link
!= BOND_LINK_UP
) {
2922 if (time_in_range(jiffies
,
2923 trans_start
- delta_in_ticks
,
2924 trans_start
+ delta_in_ticks
) &&
2925 time_in_range(jiffies
,
2926 slave
->dev
->last_rx
- delta_in_ticks
,
2927 slave
->dev
->last_rx
+ delta_in_ticks
)) {
2929 slave
->link
= BOND_LINK_UP
;
2930 slave
->state
= BOND_STATE_ACTIVE
;
2932 /* primary_slave has no meaning in round-robin
2933 * mode. the window of a slave being up and
2934 * curr_active_slave being null after enslaving
2938 pr_info("%s: link status definitely up for interface %s, ",
2943 pr_info("%s: interface %s is now up\n",
2949 /* slave->link == BOND_LINK_UP */
2951 /* not all switches will respond to an arp request
2952 * when the source ip is 0, so don't take the link down
2953 * if we don't know our ip yet
2955 if (!time_in_range(jiffies
,
2956 trans_start
- delta_in_ticks
,
2957 trans_start
+ 2 * delta_in_ticks
) ||
2958 !time_in_range(jiffies
,
2959 slave
->dev
->last_rx
- delta_in_ticks
,
2960 slave
->dev
->last_rx
+ 2 * delta_in_ticks
)) {
2962 slave
->link
= BOND_LINK_DOWN
;
2963 slave
->state
= BOND_STATE_BACKUP
;
2965 if (slave
->link_failure_count
< UINT_MAX
)
2966 slave
->link_failure_count
++;
2968 pr_info("%s: interface %s is now down.\n",
2972 if (slave
== oldcurrent
)
2977 /* note: if switch is in round-robin mode, all links
2978 * must tx arp to ensure all links rx an arp - otherwise
2979 * links may oscillate or not come up at all; if switch is
2980 * in something like xor mode, there is nothing we can
2981 * do - all replies will be rx'ed on same link causing slaves
2982 * to be unstable during low/no traffic periods
2984 if (IS_UP(slave
->dev
))
2985 bond_arp_send_all(bond
, slave
);
2990 write_lock_bh(&bond
->curr_slave_lock
);
2992 bond_select_active_slave(bond
);
2994 write_unlock_bh(&bond
->curr_slave_lock
);
2995 unblock_netpoll_tx();
2999 if (bond
->params
.arp_interval
)
3000 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3002 read_unlock(&bond
->lock
);
3006 * Called to inspect slaves for active-backup mode ARP monitor link state
3007 * changes. Sets new_link in slaves to specify what action should take
3008 * place for the slave. Returns 0 if no changes are found, >0 if changes
3009 * to link states must be committed.
3011 * Called with bond->lock held for read.
3013 static int bond_ab_arp_inspect(struct bonding
*bond
, int delta_in_ticks
)
3015 struct slave
*slave
;
3017 unsigned long trans_start
;
3019 bond_for_each_slave(bond
, slave
, i
) {
3020 slave
->new_link
= BOND_LINK_NOCHANGE
;
3022 if (slave
->link
!= BOND_LINK_UP
) {
3023 if (time_in_range(jiffies
,
3024 slave_last_rx(bond
, slave
) - delta_in_ticks
,
3025 slave_last_rx(bond
, slave
) + delta_in_ticks
)) {
3027 slave
->new_link
= BOND_LINK_UP
;
3035 * Give slaves 2*delta after being enslaved or made
3036 * active. This avoids bouncing, as the last receive
3037 * times need a full ARP monitor cycle to be updated.
3039 if (time_in_range(jiffies
,
3040 slave
->jiffies
- delta_in_ticks
,
3041 slave
->jiffies
+ 2 * delta_in_ticks
))
3045 * Backup slave is down if:
3046 * - No current_arp_slave AND
3047 * - more than 3*delta since last receive AND
3048 * - the bond has an IP address
3050 * Note: a non-null current_arp_slave indicates
3051 * the curr_active_slave went down and we are
3052 * searching for a new one; under this condition
3053 * we only take the curr_active_slave down - this
3054 * gives each slave a chance to tx/rx traffic
3055 * before being taken out
3057 if (slave
->state
== BOND_STATE_BACKUP
&&
3058 !bond
->current_arp_slave
&&
3059 !time_in_range(jiffies
,
3060 slave_last_rx(bond
, slave
) - delta_in_ticks
,
3061 slave_last_rx(bond
, slave
) + 3 * delta_in_ticks
)) {
3063 slave
->new_link
= BOND_LINK_DOWN
;
3068 * Active slave is down if:
3069 * - more than 2*delta since transmitting OR
3070 * - (more than 2*delta since receive AND
3071 * the bond has an IP address)
3073 trans_start
= dev_trans_start(slave
->dev
);
3074 if ((slave
->state
== BOND_STATE_ACTIVE
) &&
3075 (!time_in_range(jiffies
,
3076 trans_start
- delta_in_ticks
,
3077 trans_start
+ 2 * delta_in_ticks
) ||
3078 !time_in_range(jiffies
,
3079 slave_last_rx(bond
, slave
) - delta_in_ticks
,
3080 slave_last_rx(bond
, slave
) + 2 * delta_in_ticks
))) {
3082 slave
->new_link
= BOND_LINK_DOWN
;
3091 * Called to commit link state changes noted by inspection step of
3092 * active-backup mode ARP monitor.
3094 * Called with RTNL and bond->lock for read.
3096 static void bond_ab_arp_commit(struct bonding
*bond
, int delta_in_ticks
)
3098 struct slave
*slave
;
3100 unsigned long trans_start
;
3102 bond_for_each_slave(bond
, slave
, i
) {
3103 switch (slave
->new_link
) {
3104 case BOND_LINK_NOCHANGE
:
3108 trans_start
= dev_trans_start(slave
->dev
);
3109 if ((!bond
->curr_active_slave
&&
3110 time_in_range(jiffies
,
3111 trans_start
- delta_in_ticks
,
3112 trans_start
+ delta_in_ticks
)) ||
3113 bond
->curr_active_slave
!= slave
) {
3114 slave
->link
= BOND_LINK_UP
;
3115 bond
->current_arp_slave
= NULL
;
3117 pr_info("%s: link status definitely up for interface %s.\n",
3118 bond
->dev
->name
, slave
->dev
->name
);
3120 if (!bond
->curr_active_slave
||
3121 (slave
== bond
->primary_slave
))
3128 case BOND_LINK_DOWN
:
3129 if (slave
->link_failure_count
< UINT_MAX
)
3130 slave
->link_failure_count
++;
3132 slave
->link
= BOND_LINK_DOWN
;
3133 bond_set_slave_inactive_flags(slave
);
3135 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3136 bond
->dev
->name
, slave
->dev
->name
);
3138 if (slave
== bond
->curr_active_slave
) {
3139 bond
->current_arp_slave
= NULL
;
3146 pr_err("%s: impossible: new_link %d on slave %s\n",
3147 bond
->dev
->name
, slave
->new_link
,
3155 write_lock_bh(&bond
->curr_slave_lock
);
3156 bond_select_active_slave(bond
);
3157 write_unlock_bh(&bond
->curr_slave_lock
);
3158 unblock_netpoll_tx();
3161 bond_set_carrier(bond
);
3165 * Send ARP probes for active-backup mode ARP monitor.
3167 * Called with bond->lock held for read.
3169 static void bond_ab_arp_probe(struct bonding
*bond
)
3171 struct slave
*slave
;
3174 read_lock(&bond
->curr_slave_lock
);
3176 if (bond
->current_arp_slave
&& bond
->curr_active_slave
)
3177 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3178 bond
->current_arp_slave
->dev
->name
,
3179 bond
->curr_active_slave
->dev
->name
);
3181 if (bond
->curr_active_slave
) {
3182 bond_arp_send_all(bond
, bond
->curr_active_slave
);
3183 read_unlock(&bond
->curr_slave_lock
);
3187 read_unlock(&bond
->curr_slave_lock
);
3189 /* if we don't have a curr_active_slave, search for the next available
3190 * backup slave from the current_arp_slave and make it the candidate
3191 * for becoming the curr_active_slave
3194 if (!bond
->current_arp_slave
) {
3195 bond
->current_arp_slave
= bond
->first_slave
;
3196 if (!bond
->current_arp_slave
)
3200 bond_set_slave_inactive_flags(bond
->current_arp_slave
);
3202 /* search for next candidate */
3203 bond_for_each_slave_from(bond
, slave
, i
, bond
->current_arp_slave
->next
) {
3204 if (IS_UP(slave
->dev
)) {
3205 slave
->link
= BOND_LINK_BACK
;
3206 bond_set_slave_active_flags(slave
);
3207 bond_arp_send_all(bond
, slave
);
3208 slave
->jiffies
= jiffies
;
3209 bond
->current_arp_slave
= slave
;
3213 /* if the link state is up at this point, we
3214 * mark it down - this can happen if we have
3215 * simultaneous link failures and
3216 * reselect_active_interface doesn't make this
3217 * one the current slave so it is still marked
3218 * up when it is actually down
3220 if (slave
->link
== BOND_LINK_UP
) {
3221 slave
->link
= BOND_LINK_DOWN
;
3222 if (slave
->link_failure_count
< UINT_MAX
)
3223 slave
->link_failure_count
++;
3225 bond_set_slave_inactive_flags(slave
);
3227 pr_info("%s: backup interface %s is now down.\n",
3228 bond
->dev
->name
, slave
->dev
->name
);
3233 void bond_activebackup_arp_mon(struct work_struct
*work
)
3235 struct bonding
*bond
= container_of(work
, struct bonding
,
3239 read_lock(&bond
->lock
);
3241 if (bond
->kill_timers
)
3244 delta_in_ticks
= msecs_to_jiffies(bond
->params
.arp_interval
);
3246 if (bond
->slave_cnt
== 0)
3249 if (bond
->send_grat_arp
) {
3250 read_lock(&bond
->curr_slave_lock
);
3251 bond_send_gratuitous_arp(bond
);
3252 read_unlock(&bond
->curr_slave_lock
);
3255 if (bond
->send_unsol_na
) {
3256 read_lock(&bond
->curr_slave_lock
);
3257 bond_send_unsolicited_na(bond
);
3258 read_unlock(&bond
->curr_slave_lock
);
3261 if (bond_ab_arp_inspect(bond
, delta_in_ticks
)) {
3262 read_unlock(&bond
->lock
);
3264 read_lock(&bond
->lock
);
3266 bond_ab_arp_commit(bond
, delta_in_ticks
);
3268 read_unlock(&bond
->lock
);
3270 read_lock(&bond
->lock
);
3273 bond_ab_arp_probe(bond
);
3276 if (bond
->params
.arp_interval
)
3277 queue_delayed_work(bond
->wq
, &bond
->arp_work
, delta_in_ticks
);
3279 read_unlock(&bond
->lock
);
3282 /*-------------------------- netdev event handling --------------------------*/
3285 * Change device name
3287 static int bond_event_changename(struct bonding
*bond
)
3289 bond_remove_proc_entry(bond
);
3290 bond_create_proc_entry(bond
);
3292 bond_debug_reregister(bond
);
3297 static int bond_master_netdev_event(unsigned long event
,
3298 struct net_device
*bond_dev
)
3300 struct bonding
*event_bond
= netdev_priv(bond_dev
);
3303 case NETDEV_CHANGENAME
:
3304 return bond_event_changename(event_bond
);
3312 static int bond_slave_netdev_event(unsigned long event
,
3313 struct net_device
*slave_dev
)
3315 struct net_device
*bond_dev
= slave_dev
->master
;
3316 struct bonding
*bond
= netdev_priv(bond_dev
);
3319 case NETDEV_UNREGISTER
:
3321 if (bond
->setup_by_slave
)
3322 bond_release_and_destroy(bond_dev
, slave_dev
);
3324 bond_release(bond_dev
, slave_dev
);
3328 if (bond
->params
.mode
== BOND_MODE_8023AD
|| bond_is_lb(bond
)) {
3329 struct slave
*slave
;
3331 slave
= bond_get_slave_by_dev(bond
, slave_dev
);
3333 u16 old_speed
= slave
->speed
;
3334 u16 old_duplex
= slave
->duplex
;
3336 bond_update_speed_duplex(slave
);
3338 if (bond_is_lb(bond
))
3341 if (old_speed
!= slave
->speed
)
3342 bond_3ad_adapter_speed_changed(slave
);
3343 if (old_duplex
!= slave
->duplex
)
3344 bond_3ad_adapter_duplex_changed(slave
);
3351 * ... Or is it this?
3354 case NETDEV_CHANGEMTU
:
3356 * TODO: Should slaves be allowed to
3357 * independently alter their MTU? For
3358 * an active-backup bond, slaves need
3359 * not be the same type of device, so
3360 * MTUs may vary. For other modes,
3361 * slaves arguably should have the
3362 * same MTUs. To do this, we'd need to
3363 * take over the slave's change_mtu
3364 * function for the duration of their
3368 case NETDEV_CHANGENAME
:
3370 * TODO: handle changing the primary's name
3373 case NETDEV_FEAT_CHANGE
:
3374 bond_compute_features(bond
);
3384 * bond_netdev_event: handle netdev notifier chain events.
3386 * This function receives events for the netdev chain. The caller (an
3387 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3388 * locks for us to safely manipulate the slave devices (RTNL lock,
3391 static int bond_netdev_event(struct notifier_block
*this,
3392 unsigned long event
, void *ptr
)
3394 struct net_device
*event_dev
= (struct net_device
*)ptr
;
3396 pr_debug("event_dev: %s, event: %lx\n",
3397 event_dev
? event_dev
->name
: "None",
3400 if (!(event_dev
->priv_flags
& IFF_BONDING
))
3403 if (event_dev
->flags
& IFF_MASTER
) {
3404 pr_debug("IFF_MASTER\n");
3405 return bond_master_netdev_event(event
, event_dev
);
3408 if (event_dev
->flags
& IFF_SLAVE
) {
3409 pr_debug("IFF_SLAVE\n");
3410 return bond_slave_netdev_event(event
, event_dev
);
3417 * bond_inetaddr_event: handle inetaddr notifier chain events.
3419 * We keep track of device IPs primarily to use as source addresses in
3420 * ARP monitor probes (rather than spewing out broadcasts all the time).
3422 * We track one IP for the main device (if it has one), plus one per VLAN.
3424 static int bond_inetaddr_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
3426 struct in_ifaddr
*ifa
= ptr
;
3427 struct net_device
*vlan_dev
, *event_dev
= ifa
->ifa_dev
->dev
;
3428 struct bond_net
*bn
= net_generic(dev_net(event_dev
), bond_net_id
);
3429 struct bonding
*bond
;
3430 struct vlan_entry
*vlan
;
3432 list_for_each_entry(bond
, &bn
->dev_list
, bond_list
) {
3433 if (bond
->dev
== event_dev
) {
3436 bond
->master_ip
= ifa
->ifa_local
;
3439 bond
->master_ip
= bond_glean_dev_ip(bond
->dev
);
3446 list_for_each_entry(vlan
, &bond
->vlan_list
, vlan_list
) {
3449 vlan_dev
= vlan_group_get_device(bond
->vlgrp
, vlan
->vlan_id
);
3450 if (vlan_dev
== event_dev
) {
3453 vlan
->vlan_ip
= ifa
->ifa_local
;
3457 bond_glean_dev_ip(vlan_dev
);
3468 static struct notifier_block bond_netdev_notifier
= {
3469 .notifier_call
= bond_netdev_event
,
3472 static struct notifier_block bond_inetaddr_notifier
= {
3473 .notifier_call
= bond_inetaddr_event
,
3476 /*-------------------------- Packet type handling ---------------------------*/
3478 /* register to receive lacpdus on a bond */
3479 static void bond_register_lacpdu(struct bonding
*bond
)
3481 struct packet_type
*pk_type
= &(BOND_AD_INFO(bond
).ad_pkt_type
);
3483 /* initialize packet type */
3484 pk_type
->type
= PKT_TYPE_LACPDU
;
3485 pk_type
->dev
= bond
->dev
;
3486 pk_type
->func
= bond_3ad_lacpdu_recv
;
3488 dev_add_pack(pk_type
);
3491 /* unregister to receive lacpdus on a bond */
3492 static void bond_unregister_lacpdu(struct bonding
*bond
)
3494 dev_remove_pack(&(BOND_AD_INFO(bond
).ad_pkt_type
));
3497 void bond_register_arp(struct bonding
*bond
)
3499 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3504 pt
->type
= htons(ETH_P_ARP
);
3505 pt
->dev
= bond
->dev
;
3506 pt
->func
= bond_arp_rcv
;
3510 void bond_unregister_arp(struct bonding
*bond
)
3512 struct packet_type
*pt
= &bond
->arp_mon_pt
;
3514 dev_remove_pack(pt
);
3518 /*---------------------------- Hashing Policies -----------------------------*/
3521 * Hash for the output device based upon layer 2 and layer 3 data. If
3522 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3524 static int bond_xmit_hash_policy_l23(struct sk_buff
*skb
, int count
)
3526 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3527 struct iphdr
*iph
= ip_hdr(skb
);
3529 if (skb
->protocol
== htons(ETH_P_IP
)) {
3530 return ((ntohl(iph
->saddr
^ iph
->daddr
) & 0xffff) ^
3531 (data
->h_dest
[5] ^ data
->h_source
[5])) % count
;
3534 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3538 * Hash for the output device based upon layer 3 and layer 4 data. If
3539 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3540 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3542 static int bond_xmit_hash_policy_l34(struct sk_buff
*skb
, int count
)
3544 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3545 struct iphdr
*iph
= ip_hdr(skb
);
3546 __be16
*layer4hdr
= (__be16
*)((u32
*)iph
+ iph
->ihl
);
3549 if (skb
->protocol
== htons(ETH_P_IP
)) {
3550 if (!(iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) &&
3551 (iph
->protocol
== IPPROTO_TCP
||
3552 iph
->protocol
== IPPROTO_UDP
)) {
3553 layer4_xor
= ntohs((*layer4hdr
^ *(layer4hdr
+ 1)));
3555 return (layer4_xor
^
3556 ((ntohl(iph
->saddr
^ iph
->daddr
)) & 0xffff)) % count
;
3560 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3564 * Hash for the output device based upon layer 2 data
3566 static int bond_xmit_hash_policy_l2(struct sk_buff
*skb
, int count
)
3568 struct ethhdr
*data
= (struct ethhdr
*)skb
->data
;
3570 return (data
->h_dest
[5] ^ data
->h_source
[5]) % count
;
3573 /*-------------------------- Device entry points ----------------------------*/
3575 static int bond_open(struct net_device
*bond_dev
)
3577 struct bonding
*bond
= netdev_priv(bond_dev
);
3579 bond
->kill_timers
= 0;
3581 INIT_DELAYED_WORK(&bond
->mcast_work
, bond_resend_igmp_join_requests_delayed
);
3583 if (bond_is_lb(bond
)) {
3584 /* bond_alb_initialize must be called before the timer
3587 if (bond_alb_initialize(bond
, (bond
->params
.mode
== BOND_MODE_ALB
))) {
3588 /* something went wrong - fail the open operation */
3592 INIT_DELAYED_WORK(&bond
->alb_work
, bond_alb_monitor
);
3593 queue_delayed_work(bond
->wq
, &bond
->alb_work
, 0);
3596 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3597 INIT_DELAYED_WORK(&bond
->mii_work
, bond_mii_monitor
);
3598 queue_delayed_work(bond
->wq
, &bond
->mii_work
, 0);
3601 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3602 if (bond
->params
.mode
== BOND_MODE_ACTIVEBACKUP
)
3603 INIT_DELAYED_WORK(&bond
->arp_work
,
3604 bond_activebackup_arp_mon
);
3606 INIT_DELAYED_WORK(&bond
->arp_work
,
3607 bond_loadbalance_arp_mon
);
3609 queue_delayed_work(bond
->wq
, &bond
->arp_work
, 0);
3610 if (bond
->params
.arp_validate
)
3611 bond_register_arp(bond
);
3614 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3615 INIT_DELAYED_WORK(&bond
->ad_work
, bond_3ad_state_machine_handler
);
3616 queue_delayed_work(bond
->wq
, &bond
->ad_work
, 0);
3617 /* register to receive LACPDUs */
3618 bond_register_lacpdu(bond
);
3619 bond_3ad_initiate_agg_selection(bond
, 1);
3625 static int bond_close(struct net_device
*bond_dev
)
3627 struct bonding
*bond
= netdev_priv(bond_dev
);
3629 if (bond
->params
.mode
== BOND_MODE_8023AD
) {
3630 /* Unregister the receive of LACPDUs */
3631 bond_unregister_lacpdu(bond
);
3634 if (bond
->params
.arp_validate
)
3635 bond_unregister_arp(bond
);
3637 write_lock_bh(&bond
->lock
);
3639 bond
->send_grat_arp
= 0;
3640 bond
->send_unsol_na
= 0;
3642 /* signal timers not to re-arm */
3643 bond
->kill_timers
= 1;
3645 write_unlock_bh(&bond
->lock
);
3647 if (bond
->params
.miimon
) { /* link check interval, in milliseconds. */
3648 cancel_delayed_work(&bond
->mii_work
);
3651 if (bond
->params
.arp_interval
) { /* arp interval, in milliseconds. */
3652 cancel_delayed_work(&bond
->arp_work
);
3655 switch (bond
->params
.mode
) {
3656 case BOND_MODE_8023AD
:
3657 cancel_delayed_work(&bond
->ad_work
);
3661 cancel_delayed_work(&bond
->alb_work
);
3667 if (delayed_work_pending(&bond
->mcast_work
))
3668 cancel_delayed_work(&bond
->mcast_work
);
3670 if (bond_is_lb(bond
)) {
3671 /* Must be called only after all
3672 * slaves have been released
3674 bond_alb_deinitialize(bond
);
3680 static struct rtnl_link_stats64
*bond_get_stats(struct net_device
*bond_dev
,
3681 struct rtnl_link_stats64
*stats
)
3683 struct bonding
*bond
= netdev_priv(bond_dev
);
3684 struct rtnl_link_stats64 temp
;
3685 struct slave
*slave
;
3688 memset(stats
, 0, sizeof(*stats
));
3690 read_lock_bh(&bond
->lock
);
3692 bond_for_each_slave(bond
, slave
, i
) {
3693 const struct rtnl_link_stats64
*sstats
=
3694 dev_get_stats(slave
->dev
, &temp
);
3696 stats
->rx_packets
+= sstats
->rx_packets
;
3697 stats
->rx_bytes
+= sstats
->rx_bytes
;
3698 stats
->rx_errors
+= sstats
->rx_errors
;
3699 stats
->rx_dropped
+= sstats
->rx_dropped
;
3701 stats
->tx_packets
+= sstats
->tx_packets
;
3702 stats
->tx_bytes
+= sstats
->tx_bytes
;
3703 stats
->tx_errors
+= sstats
->tx_errors
;
3704 stats
->tx_dropped
+= sstats
->tx_dropped
;
3706 stats
->multicast
+= sstats
->multicast
;
3707 stats
->collisions
+= sstats
->collisions
;
3709 stats
->rx_length_errors
+= sstats
->rx_length_errors
;
3710 stats
->rx_over_errors
+= sstats
->rx_over_errors
;
3711 stats
->rx_crc_errors
+= sstats
->rx_crc_errors
;
3712 stats
->rx_frame_errors
+= sstats
->rx_frame_errors
;
3713 stats
->rx_fifo_errors
+= sstats
->rx_fifo_errors
;
3714 stats
->rx_missed_errors
+= sstats
->rx_missed_errors
;
3716 stats
->tx_aborted_errors
+= sstats
->tx_aborted_errors
;
3717 stats
->tx_carrier_errors
+= sstats
->tx_carrier_errors
;
3718 stats
->tx_fifo_errors
+= sstats
->tx_fifo_errors
;
3719 stats
->tx_heartbeat_errors
+= sstats
->tx_heartbeat_errors
;
3720 stats
->tx_window_errors
+= sstats
->tx_window_errors
;
3723 read_unlock_bh(&bond
->lock
);
3728 static int bond_do_ioctl(struct net_device
*bond_dev
, struct ifreq
*ifr
, int cmd
)
3730 struct net_device
*slave_dev
= NULL
;
3731 struct ifbond k_binfo
;
3732 struct ifbond __user
*u_binfo
= NULL
;
3733 struct ifslave k_sinfo
;
3734 struct ifslave __user
*u_sinfo
= NULL
;
3735 struct mii_ioctl_data
*mii
= NULL
;
3738 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev
->name
, cmd
);
3750 * We do this again just in case we were called by SIOCGMIIREG
3751 * instead of SIOCGMIIPHY.
3758 if (mii
->reg_num
== 1) {
3759 struct bonding
*bond
= netdev_priv(bond_dev
);
3761 read_lock(&bond
->lock
);
3762 read_lock(&bond
->curr_slave_lock
);
3763 if (netif_carrier_ok(bond
->dev
))
3764 mii
->val_out
= BMSR_LSTATUS
;
3766 read_unlock(&bond
->curr_slave_lock
);
3767 read_unlock(&bond
->lock
);
3771 case BOND_INFO_QUERY_OLD
:
3772 case SIOCBONDINFOQUERY
:
3773 u_binfo
= (struct ifbond __user
*)ifr
->ifr_data
;
3775 if (copy_from_user(&k_binfo
, u_binfo
, sizeof(ifbond
)))
3778 res
= bond_info_query(bond_dev
, &k_binfo
);
3780 copy_to_user(u_binfo
, &k_binfo
, sizeof(ifbond
)))
3784 case BOND_SLAVE_INFO_QUERY_OLD
:
3785 case SIOCBONDSLAVEINFOQUERY
:
3786 u_sinfo
= (struct ifslave __user
*)ifr
->ifr_data
;
3788 if (copy_from_user(&k_sinfo
, u_sinfo
, sizeof(ifslave
)))
3791 res
= bond_slave_info_query(bond_dev
, &k_sinfo
);
3793 copy_to_user(u_sinfo
, &k_sinfo
, sizeof(ifslave
)))
3802 if (!capable(CAP_NET_ADMIN
))
3805 slave_dev
= dev_get_by_name(dev_net(bond_dev
), ifr
->ifr_slave
);
3807 pr_debug("slave_dev=%p:\n", slave_dev
);
3812 pr_debug("slave_dev->name=%s:\n", slave_dev
->name
);
3814 case BOND_ENSLAVE_OLD
:
3815 case SIOCBONDENSLAVE
:
3816 res
= bond_enslave(bond_dev
, slave_dev
);
3818 case BOND_RELEASE_OLD
:
3819 case SIOCBONDRELEASE
:
3820 res
= bond_release(bond_dev
, slave_dev
);
3822 case BOND_SETHWADDR_OLD
:
3823 case SIOCBONDSETHWADDR
:
3824 res
= bond_sethwaddr(bond_dev
, slave_dev
);
3826 case BOND_CHANGE_ACTIVE_OLD
:
3827 case SIOCBONDCHANGEACTIVE
:
3828 res
= bond_ioctl_change_active(bond_dev
, slave_dev
);
3840 static bool bond_addr_in_mc_list(unsigned char *addr
,
3841 struct netdev_hw_addr_list
*list
,
3844 struct netdev_hw_addr
*ha
;
3846 netdev_hw_addr_list_for_each(ha
, list
)
3847 if (!memcmp(ha
->addr
, addr
, addrlen
))
3853 static void bond_set_multicast_list(struct net_device
*bond_dev
)
3855 struct bonding
*bond
= netdev_priv(bond_dev
);
3856 struct netdev_hw_addr
*ha
;
3860 * Do promisc before checking multicast_mode
3862 if ((bond_dev
->flags
& IFF_PROMISC
) && !(bond
->flags
& IFF_PROMISC
))
3864 * FIXME: Need to handle the error when one of the multi-slaves
3867 bond_set_promiscuity(bond
, 1);
3870 if (!(bond_dev
->flags
& IFF_PROMISC
) && (bond
->flags
& IFF_PROMISC
))
3871 bond_set_promiscuity(bond
, -1);
3874 /* set allmulti flag to slaves */
3875 if ((bond_dev
->flags
& IFF_ALLMULTI
) && !(bond
->flags
& IFF_ALLMULTI
))
3877 * FIXME: Need to handle the error when one of the multi-slaves
3880 bond_set_allmulti(bond
, 1);
3883 if (!(bond_dev
->flags
& IFF_ALLMULTI
) && (bond
->flags
& IFF_ALLMULTI
))
3884 bond_set_allmulti(bond
, -1);
3887 read_lock(&bond
->lock
);
3889 bond
->flags
= bond_dev
->flags
;
3891 /* looking for addresses to add to slaves' mc list */
3892 netdev_for_each_mc_addr(ha
, bond_dev
) {
3893 found
= bond_addr_in_mc_list(ha
->addr
, &bond
->mc_list
,
3894 bond_dev
->addr_len
);
3896 bond_mc_add(bond
, ha
->addr
);
3899 /* looking for addresses to delete from slaves' list */
3900 netdev_hw_addr_list_for_each(ha
, &bond
->mc_list
) {
3901 found
= bond_addr_in_mc_list(ha
->addr
, &bond_dev
->mc
,
3902 bond_dev
->addr_len
);
3904 bond_mc_del(bond
, ha
->addr
);
3907 /* save master's multicast list */
3908 __hw_addr_flush(&bond
->mc_list
);
3909 __hw_addr_add_multiple(&bond
->mc_list
, &bond_dev
->mc
,
3910 bond_dev
->addr_len
, NETDEV_HW_ADDR_T_MULTICAST
);
3912 read_unlock(&bond
->lock
);
3915 static int bond_neigh_setup(struct net_device
*dev
, struct neigh_parms
*parms
)
3917 struct bonding
*bond
= netdev_priv(dev
);
3918 struct slave
*slave
= bond
->first_slave
;
3921 const struct net_device_ops
*slave_ops
3922 = slave
->dev
->netdev_ops
;
3923 if (slave_ops
->ndo_neigh_setup
)
3924 return slave_ops
->ndo_neigh_setup(slave
->dev
, parms
);
3930 * Change the MTU of all of a master's slaves to match the master
3932 static int bond_change_mtu(struct net_device
*bond_dev
, int new_mtu
)
3934 struct bonding
*bond
= netdev_priv(bond_dev
);
3935 struct slave
*slave
, *stop_at
;
3939 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond
,
3940 (bond_dev
? bond_dev
->name
: "None"), new_mtu
);
3942 /* Can't hold bond->lock with bh disabled here since
3943 * some base drivers panic. On the other hand we can't
3944 * hold bond->lock without bh disabled because we'll
3945 * deadlock. The only solution is to rely on the fact
3946 * that we're under rtnl_lock here, and the slaves
3947 * list won't change. This doesn't solve the problem
3948 * of setting the slave's MTU while it is
3949 * transmitting, but the assumption is that the base
3950 * driver can handle that.
3952 * TODO: figure out a way to safely iterate the slaves
3953 * list, but without holding a lock around the actual
3954 * call to the base driver.
3957 bond_for_each_slave(bond
, slave
, i
) {
3958 pr_debug("s %p s->p %p c_m %p\n",
3961 slave
->dev
->netdev_ops
->ndo_change_mtu
);
3963 res
= dev_set_mtu(slave
->dev
, new_mtu
);
3966 /* If we failed to set the slave's mtu to the new value
3967 * we must abort the operation even in ACTIVE_BACKUP
3968 * mode, because if we allow the backup slaves to have
3969 * different mtu values than the active slave we'll
3970 * need to change their mtu when doing a failover. That
3971 * means changing their mtu from timer context, which
3972 * is probably not a good idea.
3974 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
3979 bond_dev
->mtu
= new_mtu
;
3984 /* unwind from head to the slave that failed */
3986 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
3989 tmp_res
= dev_set_mtu(slave
->dev
, bond_dev
->mtu
);
3991 pr_debug("unwind err %d dev %s\n",
3992 tmp_res
, slave
->dev
->name
);
4002 * Note that many devices must be down to change the HW address, and
4003 * downing the master releases all slaves. We can make bonds full of
4004 * bonding devices to test this, however.
4006 static int bond_set_mac_address(struct net_device
*bond_dev
, void *addr
)
4008 struct bonding
*bond
= netdev_priv(bond_dev
);
4009 struct sockaddr
*sa
= addr
, tmp_sa
;
4010 struct slave
*slave
, *stop_at
;
4014 if (bond
->params
.mode
== BOND_MODE_ALB
)
4015 return bond_alb_set_mac_address(bond_dev
, addr
);
4018 pr_debug("bond=%p, name=%s\n",
4019 bond
, bond_dev
? bond_dev
->name
: "None");
4022 * If fail_over_mac is set to active, do nothing and return
4023 * success. Returning an error causes ifenslave to fail.
4025 if (bond
->params
.fail_over_mac
== BOND_FOM_ACTIVE
)
4028 if (!is_valid_ether_addr(sa
->sa_data
))
4029 return -EADDRNOTAVAIL
;
4031 /* Can't hold bond->lock with bh disabled here since
4032 * some base drivers panic. On the other hand we can't
4033 * hold bond->lock without bh disabled because we'll
4034 * deadlock. The only solution is to rely on the fact
4035 * that we're under rtnl_lock here, and the slaves
4036 * list won't change. This doesn't solve the problem
4037 * of setting the slave's hw address while it is
4038 * transmitting, but the assumption is that the base
4039 * driver can handle that.
4041 * TODO: figure out a way to safely iterate the slaves
4042 * list, but without holding a lock around the actual
4043 * call to the base driver.
4046 bond_for_each_slave(bond
, slave
, i
) {
4047 const struct net_device_ops
*slave_ops
= slave
->dev
->netdev_ops
;
4048 pr_debug("slave %p %s\n", slave
, slave
->dev
->name
);
4050 if (slave_ops
->ndo_set_mac_address
== NULL
) {
4052 pr_debug("EOPNOTSUPP %s\n", slave
->dev
->name
);
4056 res
= dev_set_mac_address(slave
->dev
, addr
);
4058 /* TODO: consider downing the slave
4060 * User should expect communications
4061 * breakage anyway until ARP finish
4064 pr_debug("err %d %s\n", res
, slave
->dev
->name
);
4070 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
4074 memcpy(tmp_sa
.sa_data
, bond_dev
->dev_addr
, bond_dev
->addr_len
);
4075 tmp_sa
.sa_family
= bond_dev
->type
;
4077 /* unwind from head to the slave that failed */
4079 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
4082 tmp_res
= dev_set_mac_address(slave
->dev
, &tmp_sa
);
4084 pr_debug("unwind err %d dev %s\n",
4085 tmp_res
, slave
->dev
->name
);
4092 static int bond_xmit_roundrobin(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4094 struct bonding
*bond
= netdev_priv(bond_dev
);
4095 struct slave
*slave
, *start_at
;
4096 int i
, slave_no
, res
= 1;
4097 struct iphdr
*iph
= ip_hdr(skb
);
4099 read_lock(&bond
->lock
);
4101 if (!BOND_IS_OK(bond
))
4104 * Start with the curr_active_slave that joined the bond as the
4105 * default for sending IGMP traffic. For failover purposes one
4106 * needs to maintain some consistency for the interface that will
4107 * send the join/membership reports. The curr_active_slave found
4108 * will send all of this type of traffic.
4110 if ((iph
->protocol
== IPPROTO_IGMP
) &&
4111 (skb
->protocol
== htons(ETH_P_IP
))) {
4113 read_lock(&bond
->curr_slave_lock
);
4114 slave
= bond
->curr_active_slave
;
4115 read_unlock(&bond
->curr_slave_lock
);
4121 * Concurrent TX may collide on rr_tx_counter; we accept
4122 * that as being rare enough not to justify using an
4125 slave_no
= bond
->rr_tx_counter
++ % bond
->slave_cnt
;
4127 bond_for_each_slave(bond
, slave
, i
) {
4135 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4136 if (IS_UP(slave
->dev
) &&
4137 (slave
->link
== BOND_LINK_UP
) &&
4138 (slave
->state
== BOND_STATE_ACTIVE
)) {
4139 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4146 /* no suitable interface, frame not sent */
4149 read_unlock(&bond
->lock
);
4150 return NETDEV_TX_OK
;
4155 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4156 * the bond has a usable interface.
4158 static int bond_xmit_activebackup(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4160 struct bonding
*bond
= netdev_priv(bond_dev
);
4163 read_lock(&bond
->lock
);
4164 read_lock(&bond
->curr_slave_lock
);
4166 if (!BOND_IS_OK(bond
))
4169 if (!bond
->curr_active_slave
)
4172 res
= bond_dev_queue_xmit(bond
, skb
, bond
->curr_active_slave
->dev
);
4176 /* no suitable interface, frame not sent */
4179 read_unlock(&bond
->curr_slave_lock
);
4180 read_unlock(&bond
->lock
);
4181 return NETDEV_TX_OK
;
4185 * In bond_xmit_xor() , we determine the output device by using a pre-
4186 * determined xmit_hash_policy(), If the selected device is not enabled,
4187 * find the next active slave.
4189 static int bond_xmit_xor(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4191 struct bonding
*bond
= netdev_priv(bond_dev
);
4192 struct slave
*slave
, *start_at
;
4197 read_lock(&bond
->lock
);
4199 if (!BOND_IS_OK(bond
))
4202 slave_no
= bond
->xmit_hash_policy(skb
, bond
->slave_cnt
);
4204 bond_for_each_slave(bond
, slave
, i
) {
4212 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4213 if (IS_UP(slave
->dev
) &&
4214 (slave
->link
== BOND_LINK_UP
) &&
4215 (slave
->state
== BOND_STATE_ACTIVE
)) {
4216 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4223 /* no suitable interface, frame not sent */
4226 read_unlock(&bond
->lock
);
4227 return NETDEV_TX_OK
;
4231 * in broadcast mode, we send everything to all usable interfaces.
4233 static int bond_xmit_broadcast(struct sk_buff
*skb
, struct net_device
*bond_dev
)
4235 struct bonding
*bond
= netdev_priv(bond_dev
);
4236 struct slave
*slave
, *start_at
;
4237 struct net_device
*tx_dev
= NULL
;
4241 read_lock(&bond
->lock
);
4243 if (!BOND_IS_OK(bond
))
4246 read_lock(&bond
->curr_slave_lock
);
4247 start_at
= bond
->curr_active_slave
;
4248 read_unlock(&bond
->curr_slave_lock
);
4253 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
4254 if (IS_UP(slave
->dev
) &&
4255 (slave
->link
== BOND_LINK_UP
) &&
4256 (slave
->state
== BOND_STATE_ACTIVE
)) {
4258 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
4260 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4265 res
= bond_dev_queue_xmit(bond
, skb2
, tx_dev
);
4267 dev_kfree_skb(skb2
);
4271 tx_dev
= slave
->dev
;
4276 res
= bond_dev_queue_xmit(bond
, skb
, tx_dev
);
4280 /* no suitable interface, frame not sent */
4283 /* frame sent to all suitable interfaces */
4284 read_unlock(&bond
->lock
);
4285 return NETDEV_TX_OK
;
4288 /*------------------------- Device initialization ---------------------------*/
4290 static void bond_set_xmit_hash_policy(struct bonding
*bond
)
4292 switch (bond
->params
.xmit_policy
) {
4293 case BOND_XMIT_POLICY_LAYER23
:
4294 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l23
;
4296 case BOND_XMIT_POLICY_LAYER34
:
4297 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l34
;
4299 case BOND_XMIT_POLICY_LAYER2
:
4301 bond
->xmit_hash_policy
= bond_xmit_hash_policy_l2
;
4307 * Lookup the slave that corresponds to a qid
4309 static inline int bond_slave_override(struct bonding
*bond
,
4310 struct sk_buff
*skb
)
4313 struct slave
*slave
= NULL
;
4314 struct slave
*check_slave
;
4316 read_lock(&bond
->lock
);
4318 if (!BOND_IS_OK(bond
) || !skb
->queue_mapping
)
4321 /* Find out if any slaves have the same mapping as this skb. */
4322 bond_for_each_slave(bond
, check_slave
, i
) {
4323 if (check_slave
->queue_id
== skb
->queue_mapping
) {
4324 slave
= check_slave
;
4329 /* If the slave isn't UP, use default transmit policy. */
4330 if (slave
&& slave
->queue_id
&& IS_UP(slave
->dev
) &&
4331 (slave
->link
== BOND_LINK_UP
)) {
4332 res
= bond_dev_queue_xmit(bond
, skb
, slave
->dev
);
4336 read_unlock(&bond
->lock
);
4340 static u16
bond_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
4343 * This helper function exists to help dev_pick_tx get the correct
4344 * destination queue. Using a helper function skips the a call to
4345 * skb_tx_hash and will put the skbs in the queue we expect on their
4346 * way down to the bonding driver.
4348 return skb
->queue_mapping
;
4351 static netdev_tx_t
bond_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
4353 struct bonding
*bond
= netdev_priv(dev
);
4356 * If we risk deadlock from transmitting this in the
4357 * netpoll path, tell netpoll to queue the frame for later tx
4359 if (is_netpoll_tx_blocked(dev
))
4360 return NETDEV_TX_BUSY
;
4362 if (TX_QUEUE_OVERRIDE(bond
->params
.mode
)) {
4363 if (!bond_slave_override(bond
, skb
))
4364 return NETDEV_TX_OK
;
4367 switch (bond
->params
.mode
) {
4368 case BOND_MODE_ROUNDROBIN
:
4369 return bond_xmit_roundrobin(skb
, dev
);
4370 case BOND_MODE_ACTIVEBACKUP
:
4371 return bond_xmit_activebackup(skb
, dev
);
4373 return bond_xmit_xor(skb
, dev
);
4374 case BOND_MODE_BROADCAST
:
4375 return bond_xmit_broadcast(skb
, dev
);
4376 case BOND_MODE_8023AD
:
4377 return bond_3ad_xmit_xor(skb
, dev
);
4380 return bond_alb_xmit(skb
, dev
);
4382 /* Should never happen, mode already checked */
4383 pr_err("%s: Error: Unknown bonding mode %d\n",
4384 dev
->name
, bond
->params
.mode
);
4387 return NETDEV_TX_OK
;
4393 * set bond mode specific net device operations
4395 void bond_set_mode_ops(struct bonding
*bond
, int mode
)
4397 struct net_device
*bond_dev
= bond
->dev
;
4400 case BOND_MODE_ROUNDROBIN
:
4402 case BOND_MODE_ACTIVEBACKUP
:
4405 bond_set_xmit_hash_policy(bond
);
4407 case BOND_MODE_BROADCAST
:
4409 case BOND_MODE_8023AD
:
4410 bond_set_master_3ad_flags(bond
);
4411 bond_set_xmit_hash_policy(bond
);
4414 bond_set_master_alb_flags(bond
);
4419 /* Should never happen, mode already checked */
4420 pr_err("%s: Error: Unknown bonding mode %d\n",
4421 bond_dev
->name
, mode
);
4426 static void bond_ethtool_get_drvinfo(struct net_device
*bond_dev
,
4427 struct ethtool_drvinfo
*drvinfo
)
4429 strncpy(drvinfo
->driver
, DRV_NAME
, 32);
4430 strncpy(drvinfo
->version
, DRV_VERSION
, 32);
4431 snprintf(drvinfo
->fw_version
, 32, "%d", BOND_ABI_VERSION
);
4434 static const struct ethtool_ops bond_ethtool_ops
= {
4435 .get_drvinfo
= bond_ethtool_get_drvinfo
,
4436 .get_link
= ethtool_op_get_link
,
4437 .get_tx_csum
= ethtool_op_get_tx_csum
,
4438 .get_sg
= ethtool_op_get_sg
,
4439 .get_tso
= ethtool_op_get_tso
,
4440 .get_ufo
= ethtool_op_get_ufo
,
4441 .get_flags
= ethtool_op_get_flags
,
4444 static const struct net_device_ops bond_netdev_ops
= {
4445 .ndo_init
= bond_init
,
4446 .ndo_uninit
= bond_uninit
,
4447 .ndo_open
= bond_open
,
4448 .ndo_stop
= bond_close
,
4449 .ndo_start_xmit
= bond_start_xmit
,
4450 .ndo_select_queue
= bond_select_queue
,
4451 .ndo_get_stats64
= bond_get_stats
,
4452 .ndo_do_ioctl
= bond_do_ioctl
,
4453 .ndo_set_multicast_list
= bond_set_multicast_list
,
4454 .ndo_change_mtu
= bond_change_mtu
,
4455 .ndo_set_mac_address
= bond_set_mac_address
,
4456 .ndo_neigh_setup
= bond_neigh_setup
,
4457 .ndo_vlan_rx_register
= bond_vlan_rx_register
,
4458 .ndo_vlan_rx_add_vid
= bond_vlan_rx_add_vid
,
4459 .ndo_vlan_rx_kill_vid
= bond_vlan_rx_kill_vid
,
4460 #ifdef CONFIG_NET_POLL_CONTROLLER
4461 .ndo_netpoll_setup
= bond_netpoll_setup
,
4462 .ndo_netpoll_cleanup
= bond_netpoll_cleanup
,
4463 .ndo_poll_controller
= bond_poll_controller
,
4465 .ndo_add_slave
= bond_enslave
,
4466 .ndo_del_slave
= bond_release
,
4469 static void bond_destructor(struct net_device
*bond_dev
)
4471 struct bonding
*bond
= netdev_priv(bond_dev
);
4473 destroy_workqueue(bond
->wq
);
4474 free_netdev(bond_dev
);
4477 static void bond_setup(struct net_device
*bond_dev
)
4479 struct bonding
*bond
= netdev_priv(bond_dev
);
4481 /* initialize rwlocks */
4482 rwlock_init(&bond
->lock
);
4483 rwlock_init(&bond
->curr_slave_lock
);
4485 bond
->params
= bonding_defaults
;
4487 /* Initialize pointers */
4488 bond
->dev
= bond_dev
;
4489 INIT_LIST_HEAD(&bond
->vlan_list
);
4491 /* Initialize the device entry points */
4492 ether_setup(bond_dev
);
4493 bond_dev
->netdev_ops
= &bond_netdev_ops
;
4494 bond_dev
->ethtool_ops
= &bond_ethtool_ops
;
4495 bond_set_mode_ops(bond
, bond
->params
.mode
);
4497 bond_dev
->destructor
= bond_destructor
;
4499 /* Initialize the device options */
4500 bond_dev
->tx_queue_len
= 0;
4501 bond_dev
->flags
|= IFF_MASTER
|IFF_MULTICAST
;
4502 bond_dev
->priv_flags
|= IFF_BONDING
;
4503 bond_dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
4505 if (bond
->params
.arp_interval
)
4506 bond_dev
->priv_flags
|= IFF_MASTER_ARPMON
;
4508 /* At first, we block adding VLANs. That's the only way to
4509 * prevent problems that occur when adding VLANs over an
4510 * empty bond. The block will be removed once non-challenged
4511 * slaves are enslaved.
4513 bond_dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
4515 /* don't acquire bond device's netif_tx_lock when
4517 bond_dev
->features
|= NETIF_F_LLTX
;
4519 /* By default, we declare the bond to be fully
4520 * VLAN hardware accelerated capable. Special
4521 * care is taken in the various xmit functions
4522 * when there are slaves that are not hw accel
4525 bond_dev
->features
|= (NETIF_F_HW_VLAN_TX
|
4526 NETIF_F_HW_VLAN_RX
|
4527 NETIF_F_HW_VLAN_FILTER
);
4529 /* By default, we enable GRO on bonding devices.
4530 * Actual support requires lowlevel drivers are GRO ready.
4532 bond_dev
->features
|= NETIF_F_GRO
;
4535 static void bond_work_cancel_all(struct bonding
*bond
)
4537 write_lock_bh(&bond
->lock
);
4538 bond
->kill_timers
= 1;
4539 write_unlock_bh(&bond
->lock
);
4541 if (bond
->params
.miimon
&& delayed_work_pending(&bond
->mii_work
))
4542 cancel_delayed_work(&bond
->mii_work
);
4544 if (bond
->params
.arp_interval
&& delayed_work_pending(&bond
->arp_work
))
4545 cancel_delayed_work(&bond
->arp_work
);
4547 if (bond
->params
.mode
== BOND_MODE_ALB
&&
4548 delayed_work_pending(&bond
->alb_work
))
4549 cancel_delayed_work(&bond
->alb_work
);
4551 if (bond
->params
.mode
== BOND_MODE_8023AD
&&
4552 delayed_work_pending(&bond
->ad_work
))
4553 cancel_delayed_work(&bond
->ad_work
);
4555 if (delayed_work_pending(&bond
->mcast_work
))
4556 cancel_delayed_work(&bond
->mcast_work
);
4560 * Destroy a bonding device.
4561 * Must be under rtnl_lock when this function is called.
4563 static void bond_uninit(struct net_device
*bond_dev
)
4565 struct bonding
*bond
= netdev_priv(bond_dev
);
4566 struct vlan_entry
*vlan
, *tmp
;
4568 bond_netpoll_cleanup(bond_dev
);
4570 /* Release the bonded slaves */
4571 bond_release_all(bond_dev
);
4573 list_del(&bond
->bond_list
);
4575 bond_work_cancel_all(bond
);
4577 bond_remove_proc_entry(bond
);
4579 bond_debug_unregister(bond
);
4581 __hw_addr_flush(&bond
->mc_list
);
4583 list_for_each_entry_safe(vlan
, tmp
, &bond
->vlan_list
, vlan_list
) {
4584 list_del(&vlan
->vlan_list
);
4589 /*------------------------- Module initialization ---------------------------*/
4592 * Convert string input module parms. Accept either the
4593 * number of the mode or its string name. A bit complicated because
4594 * some mode names are substrings of other names, and calls from sysfs
4595 * may have whitespace in the name (trailing newlines, for example).
4597 int bond_parse_parm(const char *buf
, const struct bond_parm_tbl
*tbl
)
4599 int modeint
= -1, i
, rv
;
4600 char *p
, modestr
[BOND_MAX_MODENAME_LEN
+ 1] = { 0, };
4602 for (p
= (char *)buf
; *p
; p
++)
4603 if (!(isdigit(*p
) || isspace(*p
)))
4607 rv
= sscanf(buf
, "%20s", modestr
);
4609 rv
= sscanf(buf
, "%d", &modeint
);
4614 for (i
= 0; tbl
[i
].modename
; i
++) {
4615 if (modeint
== tbl
[i
].mode
)
4617 if (strcmp(modestr
, tbl
[i
].modename
) == 0)
4624 static int bond_check_params(struct bond_params
*params
)
4626 int arp_validate_value
, fail_over_mac_value
, primary_reselect_value
;
4629 * Convert string parameters.
4632 bond_mode
= bond_parse_parm(mode
, bond_mode_tbl
);
4633 if (bond_mode
== -1) {
4634 pr_err("Error: Invalid bonding mode \"%s\"\n",
4635 mode
== NULL
? "NULL" : mode
);
4640 if (xmit_hash_policy
) {
4641 if ((bond_mode
!= BOND_MODE_XOR
) &&
4642 (bond_mode
!= BOND_MODE_8023AD
)) {
4643 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4644 bond_mode_name(bond_mode
));
4646 xmit_hashtype
= bond_parse_parm(xmit_hash_policy
,
4648 if (xmit_hashtype
== -1) {
4649 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4650 xmit_hash_policy
== NULL
? "NULL" :
4658 if (bond_mode
!= BOND_MODE_8023AD
) {
4659 pr_info("lacp_rate param is irrelevant in mode %s\n",
4660 bond_mode_name(bond_mode
));
4662 lacp_fast
= bond_parse_parm(lacp_rate
, bond_lacp_tbl
);
4663 if (lacp_fast
== -1) {
4664 pr_err("Error: Invalid lacp rate \"%s\"\n",
4665 lacp_rate
== NULL
? "NULL" : lacp_rate
);
4672 params
->ad_select
= bond_parse_parm(ad_select
, ad_select_tbl
);
4673 if (params
->ad_select
== -1) {
4674 pr_err("Error: Invalid ad_select \"%s\"\n",
4675 ad_select
== NULL
? "NULL" : ad_select
);
4679 if (bond_mode
!= BOND_MODE_8023AD
) {
4680 pr_warning("ad_select param only affects 802.3ad mode\n");
4683 params
->ad_select
= BOND_AD_STABLE
;
4686 if (max_bonds
< 0) {
4687 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4688 max_bonds
, 0, INT_MAX
, BOND_DEFAULT_MAX_BONDS
);
4689 max_bonds
= BOND_DEFAULT_MAX_BONDS
;
4693 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4694 miimon
, INT_MAX
, BOND_LINK_MON_INTERV
);
4695 miimon
= BOND_LINK_MON_INTERV
;
4699 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4704 if (downdelay
< 0) {
4705 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4706 downdelay
, INT_MAX
);
4710 if ((use_carrier
!= 0) && (use_carrier
!= 1)) {
4711 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4716 if (num_grat_arp
< 0 || num_grat_arp
> 255) {
4717 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4722 if (num_unsol_na
< 0 || num_unsol_na
> 255) {
4723 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4728 /* reset values for 802.3ad */
4729 if (bond_mode
== BOND_MODE_8023AD
) {
4731 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4732 pr_warning("Forcing miimon to 100msec\n");
4737 if (tx_queues
< 1 || tx_queues
> 255) {
4738 pr_warning("Warning: tx_queues (%d) should be between "
4739 "1 and 255, resetting to %d\n",
4740 tx_queues
, BOND_DEFAULT_TX_QUEUES
);
4741 tx_queues
= BOND_DEFAULT_TX_QUEUES
;
4744 if ((all_slaves_active
!= 0) && (all_slaves_active
!= 1)) {
4745 pr_warning("Warning: all_slaves_active module parameter (%d), "
4746 "not of valid value (0/1), so it was set to "
4747 "0\n", all_slaves_active
);
4748 all_slaves_active
= 0;
4751 if (resend_igmp
< 0 || resend_igmp
> 255) {
4752 pr_warning("Warning: resend_igmp (%d) should be between "
4753 "0 and 255, resetting to %d\n",
4754 resend_igmp
, BOND_DEFAULT_RESEND_IGMP
);
4755 resend_igmp
= BOND_DEFAULT_RESEND_IGMP
;
4758 /* reset values for TLB/ALB */
4759 if ((bond_mode
== BOND_MODE_TLB
) ||
4760 (bond_mode
== BOND_MODE_ALB
)) {
4762 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4763 pr_warning("Forcing miimon to 100msec\n");
4768 if (bond_mode
== BOND_MODE_ALB
) {
4769 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",
4774 if (updelay
|| downdelay
) {
4775 /* just warn the user the up/down delay will have
4776 * no effect since miimon is zero...
4778 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4779 updelay
, downdelay
);
4782 /* don't allow arp monitoring */
4784 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4785 miimon
, arp_interval
);
4789 if ((updelay
% miimon
) != 0) {
4790 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4792 (updelay
/ miimon
) * miimon
);
4797 if ((downdelay
% miimon
) != 0) {
4798 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4800 (downdelay
/ miimon
) * miimon
);
4803 downdelay
/= miimon
;
4806 if (arp_interval
< 0) {
4807 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4808 arp_interval
, INT_MAX
, BOND_LINK_ARP_INTERV
);
4809 arp_interval
= BOND_LINK_ARP_INTERV
;
4812 for (arp_ip_count
= 0;
4813 (arp_ip_count
< BOND_MAX_ARP_TARGETS
) && arp_ip_target
[arp_ip_count
];
4815 /* not complete check, but should be good enough to
4817 if (!isdigit(arp_ip_target
[arp_ip_count
][0])) {
4818 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4819 arp_ip_target
[arp_ip_count
]);
4822 __be32 ip
= in_aton(arp_ip_target
[arp_ip_count
]);
4823 arp_target
[arp_ip_count
] = ip
;
4827 if (arp_interval
&& !arp_ip_count
) {
4828 /* don't allow arping if no arp_ip_target given... */
4829 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4835 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
) {
4836 pr_err("arp_validate only supported in active-backup mode\n");
4839 if (!arp_interval
) {
4840 pr_err("arp_validate requires arp_interval\n");
4844 arp_validate_value
= bond_parse_parm(arp_validate
,
4846 if (arp_validate_value
== -1) {
4847 pr_err("Error: invalid arp_validate \"%s\"\n",
4848 arp_validate
== NULL
? "NULL" : arp_validate
);
4852 arp_validate_value
= 0;
4855 pr_info("MII link monitoring set to %d ms\n", miimon
);
4856 } else if (arp_interval
) {
4859 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4861 arp_validate_tbl
[arp_validate_value
].modename
,
4864 for (i
= 0; i
< arp_ip_count
; i
++)
4865 pr_info(" %s", arp_ip_target
[i
]);
4869 } else if (max_bonds
) {
4870 /* miimon and arp_interval not set, we need one so things
4871 * work as expected, see bonding.txt for details
4873 pr_warning("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");
4876 if (primary
&& !USES_PRIMARY(bond_mode
)) {
4877 /* currently, using a primary only makes sense
4878 * in active backup, TLB or ALB modes
4880 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4881 primary
, bond_mode_name(bond_mode
));
4885 if (primary
&& primary_reselect
) {
4886 primary_reselect_value
= bond_parse_parm(primary_reselect
,
4888 if (primary_reselect_value
== -1) {
4889 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4891 NULL
? "NULL" : primary_reselect
);
4895 primary_reselect_value
= BOND_PRI_RESELECT_ALWAYS
;
4898 if (fail_over_mac
) {
4899 fail_over_mac_value
= bond_parse_parm(fail_over_mac
,
4901 if (fail_over_mac_value
== -1) {
4902 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4903 arp_validate
== NULL
? "NULL" : arp_validate
);
4907 if (bond_mode
!= BOND_MODE_ACTIVEBACKUP
)
4908 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4910 fail_over_mac_value
= BOND_FOM_NONE
;
4913 /* fill params struct with the proper values */
4914 params
->mode
= bond_mode
;
4915 params
->xmit_policy
= xmit_hashtype
;
4916 params
->miimon
= miimon
;
4917 params
->num_grat_arp
= num_grat_arp
;
4918 params
->num_unsol_na
= num_unsol_na
;
4919 params
->arp_interval
= arp_interval
;
4920 params
->arp_validate
= arp_validate_value
;
4921 params
->updelay
= updelay
;
4922 params
->downdelay
= downdelay
;
4923 params
->use_carrier
= use_carrier
;
4924 params
->lacp_fast
= lacp_fast
;
4925 params
->primary
[0] = 0;
4926 params
->primary_reselect
= primary_reselect_value
;
4927 params
->fail_over_mac
= fail_over_mac_value
;
4928 params
->tx_queues
= tx_queues
;
4929 params
->all_slaves_active
= all_slaves_active
;
4930 params
->resend_igmp
= resend_igmp
;
4933 strncpy(params
->primary
, primary
, IFNAMSIZ
);
4934 params
->primary
[IFNAMSIZ
- 1] = 0;
4937 memcpy(params
->arp_targets
, arp_target
, sizeof(arp_target
));
4942 static struct lock_class_key bonding_netdev_xmit_lock_key
;
4943 static struct lock_class_key bonding_netdev_addr_lock_key
;
4945 static void bond_set_lockdep_class_one(struct net_device
*dev
,
4946 struct netdev_queue
*txq
,
4949 lockdep_set_class(&txq
->_xmit_lock
,
4950 &bonding_netdev_xmit_lock_key
);
4953 static void bond_set_lockdep_class(struct net_device
*dev
)
4955 lockdep_set_class(&dev
->addr_list_lock
,
4956 &bonding_netdev_addr_lock_key
);
4957 netdev_for_each_tx_queue(dev
, bond_set_lockdep_class_one
, NULL
);
4961 * Called from registration process
4963 static int bond_init(struct net_device
*bond_dev
)
4965 struct bonding
*bond
= netdev_priv(bond_dev
);
4966 struct bond_net
*bn
= net_generic(dev_net(bond_dev
), bond_net_id
);
4968 pr_debug("Begin bond_init for %s\n", bond_dev
->name
);
4970 bond
->wq
= create_singlethread_workqueue(bond_dev
->name
);
4974 bond_set_lockdep_class(bond_dev
);
4976 netif_carrier_off(bond_dev
);
4978 bond_create_proc_entry(bond
);
4979 list_add_tail(&bond
->bond_list
, &bn
->dev_list
);
4981 bond_prepare_sysfs_group(bond
);
4983 bond_debug_register(bond
);
4985 __hw_addr_init(&bond
->mc_list
);
4989 static int bond_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
4991 if (tb
[IFLA_ADDRESS
]) {
4992 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
4994 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
4995 return -EADDRNOTAVAIL
;
5000 static struct rtnl_link_ops bond_link_ops __read_mostly
= {
5002 .priv_size
= sizeof(struct bonding
),
5003 .setup
= bond_setup
,
5004 .validate
= bond_validate
,
5007 /* Create a new bond based on the specified name and bonding parameters.
5008 * If name is NULL, obtain a suitable "bond%d" name for us.
5009 * Caller must NOT hold rtnl_lock; we need to release it here before we
5010 * set up our sysfs entries.
5012 int bond_create(struct net
*net
, const char *name
)
5014 struct net_device
*bond_dev
;
5019 bond_dev
= alloc_netdev_mq(sizeof(struct bonding
), name
? name
: "",
5020 bond_setup
, tx_queues
);
5022 pr_err("%s: eek! can't alloc netdev!\n", name
);
5027 dev_net_set(bond_dev
, net
);
5028 bond_dev
->rtnl_link_ops
= &bond_link_ops
;
5031 res
= dev_alloc_name(bond_dev
, "bond%d");
5036 * If we're given a name to register
5037 * we need to ensure that its not already
5041 if (__dev_get_by_name(net
, name
) != NULL
)
5045 res
= register_netdevice(bond_dev
);
5050 bond_destructor(bond_dev
);
5054 static int __net_init
bond_net_init(struct net
*net
)
5056 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5059 INIT_LIST_HEAD(&bn
->dev_list
);
5061 bond_create_proc_dir(bn
);
5066 static void __net_exit
bond_net_exit(struct net
*net
)
5068 struct bond_net
*bn
= net_generic(net
, bond_net_id
);
5070 bond_destroy_proc_dir(bn
);
5073 static struct pernet_operations bond_net_ops
= {
5074 .init
= bond_net_init
,
5075 .exit
= bond_net_exit
,
5077 .size
= sizeof(struct bond_net
),
5080 static int __init
bonding_init(void)
5085 pr_info("%s", bond_version
);
5087 res
= bond_check_params(&bonding_defaults
);
5091 res
= register_pernet_subsys(&bond_net_ops
);
5095 res
= rtnl_link_register(&bond_link_ops
);
5099 bond_create_debugfs();
5101 for (i
= 0; i
< max_bonds
; i
++) {
5102 res
= bond_create(&init_net
, NULL
);
5107 res
= bond_create_sysfs();
5111 register_netdevice_notifier(&bond_netdev_notifier
);
5112 register_inetaddr_notifier(&bond_inetaddr_notifier
);
5113 bond_register_ipv6_notifier();
5117 rtnl_link_unregister(&bond_link_ops
);
5119 unregister_pernet_subsys(&bond_net_ops
);
5124 static void __exit
bonding_exit(void)
5126 unregister_netdevice_notifier(&bond_netdev_notifier
);
5127 unregister_inetaddr_notifier(&bond_inetaddr_notifier
);
5128 bond_unregister_ipv6_notifier();
5130 bond_destroy_sysfs();
5131 bond_destroy_debugfs();
5133 rtnl_link_unregister(&bond_link_ops
);
5134 unregister_pernet_subsys(&bond_net_ops
);
5136 #ifdef CONFIG_NET_POLL_CONTROLLER
5138 * Make sure we don't have an imbalance on our netpoll blocking
5140 WARN_ON(atomic_read(&netpoll_block_tx
));
5144 module_init(bonding_init
);
5145 module_exit(bonding_exit
);
5146 MODULE_LICENSE("GPL");
5147 MODULE_VERSION(DRV_VERSION
);
5148 MODULE_DESCRIPTION(DRV_DESCRIPTION
", v" DRV_VERSION
);
5149 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5150 MODULE_ALIAS_RTNL_LINK("bond");